xref: /linux/drivers/net/ethernet/mellanox/mlx5/core/eswitch.c (revision 232de72bdea27c8ceada056bb3de21546441b202)
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 	/* We don't assume VFs will cleanup after themselves.
1044 	 * Calling vport change handler while vport is disabled will cleanup
1045 	 * the vport resources.
1046 	 */
1047 	esw_vport_change_handle_locked(vport);
1048 	vport->enabled_events = 0;
1049 	esw_apply_vport_rx_mode(esw, vport, false, false);
1050 	esw_vport_cleanup(esw, vport);
1051 	esw->enabled_vports--;
1052 
1053 done:
1054 	mutex_unlock(&esw->state_lock);
1055 }
1056 
1057 static int eswitch_vport_event(struct notifier_block *nb,
1058 			       unsigned long type, void *data)
1059 {
1060 	struct mlx5_eswitch *esw = mlx5_nb_cof(nb, struct mlx5_eswitch, nb);
1061 	struct mlx5_eqe *eqe = data;
1062 	struct mlx5_vport *vport;
1063 	u16 vport_num;
1064 
1065 	vport_num = be16_to_cpu(eqe->data.vport_change.vport_num);
1066 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1067 	if (!IS_ERR(vport))
1068 		queue_work(esw->work_queue, &vport->vport_change_handler);
1069 	return NOTIFY_OK;
1070 }
1071 
1072 /**
1073  * mlx5_esw_query_functions - Returns raw output about functions state
1074  * @dev:	Pointer to device to query
1075  *
1076  * mlx5_esw_query_functions() allocates and returns functions changed
1077  * raw output memory pointer from device on success. Otherwise returns ERR_PTR.
1078  * Caller must free the memory using kvfree() when valid pointer is returned.
1079  */
1080 const u32 *mlx5_esw_query_functions(struct mlx5_core_dev *dev)
1081 {
1082 	bool net_func_v1 = MLX5_CAP_GEN(dev, query_host_net_function_v1);
1083 	u32 in[MLX5_ST_SZ_DW(query_esw_functions_in)] = {};
1084 	int alloc_entries;
1085 	int outlen;
1086 	u32 *out;
1087 	int err;
1088 
1089 	if (net_func_v1) {
1090 		alloc_entries = MLX5_CAP_GEN(dev,
1091 					     query_host_net_function_num_max);
1092 		alloc_entries = max(alloc_entries, 1);
1093 		MLX5_SET(query_esw_functions_in, in, op_mod,
1094 			 MLX5_QUERY_ESW_FUNC_OP_MOD_LAYOUT_V1);
1095 		outlen = MLX5_BYTE_OFF(query_esw_functions_out,
1096 				       net_function_params) +
1097 			 alloc_entries * MLX5_UN_SZ_BYTES(net_function_params);
1098 		outlen = max_t(int, outlen,
1099 			       MLX5_ST_SZ_BYTES(query_esw_functions_out));
1100 	} else {
1101 		outlen = MLX5_ST_SZ_BYTES(query_esw_functions_out);
1102 	}
1103 
1104 	out = kvzalloc(outlen, GFP_KERNEL);
1105 	if (!out)
1106 		return ERR_PTR(-ENOMEM);
1107 
1108 	MLX5_SET(query_esw_functions_in, in, opcode,
1109 		 MLX5_CMD_OP_QUERY_ESW_FUNCTIONS);
1110 
1111 	err = mlx5_cmd_exec(dev, in, sizeof(in), out, outlen);
1112 	if (err)
1113 		goto free;
1114 
1115 	if (net_func_v1) {
1116 		int num_entries;
1117 
1118 		num_entries = MLX5_GET(query_esw_functions_out, out,
1119 				       net_function_num);
1120 		if (num_entries > alloc_entries) {
1121 			mlx5_core_warn(dev, "Got %d entries, max expected %d\n",
1122 				       num_entries, alloc_entries);
1123 			err = -EINVAL;
1124 			goto free;
1125 		}
1126 	}
1127 
1128 	return out;
1129 
1130 free:
1131 	kvfree(out);
1132 	return ERR_PTR(err);
1133 }
1134 
1135 static struct mlx5_esw_pf_info
1136 mlx5_esw_host_pf_from_host_params(const void *entry)
1137 {
1138 	return (struct mlx5_esw_pf_info) {
1139 		.pf_not_exist = MLX5_GET(host_params_context, entry,
1140 					 host_pf_not_exist),
1141 		.pf_disabled = MLX5_GET(host_params_context, entry,
1142 					host_pf_disabled),
1143 		.num_of_vfs = MLX5_GET(host_params_context, entry,
1144 				       host_num_of_vfs),
1145 		.total_vfs = MLX5_GET(host_params_context, entry,
1146 				      host_total_vfs),
1147 		.host_number = MLX5_GET(host_params_context, entry,
1148 					host_number),
1149 	};
1150 }
1151 
1152 static struct mlx5_esw_pf_info
1153 mlx5_esw_host_pf_from_net_func_params(const u8 *entry, int num_entries)
1154 {
1155 	int i;
1156 
1157 	for (i = 0; i < num_entries; i++) {
1158 		int pf_type, state;
1159 
1160 		pf_type = MLX5_GET(network_function_params, entry, pci_pf_type);
1161 		if (pf_type != MLX5_PCI_PF_TYPE_EXTERNAL_HOST_PF) {
1162 			entry += MLX5_UN_SZ_BYTES(net_function_params);
1163 			continue;
1164 		}
1165 
1166 		state = MLX5_GET(network_function_params, entry, vhca_state);
1167 
1168 		return (struct mlx5_esw_pf_info) {
1169 			.pf_disabled = state != MLX5_VHCA_STATE_IN_USE,
1170 			.num_of_vfs = MLX5_GET(network_function_params,
1171 					       entry, pci_num_vfs),
1172 			.total_vfs = MLX5_GET(network_function_params,
1173 					      entry, pci_total_vfs),
1174 			.host_number = MLX5_GET(network_function_params,
1175 						entry, host_number),
1176 			.pf_num = MLX5_GET(network_function_params, entry,
1177 					   pci_device_function),
1178 		};
1179 	}
1180 
1181 	/* No external host PF entry found */
1182 	return (struct mlx5_esw_pf_info) {
1183 		.pf_not_exist = true,
1184 		.pf_disabled = true,
1185 	};
1186 }
1187 
1188 struct mlx5_esw_pf_info
1189 mlx5_esw_get_host_pf_info(struct mlx5_core_dev *dev, const u32 *out)
1190 {
1191 	const void *entry;
1192 
1193 	entry = MLX5_ADDR_OF(query_esw_functions_out, out, net_function_params);
1194 
1195 	if (MLX5_CAP_GEN(dev, query_host_net_function_v1)) {
1196 		int num_entries = MLX5_GET(query_esw_functions_out, out,
1197 					   net_function_num);
1198 
1199 		return mlx5_esw_host_pf_from_net_func_params(entry,
1200 							     num_entries);
1201 	}
1202 
1203 	return mlx5_esw_host_pf_from_host_params(entry);
1204 }
1205 
1206 bool mlx5_esw_get_spf_disabled(struct mlx5_core_dev *dev, const u32 *out,
1207 			       u16 vhca_id)
1208 {
1209 	int num_entries;
1210 	const u8 *entry;
1211 	int i;
1212 
1213 	num_entries = MLX5_GET(query_esw_functions_out, out, net_function_num);
1214 	entry = MLX5_ADDR_OF(query_esw_functions_out, out, net_function_params);
1215 
1216 	for (i = 0; i < num_entries; i++) {
1217 		u16 entry_vhca_id = MLX5_GET(network_function_params,
1218 					     entry, vhca_id);
1219 
1220 		if (entry_vhca_id == vhca_id) {
1221 			int state;
1222 
1223 			state = MLX5_GET(network_function_params, entry,
1224 					 vhca_state);
1225 			return state != MLX5_VHCA_STATE_IN_USE;
1226 		}
1227 		entry += MLX5_UN_SZ_BYTES(net_function_params);
1228 	}
1229 
1230 	return true;
1231 }
1232 
1233 static int mlx5_esw_host_functions_enabled_query(struct mlx5_eswitch *esw)
1234 {
1235 	struct mlx5_esw_pf_info host_pf_info;
1236 	const u32 *query_host_out;
1237 
1238 	if (!mlx5_core_is_ecpf_esw_manager(esw->dev))
1239 		return 0;
1240 
1241 	query_host_out = mlx5_esw_query_functions(esw->dev);
1242 	if (IS_ERR(query_host_out))
1243 		return PTR_ERR(query_host_out);
1244 
1245 	host_pf_info = mlx5_esw_get_host_pf_info(esw->dev, query_host_out);
1246 	esw->esw_funcs.host_funcs_disabled = host_pf_info.pf_not_exist;
1247 
1248 	kvfree(query_host_out);
1249 	return 0;
1250 }
1251 
1252 static void mlx5_eswitch_event_handler_register(struct mlx5_eswitch *esw)
1253 {
1254 	if (esw->mode == MLX5_ESWITCH_OFFLOADS && mlx5_eswitch_is_funcs_handler(esw->dev)) {
1255 		MLX5_NB_INIT(&esw->esw_funcs.nb, mlx5_esw_funcs_changed_handler,
1256 			     ESW_FUNCTIONS_CHANGED);
1257 		mlx5_eq_notifier_register(esw->dev, &esw->esw_funcs.nb);
1258 	}
1259 }
1260 
1261 static void mlx5_eswitch_invalidate_wq(struct mlx5_eswitch *esw)
1262 {
1263 	atomic_inc(&esw->generation);
1264 	wake_up_all(&esw->work_queue_wait);
1265 }
1266 
1267 static void mlx5_eswitch_event_handler_unregister(struct mlx5_eswitch *esw)
1268 {
1269 	if (esw->mode == MLX5_ESWITCH_OFFLOADS &&
1270 	    mlx5_eswitch_is_funcs_handler(esw->dev))
1271 		mlx5_eq_notifier_unregister(esw->dev, &esw->esw_funcs.nb);
1272 }
1273 
1274 static void mlx5_eswitch_clear_vf_vports_info(struct mlx5_eswitch *esw)
1275 {
1276 	struct mlx5_vport *vport;
1277 	unsigned long i;
1278 
1279 	mlx5_esw_for_each_vf_vport(esw, i, vport, esw->esw_funcs.num_vfs) {
