2 * Copyright (c) 2016, Mellanox Technologies. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
33 #include <linux/etherdevice.h>
34 #include <linux/idr.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 "mlx5_core.h"
41 #include "esw/acl/ofld.h"
42 #include "esw/chains.h"
46 #include "lib/devcom.h"
49 /* There are two match-all miss flows, one for unicast dst mac and
52 #define MLX5_ESW_MISS_FLOWS (2)
53 #define UPLINK_REP_INDEX 0
55 /* Per vport tables */
57 #define MLX5_ESW_VPORT_TABLE_SIZE 128
59 /* This struct is used as a key to the hash table and we need it to be packed
60 * so hash result is consistent
62 struct mlx5_vport_key {
69 struct mlx5_vport_table {
70 struct hlist_node hlist;
71 struct mlx5_flow_table *fdb;
73 struct mlx5_vport_key key;
76 #define MLX5_ESW_VPORT_TBL_NUM_GROUPS 4
78 static struct mlx5_flow_table *
79 esw_vport_tbl_create(struct mlx5_eswitch *esw, struct mlx5_flow_namespace *ns)
81 struct mlx5_flow_table_attr ft_attr = {};
82 struct mlx5_flow_table *fdb;
84 ft_attr.autogroup.max_num_groups = MLX5_ESW_VPORT_TBL_NUM_GROUPS;
85 ft_attr.max_fte = MLX5_ESW_VPORT_TABLE_SIZE;
86 ft_attr.prio = FDB_PER_VPORT;
87 fdb = mlx5_create_auto_grouped_flow_table(ns, &ft_attr);
89 esw_warn(esw->dev, "Failed to create per vport FDB Table err %ld\n",
96 static u32 flow_attr_to_vport_key(struct mlx5_eswitch *esw,
97 struct mlx5_esw_flow_attr *attr,
98 struct mlx5_vport_key *key)
100 key->vport = attr->in_rep->vport;
101 key->chain = attr->chain;
102 key->prio = attr->prio;
103 key->vhca_id = MLX5_CAP_GEN(esw->dev, vhca_id);
104 return jhash(key, sizeof(*key), 0);
107 /* caller must hold vports.lock */
108 static struct mlx5_vport_table *
109 esw_vport_tbl_lookup(struct mlx5_eswitch *esw, struct mlx5_vport_key *skey, u32 key)
111 struct mlx5_vport_table *e;
113 hash_for_each_possible(esw->fdb_table.offloads.vports.table, e, hlist, key)
114 if (!memcmp(&e->key, skey, sizeof(*skey)))
121 esw_vport_tbl_put(struct mlx5_eswitch *esw, struct mlx5_esw_flow_attr *attr)
123 struct mlx5_vport_table *e;
124 struct mlx5_vport_key key;
127 mutex_lock(&esw->fdb_table.offloads.vports.lock);
128 hkey = flow_attr_to_vport_key(esw, attr, &key);
129 e = esw_vport_tbl_lookup(esw, &key, hkey);
130 if (!e || --e->num_rules)
134 mlx5_destroy_flow_table(e->fdb);
137 mutex_unlock(&esw->fdb_table.offloads.vports.lock);
140 static struct mlx5_flow_table *
141 esw_vport_tbl_get(struct mlx5_eswitch *esw, struct mlx5_esw_flow_attr *attr)
143 struct mlx5_core_dev *dev = esw->dev;
144 struct mlx5_flow_namespace *ns;
145 struct mlx5_flow_table *fdb;
146 struct mlx5_vport_table *e;
147 struct mlx5_vport_key skey;
150 mutex_lock(&esw->fdb_table.offloads.vports.lock);
151 hkey = flow_attr_to_vport_key(esw, attr, &skey);
152 e = esw_vport_tbl_lookup(esw, &skey, hkey);
158 e = kzalloc(sizeof(*e), GFP_KERNEL);
160 fdb = ERR_PTR(-ENOMEM);
164 ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_FDB);
166 esw_warn(dev, "Failed to get FDB namespace\n");
167 fdb = ERR_PTR(-ENOENT);
171 fdb = esw_vport_tbl_create(esw, ns);
178 hash_add(esw->fdb_table.offloads.vports.table, &e->hlist, hkey);
180 mutex_unlock(&esw->fdb_table.offloads.vports.lock);
186 mutex_unlock(&esw->fdb_table.offloads.vports.lock);
190 int mlx5_esw_vport_tbl_get(struct mlx5_eswitch *esw)
192 struct mlx5_esw_flow_attr attr = {};
193 struct mlx5_eswitch_rep rep = {};
194 struct mlx5_flow_table *fdb;
195 struct mlx5_vport *vport;
200 mlx5_esw_for_all_vports(esw, i, vport) {
201 attr.in_rep->vport = vport->vport;
202 fdb = esw_vport_tbl_get(esw, &attr);
209 mlx5_esw_vport_tbl_put(esw);
213 void mlx5_esw_vport_tbl_put(struct mlx5_eswitch *esw)
215 struct mlx5_esw_flow_attr attr = {};
216 struct mlx5_eswitch_rep rep = {};
217 struct mlx5_vport *vport;
222 mlx5_esw_for_all_vports(esw, i, vport) {
223 attr.in_rep->vport = vport->vport;
224 esw_vport_tbl_put(esw, &attr);
228 /* End: Per vport tables */
230 static struct mlx5_eswitch_rep *mlx5_eswitch_get_rep(struct mlx5_eswitch *esw,
233 int idx = mlx5_eswitch_vport_num_to_index(esw, vport_num);
235 WARN_ON(idx > esw->total_vports - 1);
236 return &esw->offloads.vport_reps[idx];
240 mlx5_eswitch_set_rule_flow_source(struct mlx5_eswitch *esw,
241 struct mlx5_flow_spec *spec,
242 struct mlx5_esw_flow_attr *attr)
244 if (MLX5_CAP_ESW_FLOWTABLE(esw->dev, flow_source) &&
245 attr && attr->in_rep && attr->in_rep->vport == MLX5_VPORT_UPLINK)
246 spec->flow_context.flow_source = MLX5_FLOW_CONTEXT_FLOW_SOURCE_UPLINK;
250 mlx5_eswitch_set_rule_source_port(struct mlx5_eswitch *esw,
251 struct mlx5_flow_spec *spec,
252 struct mlx5_esw_flow_attr *attr)
257 /* Use metadata matching because vport is not represented by single
258 * VHCA in dual-port RoCE mode, and matching on source vport may fail.
260 if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
261 misc2 = MLX5_ADDR_OF(fte_match_param, spec->match_value, misc_parameters_2);
262 MLX5_SET(fte_match_set_misc2, misc2, metadata_reg_c_0,
263 mlx5_eswitch_get_vport_metadata_for_match(attr->in_mdev->priv.eswitch,
264 attr->in_rep->vport));
266 misc2 = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, misc_parameters_2);
267 MLX5_SET(fte_match_set_misc2, misc2, metadata_reg_c_0,
268 mlx5_eswitch_get_vport_metadata_mask());
270 spec->match_criteria_enable |= MLX5_MATCH_MISC_PARAMETERS_2;
272 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value, misc_parameters);
273 MLX5_SET(fte_match_set_misc, misc, source_port, attr->in_rep->vport);
275 if (MLX5_CAP_ESW(esw->dev, merged_eswitch))
276 MLX5_SET(fte_match_set_misc, misc,
277 source_eswitch_owner_vhca_id,
278 MLX5_CAP_GEN(attr->in_mdev, vhca_id));
280 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, misc_parameters);
281 MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
282 if (MLX5_CAP_ESW(esw->dev, merged_eswitch))
283 MLX5_SET_TO_ONES(fte_match_set_misc, misc,
284 source_eswitch_owner_vhca_id);
286 spec->match_criteria_enable |= MLX5_MATCH_MISC_PARAMETERS;
290 struct mlx5_flow_handle *
291 mlx5_eswitch_add_offloaded_rule(struct mlx5_eswitch *esw,
292 struct mlx5_flow_spec *spec,
293 struct mlx5_esw_flow_attr *attr)
295 struct mlx5_flow_destination dest[MLX5_MAX_FLOW_FWD_VPORTS + 1] = {};
296 struct mlx5_flow_act flow_act = { .flags = FLOW_ACT_NO_APPEND, };
297 bool split = !!(attr->split_count);
298 struct mlx5_flow_handle *rule;
299 struct mlx5_flow_table *fdb;
302 if (esw->mode != MLX5_ESWITCH_OFFLOADS)
303 return ERR_PTR(-EOPNOTSUPP);
305 flow_act.action = attr->action;
306 /* if per flow vlan pop/push is emulated, don't set that into the firmware */
307 if (!mlx5_eswitch_vlan_actions_supported(esw->dev, 1))
308 flow_act.action &= ~(MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
309 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP);
310 else if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH) {
311 flow_act.vlan[0].ethtype = ntohs(attr->vlan_proto[0]);
312 flow_act.vlan[0].vid = attr->vlan_vid[0];
313 flow_act.vlan[0].prio = attr->vlan_prio[0];
314 if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2) {
315 flow_act.vlan[1].ethtype = ntohs(attr->vlan_proto[1]);
316 flow_act.vlan[1].vid = attr->vlan_vid[1];
317 flow_act.vlan[1].prio = attr->vlan_prio[1];
321 if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
322 struct mlx5_flow_table *ft;
325 flow_act.flags |= FLOW_ACT_IGNORE_FLOW_LEVEL;
326 dest[i].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
327 dest[i].ft = attr->dest_ft;
329 } else if (attr->flags & MLX5_ESW_ATTR_FLAG_SLOW_PATH) {
330 flow_act.flags |= FLOW_ACT_IGNORE_FLOW_LEVEL;
331 dest[i].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
332 dest[i].ft = mlx5_esw_chains_get_tc_end_ft(esw);
334 } else if (attr->dest_chain) {
335 flow_act.flags |= FLOW_ACT_IGNORE_FLOW_LEVEL;
336 ft = mlx5_esw_chains_get_table(esw, attr->dest_chain,
340 goto err_create_goto_table;
343 dest[i].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