1280 		mlx5_esw_qos_vport_qos_free(vport);
1281 		memset(&vport->info, 0, sizeof(vport->info));
1282 		vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
1283 	}
1284 }
1285 
1286 static void mlx5_eswitch_clear_ec_vf_vports_info(struct mlx5_eswitch *esw)
1287 {
1288 	struct mlx5_vport *vport;
1289 	unsigned long i;
1290 
1291 	mlx5_esw_for_each_ec_vf_vport(esw, i, vport, esw->esw_funcs.num_ec_vfs) {
1292 		mlx5_esw_qos_vport_qos_free(vport);
1293 		memset(&vport->info, 0, sizeof(vport->info));
1294 		vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
1295 	}
1296 }
1297 
1298 static int mlx5_eswitch_load_vport(struct mlx5_eswitch *esw, struct mlx5_vport *vport,
1299 				   enum mlx5_eswitch_vport_event enabled_events)
1300 {
1301 	int err;
1302 
1303 	err = mlx5_esw_vport_enable(esw, vport, enabled_events);
1304 	if (err)
1305 		return err;
1306 
1307 	err = mlx5_esw_offloads_load_rep(esw, vport);
1308 	if (err)
1309 		goto err_rep;
1310 
1311 	return err;
1312 
1313 err_rep:
1314 	mlx5_esw_vport_disable(esw, vport);
1315 	return err;
1316 }
1317 
1318 static void mlx5_eswitch_unload_vport(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
1319 {
1320 	mlx5_esw_offloads_unload_rep(esw, vport);
1321 	mlx5_esw_vport_disable(esw, vport);
1322 }
1323 
1324 static int mlx5_eswitch_load_pf_vf_vport(struct mlx5_eswitch *esw, u16 vport_num,
1325 					 enum mlx5_eswitch_vport_event enabled_events)
1326 {
1327 	struct mlx5_vport *vport;
1328 	int err;
1329 
1330 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1331 	if (IS_ERR(vport))
1332 		return PTR_ERR(vport);
1333 
1334 	err = mlx5_esw_offloads_init_pf_vf_rep(esw, vport);
1335 	if (err)
1336 		return err;
1337 
1338 	err = mlx5_eswitch_load_vport(esw, vport, enabled_events);
1339 	if (err)
1340 		goto err_load;
1341 	return 0;
1342 
1343 err_load:
1344 	mlx5_esw_offloads_cleanup_pf_vf_rep(esw, vport);
1345 	return err;
1346 }
1347 
1348 static void mlx5_eswitch_unload_pf_vf_vport(struct mlx5_eswitch *esw, u16 vport_num)
1349 {
1350 	struct mlx5_vport *vport;
1351 
1352 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1353 	if (IS_ERR(vport))
1354 		return;
1355 
1356 	mlx5_eswitch_unload_vport(esw, vport);
1357 	mlx5_esw_offloads_cleanup_pf_vf_rep(esw, vport);
1358 }
1359 
1360 int mlx5_eswitch_load_sf_vport(struct mlx5_eswitch *esw, u16 vport_num,
1361 			       enum mlx5_eswitch_vport_event enabled_events,
1362 			       struct mlx5_devlink_port *dl_port, u32 controller, u32 sfnum)
1363 {
1364 	struct mlx5_vport *vport;
1365 	int err;
1366 
1367 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1368 	if (IS_ERR(vport))
1369 		return PTR_ERR(vport);
1370 
1371 	err = mlx5_esw_offloads_init_sf_rep(esw, vport, dl_port, controller, sfnum);
1372 	if (err)
1373 		return err;
1374 
1375 	err = mlx5_eswitch_load_vport(esw, vport, enabled_events);
1376 	if (err)
1377 		goto err_load;
1378 
1379 	return 0;
1380 
1381 err_load:
1382 	mlx5_esw_offloads_cleanup_sf_rep(esw, vport);
1383 	return err;
1384 }
1385 
1386 void mlx5_eswitch_unload_sf_vport(struct mlx5_eswitch *esw, u16 vport_num)
1387 {
1388 	struct mlx5_vport *vport;
1389 
1390 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1391 	if (IS_ERR(vport))
1392 		return;
1393 
1394 	mlx5_eswitch_unload_vport(esw, vport);
1395 	mlx5_esw_offloads_cleanup_sf_rep(esw, vport);
1396 }
1397 
1398 void mlx5_eswitch_unload_vf_vports(struct mlx5_eswitch *esw, u16 num_vfs)
1399 {
1400 	struct mlx5_vport *vport;
1401 	unsigned long i;
1402 
1403 	mlx5_esw_for_each_vf_vport(esw, i, vport, num_vfs) {
1404 		/* Adjacent VFs are unloaded separately */
1405 		if (!vport->enabled || vport->adjacent)
1406 			continue;
1407 		mlx5_eswitch_unload_pf_vf_vport(esw, vport->vport);
1408 	}
1409 }
1410 
1411 static void mlx5_eswitch_unload_ec_vf_vports(struct mlx5_eswitch *esw,
1412 					     u16 num_ec_vfs)
1413 {
1414 	struct mlx5_vport *vport;
1415 	unsigned long i;
1416 
1417 	mlx5_esw_for_each_ec_vf_vport(esw, i, vport, num_ec_vfs) {
1418 		if (!vport->enabled)
1419 			continue;
1420 		mlx5_eswitch_unload_pf_vf_vport(esw, vport->vport);
1421 	}
1422 }
1423 
1424 static void mlx5_eswitch_unload_adj_vf_vports(struct mlx5_eswitch *esw)
1425 {
1426 	struct mlx5_vport *vport;
1427 	unsigned long i;
1428 
1429 	mlx5_esw_for_each_vf_vport(esw, i, vport, U16_MAX) {
1430 		if (!vport->enabled || !vport->adjacent)
1431 			continue;
1432 		mlx5_eswitch_unload_pf_vf_vport(esw, vport->vport);
1433 	}
1434 }
1435 
1436 static int
1437 mlx5_eswitch_load_adj_vf_vports(struct mlx5_eswitch *esw,
1438 				enum mlx5_eswitch_vport_event enabled_events)
1439 {
1440 	struct mlx5_vport *vport;
1441 	unsigned long i;
1442 	int err;
1443 
1444 	mlx5_esw_for_each_vf_vport(esw, i, vport, U16_MAX) {
1445 		if (!vport->adjacent)
1446 			continue;
1447 		err = mlx5_eswitch_load_pf_vf_vport(esw, vport->vport,
1448 						    enabled_events);
1449 		if (err)
1450 			goto unload_adj_vf_vport;
1451 	}
1452 
1453 	return 0;
1454 
1455 unload_adj_vf_vport:
1456 	mlx5_eswitch_unload_adj_vf_vports(esw);
1457 	return err;
1458 }
1459 
1460 int mlx5_eswitch_load_vf_vports(struct mlx5_eswitch *esw, u16 num_vfs,
1461 				enum mlx5_eswitch_vport_event enabled_events)
1462 {
1463 	struct mlx5_vport *vport;
1464 	unsigned long i;
1465 	int err;
1466 
1467 	mlx5_esw_for_each_vf_vport(esw, i, vport, num_vfs) {
1468 		err = mlx5_eswitch_load_pf_vf_vport(esw, vport->vport, enabled_events);
1469 		if (err)
1470 			goto vf_err;
1471 	}
1472 
1473 	return 0;
1474 
1475 vf_err:
1476 	mlx5_eswitch_unload_vf_vports(esw, num_vfs);
1477 	return err;
1478 }
1479 
1480 static int mlx5_eswitch_load_ec_vf_vports(struct mlx5_eswitch *esw, u16 num_ec_vfs,
1481 					  enum mlx5_eswitch_vport_event enabled_events)
1482 {
1483 	struct mlx5_vport *vport;
1484 	unsigned long i;
1485 	int err;
1486 
1487 	mlx5_esw_for_each_ec_vf_vport(esw, i, vport, num_ec_vfs) {
1488 		err = mlx5_eswitch_load_pf_vf_vport(esw, vport->vport, enabled_events);
1489 		if (err)
1490 			goto vf_err;
1491 	}
1492 
1493 	return 0;
1494 
1495 vf_err:
1496 	mlx5_eswitch_unload_ec_vf_vports(esw, num_ec_vfs);
1497 	return err;
1498 }
1499 
1500 int mlx5_esw_pf_enable_hca(struct mlx5_core_dev *dev, u16 vport_num)
1501 {
1502 	struct mlx5_eswitch *esw = dev->priv.eswitch;
1503 	struct mlx5_vport *vport;
1504 	int err;
1505 
1506 	if (!mlx5_core_is_ecpf(dev) || !mlx5_esw_allowed(esw))
1507 		return 0;
1508 
1509 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1510 	if (IS_ERR(vport))
1511 		return PTR_ERR(vport);
1512 
1513 	/* Once vport and representor are ready, take the PF out of
1514 	 * initializing state. Enabling HCA clears the iser->initializing
1515 	 * bit and PF driver loading can progress.
1516 	 */
1517 	err = mlx5_cmd_pf_enable_hca(dev, vport_num);
1518 	if (err)
1519 		return err;
1520 
1521 	vport->pf_activated = true;
1522 
1523 	return 0;
1524 }
1525 
1526 int mlx5_esw_pf_disable_hca(struct mlx5_core_dev *dev, u16 vport_num)
1527 {
1528 	struct mlx5_eswitch *esw = dev->priv.eswitch;
1529 	struct mlx5_vport *vport;
1530 	int err;
1531 
1532 	if (!mlx5_core_is_ecpf(dev) || !mlx5_esw_allowed(esw))
1533 		return 0;
1534 
1535 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1536 	if (IS_ERR(vport))
1537 		return PTR_ERR(vport);
1538 
1539 	err = mlx5_cmd_pf_disable_hca(dev, vport_num);
1540 	if (err)
1541 		return err;
1542 
1543 	vport->pf_activated = false;
1544 
1545 	return 0;
1546 }
1547 
1548 int mlx5_esw_host_pf_enable_hca(struct mlx5_core_dev *dev)
1549 {
1550 	return mlx5_esw_pf_enable_hca(dev, MLX5_VPORT_HOST_PF);
1551 }
1552 
1553 int mlx5_esw_host_pf_disable_hca(struct mlx5_core_dev *dev)
1554 {
1555 	return mlx5_esw_pf_disable_hca(dev, MLX5_VPORT_HOST_PF);
1556 }
1557 
1558 /* mlx5_eswitch_enable_pf_vf_vports() enables vports of PF, ECPF and VFs
1559  * whichever are present on the eswitch.
1560  */
1561 int
1562 mlx5_eswitch_enable_pf_vf_vports(struct mlx5_eswitch *esw,
1563 				 enum mlx5_eswitch_vport_event enabled_events)
1564 {
1565 	struct mlx5_esw_functions *esw_funcs = &esw->esw_funcs;
1566 	bool pf_needed;
1567 	u16 vport_num;
1568 	int ret;
1569 	int i;
1570 
1571 	pf_needed = mlx5_core_is_ecpf_esw_manager(esw->dev) ||
1572 		    esw->mode == MLX5_ESWITCH_LEGACY;
1573 
1574 	/* Enable PF vport */
1575 	if (pf_needed && mlx5_esw_host_functions_enabled(esw->dev)) {
1576 		ret = mlx5_eswitch_load_pf_vf_vport(esw, MLX5_VPORT_HOST_PF,
1577 						    enabled_events);
1578 		if (ret)
1579 			return ret;
1580 	}
1581 
1582 	if (mlx5_esw_host_functions_enabled(esw->dev)) {
1583 		/* Enable external host PF HCA */
1584 		ret = mlx5_esw_host_pf_enable_hca(esw->dev);
1585 		if (ret)
1586 			goto pf_hca_err;
1587 	}
1588 
1589 	/* Enable ECPF vport */
1590 	if (mlx5_ecpf_vport_exists(esw->dev)) {
1591 		ret = mlx5_eswitch_load_pf_vf_vport(esw, MLX5_VPORT_ECPF, enabled_events);
1592 		if (ret)
1593 			goto ecpf_err;
1594 	}
1595 
1596 	/* Enable ECVF vports */
1597 	if (mlx5_core_ec_sriov_enabled(esw->dev)) {
1598 		ret = mlx5_eswitch_load_ec_vf_vports(esw,
1599 						     esw_funcs->num_ec_vfs,
1600 						     enabled_events);
1601 		if (ret)
1602 			goto ec_vf_err;
1603 	}
1604 
1605 	/* Enable VF vports */
1606 	ret = mlx5_eswitch_load_vf_vports(esw, esw_funcs->num_vfs,
1607 					  enabled_events);
1608 	if (ret)
1609 		goto vf_err;
1610 
1611 	/* Enable adjacent VF vports */
1612 	ret = mlx5_eswitch_load_adj_vf_vports(esw, enabled_events);
1613 	if (ret)
1614 		goto unload_vf_vports;
1615 
1616 	/* Enable satellite PF vports */
1617 	for (i = 0; i < esw_funcs->num_spfs; i++) {
1618 		vport_num = esw_funcs->spfs[i].vport_num;
1619 
1620 		ret = mlx5_eswitch_load_pf_vf_vport(esw, vport_num,
1621 						    enabled_events);
1622 		if (ret)
1623 			goto spf_err;
1624 
1625 		ret = mlx5_esw_pf_enable_hca(esw->dev, vport_num);
1626 		if (ret) {
1627 			mlx5_eswitch_unload_pf_vf_vport(esw, vport_num);
1628 			goto spf_err;
1629 		}
1630 	}
1631 
1632 	return 0;
1633 
1634 spf_err:
1635 	while (i-- > 0) {
1636 		vport_num = esw_funcs->spfs[i].vport_num;
1637 		mlx5_esw_pf_disable_hca(esw->dev, vport_num);
1638 		mlx5_eswitch_unload_pf_vf_vport(esw, vport_num);
1639 	}
1640 	mlx5_eswitch_unload_adj_vf_vports(esw);
1641 unload_vf_vports:
1642 	mlx5_eswitch_unload_vf_vports(esw, esw_funcs->num_vfs);
1643 vf_err:
1644 	if (mlx5_core_ec_sriov_enabled(esw->dev))
1645 		mlx5_eswitch_unload_ec_vf_vports(esw, esw_funcs->num_ec_vfs);
1646 ec_vf_err:
1647 	if (mlx5_ecpf_vport_exists(esw->dev))
1648 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_ECPF);
1649 ecpf_err:
1650 	if (mlx5_esw_host_functions_enabled(esw->dev))
1651 		mlx5_esw_host_pf_disable_hca(esw->dev);
1652 pf_hca_err:
1653 	if (pf_needed && mlx5_esw_host_functions_enabled(esw->dev))
1654 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_HOST_PF);
1655 	return ret;
1656 }
1657 
1658 /* mlx5_eswitch_disable_pf_vf_vports() disables vports of PF, ECPF and VFs
1659  * whichever are previously enabled on the eswitch.
1660  */
1661 void mlx5_eswitch_disable_pf_vf_vports(struct mlx5_eswitch *esw)
1662 {
1663 	struct mlx5_esw_functions *esw_funcs = &esw->esw_funcs;
1664 	u16 vport_num;
1665 	int i;
1666 
1667 	for (i = 0; i < esw_funcs->num_spfs; i++) {
1668 		vport_num = esw_funcs->spfs[i].vport_num;
1669 		mlx5_esw_pf_disable_hca(esw->dev, vport_num);
1670 		mlx5_eswitch_unload_pf_vf_vport(esw, vport_num);
1671 	}
1672 
1673 	mlx5_eswitch_unload_adj_vf_vports(esw);
1674 
1675 	mlx5_eswitch_unload_vf_vports(esw, esw_funcs->num_vfs);
1676 
1677 	if (mlx5_core_ec_sriov_enabled(esw->dev))
1678 		mlx5_eswitch_unload_ec_vf_vports(esw, esw_funcs->num_ec_vfs);
1679 
1680 	if (mlx5_ecpf_vport_exists(esw->dev)) {
1681 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_ECPF);
1682 	}
1683 
1684 	if (mlx5_esw_host_functions_enabled(esw->dev))
1685 		mlx5_esw_host_pf_disable_hca(esw->dev);
1686 
1687 	if ((mlx5_core_is_ecpf_esw_manager(esw->dev) ||
1688 	     esw->mode == MLX5_ESWITCH_LEGACY) &&
1689 	    mlx5_esw_host_functions_enabled(esw->dev))
1690 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_HOST_PF);
1691 }
1692 
1693 static void mlx5_eswitch_get_devlink_param(struct mlx5_eswitch *esw)
1694 {
1695 	struct devlink *devlink = priv_to_devlink(esw->dev);
1696 	union devlink_param_value val;
1697 	int err;
1698 
1699 	err = devl_param_driverinit_value_get(devlink,
1700 					      MLX5_DEVLINK_PARAM_ID_ESW_LARGE_GROUP_NUM,
1701 					      &val);
1702 	if (!err) {
1703 		esw->params.large_group_num = val.vu32;
1704 	} else {
1705 		esw_warn(esw->dev,
1706 			 "Devlink can't get param fdb_large_groups, uses default (%d).\n",
1707 			 ESW_OFFLOADS_DEFAULT_NUM_GROUPS);
1708 		esw->params.large_group_num = ESW_OFFLOADS_DEFAULT_NUM_GROUPS;
1709 	}
1710 }
1711 
1712 static void
1713 mlx5_eswitch_update_num_of_vfs(struct mlx5_eswitch *esw, int num_vfs)
1714 {
1715 	struct mlx5_esw_pf_info host_pf_info;
1716 	const u32 *out;
1717 
1718 	if (num_vfs < 0)
1719 		return;
1720 
1721 	if (!mlx5_core_is_ecpf_esw_manager(esw->dev)) {
1722 		esw->esw_funcs.num_vfs = num_vfs;
1723 		return;
1724 	}
1725 
1726 	out = mlx5_esw_query_functions(esw->dev);
1727 	if (IS_ERR(out))
1728 		return;
1729 
1730 	host_pf_info = mlx5_esw_get_host_pf_info(esw->dev, out);
1731 	esw->esw_funcs.num_vfs = host_pf_info.num_of_vfs;
1732 	if (mlx5_core_ec_sriov_enabled(esw->dev))
1733 		esw->esw_funcs.num_ec_vfs = num_vfs;
1734 
1735 	kvfree(out);
1736 }
1737 
1738 static void mlx5_esw_mode_change_notify(struct mlx5_eswitch *esw, u16 mode)
1739 {
1740 	struct mlx5_esw_event_info info = {};
1741 
1742 	info.new_mode = mode;
1743 
1744 	blocking_notifier_call_chain(&esw->dev->priv.esw_n_head, 0, &info);
1745 }
1746 
1747 static int mlx5_esw_egress_acls_init(struct mlx5_core_dev *dev)
1748 {
1749 	struct mlx5_flow_steering *steering = dev->priv.steering;
1750 	int total_vports = mlx5_eswitch_get_total_vports(dev);
1751 	int err;
1752 	int i;
1753 
1754 	for (i = 0; i < total_vports; i++) {
1755 		err = mlx5_fs_vport_egress_acl_ns_add(steering, i);
1756 		if (err)
1757 			goto acl_ns_remove;
1758 	}
1759 	return 0;
1760 
1761 acl_ns_remove:
1762 	while (i--)
1763 		mlx5_fs_vport_egress_acl_ns_remove(steering, i);
1764 	return err;
1765 }
1766 
1767 static void mlx5_esw_egress_acls_cleanup(struct mlx5_core_dev *dev)
1768 {
1769 	struct mlx5_flow_steering *steering = dev->priv.steering;
1770 	int total_vports = mlx5_eswitch_get_total_vports(dev);
1771 	int i;
1772 
1773 	for (i = total_vports - 1; i >= 0; i--)
1774 		mlx5_fs_vport_egress_acl_ns_remove(steering, i);
1775 }
1776 
1777 static int mlx5_esw_ingress_acls_init(struct mlx5_core_dev *dev)
1778 {
1779 	struct mlx5_flow_steering *steering = dev->priv.steering;
1780 	int total_vports = mlx5_eswitch_get_total_vports(dev);
1781 	int err;
1782 	int i;
1783 
1784 	for (i = 0; i < total_vports; i++) {
1785 		err = mlx5_fs_vport_ingress_acl_ns_add(steering, i);
1786 		if (err)
1787 			goto acl_ns_remove;
1788 	}
1789 	return 0;
1790 
1791 acl_ns_remove:
1792 	while (i--)
1793 		mlx5_fs_vport_ingress_acl_ns_remove(steering, i);
1794 	return err;
1795 }
1796 
1797 static void mlx5_esw_ingress_acls_cleanup(struct mlx5_core_dev *dev)
1798 {
1799 	struct mlx5_flow_steering *steering = dev->priv.steering;
1800 	int total_vports = mlx5_eswitch_get_total_vports(dev);
1801 	int i;
1802 
1803 	for (i = total_vports - 1; i >= 0; i--)
1804 		mlx5_fs_vport_ingress_acl_ns_remove(steering, i);
1805 }
1806 
1807 static int mlx5_esw_acls_ns_init(struct mlx5_eswitch *esw)
1808 {
1809 	struct mlx5_core_dev *dev = esw->dev;
1810 	int err;
1811 
1812 	if (esw->flags & MLX5_ESWITCH_VPORT_ACL_NS_CREATED)
1813 		return 0;
1814 
1815 	if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
1816 		err = mlx5_esw_egress_acls_init(dev);
1817 		if (err)
1818 			return err;
1819 	} else {
1820 		esw_warn(dev, "egress ACL is not supported by FW\n");
1821 	}
1822 
1823 	if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
1824 		err = mlx5_esw_ingress_acls_init(dev);
1825 		if (err)
1826 			goto err;
1827 	} else {
1828 		esw_warn(dev, "ingress ACL is not supported by FW\n");
1829 	}
1830 	esw->flags |= MLX5_ESWITCH_VPORT_ACL_NS_CREATED;
1831 	return 0;
1832 
1833 err:
1834 	if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support))
1835 		mlx5_esw_egress_acls_cleanup(dev);
1836 	return err;
1837 }
1838 
1839 static void mlx5_esw_acls_ns_cleanup(struct mlx5_eswitch *esw)
1840 {
1841 	struct mlx5_core_dev *dev = esw->dev;
1842 
1843 	esw->flags &= ~MLX5_ESWITCH_VPORT_ACL_NS_CREATED;
1844 	if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support))
1845 		mlx5_esw_ingress_acls_cleanup(dev);
1846 	if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support))
1847 		mlx5_esw_egress_acls_cleanup(dev);
1848 }
1849 
1850 /**
1851  * mlx5_eswitch_enable_locked - Enable eswitch
1852  * @esw:	Pointer to eswitch
1853  * @num_vfs:	Enable eswitch for given number of VFs. This is optional.