347 for (j = attr->split_count; j < attr->out_count; j++) {
348 dest[i].type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
349 dest[i].vport.num = attr->dests[j].rep->vport;
350 dest[i].vport.vhca_id =
351 MLX5_CAP_GEN(attr->dests[j].mdev, vhca_id);
352 if (MLX5_CAP_ESW(esw->dev, merged_eswitch))
353 dest[i].vport.flags |=
354 MLX5_FLOW_DEST_VPORT_VHCA_ID;
355 if (attr->dests[j].flags & MLX5_ESW_DEST_ENCAP) {
356 flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT;
357 flow_act.pkt_reformat = attr->dests[j].pkt_reformat;
358 dest[i].vport.flags |= MLX5_FLOW_DEST_VPORT_REFORMAT_ID;
359 dest[i].vport.pkt_reformat =
360 attr->dests[j].pkt_reformat;
367 if (attr->decap_pkt_reformat)
368 flow_act.pkt_reformat = attr->decap_pkt_reformat;
370 if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
371 dest[i].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
372 dest[i].counter_id = mlx5_fc_id(attr->counter);
376 if (attr->outer_match_level != MLX5_MATCH_NONE)
377 spec->match_criteria_enable |= MLX5_MATCH_OUTER_HEADERS;
378 if (attr->inner_match_level != MLX5_MATCH_NONE)
379 spec->match_criteria_enable |= MLX5_MATCH_INNER_HEADERS;
381 if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
382 flow_act.modify_hdr = attr->modify_hdr;
385 fdb = esw_vport_tbl_get(esw, attr);
387 if (attr->chain || attr->prio)
388 fdb = mlx5_esw_chains_get_table(esw, attr->chain,
393 if (!(attr->flags & MLX5_ESW_ATTR_FLAG_NO_IN_PORT))
394 mlx5_eswitch_set_rule_source_port(esw, spec, attr);
397 rule = ERR_CAST(fdb);
401 mlx5_eswitch_set_rule_flow_source(esw, spec, attr);
403 if (mlx5_eswitch_termtbl_required(esw, attr, &flow_act, spec))
404 rule = mlx5_eswitch_add_termtbl_rule(esw, fdb, spec, attr,
407 rule = mlx5_add_flow_rules(fdb, spec, &flow_act, dest, i);
411 atomic64_inc(&esw->offloads.num_flows);
417 esw_vport_tbl_put(esw, attr);
418 else if (attr->chain || attr->prio)
419 mlx5_esw_chains_put_table(esw, attr->chain, attr->prio, 0);
421 if (!(attr->flags & MLX5_ESW_ATTR_FLAG_SLOW_PATH) && attr->dest_chain)
422 mlx5_esw_chains_put_table(esw, attr->dest_chain, 1, 0);
423 err_create_goto_table:
427 struct mlx5_flow_handle *
428 mlx5_eswitch_add_fwd_rule(struct mlx5_eswitch *esw,
429 struct mlx5_flow_spec *spec,
430 struct mlx5_esw_flow_attr *attr)
432 struct mlx5_flow_destination dest[MLX5_MAX_FLOW_FWD_VPORTS + 1] = {};
433 struct mlx5_flow_act flow_act = { .flags = FLOW_ACT_NO_APPEND, };
434 struct mlx5_flow_table *fast_fdb;
435 struct mlx5_flow_table *fwd_fdb;
436 struct mlx5_flow_handle *rule;
439 fast_fdb = mlx5_esw_chains_get_table(esw, attr->chain, attr->prio, 0);
440 if (IS_ERR(fast_fdb)) {
441 rule = ERR_CAST(fast_fdb);
445 fwd_fdb = esw_vport_tbl_get(esw, attr);
446 if (IS_ERR(fwd_fdb)) {
447 rule = ERR_CAST(fwd_fdb);
451 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
452 for (i = 0; i < attr->split_count; i++) {
453 dest[i].type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
454 dest[i].vport.num = attr->dests[i].rep->vport;
455 dest[i].vport.vhca_id =
456 MLX5_CAP_GEN(attr->dests[i].mdev, vhca_id);
457 if (MLX5_CAP_ESW(esw->dev, merged_eswitch))
458 dest[i].vport.flags |= MLX5_FLOW_DEST_VPORT_VHCA_ID;
459 if (attr->dests[i].flags & MLX5_ESW_DEST_ENCAP) {
460 dest[i].vport.flags |= MLX5_FLOW_DEST_VPORT_REFORMAT_ID;
461 dest[i].vport.pkt_reformat = attr->dests[i].pkt_reformat;
464 dest[i].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
465 dest[i].ft = fwd_fdb,
468 mlx5_eswitch_set_rule_source_port(esw, spec, attr);
469 mlx5_eswitch_set_rule_flow_source(esw, spec, attr);
471 if (attr->outer_match_level != MLX5_MATCH_NONE)
472 spec->match_criteria_enable |= MLX5_MATCH_OUTER_HEADERS;
474 flow_act.flags |= FLOW_ACT_IGNORE_FLOW_LEVEL;
475 rule = mlx5_add_flow_rules(fast_fdb, spec, &flow_act, dest, i);
480 atomic64_inc(&esw->offloads.num_flows);
484 esw_vport_tbl_put(esw, attr);
486 mlx5_esw_chains_put_table(esw, attr->chain, attr->prio, 0);
492 __mlx5_eswitch_del_rule(struct mlx5_eswitch *esw,
493 struct mlx5_flow_handle *rule,
494 struct mlx5_esw_flow_attr *attr,
497 bool split = (attr->split_count > 0);
500 mlx5_del_flow_rules(rule);
502 if (!(attr->flags & MLX5_ESW_ATTR_FLAG_SLOW_PATH)) {
503 /* unref the term table */
504 for (i = 0; i < MLX5_MAX_FLOW_FWD_VPORTS; i++) {
505 if (attr->dests[i].termtbl)
506 mlx5_eswitch_termtbl_put(esw, attr->dests[i].termtbl);
510 atomic64_dec(&esw->offloads.num_flows);
513 esw_vport_tbl_put(esw, attr);
514 mlx5_esw_chains_put_table(esw, attr->chain, attr->prio, 0);
517 esw_vport_tbl_put(esw, attr);
518 else if (attr->chain || attr->prio)
519 mlx5_esw_chains_put_table(esw, attr->chain, attr->prio,
521 if (attr->dest_chain)
522 mlx5_esw_chains_put_table(esw, attr->dest_chain, 1, 0);
527 mlx5_eswitch_del_offloaded_rule(struct mlx5_eswitch *esw,
528 struct mlx5_flow_handle *rule,
529 struct mlx5_esw_flow_attr *attr)
531 __mlx5_eswitch_del_rule(esw, rule, attr, false);
535 mlx5_eswitch_del_fwd_rule(struct mlx5_eswitch *esw,
536 struct mlx5_flow_handle *rule,
537 struct mlx5_esw_flow_attr *attr)
539 __mlx5_eswitch_del_rule(esw, rule, attr, true);
542 static int esw_set_global_vlan_pop(struct mlx5_eswitch *esw, u8 val)
544 struct mlx5_eswitch_rep *rep;
547 esw_debug(esw->dev, "%s applying global %s policy\n", __func__, val ? "pop" : "none");
548 mlx5_esw_for_each_host_func_rep(esw, i, rep, esw->esw_funcs.num_vfs) {
549 if (atomic_read(&rep->rep_data[REP_ETH].state) != REP_LOADED)
552 err = __mlx5_eswitch_set_vport_vlan(esw, rep->vport, 0, 0, val);
561 static struct mlx5_eswitch_rep *
562 esw_vlan_action_get_vport(struct mlx5_esw_flow_attr *attr, bool push, bool pop)
564 struct mlx5_eswitch_rep *in_rep, *out_rep, *vport = NULL;
566 in_rep = attr->in_rep;
567 out_rep = attr->dests[0].rep;
579 static int esw_add_vlan_action_check(struct mlx5_esw_flow_attr *attr,
580 bool push, bool pop, bool fwd)
582 struct mlx5_eswitch_rep *in_rep, *out_rep;
584 if ((push || pop) && !fwd)
587 in_rep = attr->in_rep;
588 out_rep = attr->dests[0].rep;
590 if (push && in_rep->vport == MLX5_VPORT_UPLINK)
593 if (pop && out_rep->vport == MLX5_VPORT_UPLINK)
596 /* vport has vlan push configured, can't offload VF --> wire rules w.o it */
597 if (!push && !pop && fwd)
598 if (in_rep->vlan && out_rep->vport == MLX5_VPORT_UPLINK)
601 /* protects against (1) setting rules with different vlans to push and
602 * (2) setting rules w.o vlans (attr->vlan = 0) && w. vlans to push (!= 0)
604 if (push && in_rep->vlan_refcount && (in_rep->vlan != attr->vlan_vid[0]))
613 int mlx5_eswitch_add_vlan_action(struct mlx5_eswitch *esw,
614 struct mlx5_esw_flow_attr *attr)
616 struct offloads_fdb *offloads = &esw->fdb_table.offloads;
617 struct mlx5_eswitch_rep *vport = NULL;
621 /* nop if we're on the vlan push/pop non emulation mode */
622 if (mlx5_eswitch_vlan_actions_supported(esw->dev, 1))
625 push = !!(attr->action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH);
626 pop = !!(attr->action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP);
627 fwd = !!((attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
630 mutex_lock(&esw->state_lock);
632 err = esw_add_vlan_action_check(attr, push, pop, fwd);
636 attr->flags &= ~MLX5_ESW_ATTR_FLAG_VLAN_HANDLED;
638 vport = esw_vlan_action_get_vport(attr, push, pop);
640 if (!push && !pop && fwd) {
641 /* tracks VF --> wire rules without vlan push action */
642 if (attr->dests[0].rep->vport == MLX5_VPORT_UPLINK) {
643 vport->vlan_refcount++;
644 attr->flags |= MLX5_ESW_ATTR_FLAG_VLAN_HANDLED;
653 if (!(offloads->vlan_push_pop_refcount)) {
654 /* it's the 1st vlan rule, apply global vlan pop policy */
655 err = esw_set_global_vlan_pop(esw, SET_VLAN_STRIP);
659 offloads->vlan_push_pop_refcount++;
662 if (vport->vlan_refcount)
665 err = __mlx5_eswitch_set_vport_vlan(esw, vport->vport, attr->vlan_vid[0], 0,
666 SET_VLAN_INSERT | SET_VLAN_STRIP);
669 vport->vlan = attr->vlan_vid[0];
671 vport->vlan_refcount++;
675 attr->flags |= MLX5_ESW_ATTR_FLAG_VLAN_HANDLED;