1854  *		Valid value are 0, > 0 and MLX5_ESWITCH_IGNORE_NUM_VFS.
1855  *		Caller should pass num_vfs > 0 when enabling eswitch for
1856  *		vf vports. Caller should pass num_vfs = 0, when eswitch
1857  *		is enabled without sriov VFs or when caller
1858  *		is unaware of the sriov state of the host PF on ECPF based
1859  *		eswitch. Caller should pass < 0 when num_vfs should be
1860  *		completely ignored. This is typically the case when eswitch
1861  *		is enabled without sriov regardless of PF/ECPF system.
1862  * mlx5_eswitch_enable_locked() Enables eswitch in either legacy or offloads
1863  * mode. If num_vfs >=0 is provided, it setup VF related eswitch vports.
1864  * It returns 0 on success or error code on failure.
1865  */
1866 int mlx5_eswitch_enable_locked(struct mlx5_eswitch *esw, int num_vfs)
1867 {
1868 	int err;
1869 
1870 	devl_assert_locked(priv_to_devlink(esw->dev));
1871 
1872 	if (!MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, ft_support)) {
1873 		esw_warn(esw->dev, "FDB is not supported, aborting ...\n");
1874 		return -EOPNOTSUPP;
1875 	}
1876 
1877 	mlx5_eswitch_get_devlink_param(esw);
1878 
1879 	err = mlx5_esw_acls_ns_init(esw);
1880 	if (err)
1881 		return err;
1882 
1883 	mlx5_eswitch_update_num_of_vfs(esw, num_vfs);
1884 
1885 	MLX5_NB_INIT(&esw->nb, eswitch_vport_event, NIC_VPORT_CHANGE);
1886 	mlx5_eq_notifier_register(esw->dev, &esw->nb);
1887 
1888 	err = mlx5_esw_qos_init(esw);
1889 	if (err)
1890 		goto err_esw_init;
1891 
1892 	if (esw->mode == MLX5_ESWITCH_LEGACY) {
1893 		err = esw_legacy_enable(esw);
1894 	} else {
1895 		err = esw_offloads_enable(esw);
1896 	}
1897 
1898 	if (err)
1899 		goto err_esw_init;
1900 
1901 	esw->fdb_table.flags |= MLX5_ESW_FDB_CREATED;
1902 
1903 	mlx5_eswitch_event_handler_register(esw);
1904 
1905 	esw_info(esw->dev, "Enable: mode(%s), nvfs(%d), necvfs(%d), active vports(%d)\n",
1906 		 esw->mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
1907 		 esw->esw_funcs.num_vfs, esw->esw_funcs.num_ec_vfs, esw->enabled_vports);
1908 
1909 	mlx5_esw_mode_change_notify(esw, esw->mode);
1910 
1911 	return 0;
1912 
1913 err_esw_init:
1914 	mlx5_eq_notifier_unregister(esw->dev, &esw->nb);
1915 	mlx5_esw_acls_ns_cleanup(esw);
1916 	return err;
1917 }
1918 
1919 /**
1920  * mlx5_eswitch_enable - Enable eswitch
1921  * @esw:	Pointer to eswitch
1922  * @num_vfs:	Enable eswitch switch for given number of VFs.
1923  *		Caller must pass num_vfs > 0 when enabling eswitch for
1924  *		vf vports.
1925  * mlx5_eswitch_enable() returns 0 on success or error code on failure.
1926  */
1927 int mlx5_eswitch_enable(struct mlx5_eswitch *esw, int num_vfs)
1928 {
1929 	bool toggle_lag;
1930 	int ret = 0;
1931 
1932 	if (!mlx5_esw_allowed(esw))
1933 		return 0;
1934 
1935 	devl_assert_locked(priv_to_devlink(esw->dev));
1936 
1937 	toggle_lag = !mlx5_esw_is_fdb_created(esw);
1938 
1939 	if (toggle_lag)
1940 		mlx5_lag_disable_change(esw->dev);
1941 
1942 	mlx5_eswitch_invalidate_wq(esw);
1943 	mlx5_esw_reps_block(esw);
1944 
1945 	if (!mlx5_esw_is_fdb_created(esw)) {
1946 		ret = mlx5_eswitch_enable_locked(esw, num_vfs);
1947 	} else {
1948 		enum mlx5_eswitch_vport_event vport_events;
1949 
1950 		vport_events = (esw->mode == MLX5_ESWITCH_LEGACY) ?
1951 					MLX5_LEGACY_SRIOV_VPORT_EVENTS : MLX5_VPORT_UC_ADDR_CHANGE;
1952 		/* If this is the ECPF the number of host VFs is managed via the
1953 		 * eswitch function change event handler, and any num_vfs provided
1954 		 * here are intended to be EC VFs.
1955 		 */
1956 		if (!mlx5_core_is_ecpf(esw->dev)) {
1957 			ret = mlx5_eswitch_load_vf_vports(esw, num_vfs, vport_events);
1958 			if (!ret)
1959 				esw->esw_funcs.num_vfs = num_vfs;
1960 		} else if (mlx5_core_ec_sriov_enabled(esw->dev)) {
1961 			ret = mlx5_eswitch_load_ec_vf_vports(esw, num_vfs, vport_events);
1962 			if (!ret)
1963 				esw->esw_funcs.num_ec_vfs = num_vfs;
1964 		}
1965 	}
1966 
1967 	mlx5_esw_reps_unblock(esw);
1968 
1969 	if (toggle_lag)
1970 		mlx5_lag_enable_change(esw->dev);
1971 
1972 	return ret;
1973 }
1974 
1975 /* When disabling sriov, free driver level resources. */
1976 void mlx5_eswitch_disable_sriov(struct mlx5_eswitch *esw, bool clear_vf)
1977 {
1978 	if (!mlx5_esw_allowed(esw))
1979 		return;
1980 
1981 	devl_assert_locked(priv_to_devlink(esw->dev));
1982 	/* If driver is unloaded, this function is called twice by remove_one()
1983 	 * and mlx5_unload(). Prevent the second call.
1984 	 */
1985 	if (!esw->esw_funcs.num_vfs && !esw->esw_funcs.num_ec_vfs && !clear_vf)
1986 		return;
1987 
1988 	esw_info(esw->dev, "Unload vfs: mode(%s), nvfs(%d), necvfs(%d), active vports(%d)\n",
1989 		 esw->mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
1990 		 esw->esw_funcs.num_vfs, esw->esw_funcs.num_ec_vfs, esw->enabled_vports);
1991 
1992 	mlx5_eswitch_invalidate_wq(esw);
1993 	mlx5_esw_reps_block(esw);
1994 
1995 	if (!mlx5_core_is_ecpf(esw->dev)) {
1996 		mlx5_eswitch_unload_vf_vports(esw, esw->esw_funcs.num_vfs);
1997 		if (clear_vf)
1998 			mlx5_eswitch_clear_vf_vports_info(esw);
1999 	} else if (mlx5_core_ec_sriov_enabled(esw->dev)) {
2000 		mlx5_eswitch_unload_ec_vf_vports(esw, esw->esw_funcs.num_ec_vfs);
2001 		if (clear_vf)
2002 			mlx5_eswitch_clear_ec_vf_vports_info(esw);
2003 	}
2004 
2005 	mlx5_esw_reps_unblock(esw);
2006 
2007 	if (esw->mode == MLX5_ESWITCH_OFFLOADS) {
2008 		struct devlink *devlink = priv_to_devlink(esw->dev);
2009 
2010 		devl_rate_nodes_destroy(devlink);
2011 	}
2012 	/* Destroy legacy fdb when disabling sriov in legacy mode. */
2013 	if (esw->mode == MLX5_ESWITCH_LEGACY)
2014 		mlx5_eswitch_disable_locked(esw);
2015 
2016 	if (!mlx5_core_is_ecpf(esw->dev))
2017 		esw->esw_funcs.num_vfs = 0;
2018 	else
2019 		esw->esw_funcs.num_ec_vfs = 0;
2020 }
2021 
2022 /* Free resources for corresponding eswitch mode. It is called by devlink
2023  * when changing eswitch mode or modprobe when unloading driver.
2024  */
2025 void mlx5_eswitch_disable_locked(struct mlx5_eswitch *esw)
2026 {
2027 	struct devlink *devlink = priv_to_devlink(esw->dev);
2028 
2029 	/* Notify eswitch users that it is exiting from current mode.
2030 	 * So that it can do necessary cleanup before the eswitch is disabled.
2031 	 */
2032 	mlx5_esw_mode_change_notify(esw, MLX5_ESWITCH_LEGACY);
2033 
2034 	mlx5_eq_notifier_unregister(esw->dev, &esw->nb);
2035 	mlx5_eswitch_event_handler_unregister(esw);
2036 	mlx5_eswitch_invalidate_wq(esw);
2037 
2038 	esw_info(esw->dev, "Disable: mode(%s), nvfs(%d), necvfs(%d), active vports(%d)\n",
2039 		 esw->mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
2040 		 esw->esw_funcs.num_vfs, esw->esw_funcs.num_ec_vfs, esw->enabled_vports);
2041 
2042 	if (esw->fdb_table.flags & MLX5_ESW_FDB_CREATED) {
2043 		esw->fdb_table.flags &= ~MLX5_ESW_FDB_CREATED;
2044 		if (esw->mode == MLX5_ESWITCH_OFFLOADS)
2045 			esw_offloads_disable(esw);
2046 		else if (esw->mode == MLX5_ESWITCH_LEGACY)
2047 			esw_legacy_disable(esw);
2048 		mlx5_esw_acls_ns_cleanup(esw);
2049 	}
2050 
2051 	if (esw->mode == MLX5_ESWITCH_OFFLOADS)
2052 		devl_rate_nodes_destroy(devlink);
2053 }
2054 
2055 void mlx5_eswitch_disable(struct mlx5_eswitch *esw)
2056 {
2057 	if (!mlx5_esw_allowed(esw))
2058 		return;
2059 
2060 	devl_assert_locked(priv_to_devlink(esw->dev));
2061 	mlx5_lag_disable_change(esw->dev);
2062 
2063 	mlx5_esw_reps_block(esw);
2064 	mlx5_eswitch_disable_locked(esw);
2065 	mlx5_esw_reps_unblock(esw);
2066 
2067 	esw->mode = MLX5_ESWITCH_LEGACY;
2068 	mlx5_sd_eswitch_mode_set(esw->dev, MLX5_ESWITCH_LEGACY);
2069 	mlx5_lag_enable_change(esw->dev);
2070 }
2071 
2072 static int mlx5_esw_sf_max_pf_functions(struct mlx5_core_dev *dev,
2073 					u16 vport_num, u16 *max_sfs,
2074 					u16 *sf_base_id)
2075 {
2076 	int query_out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
2077 	void *query_ctx;
2078 	void *hca_caps;
2079 	int err;
2080 
2081 	query_ctx = kzalloc(query_out_sz, GFP_KERNEL);
2082 	if (!query_ctx)
2083 		return -ENOMEM;
2084 
2085 	err = mlx5_vport_get_other_func_general_cap(dev, vport_num, query_ctx);
2086 	if (err)
2087 		goto out_free;
2088 
2089 	hca_caps = MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability);
2090 	*max_sfs = MLX5_GET(cmd_hca_cap, hca_caps, max_num_sf);
2091 	*sf_base_id = MLX5_GET(cmd_hca_cap, hca_caps, sf_base_id);
2092 
2093 out_free:
2094 	kfree(query_ctx);
2095 	return err;
2096 }
2097 
2098 int mlx5_esw_sf_max_hpf_functions(struct mlx5_core_dev *dev, u16 *max_sfs,
2099 				  u16 *sf_base_id)
2100 {
2101 	if (!mlx5_core_is_ecpf(dev) ||
2102 	    !mlx5_esw_host_functions_enabled(dev)) {
2103 		*max_sfs = 0;
2104 		return 0;
2105 	}
2106 
2107 	return mlx5_esw_sf_max_pf_functions(dev, MLX5_VPORT_HOST_PF, max_sfs,
2108 					    sf_base_id);
2109 }
2110 
2111 int mlx5_esw_sf_max_spf_functions(struct mlx5_core_dev *dev, int spf_idx,
2112 				  u16 *max_sfs, u16 *sf_base_id)
2113 {
2114 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2115 	u16 vport_num;
2116 
2117 	if (!mlx5_esw_allowed(esw)) {
2118 		*max_sfs = 0;
2119 		return 0;
2120 	}
2121 
2122 	if (spf_idx >= esw->esw_funcs.num_spfs)
2123 		return -EINVAL;
2124 
2125 	vport_num = esw->esw_funcs.spfs[spf_idx].vport_num;
2126 	return mlx5_esw_sf_max_pf_functions(dev, vport_num, max_sfs,
2127 					    sf_base_id);
2128 }
2129 
2130 int mlx5_esw_get_num_spfs(struct mlx5_core_dev *dev)
2131 {
2132 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2133 
2134 	if (!mlx5_esw_allowed(esw))
2135 		return 0;
2136 
2137 	return esw->esw_funcs.num_spfs;
2138 }
2139 
2140 int mlx5_esw_spf_get_host_number(struct mlx5_core_dev *dev, int spf_idx,
2141 				 u16 *host_number)
2142 {
2143 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2144 
2145 	if (!mlx5_esw_allowed(esw))
2146 		return -EPERM;
2147 
2148 	if (spf_idx >= esw->esw_funcs.num_spfs)
2149 		return -EINVAL;
2150 
2151 	*host_number = esw->esw_funcs.spfs[spf_idx].host_number;
2152 	return 0;
2153 }
2154 
2155 u16 mlx5_esw_get_hpf_host_number(struct mlx5_core_dev *dev)
2156 {
2157 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2158 
2159 	if (!mlx5_esw_allowed(esw))
2160 		return 0;
2161 
2162 	return esw->esw_funcs.hpf_host_number;
2163 }
2164 
2165 u16 mlx5_esw_get_hpf_pf_num(struct mlx5_core_dev *dev)
2166 {
2167 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2168 
2169 	if (mlx5_core_is_ecpf_esw_manager(dev) &&
2170 	    MLX5_CAP_GEN(dev, query_host_net_function_v1))
2171 		return esw->esw_funcs.hpf_pf_num;
2172 
2173 	return PCI_FUNC(dev->pdev->devfn);
2174 }
2175 
2176 u16 mlx5_esw_sf_controller_to_pfnum(struct mlx5_core_dev *dev, u32 controller)
2177 {
2178 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2179 	struct mlx5_esw_functions *esw_funcs;
2180 	int i;
2181 
2182 	if (!controller)
2183 		return PCI_FUNC(dev->pdev->devfn);
2184 
2185 	esw_funcs = &esw->esw_funcs;
2186 	for (i = 0; i < esw_funcs->num_spfs; i++)
2187 		if (controller == esw_funcs->spfs[i].host_number + 1)
2188 			return esw_funcs->spfs[i].pf_num;
2189 
2190 	return mlx5_esw_get_hpf_pf_num(dev);
2191 }
2192 
2193 bool mlx5_esw_has_spf_sfs(struct mlx5_core_dev *dev)
2194 {
2195 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2196 
2197 	if (!mlx5_esw_allowed(esw))
2198 		return false;
2199 
2200 	return esw->esw_funcs.has_spf_sfs;
2201 }
2202 
2203 static int mlx5_esw_hpf_info_init(struct mlx5_eswitch *esw)
2204 {
2205 	struct mlx5_esw_pf_info host_pf_info;
2206 	const u32 *query_host_out;
2207 
2208 	if (!mlx5_core_is_ecpf_esw_manager(esw->dev))
2209 		return 0;
2210 
2211 	query_host_out = mlx5_esw_query_functions(esw->dev);
2212 	if (IS_ERR(query_host_out))
2213 		return PTR_ERR(query_host_out);
2214 
2215 	/* Mark non local controller with non zero controller number. */
2216 	host_pf_info = mlx5_esw_get_host_pf_info(esw->dev, query_host_out);
2217 	esw->esw_funcs.hpf_host_number = host_pf_info.host_number;
2218 	esw->esw_funcs.hpf_pf_num = host_pf_info.pf_num;
2219 	kvfree(query_host_out);
2220 	return 0;
2221 }
2222 
2223 int mlx5_esw_vport_alloc(struct mlx5_eswitch *esw, int index, u16 vport_num)
2224 {
2225 	struct mlx5_vport *vport;
2226 	int err;
2227 
2228 	vport = kzalloc_obj(*vport);
2229 	if (!vport)
2230 		return -ENOMEM;
2231 
2232 	vport->dev = esw->dev;
2233 	vport->vport = vport_num;
2234 	vport->index = index;
2235 	vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
2236 	vport->vhca_id = MLX5_VHCA_ID_INVALID;
2237 	INIT_WORK(&vport->vport_change_handler, esw_vport_change_handler);
2238 	err = xa_insert(&esw->vports, vport_num, vport, GFP_KERNEL);
2239 	if (err)
2240 		goto insert_err;
2241 
2242 	esw->total_vports++;
2243 	return 0;
2244 
2245 insert_err:
2246 	kfree(vport);
2247 	return err;
2248 }
2249 
2250 void mlx5_esw_vport_free(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
2251 {
2252 	esw->total_vports--;
2253 	xa_erase(&esw->vports, vport->vport);
2254 	kfree(vport);
2255 }
2256 
2257 static void mlx5_esw_spfs_cleanup(struct mlx5_eswitch *esw)
2258 {
2259 	struct mlx5_esw_functions *esw_funcs = &esw->esw_funcs;
2260 	int i;
2261 
2262 	for (i = 0; i < esw_funcs->num_spfs; i++)
2263 		mlx5_esw_destroy_esw_vport(esw->dev,
2264 					   esw_funcs->spfs[i].vport_num);
2265 
2266 	kfree(esw_funcs->spfs);
2267 	esw_funcs->spfs = NULL;
2268 	esw_funcs->num_spfs = 0;
2269 }
2270 
2271 static int mlx5_esw_spfs_init(struct mlx5_eswitch *esw)
2272 {
2273 	struct mlx5_esw_functions *esw_funcs = &esw->esw_funcs;
2274 	struct mlx5_core_dev *dev = esw->dev;
2275 	int num_entries;
2276 	const u8 *entry;
2277 	const u32 *out;
2278 	int err = 0;
2279 	int pf_type;
2280 	u16 vhca_id;
2281 	int i;
2282 
2283 	if (!MLX5_CAP_GEN(dev, query_host_net_function_v1))
2284 		return 0;
2285 
2286 	out = mlx5_esw_query_functions(dev);
2287 	if (IS_ERR(out))
2288 		return PTR_ERR(out);
2289 
2290 	num_entries = MLX5_GET(query_esw_functions_out, out, net_function_num);
2291 	if (!num_entries)
2292 		goto out_free;
2293 
2294 	esw_funcs->spfs = kcalloc(num_entries, sizeof(*esw_funcs->spfs),
2295 				  GFP_KERNEL);
2296 	if (!esw_funcs->spfs) {
2297 		err = -ENOMEM;
2298 		goto out_free;
2299 	}
2300 
2301 	entry = MLX5_ADDR_OF(query_esw_functions_out, out, net_function_params);
2302 
2303 	for (i = 0; i < num_entries; i++) {
2304 		u16 vport_num;
2305 
2306 		pf_type = MLX5_GET(network_function_params, entry, pci_pf_type);
2307 		if (pf_type != MLX5_PCI_PF_TYPE_SATELLITE_PF) {
2308 			entry += MLX5_UN_SZ_BYTES(net_function_params);
2309 			continue;
2310 		}
2311 