677 mutex_unlock(&esw->state_lock);
681 int mlx5_eswitch_del_vlan_action(struct mlx5_eswitch *esw,
682 struct mlx5_esw_flow_attr *attr)
684 struct offloads_fdb *offloads = &esw->fdb_table.offloads;
685 struct mlx5_eswitch_rep *vport = NULL;
689 /* nop if we're on the vlan push/pop non emulation mode */
690 if (mlx5_eswitch_vlan_actions_supported(esw->dev, 1))
693 if (!(attr->flags & MLX5_ESW_ATTR_FLAG_VLAN_HANDLED))
696 push = !!(attr->action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH);
697 pop = !!(attr->action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP);
698 fwd = !!(attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST);
700 mutex_lock(&esw->state_lock);
702 vport = esw_vlan_action_get_vport(attr, push, pop);
704 if (!push && !pop && fwd) {
705 /* tracks VF --> wire rules without vlan push action */
706 if (attr->dests[0].rep->vport == MLX5_VPORT_UPLINK)
707 vport->vlan_refcount--;
713 vport->vlan_refcount--;
714 if (vport->vlan_refcount)
715 goto skip_unset_push;
718 err = __mlx5_eswitch_set_vport_vlan(esw, vport->vport,
719 0, 0, SET_VLAN_STRIP);
725 offloads->vlan_push_pop_refcount--;
726 if (offloads->vlan_push_pop_refcount)
729 /* no more vlan rules, stop global vlan pop policy */
730 err = esw_set_global_vlan_pop(esw, 0);
733 mutex_unlock(&esw->state_lock);
737 struct mlx5_flow_handle *
738 mlx5_eswitch_add_send_to_vport_rule(struct mlx5_eswitch *esw, u16 vport,
741 struct mlx5_flow_act flow_act = {0};
742 struct mlx5_flow_destination dest = {};
743 struct mlx5_flow_handle *flow_rule;
744 struct mlx5_flow_spec *spec;
747 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
749 flow_rule = ERR_PTR(-ENOMEM);
753 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value, misc_parameters);
754 MLX5_SET(fte_match_set_misc, misc, source_sqn, sqn);
755 /* source vport is the esw manager */
756 MLX5_SET(fte_match_set_misc, misc, source_port, esw->manager_vport);
758 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, misc_parameters);
759 MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_sqn);
760 MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
762 spec->match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS;
763 dest.type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
764 dest.vport.num = vport;
765 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
767 flow_rule = mlx5_add_flow_rules(esw->fdb_table.offloads.slow_fdb,
768 spec, &flow_act, &dest, 1);
769 if (IS_ERR(flow_rule))
770 esw_warn(esw->dev, "FDB: Failed to add send to vport rule err %ld\n", PTR_ERR(flow_rule));
775 EXPORT_SYMBOL(mlx5_eswitch_add_send_to_vport_rule);
777 void mlx5_eswitch_del_send_to_vport_rule(struct mlx5_flow_handle *rule)
779 mlx5_del_flow_rules(rule);
782 static bool mlx5_eswitch_reg_c1_loopback_supported(struct mlx5_eswitch *esw)
784 return MLX5_CAP_ESW_FLOWTABLE(esw->dev, fdb_to_vport_reg_c_id) &
785 MLX5_FDB_TO_VPORT_REG_C_1;
788 static int esw_set_passing_vport_metadata(struct mlx5_eswitch *esw, bool enable)
790 u32 out[MLX5_ST_SZ_DW(query_esw_vport_context_out)] = {};
791 u32 min[MLX5_ST_SZ_DW(modify_esw_vport_context_in)] = {};
792 u32 in[MLX5_ST_SZ_DW(query_esw_vport_context_in)] = {};
796 if (!mlx5_eswitch_reg_c1_loopback_supported(esw) &&
797 !mlx5_eswitch_vport_match_metadata_enabled(esw))
800 MLX5_SET(query_esw_vport_context_in, in, opcode,
801 MLX5_CMD_OP_QUERY_ESW_VPORT_CONTEXT);
802 err = mlx5_cmd_exec_inout(esw->dev, query_esw_vport_context, in, out);
806 curr = MLX5_GET(query_esw_vport_context_out, out,
807 esw_vport_context.fdb_to_vport_reg_c_id);
808 wanted = MLX5_FDB_TO_VPORT_REG_C_0;
809 if (mlx5_eswitch_reg_c1_loopback_supported(esw))
810 wanted |= MLX5_FDB_TO_VPORT_REG_C_1;
817 MLX5_SET(modify_esw_vport_context_in, min,
818 esw_vport_context.fdb_to_vport_reg_c_id, curr);
819 MLX5_SET(modify_esw_vport_context_in, min,
820 field_select.fdb_to_vport_reg_c_id, 1);
822 err = mlx5_eswitch_modify_esw_vport_context(esw->dev, 0, false, min);
824 if (enable && (curr & MLX5_FDB_TO_VPORT_REG_C_1))
825 esw->flags |= MLX5_ESWITCH_REG_C1_LOOPBACK_ENABLED;
827 esw->flags &= ~MLX5_ESWITCH_REG_C1_LOOPBACK_ENABLED;
833 static void peer_miss_rules_setup(struct mlx5_eswitch *esw,
834 struct mlx5_core_dev *peer_dev,
835 struct mlx5_flow_spec *spec,
836 struct mlx5_flow_destination *dest)
840 if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
841 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
843 MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
844 mlx5_eswitch_get_vport_metadata_mask());
846 spec->match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS_2;
848 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
851 MLX5_SET(fte_match_set_misc, misc, source_eswitch_owner_vhca_id,
852 MLX5_CAP_GEN(peer_dev, vhca_id));
854 spec->match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS;
856 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
858 MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
859 MLX5_SET_TO_ONES(fte_match_set_misc, misc,
860 source_eswitch_owner_vhca_id);
863 dest->type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
864 dest->vport.num = peer_dev->priv.eswitch->manager_vport;
865 dest->vport.vhca_id = MLX5_CAP_GEN(peer_dev, vhca_id);
866 dest->vport.flags |= MLX5_FLOW_DEST_VPORT_VHCA_ID;
869 static void esw_set_peer_miss_rule_source_port(struct mlx5_eswitch *esw,
870 struct mlx5_eswitch *peer_esw,
871 struct mlx5_flow_spec *spec,
876 if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
877 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
879 MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
880 mlx5_eswitch_get_vport_metadata_for_match(peer_esw,
883 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
885 MLX5_SET(fte_match_set_misc, misc, source_port, vport);
889 static int esw_add_fdb_peer_miss_rules(struct mlx5_eswitch *esw,
890 struct mlx5_core_dev *peer_dev)
892 struct mlx5_flow_destination dest = {};
893 struct mlx5_flow_act flow_act = {0};
894 struct mlx5_flow_handle **flows;
895 struct mlx5_flow_handle *flow;
896 struct mlx5_flow_spec *spec;
897 /* total vports is the same for both e-switches */
898 int nvports = esw->total_vports;
902 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
906 peer_miss_rules_setup(esw, peer_dev, spec, &dest);
908 flows = kvzalloc(nvports * sizeof(*flows), GFP_KERNEL);
911 goto alloc_flows_err;
914 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
915 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
918 if (mlx5_core_is_ecpf_esw_manager(esw->dev)) {
919 esw_set_peer_miss_rule_source_port(esw, peer_dev->priv.eswitch,
920 spec, MLX5_VPORT_PF);
922 flow = mlx5_add_flow_rules(esw->fdb_table.offloads.slow_fdb,
923 spec, &flow_act, &dest, 1);
926 goto add_pf_flow_err;
928 flows[MLX5_VPORT_PF] = flow;
931 if (mlx5_ecpf_vport_exists(esw->dev)) {
932 MLX5_SET(fte_match_set_misc, misc, source_port, MLX5_VPORT_ECPF);
933 flow = mlx5_add_flow_rules(esw->fdb_table.offloads.slow_fdb,
934 spec, &flow_act, &dest, 1);
937 goto add_ecpf_flow_err;
939 flows[mlx5_eswitch_ecpf_idx(esw)] = flow;
942 mlx5_esw_for_each_vf_vport_num(esw, i, mlx5_core_max_vfs(esw->dev)) {
943 esw_set_peer_miss_rule_source_port(esw,
944 peer_dev->priv.eswitch,
947 flow = mlx5_add_flow_rules(esw->fdb_table.offloads.slow_fdb,
948 spec, &flow_act, &dest, 1);
951 goto add_vf_flow_err;
956 esw->fdb_table.offloads.peer_miss_rules = flows;
963 mlx5_esw_for_each_vf_vport_num_reverse(esw, i, nvports)
964 mlx5_del_flow_rules(flows[i]);
966 if (mlx5_ecpf_vport_exists(esw->dev))
967 mlx5_del_flow_rules(flows[mlx5_eswitch_ecpf_idx(esw)]);
969 if (mlx5_core_is_ecpf_esw_manager(esw->dev))
970 mlx5_del_flow_rules(flows[MLX5_VPORT_PF]);
972 esw_warn(esw->dev, "FDB: Failed to add peer miss flow rule err %d\n", err);
979 static void esw_del_fdb_peer_miss_rules(struct mlx5_eswitch *esw)
981 struct mlx5_flow_handle **flows;
984 flows = esw->fdb_table.offloads.peer_miss_rules;