2312 		if (!MLX5_GET(network_function_params, entry,
2313 			      esw_vport_manual)) {
2314 			esw_warn(dev, "Satellite PF without esw_vport_manual is not supported\n");
2315 			entry += MLX5_UN_SZ_BYTES(net_function_params);
2316 			continue;
2317 		}
2318 
2319 		vhca_id = MLX5_GET(network_function_params, entry, vhca_id);
2320 
2321 		err = mlx5_esw_create_esw_vport(dev, vhca_id, &vport_num);
2322 		if (err) {
2323 			esw_warn(dev, "Failed to create satellite PF vport for vhca_id 0x%x, err %d\n",
2324 				 vhca_id, err);
2325 			goto spfs_cleanup;
2326 		}
2327 
2328 		esw_funcs->spfs[esw_funcs->num_spfs].vport_num = vport_num;
2329 		esw_funcs->spfs[esw_funcs->num_spfs].vhca_id = vhca_id;
2330 		esw_funcs->spfs[esw_funcs->num_spfs].host_number =
2331 			MLX5_GET(network_function_params, entry, host_number);
2332 		esw_funcs->spfs[esw_funcs->num_spfs].pf_num =
2333 			MLX5_GET(network_function_params, entry,
2334 				 pci_device_function);
2335 		esw_funcs->num_spfs++;
2336 
2337 		entry += MLX5_UN_SZ_BYTES(net_function_params);
2338 	}
2339 
2340 	if (!esw_funcs->num_spfs) {
2341 		kfree(esw_funcs->spfs);
2342 		esw_funcs->spfs = NULL;
2343 	}
2344 
2345 	kvfree(out);
2346 	return 0;
2347 
2348 spfs_cleanup:
2349 	mlx5_esw_spfs_cleanup(esw);
2350 out_free:
2351 	kvfree(out);
2352 	return err;
2353 }
2354 
2355 static void mlx5_esw_vports_cleanup(struct mlx5_eswitch *esw)
2356 {
2357 	struct mlx5_vport *vport;
2358 	unsigned long i;
2359 
2360 	mlx5_esw_spfs_cleanup(esw);
2361 	esw->esw_funcs.has_spf_sfs = false;
2362 	mlx5_esw_for_each_vport(esw, i, vport)
2363 		mlx5_esw_vport_free(esw, vport);
2364 	xa_destroy(&esw->vports);
2365 }
2366 
2367 static int mlx5_esw_vports_init(struct mlx5_eswitch *esw)
2368 {
2369 	struct mlx5_core_dev *dev = esw->dev;
2370 	u16 max_sfs, base_sf_num;
2371 	int idx = 0;
2372 	int err;
2373 	int i;
2374 
2375 	xa_init(&esw->vports);
2376 
2377 	err = mlx5_esw_hpf_info_init(esw);
2378 	if (err)
2379 		goto err;
2380 
2381 	if (mlx5_esw_host_functions_enabled(dev)) {
2382 		err = mlx5_esw_vport_alloc(esw, idx, MLX5_VPORT_HOST_PF);
2383 		if (err)
2384 			goto err;
2385 		if (esw->first_host_vport == MLX5_VPORT_HOST_PF)
2386 			xa_set_mark(&esw->vports, idx, MLX5_ESW_VPT_HOST_FN);
2387 		idx++;
2388 		for (i = 0; i < mlx5_core_max_vfs(dev); i++) {
2389 			err = mlx5_esw_vport_alloc(esw, idx, idx);
2390 			if (err)
2391 				goto err;
2392 			xa_set_mark(&esw->vports, idx, MLX5_ESW_VPT_VF);
2393 			xa_set_mark(&esw->vports, idx, MLX5_ESW_VPT_HOST_FN);
2394 			idx++;
2395 		}
2396 	}
2397 
2398 	base_sf_num = mlx5_sf_start_function_id(dev);
2399 	for (i = 0; i < mlx5_sf_max_functions(dev); i++) {
2400 		err = mlx5_esw_vport_alloc(esw, idx, base_sf_num + i);
2401 		if (err)
2402 			goto err;
2403 		xa_set_mark(&esw->vports, base_sf_num + i, MLX5_ESW_VPT_SF);
2404 		idx++;
2405 	}
2406 
2407 	err = mlx5_esw_sf_max_hpf_functions(dev, &max_sfs, &base_sf_num);
2408 	if (err)
2409 		goto err;
2410 	for (i = 0; i < max_sfs; i++) {
2411 		err = mlx5_esw_vport_alloc(esw, idx, base_sf_num + i);
2412 		if (err)
2413 			goto err;
2414 		xa_set_mark(&esw->vports, base_sf_num + i, MLX5_ESW_VPT_SF);
2415 		idx++;
2416 	}
2417 
2418 	err = mlx5_esw_spfs_init(esw);
2419 	if (err)
2420 		goto err;
2421 
2422 	for (i = 0; i < esw->esw_funcs.num_spfs; i++) {
2423 		struct mlx5_vport *vport;
2424 		u16 vport_num;
2425 
2426 		vport_num = esw->esw_funcs.spfs[i].vport_num;
2427 		err = mlx5_esw_vport_alloc(esw, idx++, vport_num);
2428 		if (err)
2429 			goto err;
2430 		vport = mlx5_eswitch_get_vport(esw, vport_num);
2431 		vport->vhca_id = esw->esw_funcs.spfs[i].vhca_id;
2432 
2433 		err = mlx5_esw_sf_max_spf_functions(dev, i,
2434 						    &max_sfs, &base_sf_num);
2435 		if (err)
2436 			goto err;
2437 		if (max_sfs)
2438 			esw->esw_funcs.has_spf_sfs = true;
2439 		for (int j = 0; j < max_sfs; j++) {
2440 			err = mlx5_esw_vport_alloc(esw, idx,
2441 						   base_sf_num + j);
2442 			if (err)
2443 				goto err;
2444 			xa_set_mark(&esw->vports, base_sf_num + j,
2445 				    MLX5_ESW_VPT_SF);
2446 			idx++;
2447 		}
2448 	}
2449 
2450 	if (mlx5_core_ec_sriov_enabled(esw->dev)) {
2451 		int ec_vf_base_num = mlx5_core_ec_vf_vport_base(dev);
2452 
2453 		for (i = 0; i < mlx5_core_max_ec_vfs(esw->dev); i++) {
2454 			err = mlx5_esw_vport_alloc(esw, idx, ec_vf_base_num + i);
2455 			if (err)
2456 				goto err;
2457 			idx++;
2458 		}
2459 	}
2460 
2461 	if (mlx5_ecpf_vport_exists(dev) ||
2462 	    mlx5_core_is_ecpf_esw_manager(dev)) {
2463 		err = mlx5_esw_vport_alloc(esw, idx, MLX5_VPORT_ECPF);
2464 		if (err)
2465 			goto err;
2466 		idx++;
2467 	}
2468 	err = mlx5_esw_vport_alloc(esw, idx, MLX5_VPORT_UPLINK);
2469 	if (err)
2470 		goto err;
2471 
2472 	/* Adjacent vports or other dynamically create vports will use this */
2473 	esw->last_vport_idx = ++idx;
2474 	return 0;
2475 
2476 err:
2477 	mlx5_esw_vports_cleanup(esw);
2478 	return err;
2479 }
2480 
2481 static int mlx5_devlink_esw_multiport_set(struct devlink *devlink, u32 id,
2482 					  struct devlink_param_gset_ctx *ctx,
2483 					  struct netlink_ext_ack *extack)
2484 {
2485 	struct mlx5_core_dev *dev = devlink_priv(devlink);
2486 
2487 	if (!MLX5_ESWITCH_MANAGER(dev))
2488 		return -EOPNOTSUPP;
2489 
2490 	if (ctx->val.vbool)
2491 		return mlx5_lag_mpesw_enable(dev);
2492 
2493 	mlx5_lag_mpesw_disable(dev);
2494 	return 0;
2495 }
2496 
2497 static int mlx5_devlink_esw_multiport_get(struct devlink *devlink, u32 id,
2498 					  struct devlink_param_gset_ctx *ctx,
2499 					  struct netlink_ext_ack *extack)
2500 {
2501 	struct mlx5_core_dev *dev = devlink_priv(devlink);
2502 
2503 	ctx->val.vbool = mlx5_lag_is_mpesw(dev);
2504 	return 0;
2505 }
2506 
2507 static const struct devlink_param mlx5_eswitch_params[] = {
2508 	DEVLINK_PARAM_DRIVER(MLX5_DEVLINK_PARAM_ID_ESW_MULTIPORT,
2509 			     "esw_multiport", DEVLINK_PARAM_TYPE_BOOL,
2510 			     BIT(DEVLINK_PARAM_CMODE_RUNTIME),
2511 			     mlx5_devlink_esw_multiport_get,
2512 			     mlx5_devlink_esw_multiport_set, NULL),
2513 };
2514 
2515 int mlx5_eswitch_init(struct mlx5_core_dev *dev)
2516 {
2517 	struct mlx5_eswitch *esw;
2518 	int err;
2519 
2520 	if (!MLX5_VPORT_MANAGER(dev) && !MLX5_ESWITCH_MANAGER(dev))
2521 		return 0;
2522 
2523 	esw = kzalloc_obj(*esw);
2524 	if (!esw)
2525 		return -ENOMEM;
2526 
2527 	err = devl_params_register(priv_to_devlink(dev), mlx5_eswitch_params,
2528 				   ARRAY_SIZE(mlx5_eswitch_params));
2529 	if (err)
2530 		goto free_esw;
2531 
2532 	esw->dev = dev;
2533 	dev->priv.eswitch = esw;
2534 	esw->manager_vport = mlx5_eswitch_manager_vport(dev);
2535 	esw->first_host_vport = mlx5_eswitch_first_host_vport_num(dev);
2536 
2537 	esw->debugfs_root = debugfs_create_dir("esw", mlx5_debugfs_get_dev_root(dev));
2538 	esw->work_queue = create_singlethread_workqueue("mlx5_esw_wq");
2539 	if (!esw->work_queue) {
2540 		err = -ENOMEM;
2541 		goto abort;
2542 	}
2543 
2544 	err = mlx5_esw_host_functions_enabled_query(esw);
2545 	if (err)
2546 		goto abort;
2547 
2548 	err = mlx5_esw_vports_init(esw);
2549 	if (err)
2550 		goto abort;
2551 
2552 	err = esw_offloads_init(esw);
2553 	if (err)
2554 		goto reps_err;
2555 
2556 	esw->mode = MLX5_ESWITCH_LEGACY;
2557 	err = mlx5_esw_qos_init(esw);
2558 	if (err)
2559 		goto reps_err;
2560 
2561 	mutex_init(&esw->offloads.encap_tbl_lock);
2562 	hash_init(esw->offloads.encap_tbl);
2563 	mutex_init(&esw->offloads.decap_tbl_lock);
2564 	hash_init(esw->offloads.decap_tbl);
2565 	mlx5e_mod_hdr_tbl_init(&esw->offloads.mod_hdr);
2566 	atomic64_set(&esw->offloads.num_flows, 0);
2567 	ida_init(&esw->offloads.vport_metadata_ida);
2568 	xa_init_flags(&esw->offloads.vhca_map, XA_FLAGS_ALLOC);
2569 	xa_init(&esw->vhca_type_map);
2570 	mutex_init(&esw->state_lock);
2571 	init_rwsem(&esw->mode_lock);
2572 	refcount_set(&esw->qos.refcnt, 0);
2573 	init_waitqueue_head(&esw->work_queue_wait);
2574 	atomic_set(&esw->generation, 0);
2575 
2576 	esw->enabled_vports = 0;
2577 	esw->offloads.inline_mode = MLX5_INLINE_MODE_NONE;
2578 	if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, reformat) &&
2579 	    MLX5_CAP_ESW_FLOWTABLE_FDB(dev, decap))
2580 		esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_BASIC;
2581 	else
2582 		esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_NONE;
2583 
2584 	esw_info(dev,
2585 		 "Total vports %d, per vport: max uc(%d) max mc(%d)\n",
2586 		 esw->total_vports,
2587 		 MLX5_MAX_UC_PER_VPORT(dev),
2588 		 MLX5_MAX_MC_PER_VPORT(dev));
2589 	return 0;
2590 
2591 reps_err:
2592 	mlx5_esw_vports_cleanup(esw);
2593 	dev->priv.eswitch = NULL;
2594 abort:
2595 	if (esw->work_queue)
2596 		destroy_workqueue(esw->work_queue);
2597 	debugfs_remove_recursive(esw->debugfs_root);
2598 	devl_params_unregister(priv_to_devlink(dev), mlx5_eswitch_params,
2599 			       ARRAY_SIZE(mlx5_eswitch_params));
2600 free_esw:
2601 	kfree(esw);
2602 	return err;
2603 }
2604 
2605 void mlx5_eswitch_cleanup(struct mlx5_eswitch *esw)
2606 {
2607 	if (!esw)
2608 		return;
2609 
2610 	esw_info(esw->dev, "cleanup\n");
2611 
2612 	mlx5_eswitch_invalidate_wq(esw);
2613 	destroy_workqueue(esw->work_queue);
2614 	mlx5_esw_qos_cleanup(esw);
2615 	WARN_ON(refcount_read(&esw->qos.refcnt));
2616 	mutex_destroy(&esw->state_lock);
2617 	WARN_ON(!xa_empty(&esw->offloads.vhca_map));
2618 	xa_destroy(&esw->offloads.vhca_map);
2619 	xa_destroy(&esw->vhca_type_map);
2620 	ida_destroy(&esw->offloads.vport_metadata_ida);
2621 	mlx5e_mod_hdr_tbl_destroy(&esw->offloads.mod_hdr);
2622 	mutex_destroy(&esw->offloads.encap_tbl_lock);
2623 	mutex_destroy(&esw->offloads.decap_tbl_lock);
2624 	esw_offloads_cleanup(esw);
2625 	esw->dev->priv.eswitch = NULL;
2626 	mlx5_esw_vports_cleanup(esw);
2627 	debugfs_remove_recursive(esw->debugfs_root);
2628 	devl_params_unregister(priv_to_devlink(esw->dev), mlx5_eswitch_params,
2629 			       ARRAY_SIZE(mlx5_eswitch_params));
2630 	kfree(esw);
2631 }
2632 
2633 /* Vport Administration */
2634 static int
2635 mlx5_esw_set_vport_mac_locked(struct mlx5_eswitch *esw,
2636 			      struct mlx5_vport *evport, const u8 *mac)
2637 {
2638 	u16 vport_num = evport->vport;
2639 	u64 node_guid;
2640 	int err = 0;
2641 
2642 	if (is_multicast_ether_addr(mac))
2643 		return -EINVAL;
2644 
2645 	if (evport->info.spoofchk && !is_valid_ether_addr(mac))
2646 		mlx5_core_warn(esw->dev,
2647 			       "Set invalid MAC while spoofchk is on, vport(%d)\n",
2648 			       vport_num);
2649 
2650 	err = mlx5_modify_nic_vport_mac_address(esw->dev, vport_num, mac);
2651 	if (err) {
2652 		mlx5_core_warn(esw->dev,
2653 			       "Failed to mlx5_modify_nic_vport_mac vport(%d) err=(%d)\n",
2654 			       vport_num, err);
2655 		return err;
2656 	}
2657 
2658 	node_guid_gen_from_mac(&node_guid, mac);
2659 	err = mlx5_modify_nic_vport_node_guid(esw->dev, vport_num, node_guid);
2660 	if (err)
2661 		mlx5_core_warn(esw->dev,
2662 			       "Failed to set vport %d node guid, err = %d. RDMA_CM will not function properly for this VF.\n",
2663 			       vport_num, err);
2664 
2665 	ether_addr_copy(evport->info.mac, mac);
2666 	evport->info.node_guid = node_guid;
2667 	if (evport->enabled && esw->mode == MLX5_ESWITCH_LEGACY)
2668 		err = esw_acl_ingress_lgcy_setup(esw, evport);
2669 
2670 	return err;
2671 }
2672 
2673 int mlx5_eswitch_set_vport_mac(struct mlx5_eswitch *esw,
2674 			       u16 vport, const u8 *mac)
2675 {
2676 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2677 	int err = 0;
2678 
2679 	if (IS_ERR(evport))
2680 		return PTR_ERR(evport);
2681 
2682 	mutex_lock(&esw->state_lock);
2683 	err = mlx5_esw_set_vport_mac_locked(esw, evport, mac);
2684 	mutex_unlock(&esw->state_lock);
2685 	return err;
2686 }
2687 
2688 static bool mlx5_esw_check_port_type(struct mlx5_eswitch *esw, u16 vport_num, xa_mark_t mark)
2689 {
2690 	return xa_get_mark(&esw->vports, vport_num, mark);
2691 }
2692 
2693 bool mlx5_eswitch_is_vf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2694 {
2695 	return mlx5_esw_check_port_type(esw, vport_num, MLX5_ESW_VPT_VF);
2696 }
2697 
2698 int mlx5_esw_spf_vport_to_idx(struct mlx5_eswitch *esw, u16 vport_num)
2699 {
2700 	struct mlx5_esw_functions *esw_funcs = &esw->esw_funcs;
2701 	int i;
2702 
2703 	for (i = 0; i < esw_funcs->num_spfs; i++) {
2704 		if (esw_funcs->spfs[i].vport_num == vport_num)
2705 			return i;
2706 	}
2707 
2708 	return -ENOENT;
2709 }
2710 
2711 bool mlx5_esw_is_spf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2712 {
2713 	return mlx5_esw_spf_vport_to_idx(esw, vport_num) >= 0;
2714 }
2715 
2716 bool mlx5_eswitch_is_pf_vf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2717 {
2718 	return vport_num == MLX5_VPORT_HOST_PF ||
2719 		mlx5_eswitch_is_vf_vport(esw, vport_num) ||
2720 		mlx5_esw_is_spf_vport(esw, vport_num);
2721 }
2722 
2723 bool mlx5_esw_is_sf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2724 {
2725 	return mlx5_esw_check_port_type(esw, vport_num, MLX5_ESW_VPT_SF);
2726 }
2727 
2728 int mlx5_eswitch_set_vport_state(struct mlx5_eswitch *esw,
2729 				 u16 vport, int link_state)
2730 {
2731 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2732 	int opmod = MLX5_VPORT_STATE_OP_MOD_ESW_VPORT;
2733 	int other_vport = 1;
2734 	int err = 0;
2735 
2736 	if (!mlx5_esw_allowed(esw))
2737 		return -EPERM;
2738 	if (IS_ERR(evport))
2739 		return PTR_ERR(evport);
2740 
2741 	if (vport == MLX5_VPORT_UPLINK) {
2742 		opmod = MLX5_VPORT_STATE_OP_MOD_UPLINK;
2743 		other_vport = 0;
2744 		vport = 0;
2745 	}
2746 	mutex_lock(&esw->state_lock);
2747 	if (esw->mode != MLX5_ESWITCH_LEGACY) {
2748 		err = -EOPNOTSUPP;
2749 		goto unlock;
2750 	}
2751 
2752 	err = mlx5_modify_vport_admin_state(esw->dev, opmod, vport, other_vport, link_state);
2753 	if (err) {
2754 		mlx5_core_warn(esw->dev, "Failed to set vport %d link state, opmod = %d, err = %d",
2755 			       vport, opmod, err);
2756 		goto unlock;
2757 	}
2758 
2759 	evport->info.link_state = link_state;
2760 
2761 unlock:
2762 	mutex_unlock(&esw->state_lock);
2763 	return err;
2764 }
2765 
2766 int mlx5_eswitch_get_vport_config(struct mlx5_eswitch *esw,
2767 				  u16 vport, struct ifla_vf_info *ivi)
2768 {
2769 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2770 	u32 max_rate, min_rate;
2771 
2772 	if (IS_ERR(evport))
2773 		return PTR_ERR(evport);
2774 
2775 	memset(ivi, 0, sizeof(*ivi));
2776 	ivi->vf = vport - 1;
2777 
2778 	mutex_lock(&esw->state_lock);
2779 
2780 	mlx5_query_nic_vport_mac_address(esw->dev, vport, true,
2781 					 evport->info.mac);
2782 	ether_addr_copy(ivi->mac, evport->info.mac);
2783 	ivi->linkstate = evport->info.link_state;
2784 	ivi->vlan = evport->info.vlan;
2785 	ivi->qos = evport->info.qos;
2786 	ivi->spoofchk = evport->info.spoofchk;
2787 	ivi->trusted = evport->info.trusted;
2788 
2789 	if (mlx5_esw_qos_get_vport_rate(evport, &max_rate, &min_rate)) {
2790 		ivi->max_tx_rate = max_rate;
2791 		ivi->min_tx_rate = min_rate;
2792 	}
2793 	mutex_unlock(&esw->state_lock);
2794 
2795 	return 0;
2796 }
2797 
2798 int __mlx5_eswitch_set_vport_vlan(struct mlx5_eswitch *esw,
2799 				  u16 vport, u16 vlan, u8 qos, u8 set_flags)
2800 {
2801 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2802 	bool vst_mode_steering = esw_vst_mode_is_steering(esw);
2803 	int err = 0;
2804 
2805 	if (IS_ERR(evport))
2806 		return PTR_ERR(evport);
2807 	if (vlan > 4095 || qos > 7)
2808 		return -EINVAL;
2809 
2810 	if (esw->mode == MLX5_ESWITCH_OFFLOADS || !vst_mode_steering) {
2811 		err = modify_esw_vport_cvlan(esw->dev, vport, vlan, qos, set_flags);
2812 		if (err)
2813 			return err;
2814 	}
2815 
2816 	evport->info.vlan = vlan;
2817 	evport->info.qos = qos;
2818 	if (evport->enabled && esw->mode == MLX5_ESWITCH_LEGACY) {
2819 		err = esw_acl_ingress_lgcy_setup(esw, evport);
2820 		if (err)
2821 			return err;
2822 		err = esw_acl_egress_lgcy_setup(esw, evport);
2823 	}
2824 
2825 	return err;
2826 }
2827 
2828 int mlx5_eswitch_get_vport_stats(struct mlx5_eswitch *esw,
2829 				 u16 vport_num,
2830 				 struct ifla_vf_stats *vf_stats)
2831 {
2832 	struct mlx5_vport *vport = mlx5_eswitch_get_vport(esw, vport_num);
2833 	int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
2834 	u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)] = {};
2835 	struct mlx5_vport_drop_stats stats = {};
2836 	int err = 0;
2837 	u32 *out;
2838 
2839 	if (IS_ERR(vport))
2840 		return PTR_ERR(vport);
2841 
2842 	out = kvzalloc(outlen, GFP_KERNEL);
2843 	if (!out)
2844 		return -ENOMEM;
2845 
2846 	MLX5_SET(query_vport_counter_in, in, opcode,
2847 		 MLX5_CMD_OP_QUERY_VPORT_COUNTER);