986 mlx5_esw_for_each_vf_vport_num_reverse(esw, i,
987 mlx5_core_max_vfs(esw->dev))
988 mlx5_del_flow_rules(flows[i]);
990 if (mlx5_ecpf_vport_exists(esw->dev))
991 mlx5_del_flow_rules(flows[mlx5_eswitch_ecpf_idx(esw)]);
993 if (mlx5_core_is_ecpf_esw_manager(esw->dev))
994 mlx5_del_flow_rules(flows[MLX5_VPORT_PF]);
999 static int esw_add_fdb_miss_rule(struct mlx5_eswitch *esw)
1001 struct mlx5_flow_act flow_act = {0};
1002 struct mlx5_flow_destination dest = {};
1003 struct mlx5_flow_handle *flow_rule = NULL;
1004 struct mlx5_flow_spec *spec;
1011 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
1017 spec->match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
1018 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1020 dmac_c = MLX5_ADDR_OF(fte_match_param, headers_c,
1021 outer_headers.dmac_47_16);
1024 dest.type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
1025 dest.vport.num = esw->manager_vport;
1026 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1028 flow_rule = mlx5_add_flow_rules(esw->fdb_table.offloads.slow_fdb,
1029 spec, &flow_act, &dest, 1);
1030 if (IS_ERR(flow_rule)) {
1031 err = PTR_ERR(flow_rule);
1032 esw_warn(esw->dev, "FDB: Failed to add unicast miss flow rule err %d\n", err);
1036 esw->fdb_table.offloads.miss_rule_uni = flow_rule;
1038 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1040 dmac_v = MLX5_ADDR_OF(fte_match_param, headers_v,
1041 outer_headers.dmac_47_16);
1043 flow_rule = mlx5_add_flow_rules(esw->fdb_table.offloads.slow_fdb,
1044 spec, &flow_act, &dest, 1);
1045 if (IS_ERR(flow_rule)) {
1046 err = PTR_ERR(flow_rule);
1047 esw_warn(esw->dev, "FDB: Failed to add multicast miss flow rule err %d\n", err);
1048 mlx5_del_flow_rules(esw->fdb_table.offloads.miss_rule_uni);
1052 esw->fdb_table.offloads.miss_rule_multi = flow_rule;
1059 struct mlx5_flow_handle *
1060 esw_add_restore_rule(struct mlx5_eswitch *esw, u32 tag)
1062 struct mlx5_flow_act flow_act = { .flags = FLOW_ACT_NO_APPEND, };
1063 struct mlx5_flow_table *ft = esw->offloads.ft_offloads_restore;
1064 struct mlx5_flow_context *flow_context;
1065 struct mlx5_flow_handle *flow_rule;
1066 struct mlx5_flow_destination dest;
1067 struct mlx5_flow_spec *spec;
1070 if (!mlx5_eswitch_reg_c1_loopback_supported(esw))
1071 return ERR_PTR(-EOPNOTSUPP);
1073 spec = kzalloc(sizeof(*spec), GFP_KERNEL);
1075 return ERR_PTR(-ENOMEM);
1077 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
1079 MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
1080 ESW_CHAIN_TAG_METADATA_MASK);
1081 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
1083 MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0, tag);
1084 spec->match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS_2;
1085 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
1086 MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
1087 flow_act.modify_hdr = esw->offloads.restore_copy_hdr_id;
1089 flow_context = &spec->flow_context;
1090 flow_context->flags |= FLOW_CONTEXT_HAS_TAG;
1091 flow_context->flow_tag = tag;
1092 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1093 dest.ft = esw->offloads.ft_offloads;
1095 flow_rule = mlx5_add_flow_rules(ft, spec, &flow_act, &dest, 1);
1098 if (IS_ERR(flow_rule))
1100 "Failed to create restore rule for tag: %d, err(%d)\n",
1101 tag, (int)PTR_ERR(flow_rule));
1107 esw_get_max_restore_tag(struct mlx5_eswitch *esw)
1109 return ESW_CHAIN_TAG_METADATA_MASK;
1112 #define MAX_PF_SQ 256
1113 #define MAX_SQ_NVPORTS 32
1115 static void esw_set_flow_group_source_port(struct mlx5_eswitch *esw,
1118 void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1122 if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
1123 MLX5_SET(create_flow_group_in, flow_group_in,
1124 match_criteria_enable,
1125 MLX5_MATCH_MISC_PARAMETERS_2);
1127 MLX5_SET(fte_match_param, match_criteria,
1128 misc_parameters_2.metadata_reg_c_0,
1129 mlx5_eswitch_get_vport_metadata_mask());
1131 MLX5_SET(create_flow_group_in, flow_group_in,
1132 match_criteria_enable,
1133 MLX5_MATCH_MISC_PARAMETERS);
1135 MLX5_SET_TO_ONES(fte_match_param, match_criteria,
1136 misc_parameters.source_port);
1140 static int esw_create_offloads_fdb_tables(struct mlx5_eswitch *esw)
1142 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1143 struct mlx5_flow_table_attr ft_attr = {};
1144 struct mlx5_core_dev *dev = esw->dev;
1145 struct mlx5_flow_namespace *root_ns;
1146 struct mlx5_flow_table *fdb = NULL;
1147 u32 flags = 0, *flow_group_in;
1148 int table_size, ix, err = 0;
1149 struct mlx5_flow_group *g;
1150 void *match_criteria;
1153 esw_debug(esw->dev, "Create offloads FDB Tables\n");
1155 flow_group_in = kvzalloc(inlen, GFP_KERNEL);
1159 root_ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_FDB);
1161 esw_warn(dev, "Failed to get FDB flow namespace\n");
1165 esw->fdb_table.offloads.ns = root_ns;
1166 err = mlx5_flow_namespace_set_mode(root_ns,
1167 esw->dev->priv.steering->mode);
1169 esw_warn(dev, "Failed to set FDB namespace steering mode\n");
1173 table_size = esw->total_vports * MAX_SQ_NVPORTS + MAX_PF_SQ +
1174 MLX5_ESW_MISS_FLOWS + esw->total_vports;
1176 /* create the slow path fdb with encap set, so further table instances
1177 * can be created at run time while VFs are probed if the FW allows that.
1179 if (esw->offloads.encap != DEVLINK_ESWITCH_ENCAP_MODE_NONE)
1180 flags |= (MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT |
1181 MLX5_FLOW_TABLE_TUNNEL_EN_DECAP);
1183 ft_attr.flags = flags;
1184 ft_attr.max_fte = table_size;
1185 ft_attr.prio = FDB_SLOW_PATH;
1187 fdb = mlx5_create_flow_table(root_ns, &ft_attr);
1190 esw_warn(dev, "Failed to create slow path FDB Table err %d\n", err);
1193 esw->fdb_table.offloads.slow_fdb = fdb;
1195 err = mlx5_esw_chains_create(esw);
1197 esw_warn(dev, "Failed to create fdb chains err(%d)\n", err);
1198 goto fdb_chains_err;
1201 /* create send-to-vport group */
1202 MLX5_SET(create_flow_group_in, flow_group_in, match_criteria_enable,
1203 MLX5_MATCH_MISC_PARAMETERS);
1205 match_criteria = MLX5_ADDR_OF(create_flow_group_in, flow_group_in, match_criteria);
1207 MLX5_SET_TO_ONES(fte_match_param, match_criteria, misc_parameters.source_sqn);
1208 MLX5_SET_TO_ONES(fte_match_param, match_criteria, misc_parameters.source_port);
1210 ix = esw->total_vports * MAX_SQ_NVPORTS + MAX_PF_SQ;
1211 MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, 0);
1212 MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index, ix - 1);
1214 g = mlx5_create_flow_group(fdb, flow_group_in);
1217 esw_warn(dev, "Failed to create send-to-vport flow group err(%d)\n", err);
1218 goto send_vport_err;
1220 esw->fdb_table.offloads.send_to_vport_grp = g;
1222 /* create peer esw miss group */
1223 memset(flow_group_in, 0, inlen);
1225 esw_set_flow_group_source_port(esw, flow_group_in);
1227 if (!mlx5_eswitch_vport_match_metadata_enabled(esw)) {
1228 match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1232 MLX5_SET_TO_ONES(fte_match_param, match_criteria,
1233 misc_parameters.source_eswitch_owner_vhca_id);
1235 MLX5_SET(create_flow_group_in, flow_group_in,
1236 source_eswitch_owner_vhca_id_valid, 1);
1239 MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, ix);
1240 MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index,
1241 ix + esw->total_vports - 1);
1242 ix += esw->total_vports;
1244 g = mlx5_create_flow_group(fdb, flow_group_in);
1247 esw_warn(dev, "Failed to create peer miss flow group err(%d)\n", err);
1250 esw->fdb_table.offloads.peer_miss_grp = g;
1252 /* create miss group */
1253 memset(flow_group_in, 0, inlen);
1254 MLX5_SET(create_flow_group_in, flow_group_in, match_criteria_enable,
1255 MLX5_MATCH_OUTER_HEADERS);
1256 match_criteria = MLX5_ADDR_OF(create_flow_group_in, flow_group_in,
1258 dmac = MLX5_ADDR_OF(fte_match_param, match_criteria,
1259 outer_headers.dmac_47_16);
1262 MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, ix);
1263 MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index,
1264 ix + MLX5_ESW_MISS_FLOWS);
1266 g = mlx5_create_flow_group(fdb, flow_group_in);
1269 esw_warn(dev, "Failed to create miss flow group err(%d)\n", err);
1272 esw->fdb_table.offloads.miss_grp = g;
1274 err = esw_add_fdb_miss_rule(esw);
1278 kvfree(flow_group_in);
1282 mlx5_destroy_flow_group(esw->fdb_table.offloads.miss_grp);
1284 mlx5_destroy_flow_group(esw->fdb_table.offloads.peer_miss_grp);
1286 mlx5_destroy_flow_group(esw->fdb_table.offloads.send_to_vport_grp);
1288 mlx5_esw_chains_destroy(esw);
1290 mlx5_destroy_flow_table(esw->fdb_table.offloads.slow_fdb);
1292 /* Holds true only as long as DMFS is the default */
1293 mlx5_flow_namespace_set_mode(root_ns, MLX5_FLOW_STEERING_MODE_DMFS);
1295 kvfree(flow_group_in);
1299 static void esw_destroy_offloads_fdb_tables(struct mlx5_eswitch *esw)
1301 if (!esw->fdb_table.offloads.slow_fdb)
1304 esw_debug(esw->dev, "Destroy offloads FDB Tables\n");
1305 mlx5_del_flow_rules(esw->fdb_table.offloads.miss_rule_multi);
1306 mlx5_del_flow_rules(esw->fdb_table.offloads.miss_rule_uni);
1307 mlx5_destroy_flow_group(esw->fdb_table.offloads.send_to_vport_grp);
1308 mlx5_destroy_flow_group(esw->fdb_table.offloads.peer_miss_grp);
1309 mlx5_destroy_flow_group(esw->fdb_table.offloads.miss_grp);
1311 mlx5_esw_chains_destroy(esw);
1312 mlx5_destroy_flow_table(esw->fdb_table.offloads.slow_fdb);
1313 /* Holds true only as long as DMFS is the default */
1314 mlx5_flow_namespace_set_mode(esw->fdb_table.offloads.ns,
1315 MLX5_FLOW_STEERING_MODE_DMFS);
1318 static int esw_create_offloads_table(struct mlx5_eswitch *esw)
1320 struct mlx5_flow_table_attr ft_attr = {};
1321 struct mlx5_core_dev *dev = esw->dev;
1322 struct mlx5_flow_table *ft_offloads;
1323 struct mlx5_flow_namespace *ns;
1326 ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_OFFLOADS);
1328 esw_warn(esw->dev, "Failed to get offloads flow namespace\n");
1332 ft_attr.max_fte = esw->total_vports + MLX5_ESW_MISS_FLOWS;
1335 ft_offloads = mlx5_create_flow_table(ns, &ft_attr);
1336 if (IS_ERR(ft_offloads)) {
1337 err = PTR_ERR(ft_offloads);
1338 esw_warn(esw->dev, "Failed to create offloads table, err %d\n", err);
1342 esw->offloads.ft_offloads = ft_offloads;
1346 static void esw_destroy_offloads_table(struct mlx5_eswitch *esw)
1348 struct mlx5_esw_offload *offloads = &esw->offloads;
1350 mlx5_destroy_flow_table(offloads->ft_offloads);
1353 static int esw_create_vport_rx_group(struct mlx5_eswitch *esw)
1355 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1356 struct mlx5_flow_group *g;
1361 nvports = esw->total_vports + MLX5_ESW_MISS_FLOWS;
1362 flow_group_in = kvzalloc(inlen, GFP_KERNEL);
1366 /* create vport rx group */
1367 esw_set_flow_group_source_port(esw, flow_group_in);
1369 MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, 0);
1370 MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index, nvports - 1);
1372 g = mlx5_create_flow_group(esw->offloads.ft_offloads, flow_group_in);
1376 mlx5_core_warn(esw->dev, "Failed to create vport rx group err %d\n", err);
1380 esw->offloads.vport_rx_group = g;
1382 kvfree(flow_group_in);
1386 static void esw_destroy_vport_rx_group(struct mlx5_eswitch *esw)
1388 mlx5_destroy_flow_group(esw->offloads.vport_rx_group);
1391 struct mlx5_flow_handle *
1392 mlx5_eswitch_create_vport_rx_rule(struct mlx5_eswitch *esw, u16 vport,
1393 struct mlx5_flow_destination *dest)
1395 struct mlx5_flow_act flow_act = {0};
1396 struct mlx5_flow_handle *flow_rule;
1397 struct mlx5_flow_spec *spec;
1400 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
1402 flow_rule = ERR_PTR(-ENOMEM);
1406 if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
1407 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value, misc_parameters_2);
1408 MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
1409 mlx5_eswitch_get_vport_metadata_for_match(esw, vport));
1411 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, misc_parameters_2);
1412 MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
1413 mlx5_eswitch_get_vport_metadata_mask());
1415 spec->match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS_2;
1417 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value, misc_parameters);
1418 MLX5_SET(fte_match_set_misc, misc, source_port, vport);
1420 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, misc_parameters);
1421 MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
1423 spec->match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS;
1426 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1427 flow_rule = mlx5_add_flow_rules(esw->offloads.ft_offloads, spec,
1428 &flow_act, dest, 1);
1429 if (IS_ERR(flow_rule)) {
1430 esw_warn(esw->dev, "fs offloads: Failed to add vport rx rule err %ld\n", PTR_ERR(flow_rule));
1440 static int mlx5_eswitch_inline_mode_get(const struct mlx5_eswitch *esw, u8 *mode)
1442 u8 prev_mlx5_mode, mlx5_mode = MLX5_INLINE_MODE_L2;
1443 struct mlx5_core_dev *dev = esw->dev;
1446 if (!MLX5_CAP_GEN(dev, vport_group_manager))
1449 if (esw->mode == MLX5_ESWITCH_NONE)
1452 switch (MLX5_CAP_ETH(dev, wqe_inline_mode)) {
1453 case MLX5_CAP_INLINE_MODE_NOT_REQUIRED:
1454 mlx5_mode = MLX5_INLINE_MODE_NONE;
1456 case MLX5_CAP_INLINE_MODE_L2:
1457 mlx5_mode = MLX5_INLINE_MODE_L2;
1459 case MLX5_CAP_INLINE_MODE_VPORT_CONTEXT:
1464 mlx5_query_nic_vport_min_inline(dev, esw->first_host_vport, &prev_mlx5_mode);
1465 mlx5_esw_for_each_host_func_vport(esw, vport, esw->esw_funcs.num_vfs) {
1466 mlx5_query_nic_vport_min_inline(dev, vport, &mlx5_mode);
1467 if (prev_mlx5_mode != mlx5_mode)
1469 prev_mlx5_mode = mlx5_mode;
1477 static void esw_destroy_restore_table(struct mlx5_eswitch *esw)
1479 struct mlx5_esw_offload *offloads = &esw->offloads;
1481 if (!mlx5_eswitch_reg_c1_loopback_supported(esw))
1484 mlx5_modify_header_dealloc(esw->dev, offloads->restore_copy_hdr_id);
1485 mlx5_destroy_flow_group(offloads->restore_group);
1486 mlx5_destroy_flow_table(offloads->ft_offloads_restore);
1489 static int esw_create_restore_table(struct mlx5_eswitch *esw)
1491 u8 modact[MLX5_UN_SZ_BYTES(set_add_copy_action_in_auto)] = {};
1492 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1493 struct mlx5_flow_table_attr ft_attr = {};
1494 struct mlx5_core_dev *dev = esw->dev;
1495 struct mlx5_flow_namespace *ns;
1496 struct mlx5_modify_hdr *mod_hdr;
1497 void *match_criteria, *misc;
1498 struct mlx5_flow_table *ft;
1499 struct mlx5_flow_group *g;
1503 if (!mlx5_eswitch_reg_c1_loopback_supported(esw))
1506 ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_OFFLOADS);
1508 esw_warn(esw->dev, "Failed to get offloads flow namespace\n");
1512 flow_group_in = kvzalloc(inlen, GFP_KERNEL);
1513 if (!flow_group_in) {
1518 ft_attr.max_fte = 1 << ESW_CHAIN_TAG_METADATA_BITS;
1519 ft = mlx5_create_flow_table(ns, &ft_attr);
1522 esw_warn(esw->dev, "Failed to create restore table, err %d\n",
1527 memset(flow_group_in, 0, inlen);
1528 match_criteria = MLX5_ADDR_OF(create_flow_group_in, flow_group_in,
1530 misc = MLX5_ADDR_OF(fte_match_param, match_criteria,
1533 MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
1534 ESW_CHAIN_TAG_METADATA_MASK);
1535 MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, 0);
1536 MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index,
1537 ft_attr.max_fte - 1);
1538 MLX5_SET(create_flow_group_in, flow_group_in, match_criteria_enable,
1539 MLX5_MATCH_MISC_PARAMETERS_2);
1540 g = mlx5_create_flow_group(ft, flow_group_in);
1543 esw_warn(dev, "Failed to create restore flow group, err: %d\n",
1548 MLX5_SET(copy_action_in, modact, action_type, MLX5_ACTION_TYPE_COPY);
1549 MLX5_SET(copy_action_in, modact, src_field,
1550 MLX5_ACTION_IN_FIELD_METADATA_REG_C_1);
1551 MLX5_SET(copy_action_in, modact, dst_field,
1552 MLX5_ACTION_IN_FIELD_METADATA_REG_B);
1553 mod_hdr = mlx5_modify_header_alloc(esw->dev,
1554 MLX5_FLOW_NAMESPACE_KERNEL, 1,
1556 if (IS_ERR(mod_hdr)) {
1557 err = PTR_ERR(mod_hdr);
1558 esw_warn(dev, "Failed to create restore mod header, err: %d\n",