2848 	MLX5_SET(query_vport_counter_in, in, op_mod, 0);
2849 	MLX5_SET(query_vport_counter_in, in, vport_number, vport->vport);
2850 	MLX5_SET(query_vport_counter_in, in, other_vport, 1);
2851 
2852 	err = mlx5_cmd_exec_inout(esw->dev, query_vport_counter, in, out);
2853 	if (err)
2854 		goto free_out;
2855 
2856 	#define MLX5_GET_CTR(p, x) \
2857 		MLX5_GET64(query_vport_counter_out, p, x)
2858 
2859 	memset(vf_stats, 0, sizeof(*vf_stats));
2860 	vf_stats->rx_packets =
2861 		MLX5_GET_CTR(out, received_eth_unicast.packets) +
2862 		MLX5_GET_CTR(out, received_ib_unicast.packets) +
2863 		MLX5_GET_CTR(out, received_eth_multicast.packets) +
2864 		MLX5_GET_CTR(out, received_ib_multicast.packets) +
2865 		MLX5_GET_CTR(out, received_eth_broadcast.packets);
2866 
2867 	vf_stats->rx_bytes =
2868 		MLX5_GET_CTR(out, received_eth_unicast.octets) +
2869 		MLX5_GET_CTR(out, received_ib_unicast.octets) +
2870 		MLX5_GET_CTR(out, received_eth_multicast.octets) +
2871 		MLX5_GET_CTR(out, received_ib_multicast.octets) +
2872 		MLX5_GET_CTR(out, received_eth_broadcast.octets);
2873 
2874 	vf_stats->tx_packets =
2875 		MLX5_GET_CTR(out, transmitted_eth_unicast.packets) +
2876 		MLX5_GET_CTR(out, transmitted_ib_unicast.packets) +
2877 		MLX5_GET_CTR(out, transmitted_eth_multicast.packets) +
2878 		MLX5_GET_CTR(out, transmitted_ib_multicast.packets) +
2879 		MLX5_GET_CTR(out, transmitted_eth_broadcast.packets);
2880 
2881 	vf_stats->tx_bytes =
2882 		MLX5_GET_CTR(out, transmitted_eth_unicast.octets) +
2883 		MLX5_GET_CTR(out, transmitted_ib_unicast.octets) +
2884 		MLX5_GET_CTR(out, transmitted_eth_multicast.octets) +
2885 		MLX5_GET_CTR(out, transmitted_ib_multicast.octets) +
2886 		MLX5_GET_CTR(out, transmitted_eth_broadcast.octets);
2887 
2888 	vf_stats->multicast =
2889 		MLX5_GET_CTR(out, received_eth_multicast.packets) +
2890 		MLX5_GET_CTR(out, received_ib_multicast.packets);
2891 
2892 	vf_stats->broadcast =
2893 		MLX5_GET_CTR(out, received_eth_broadcast.packets);
2894 
2895 	err = mlx5_esw_query_vport_drop_stats(esw->dev, vport, &stats);
2896 	if (err)
2897 		goto free_out;
2898 	vf_stats->rx_dropped = stats.rx_dropped;
2899 	vf_stats->tx_dropped = stats.tx_dropped;
2900 
2901 free_out:
2902 	kvfree(out);
2903 	return err;
2904 }
2905 
2906 u8 mlx5_eswitch_mode(const struct mlx5_core_dev *dev)
2907 {
2908 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2909 
2910 	return mlx5_esw_allowed(esw) ? esw->mode : MLX5_ESWITCH_LEGACY;
2911 }
2912 EXPORT_SYMBOL_GPL(mlx5_eswitch_mode);
2913 
2914 enum devlink_eswitch_encap_mode
2915 mlx5_eswitch_get_encap_mode(const struct mlx5_core_dev *dev)
2916 {
2917 	struct mlx5_eswitch *esw;
2918 
2919 	esw = dev->priv.eswitch;
2920 	return (mlx5_eswitch_mode(dev) == MLX5_ESWITCH_OFFLOADS)  ? esw->offloads.encap :
2921 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
2922 }
2923 EXPORT_SYMBOL(mlx5_eswitch_get_encap_mode);
2924 
2925 bool mlx5_esw_multipath_prereq(struct mlx5_core_dev *dev0,
2926 			       struct mlx5_core_dev *dev1)
2927 {
2928 	return (dev0->priv.eswitch->mode == MLX5_ESWITCH_OFFLOADS &&
2929 		dev1->priv.eswitch->mode == MLX5_ESWITCH_OFFLOADS);
2930 }
2931 
2932 int mlx5_esw_event_notifier_register(struct mlx5_core_dev *dev,
2933 				     struct notifier_block *nb)
2934 {
2935 	return blocking_notifier_chain_register(&dev->priv.esw_n_head, nb);
2936 }
2937 
2938 void mlx5_esw_event_notifier_unregister(struct mlx5_core_dev *dev,
2939 					struct notifier_block *nb)
2940 {
2941 	blocking_notifier_chain_unregister(&dev->priv.esw_n_head, nb);
2942 }
2943 
2944 /**
2945  * mlx5_esw_hold() - Try to take a read lock on esw mode lock.
2946  * @mdev: mlx5 core device.
2947  *
2948  * Should be called by esw resources callers.
2949  *
2950  * Return: true on success or false.
2951  */
2952 bool mlx5_esw_hold(struct mlx5_core_dev *mdev)
2953 {
2954 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2955 
2956 	/* e.g. VF doesn't have eswitch so nothing to do */
2957 	if (!mlx5_esw_allowed(esw))
2958 		return true;
2959 
2960 	if (down_read_trylock(&esw->mode_lock) != 0) {
2961 		if (esw->eswitch_operation_in_progress) {
2962 			up_read(&esw->mode_lock);
2963 			return false;
2964 		}
2965 		return true;
2966 	}
2967 
2968 	return false;
2969 }
2970 
2971 /**
2972  * mlx5_esw_release() - Release a read lock on esw mode lock.
2973  * @mdev: mlx5 core device.
2974  */
2975 void mlx5_esw_release(struct mlx5_core_dev *mdev)
2976 {
2977 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2978 
2979 	if (mlx5_esw_allowed(esw))
2980 		up_read(&esw->mode_lock);
2981 }
2982 
2983 /**
2984  * mlx5_esw_get() - Increase esw user count.
2985  * @mdev: mlx5 core device.
2986  */
2987 void mlx5_esw_get(struct mlx5_core_dev *mdev)
2988 {
2989 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2990 
2991 	if (mlx5_esw_allowed(esw))
2992 		atomic64_inc(&esw->user_count);
2993 }
2994 
2995 /**
2996  * mlx5_esw_put() - Decrease esw user count.
2997  * @mdev: mlx5 core device.
2998  */
2999 void mlx5_esw_put(struct mlx5_core_dev *mdev)
3000 {
3001 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
3002 
3003 	if (mlx5_esw_allowed(esw))
3004 		atomic64_dec_if_positive(&esw->user_count);
3005 }
3006 
3007 /**
3008  * mlx5_esw_try_lock() - Take a write lock on esw mode lock.
3009  * @esw: eswitch device.
3010  *
3011  * Should be called by esw mode change routine.
3012  *
3013  * Return:
3014  * * 0       - esw mode if successfully locked and refcount is 0.
3015  * * -EBUSY  - refcount is not 0.
3016  * * -EINVAL - In the middle of switching mode or lock is already held.
3017  */
3018 int mlx5_esw_try_lock(struct mlx5_eswitch *esw)
3019 {
3020 	if (down_write_trylock(&esw->mode_lock) == 0)
3021 		return -EINVAL;
3022 
3023 	if (esw->eswitch_operation_in_progress ||
3024 	    atomic64_read(&esw->user_count) > 0) {
3025 		up_write(&esw->mode_lock);
3026 		return -EBUSY;
3027 	}
3028 
3029 	return esw->mode;
3030 }
3031 
3032 int mlx5_esw_lock(struct mlx5_eswitch *esw)
3033 {
3034 	down_write(&esw->mode_lock);
3035 
3036 	if (esw->eswitch_operation_in_progress) {
3037 		up_write(&esw->mode_lock);
3038 		return -EBUSY;
3039 	}
3040 
3041 	return 0;
3042 }
3043 
3044 /**
3045  * mlx5_esw_unlock() - Release write lock on esw mode lock
3046  * @esw: eswitch device.
3047  */
3048 void mlx5_esw_unlock(struct mlx5_eswitch *esw)
3049 {
3050 	up_write(&esw->mode_lock);
3051 }
3052 
3053 /**
3054  * mlx5_eswitch_get_total_vports - Get total vports of the eswitch
3055  *
3056  * @dev: Pointer to core device
3057  *
3058  * mlx5_eswitch_get_total_vports returns total number of eswitch vports.
3059  */
3060 u16 mlx5_eswitch_get_total_vports(const struct mlx5_core_dev *dev)
3061 {
3062 	struct mlx5_eswitch *esw;
3063 
3064 	esw = dev->priv.eswitch;
3065 	return mlx5_esw_allowed(esw) ? esw->total_vports : 0;
3066 }
3067 EXPORT_SYMBOL_GPL(mlx5_eswitch_get_total_vports);
3068 
3069 /**
3070  * mlx5_eswitch_get_core_dev - Get the mdev device
3071  * @esw : eswitch device.
3072  *
3073  * Return the mellanox core device which manages the eswitch.
3074  */
3075 struct mlx5_core_dev *mlx5_eswitch_get_core_dev(struct mlx5_eswitch *esw)
3076 {
3077 	return mlx5_esw_allowed(esw) ? esw->dev : NULL;
3078 }
3079 EXPORT_SYMBOL(mlx5_eswitch_get_core_dev);
3080 
3081 bool mlx5_eswitch_block_ipsec(struct mlx5_core_dev *dev)
3082 {
3083 	struct mlx5_eswitch *esw = dev->priv.eswitch;
3084 
3085 	if (!mlx5_esw_allowed(esw))
3086 		return true;
3087 
3088 	mutex_lock(&esw->state_lock);
3089 	if (esw->enabled_ipsec_vf_count) {
3090 		mutex_unlock(&esw->state_lock);
3091 		return false;
3092 	}
3093 
3094 	dev->num_ipsec_offloads++;
3095 	mutex_unlock(&esw->state_lock);
3096 	return true;
3097 }
3098 
3099 void mlx5_eswitch_unblock_ipsec(struct mlx5_core_dev *dev)
3100 {
3101 	struct mlx5_eswitch *esw = dev->priv.eswitch;
3102 
3103 	if (!mlx5_esw_allowed(esw))
3104 		/* Failure means no eswitch => core dev is not a PF */
3105 		return;
3106 
3107 	mutex_lock(&esw->state_lock);
3108 	dev->num_ipsec_offloads--;
3109 	mutex_unlock(&esw->state_lock);
3110 }
3111 
3112 bool mlx5_esw_host_functions_enabled(const struct mlx5_core_dev *dev)
3113 {
3114 	if (!dev->priv.eswitch)
3115 		return true;
3116 
3117 	return !dev->priv.eswitch->esw_funcs.host_funcs_disabled;
3118 }
3119