1563 esw->offloads.ft_offloads_restore = ft;
1564 esw->offloads.restore_group = g;
1565 esw->offloads.restore_copy_hdr_id = mod_hdr;
1567 kvfree(flow_group_in);
1572 mlx5_destroy_flow_group(g);
1574 mlx5_destroy_flow_table(ft);
1576 kvfree(flow_group_in);
1581 static int esw_offloads_start(struct mlx5_eswitch *esw,
1582 struct netlink_ext_ack *extack)
1586 mlx5_eswitch_disable_locked(esw, false);
1587 err = mlx5_eswitch_enable_locked(esw, MLX5_ESWITCH_OFFLOADS,
1588 esw->dev->priv.sriov.num_vfs);
1590 NL_SET_ERR_MSG_MOD(extack,
1591 "Failed setting eswitch to offloads");
1592 err1 = mlx5_eswitch_enable_locked(esw, MLX5_ESWITCH_LEGACY,
1593 MLX5_ESWITCH_IGNORE_NUM_VFS);
1595 NL_SET_ERR_MSG_MOD(extack,
1596 "Failed setting eswitch back to legacy");
1599 if (esw->offloads.inline_mode == MLX5_INLINE_MODE_NONE) {
1600 if (mlx5_eswitch_inline_mode_get(esw,
1601 &esw->offloads.inline_mode)) {
1602 esw->offloads.inline_mode = MLX5_INLINE_MODE_L2;
1603 NL_SET_ERR_MSG_MOD(extack,
1604 "Inline mode is different between vports");
1610 void esw_offloads_cleanup_reps(struct mlx5_eswitch *esw)
1612 kfree(esw->offloads.vport_reps);
1615 int esw_offloads_init_reps(struct mlx5_eswitch *esw)
1617 int total_vports = esw->total_vports;
1618 struct mlx5_eswitch_rep *rep;
1622 esw->offloads.vport_reps = kcalloc(total_vports,
1623 sizeof(struct mlx5_eswitch_rep),
1625 if (!esw->offloads.vport_reps)
1628 mlx5_esw_for_all_reps(esw, vport_index, rep) {
1629 rep->vport = mlx5_eswitch_index_to_vport_num(esw, vport_index);
1630 rep->vport_index = vport_index;
1632 for (rep_type = 0; rep_type < NUM_REP_TYPES; rep_type++)
1633 atomic_set(&rep->rep_data[rep_type].state,
1640 static void __esw_offloads_unload_rep(struct mlx5_eswitch *esw,
1641 struct mlx5_eswitch_rep *rep, u8 rep_type)
1643 if (atomic_cmpxchg(&rep->rep_data[rep_type].state,
1644 REP_LOADED, REP_REGISTERED) == REP_LOADED)
1645 esw->offloads.rep_ops[rep_type]->unload(rep);
1648 static void __unload_reps_all_vport(struct mlx5_eswitch *esw, u8 rep_type)
1650 struct mlx5_eswitch_rep *rep;
1653 mlx5_esw_for_each_vf_rep_reverse(esw, i, rep, esw->esw_funcs.num_vfs)
1654 __esw_offloads_unload_rep(esw, rep, rep_type);
1656 if (mlx5_ecpf_vport_exists(esw->dev)) {
1657 rep = mlx5_eswitch_get_rep(esw, MLX5_VPORT_ECPF);
1658 __esw_offloads_unload_rep(esw, rep, rep_type);
1661 if (mlx5_core_is_ecpf_esw_manager(esw->dev)) {
1662 rep = mlx5_eswitch_get_rep(esw, MLX5_VPORT_PF);
1663 __esw_offloads_unload_rep(esw, rep, rep_type);
1666 rep = mlx5_eswitch_get_rep(esw, MLX5_VPORT_UPLINK);
1667 __esw_offloads_unload_rep(esw, rep, rep_type);
1670 int esw_offloads_load_rep(struct mlx5_eswitch *esw, u16 vport_num)
1672 struct mlx5_eswitch_rep *rep;
1676 if (esw->mode != MLX5_ESWITCH_OFFLOADS)
1679 rep = mlx5_eswitch_get_rep(esw, vport_num);
1680 for (rep_type = 0; rep_type < NUM_REP_TYPES; rep_type++)
1681 if (atomic_cmpxchg(&rep->rep_data[rep_type].state,
1682 REP_REGISTERED, REP_LOADED) == REP_REGISTERED) {
1683 err = esw->offloads.rep_ops[rep_type]->load(esw->dev, rep);
1691 atomic_set(&rep->rep_data[rep_type].state, REP_REGISTERED);
1692 for (--rep_type; rep_type >= 0; rep_type--)
1693 __esw_offloads_unload_rep(esw, rep, rep_type);
1697 void esw_offloads_unload_rep(struct mlx5_eswitch *esw, u16 vport_num)
1699 struct mlx5_eswitch_rep *rep;
1702 if (esw->mode != MLX5_ESWITCH_OFFLOADS)
1705 rep = mlx5_eswitch_get_rep(esw, vport_num);
1706 for (rep_type = NUM_REP_TYPES - 1; rep_type >= 0; rep_type--)
1707 __esw_offloads_unload_rep(esw, rep, rep_type);
1710 #define ESW_OFFLOADS_DEVCOM_PAIR (0)
1711 #define ESW_OFFLOADS_DEVCOM_UNPAIR (1)
1713 static int mlx5_esw_offloads_pair(struct mlx5_eswitch *esw,
1714 struct mlx5_eswitch *peer_esw)
1718 err = esw_add_fdb_peer_miss_rules(esw, peer_esw->dev);
1725 static void mlx5_esw_offloads_unpair(struct mlx5_eswitch *esw)
1727 #if IS_ENABLED(CONFIG_MLX5_CLS_ACT)
1728 mlx5e_tc_clean_fdb_peer_flows(esw);
1730 esw_del_fdb_peer_miss_rules(esw);
1733 static int mlx5_esw_offloads_set_ns_peer(struct mlx5_eswitch *esw,
1734 struct mlx5_eswitch *peer_esw,
1737 struct mlx5_flow_root_namespace *peer_ns;
1738 struct mlx5_flow_root_namespace *ns;
1741 peer_ns = peer_esw->dev->priv.steering->fdb_root_ns;
1742 ns = esw->dev->priv.steering->fdb_root_ns;
1745 err = mlx5_flow_namespace_set_peer(ns, peer_ns);
1749 err = mlx5_flow_namespace_set_peer(peer_ns, ns);
1751 mlx5_flow_namespace_set_peer(ns, NULL);
1755 mlx5_flow_namespace_set_peer(ns, NULL);
1756 mlx5_flow_namespace_set_peer(peer_ns, NULL);
1762 static int mlx5_esw_offloads_devcom_event(int event,
1766 struct mlx5_eswitch *esw = my_data;
1767 struct mlx5_devcom *devcom = esw->dev->priv.devcom;
1768 struct mlx5_eswitch *peer_esw = event_data;
1772 case ESW_OFFLOADS_DEVCOM_PAIR:
1773 if (mlx5_eswitch_vport_match_metadata_enabled(esw) !=
1774 mlx5_eswitch_vport_match_metadata_enabled(peer_esw))
1777 err = mlx5_esw_offloads_set_ns_peer(esw, peer_esw, true);
1780 err = mlx5_esw_offloads_pair(esw, peer_esw);
1784 err = mlx5_esw_offloads_pair(peer_esw, esw);
1788 mlx5_devcom_set_paired(devcom, MLX5_DEVCOM_ESW_OFFLOADS, true);
1791 case ESW_OFFLOADS_DEVCOM_UNPAIR:
1792 if (!mlx5_devcom_is_paired(devcom, MLX5_DEVCOM_ESW_OFFLOADS))
1795 mlx5_devcom_set_paired(devcom, MLX5_DEVCOM_ESW_OFFLOADS, false);
1796 mlx5_esw_offloads_unpair(peer_esw);
1797 mlx5_esw_offloads_unpair(esw);
1798 mlx5_esw_offloads_set_ns_peer(esw, peer_esw, false);
1805 mlx5_esw_offloads_unpair(esw);
1807 mlx5_esw_offloads_set_ns_peer(esw, peer_esw, false);
1809 mlx5_core_err(esw->dev, "esw offloads devcom event failure, event %u err %d",
1814 static void esw_offloads_devcom_init(struct mlx5_eswitch *esw)
1816 struct mlx5_devcom *devcom = esw->dev->priv.devcom;
1818 INIT_LIST_HEAD(&esw->offloads.peer_flows);
1819 mutex_init(&esw->offloads.peer_mutex);
1821 if (!MLX5_CAP_ESW(esw->dev, merged_eswitch))
1824 mlx5_devcom_register_component(devcom,
1825 MLX5_DEVCOM_ESW_OFFLOADS,
1826 mlx5_esw_offloads_devcom_event,
1829 mlx5_devcom_send_event(devcom,
1830 MLX5_DEVCOM_ESW_OFFLOADS,
1831 ESW_OFFLOADS_DEVCOM_PAIR, esw);
1834 static void esw_offloads_devcom_cleanup(struct mlx5_eswitch *esw)
1836 struct mlx5_devcom *devcom = esw->dev->priv.devcom;
1838 if (!MLX5_CAP_ESW(esw->dev, merged_eswitch))
1841 mlx5_devcom_send_event(devcom, MLX5_DEVCOM_ESW_OFFLOADS,
1842 ESW_OFFLOADS_DEVCOM_UNPAIR, esw);
1844 mlx5_devcom_unregister_component(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
1848 esw_check_vport_match_metadata_supported(const struct mlx5_eswitch *esw)
1850 if (!MLX5_CAP_ESW(esw->dev, esw_uplink_ingress_acl))
1853 if (!(MLX5_CAP_ESW_FLOWTABLE(esw->dev, fdb_to_vport_reg_c_id) &
1854 MLX5_FDB_TO_VPORT_REG_C_0))
1857 if (!MLX5_CAP_ESW_FLOWTABLE(esw->dev, flow_source))
1860 if (mlx5_core_is_ecpf_esw_manager(esw->dev) ||
1861 mlx5_ecpf_vport_exists(esw->dev))
1868 esw_check_vport_match_metadata_mandatory(const struct mlx5_eswitch *esw)
1870 return mlx5_core_mp_enabled(esw->dev);
1873 static bool esw_use_vport_metadata(const struct mlx5_eswitch *esw)
1875 return esw_check_vport_match_metadata_mandatory(esw) &&
1876 esw_check_vport_match_metadata_supported(esw);
1879 u32 mlx5_esw_match_metadata_alloc(struct mlx5_eswitch *esw)
1881 u32 num_vports = GENMASK(ESW_VPORT_BITS - 1, 0) - 1;
1882 u32 vhca_id_mask = GENMASK(ESW_VHCA_ID_BITS - 1, 0);
1883 u32 vhca_id = MLX5_CAP_GEN(esw->dev, vhca_id);
1888 /* Make sure the vhca_id fits the ESW_VHCA_ID_BITS */
1889 WARN_ON_ONCE(vhca_id >= BIT(ESW_VHCA_ID_BITS));
1891 /* Trim vhca_id to ESW_VHCA_ID_BITS */
1892 vhca_id &= vhca_id_mask;
1894 start = (vhca_id << ESW_VPORT_BITS);
1895 end = start + num_vports;
1897 start += 1; /* zero is reserved/invalid metadata */
1898 id = ida_alloc_range(&esw->offloads.vport_metadata_ida, start, end, GFP_KERNEL);
1900 return (id < 0) ? 0 : id;
1903 void mlx5_esw_match_metadata_free(struct mlx5_eswitch *esw, u32 metadata)
1905 ida_free(&esw->offloads.vport_metadata_ida, metadata);
1908 static int esw_offloads_vport_metadata_setup(struct mlx5_eswitch *esw,
1909 struct mlx5_vport *vport)
1911 if (vport->vport == MLX5_VPORT_UPLINK)
1914 vport->default_metadata = mlx5_esw_match_metadata_alloc(esw);
1915 vport->metadata = vport->default_metadata;
1916 return vport->metadata ? 0 : -ENOSPC;
1919 static void esw_offloads_vport_metadata_cleanup(struct mlx5_eswitch *esw,
1920 struct mlx5_vport *vport)
1922 if (vport->vport == MLX5_VPORT_UPLINK || !vport->default_metadata)
1925 WARN_ON(vport->metadata != vport->default_metadata);
1926 mlx5_esw_match_metadata_free(esw, vport->default_metadata);
1930 esw_vport_create_offloads_acl_tables(struct mlx5_eswitch *esw,
1931 struct mlx5_vport *vport)
1935 err = esw_offloads_vport_metadata_setup(esw, vport);
1939 err = esw_acl_ingress_ofld_setup(esw, vport);
1943 if (mlx5_eswitch_is_vf_vport(esw, vport->vport)) {
1944 err = esw_acl_egress_ofld_setup(esw, vport);
1952 esw_acl_ingress_ofld_cleanup(esw, vport);
1954 esw_offloads_vport_metadata_cleanup(esw, vport);
1960 esw_vport_destroy_offloads_acl_tables(struct mlx5_eswitch *esw,
1961 struct mlx5_vport *vport)
1963 esw_acl_egress_ofld_cleanup(vport);
1964 esw_acl_ingress_ofld_cleanup(esw, vport);
1965 esw_offloads_vport_metadata_cleanup(esw, vport);
1968 static int esw_create_uplink_offloads_acl_tables(struct mlx5_eswitch *esw)
1970 struct mlx5_vport *vport;
1973 if (esw_use_vport_metadata(esw))
1974 esw->flags |= MLX5_ESWITCH_VPORT_MATCH_METADATA;
1976 vport = mlx5_eswitch_get_vport(esw, MLX5_VPORT_UPLINK);
1977 err = esw_vport_create_offloads_acl_tables(esw, vport);
1979 esw->flags &= ~MLX5_ESWITCH_VPORT_MATCH_METADATA;
1983 static void esw_destroy_uplink_offloads_acl_tables(struct mlx5_eswitch *esw)
1985 struct mlx5_vport *vport;
1987 vport = mlx5_eswitch_get_vport(esw, MLX5_VPORT_UPLINK);
1988 esw_vport_destroy_offloads_acl_tables(esw, vport);
1989 esw->flags &= ~MLX5_ESWITCH_VPORT_MATCH_METADATA;
1992 static int esw_offloads_steering_init(struct mlx5_eswitch *esw)
1996 memset(&esw->fdb_table.offloads, 0, sizeof(struct offloads_fdb));
1997 mutex_init(&esw->fdb_table.offloads.vports.lock);
1998 hash_init(esw->fdb_table.offloads.vports.table);
2000 err = esw_create_uplink_offloads_acl_tables(esw);
2002 goto create_acl_err;
2004 err = esw_create_offloads_table(esw);
2006 goto create_offloads_err;
2008 err = esw_create_restore_table(esw);
2010 goto create_restore_err;
2012 err = esw_create_offloads_fdb_tables(esw);
2014 goto create_fdb_err;
2016 err = esw_create_vport_rx_group(esw);
2023 esw_destroy_offloads_fdb_tables(esw);
2025 esw_destroy_restore_table(esw);
2027 esw_destroy_offloads_table(esw);
2028 create_offloads_err:
2029 esw_destroy_uplink_offloads_acl_tables(esw);
2031 mutex_destroy(&esw->fdb_table.offloads.vports.lock);
2035 static void esw_offloads_steering_cleanup(struct mlx5_eswitch *esw)
2037 esw_destroy_vport_rx_group(esw);
2038 esw_destroy_offloads_fdb_tables(esw);
2039 esw_destroy_restore_table(esw);
2040 esw_destroy_offloads_table(esw);
2041 esw_destroy_uplink_offloads_acl_tables(esw);
2042 mutex_destroy(&esw->fdb_table.offloads.vports.lock);
2046 esw_vfs_changed_event_handler(struct mlx5_eswitch *esw, const u32 *out)
2048 bool host_pf_disabled;
2051 new_num_vfs = MLX5_GET(query_esw_functions_out, out,
2052 host_params_context.host_num_of_vfs);
2053 host_pf_disabled = MLX5_GET(query_esw_functions_out, out,
2054 host_params_context.host_pf_disabled);
2056 if (new_num_vfs == esw->esw_funcs.num_vfs || host_pf_disabled)
2059 /* Number of VFs can only change from "0 to x" or "x to 0". */
2060 if (esw->esw_funcs.num_vfs > 0) {
2061 mlx5_eswitch_unload_vf_vports(esw, esw->esw_funcs.num_vfs);
2065 err = mlx5_eswitch_load_vf_vports(esw, new_num_vfs,
2066 MLX5_VPORT_UC_ADDR_CHANGE);
2070 esw->esw_funcs.num_vfs = new_num_vfs;
2073 static void esw_functions_changed_event_handler(struct work_struct *work)
2075 struct mlx5_host_work *host_work;
2076 struct mlx5_eswitch *esw;
2079 host_work = container_of(work, struct mlx5_host_work, work);
2080 esw = host_work->esw;
2082 out = mlx5_esw_query_functions(esw->dev);
2086 esw_vfs_changed_event_handler(esw, out);
2092 int mlx5_esw_funcs_changed_handler(struct notifier_block *nb, unsigned long type, void *data)
2094 struct mlx5_esw_functions *esw_funcs;
2095 struct mlx5_host_work *host_work;
2096 struct mlx5_eswitch *esw;
2098 host_work = kzalloc(sizeof(*host_work), GFP_ATOMIC);
2102 esw_funcs = mlx5_nb_cof(nb, struct mlx5_esw_functions, nb);
2103 esw = container_of(esw_funcs, struct mlx5_eswitch, esw_funcs);
2105 host_work->esw = esw;
2107 INIT_WORK(&host_work->work, esw_functions_changed_event_handler);
2108 queue_work(esw->work_queue, &host_work->work);
2113 int esw_offloads_enable(struct mlx5_eswitch *esw)
2115 struct mlx5_vport *vport;
2118 if (MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, reformat) &&
2119 MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, decap))
2120 esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_BASIC;
2122 esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_NONE;
2124 mutex_init(&esw->offloads.termtbl_mutex);
2125 mlx5_rdma_enable_roce(esw->dev);
2127 err = esw_set_passing_vport_metadata(esw, true);
2129 goto err_vport_metadata;
2131 err = esw_offloads_steering_init(esw);
2133 goto err_steering_init;
2135 /* Representor will control the vport link state */
2136 mlx5_esw_for_each_vf_vport(esw, i, vport, esw->esw_funcs.num_vfs)
2137 vport->info.link_state = MLX5_VPORT_ADMIN_STATE_DOWN;
2139 /* Uplink vport rep must load first. */
2140 err = esw_offloads_load_rep(esw, MLX5_VPORT_UPLINK);
2144 err = mlx5_eswitch_enable_pf_vf_vports(esw, MLX5_VPORT_UC_ADDR_CHANGE);
2148 esw_offloads_devcom_init(esw);
2153 esw_offloads_unload_rep(esw, MLX5_VPORT_UPLINK);
2155 esw_offloads_steering_cleanup(esw);
2157 esw_set_passing_vport_metadata(esw, false);
2159 mlx5_rdma_disable_roce(esw->dev);
2160 mutex_destroy(&esw->offloads.termtbl_mutex);
2164 static int esw_offloads_stop(struct mlx5_eswitch *esw,
2165 struct netlink_ext_ack *extack)
2169 mlx5_eswitch_disable_locked(esw, false);
2170 err = mlx5_eswitch_enable_locked(esw, MLX5_ESWITCH_LEGACY,
2171 MLX5_ESWITCH_IGNORE_NUM_VFS);
2173 NL_SET_ERR_MSG_MOD(extack, "Failed setting eswitch to legacy");
2174 err1 = mlx5_eswitch_enable_locked(esw, MLX5_ESWITCH_OFFLOADS,
2175 MLX5_ESWITCH_IGNORE_NUM_VFS);
2177 NL_SET_ERR_MSG_MOD(extack,
2178 "Failed setting eswitch back to offloads");
2185 void esw_offloads_disable(struct mlx5_eswitch *esw)
2187 esw_offloads_devcom_cleanup(esw);
2188 mlx5_eswitch_disable_pf_vf_vports(esw);
2189 esw_offloads_unload_rep(esw, MLX5_VPORT_UPLINK);
2190 esw_set_passing_vport_metadata(esw, false);
2191 esw_offloads_steering_cleanup(esw);
2192 mlx5_rdma_disable_roce(esw->dev);
2193 mutex_destroy(&esw->offloads.termtbl_mutex);
2194 esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_NONE;
2197 static int esw_mode_from_devlink(u16 mode, u16 *mlx5_mode)
2200 case DEVLINK_ESWITCH_MODE_LEGACY:
2201 *mlx5_mode = MLX5_ESWITCH_LEGACY;
2203 case DEVLINK_ESWITCH_MODE_SWITCHDEV:
2204 *mlx5_mode = MLX5_ESWITCH_OFFLOADS;
2213 static int esw_mode_to_devlink(u16 mlx5_mode, u16 *mode)
2215 switch (mlx5_mode) {
2216 case MLX5_ESWITCH_LEGACY:
2217 *mode = DEVLINK_ESWITCH_MODE_LEGACY;
2219 case MLX5_ESWITCH_OFFLOADS:
2220 *mode = DEVLINK_ESWITCH_MODE_SWITCHDEV;
2229 static int esw_inline_mode_from_devlink(u8 mode, u8 *mlx5_mode)
2232 case DEVLINK_ESWITCH_INLINE_MODE_NONE:
2233 *mlx5_mode = MLX5_INLINE_MODE_NONE;
2235 case DEVLINK_ESWITCH_INLINE_MODE_LINK:
2236 *mlx5_mode = MLX5_INLINE_MODE_L2;
2238 case DEVLINK_ESWITCH_INLINE_MODE_NETWORK:
2239 *mlx5_mode = MLX5_INLINE_MODE_IP;
2241 case DEVLINK_ESWITCH_INLINE_MODE_TRANSPORT:
2242 *mlx5_mode = MLX5_INLINE_MODE_TCP_UDP;
2251 static int esw_inline_mode_to_devlink(u8 mlx5_mode, u8 *mode)
2253 switch (mlx5_mode) {
2254 case MLX5_INLINE_MODE_NONE:
2255 *mode = DEVLINK_ESWITCH_INLINE_MODE_NONE;
2257 case MLX5_INLINE_MODE_L2:
2258 *mode = DEVLINK_ESWITCH_INLINE_MODE_LINK;
2260 case MLX5_INLINE_MODE_IP:
2261 *mode = DEVLINK_ESWITCH_INLINE_MODE_NETWORK;
2263 case MLX5_INLINE_MODE_TCP_UDP:
2264 *mode = DEVLINK_ESWITCH_INLINE_MODE_TRANSPORT;
2273 static int eswitch_devlink_esw_mode_check(const struct mlx5_eswitch *esw)
2275 /* devlink commands in NONE eswitch mode are currently supported only
2278 return (esw->mode == MLX5_ESWITCH_NONE &&
2279 !mlx5_core_is_ecpf_esw_manager(esw->dev)) ? -EOPNOTSUPP : 0;
2282 int mlx5_devlink_eswitch_mode_set(struct devlink *devlink, u16 mode,
2283 struct netlink_ext_ack *extack)
2285 u16 cur_mlx5_mode, mlx5_mode = 0;
2286 struct mlx5_eswitch *esw;
2289 esw = mlx5_devlink_eswitch_get(devlink);
2291 return PTR_ERR(esw);
2293 if (esw_mode_from_devlink(mode, &mlx5_mode))
2296 mutex_lock(&esw->mode_lock);
2297 cur_mlx5_mode = esw->mode;
2298 if (cur_mlx5_mode == mlx5_mode)
2301 if (mode == DEVLINK_ESWITCH_MODE_SWITCHDEV)
2302 err = esw_offloads_start(esw, extack);
2303 else if (mode == DEVLINK_ESWITCH_MODE_LEGACY)
2304 err = esw_offloads_stop(esw, extack);
2309 mutex_unlock(&esw->mode_lock);
2313 int mlx5_devlink_eswitch_mode_get(struct devlink *devlink, u16 *mode)
2315 struct mlx5_eswitch *esw;
2318 esw = mlx5_devlink_eswitch_get(devlink);
2320 return PTR_ERR(esw);
2322 mutex_lock(&esw->mode_lock);
2323 err = eswitch_devlink_esw_mode_check(esw);
2327 err = esw_mode_to_devlink(esw->mode, mode);
2329 mutex_unlock(&esw->mode_lock);
2333 int mlx5_devlink_eswitch_inline_mode_set(struct devlink *devlink, u8 mode,
2334 struct netlink_ext_ack *extack)
2336 struct mlx5_core_dev *dev = devlink_priv(devlink);
2337 int err, vport, num_vport;
2338 struct mlx5_eswitch *esw;
2341 esw = mlx5_devlink_eswitch_get(devlink);
2343 return PTR_ERR(esw);
2345 mutex_lock(&esw->mode_lock);
2346 err = eswitch_devlink_esw_mode_check(esw);
2350 switch (MLX5_CAP_ETH(dev, wqe_inline_mode)) {
2351 case MLX5_CAP_INLINE_MODE_NOT_REQUIRED:
2352 if (mode == DEVLINK_ESWITCH_INLINE_MODE_NONE)
2355 case MLX5_CAP_INLINE_MODE_L2:
2356 NL_SET_ERR_MSG_MOD(extack, "Inline mode can't be set");
2359 case MLX5_CAP_INLINE_MODE_VPORT_CONTEXT:
2363 if (atomic64_read(&esw->offloads.num_flows) > 0) {
2364 NL_SET_ERR_MSG_MOD(extack,
2365 "Can't set inline mode when flows are configured");
2370 err = esw_inline_mode_from_devlink(mode, &mlx5_mode);
2374 mlx5_esw_for_each_host_func_vport(esw, vport, esw->esw_funcs.num_vfs) {
2375 err = mlx5_modify_nic_vport_min_inline(dev, vport, mlx5_mode);
2377 NL_SET_ERR_MSG_MOD(extack,
2378 "Failed to set min inline on vport");
2379 goto revert_inline_mode;
2383 esw->offloads.inline_mode = mlx5_mode;
2384 mutex_unlock(&esw->mode_lock);
2388 num_vport = --vport;
2389 mlx5_esw_for_each_host_func_vport_reverse(esw, vport, num_vport)
2390 mlx5_modify_nic_vport_min_inline(dev,
2392 esw->offloads.inline_mode);
2394 mutex_unlock(&esw->mode_lock);
2398 int mlx5_devlink_eswitch_inline_mode_get(struct devlink *devlink, u8 *mode)
2400 struct mlx5_eswitch *esw;
2403 esw = mlx5_devlink_eswitch_get(devlink);
2405 return PTR_ERR(esw);
2407 mutex_lock(&esw->mode_lock);
2408 err = eswitch_devlink_esw_mode_check(esw);
2412 err = esw_inline_mode_to_devlink(esw->offloads.inline_mode, mode);
2414 mutex_unlock(&esw->mode_lock);
2418 int mlx5_devlink_eswitch_encap_mode_set(struct devlink *devlink,
2419 enum devlink_eswitch_encap_mode encap,
2420 struct netlink_ext_ack *extack)
2422 struct mlx5_core_dev *dev = devlink_priv(devlink);
2423 struct mlx5_eswitch *esw;
2426 esw = mlx5_devlink_eswitch_get(devlink);
2428 return PTR_ERR(esw);
2430 mutex_lock(&esw->mode_lock);
2431 err = eswitch_devlink_esw_mode_check(esw);
2435 if (encap != DEVLINK_ESWITCH_ENCAP_MODE_NONE &&
2436 (!MLX5_CAP_ESW_FLOWTABLE_FDB(dev, reformat) ||
2437 !MLX5_CAP_ESW_FLOWTABLE_FDB(dev, decap))) {
2442 if (encap && encap != DEVLINK_ESWITCH_ENCAP_MODE_BASIC) {
2447 if (esw->mode == MLX5_ESWITCH_LEGACY) {
2448 esw->offloads.encap = encap;
2452 if (esw->offloads.encap == encap)
2455 if (atomic64_read(&esw->offloads.num_flows) > 0) {
2456 NL_SET_ERR_MSG_MOD(extack,
2457 "Can't set encapsulation when flows are configured");
2462 esw_destroy_offloads_fdb_tables(esw);
2464 esw->offloads.encap = encap;
2466 err = esw_create_offloads_fdb_tables(esw);
2469 NL_SET_ERR_MSG_MOD(extack,
2470 "Failed re-creating fast FDB table");
2471 esw->offloads.encap = !encap;
2472 (void)esw_create_offloads_fdb_tables(esw);
2476 mutex_unlock(&esw->mode_lock);
2480 int mlx5_devlink_eswitch_encap_mode_get(struct devlink *devlink,
2481 enum devlink_eswitch_encap_mode *encap)
2483 struct mlx5_eswitch *esw;
2486 esw = mlx5_devlink_eswitch_get(devlink);
2488 return PTR_ERR(esw);
2491 mutex_lock(&esw->mode_lock);
2492 err = eswitch_devlink_esw_mode_check(esw);
2496 *encap = esw->offloads.encap;
2498 mutex_unlock(&esw->mode_lock);
2503 mlx5_eswitch_vport_has_rep(const struct mlx5_eswitch *esw, u16 vport_num)
2505 /* Currently, only ECPF based device has representor for host PF. */
2506 if (vport_num == MLX5_VPORT_PF &&
2507 !mlx5_core_is_ecpf_esw_manager(esw->dev))
2510 if (vport_num == MLX5_VPORT_ECPF &&
2511 !mlx5_ecpf_vport_exists(esw->dev))
2517 void mlx5_eswitch_register_vport_reps(struct mlx5_eswitch *esw,
2518 const struct mlx5_eswitch_rep_ops *ops,
2521 struct mlx5_eswitch_rep_data *rep_data;
2522 struct mlx5_eswitch_rep *rep;
2525 esw->offloads.rep_ops[rep_type] = ops;
2526 mlx5_esw_for_all_reps(esw, i, rep) {
2527 if (likely(mlx5_eswitch_vport_has_rep(esw, i))) {
2528 rep_data = &rep->rep_data[rep_type];
2529 atomic_set(&rep_data->state, REP_REGISTERED);
2533 EXPORT_SYMBOL(mlx5_eswitch_register_vport_reps);
2535 void mlx5_eswitch_unregister_vport_reps(struct mlx5_eswitch *esw, u8 rep_type)
2537 struct mlx5_eswitch_rep *rep;
2540 if (esw->mode == MLX5_ESWITCH_OFFLOADS)
2541 __unload_reps_all_vport(esw, rep_type);
2543 mlx5_esw_for_all_reps(esw, i, rep)
2544 atomic_set(&rep->rep_data[rep_type].state, REP_UNREGISTERED);
2546 EXPORT_SYMBOL(mlx5_eswitch_unregister_vport_reps);
2548 void *mlx5_eswitch_get_uplink_priv(struct mlx5_eswitch *esw, u8 rep_type)
2550 struct mlx5_eswitch_rep *rep;
2552 rep = mlx5_eswitch_get_rep(esw, MLX5_VPORT_UPLINK);
2553 return rep->rep_data[rep_type].priv;
2556 void *mlx5_eswitch_get_proto_dev(struct mlx5_eswitch *esw,
2560 struct mlx5_eswitch_rep *rep;
2562 rep = mlx5_eswitch_get_rep(esw, vport);
2564 if (atomic_read(&rep->rep_data[rep_type].state) == REP_LOADED &&
2565 esw->offloads.rep_ops[rep_type]->get_proto_dev)
2566 return esw->offloads.rep_ops[rep_type]->get_proto_dev(rep);
2569 EXPORT_SYMBOL(mlx5_eswitch_get_proto_dev);
2571 void *mlx5_eswitch_uplink_get_proto_dev(struct mlx5_eswitch *esw, u8 rep_type)
2573 return mlx5_eswitch_get_proto_dev(esw, MLX5_VPORT_UPLINK, rep_type);
2575 EXPORT_SYMBOL(mlx5_eswitch_uplink_get_proto_dev);
2577 struct mlx5_eswitch_rep *mlx5_eswitch_vport_rep(struct mlx5_eswitch *esw,
2580 return mlx5_eswitch_get_rep(esw, vport);
2582 EXPORT_SYMBOL(mlx5_eswitch_vport_rep);
2584 bool mlx5_eswitch_is_vf_vport(const struct mlx5_eswitch *esw, u16 vport_num)
2586 return vport_num >= MLX5_VPORT_FIRST_VF &&
2587 vport_num <= esw->dev->priv.sriov.max_vfs;
2590 bool mlx5_eswitch_reg_c1_loopback_enabled(const struct mlx5_eswitch *esw)
2592 return !!(esw->flags & MLX5_ESWITCH_REG_C1_LOOPBACK_ENABLED);
2594 EXPORT_SYMBOL(mlx5_eswitch_reg_c1_loopback_enabled);
2596 bool mlx5_eswitch_vport_match_metadata_enabled(const struct mlx5_eswitch *esw)
2598 return !!(esw->flags & MLX5_ESWITCH_VPORT_MATCH_METADATA);
2600 EXPORT_SYMBOL(mlx5_eswitch_vport_match_metadata_enabled);
2602 u32 mlx5_eswitch_get_vport_metadata_for_match(struct mlx5_eswitch *esw,
2605 struct mlx5_vport *vport = mlx5_eswitch_get_vport(esw, vport_num);
2607 if (WARN_ON_ONCE(IS_ERR(vport)))
2610 return vport->metadata << (32 - ESW_SOURCE_PORT_METADATA_BITS);
2612 EXPORT_SYMBOL(mlx5_eswitch_get_vport_metadata_for_match);