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/mlx5/driver.h>
35 #include <linux/mlx5/mlx5_ifc.h>
36 #include <linux/mlx5/vport.h>
37 #include <linux/mlx5/fs.h>
38 #include "mlx5_core.h"
43 #include "lib/devcom.h"
46 /* There are two match-all miss flows, one for unicast dst mac and
49 #define MLX5_ESW_MISS_FLOWS (2)
51 #define fdb_prio_table(esw, chain, prio, level) \
52 (esw)->fdb_table.offloads.fdb_prio[(chain)][(prio)][(level)]
54 #define UPLINK_REP_INDEX 0
56 static struct mlx5_eswitch_rep *mlx5_eswitch_get_rep(struct mlx5_eswitch *esw,
59 int idx = mlx5_eswitch_vport_num_to_index(esw, vport_num);
61 WARN_ON(idx > esw->total_vports - 1);
62 return &esw->offloads.vport_reps[idx];
65 static struct mlx5_flow_table *
66 esw_get_prio_table(struct mlx5_eswitch *esw, u32 chain, u16 prio, int level);
68 esw_put_prio_table(struct mlx5_eswitch *esw, u32 chain, u16 prio, int level);
70 bool mlx5_eswitch_prios_supported(struct mlx5_eswitch *esw)
72 return (!!(esw->fdb_table.flags & ESW_FDB_CHAINS_AND_PRIOS_SUPPORTED));
75 u32 mlx5_eswitch_get_chain_range(struct mlx5_eswitch *esw)
77 if (esw->fdb_table.flags & ESW_FDB_CHAINS_AND_PRIOS_SUPPORTED)
83 u16 mlx5_eswitch_get_prio_range(struct mlx5_eswitch *esw)
85 if (esw->fdb_table.flags & ESW_FDB_CHAINS_AND_PRIOS_SUPPORTED)
92 mlx5_eswitch_set_rule_source_port(struct mlx5_eswitch *esw,
93 struct mlx5_flow_spec *spec,
94 struct mlx5_esw_flow_attr *attr)
99 /* Use metadata matching because vport is not represented by single
100 * VHCA in dual-port RoCE mode, and matching on source vport may fail.
102 if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
103 misc2 = MLX5_ADDR_OF(fte_match_param, spec->match_value, misc_parameters_2);
104 MLX5_SET(fte_match_set_misc2, misc2, metadata_reg_c_0,
105 mlx5_eswitch_get_vport_metadata_for_match(attr->in_mdev->priv.eswitch,
106 attr->in_rep->vport));
108 misc2 = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, misc_parameters_2);
109 MLX5_SET_TO_ONES(fte_match_set_misc2, misc2, metadata_reg_c_0);
111 spec->match_criteria_enable |= MLX5_MATCH_MISC_PARAMETERS_2;
112 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, misc_parameters);
113 if (memchr_inv(misc, 0, MLX5_ST_SZ_BYTES(fte_match_set_misc)))
114 spec->match_criteria_enable |= MLX5_MATCH_MISC_PARAMETERS;
116 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value, misc_parameters);
117 MLX5_SET(fte_match_set_misc, misc, source_port, attr->in_rep->vport);
119 if (MLX5_CAP_ESW(esw->dev, merged_eswitch))
120 MLX5_SET(fte_match_set_misc, misc,
121 source_eswitch_owner_vhca_id,
122 MLX5_CAP_GEN(attr->in_mdev, vhca_id));
124 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, misc_parameters);
125 MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
126 if (MLX5_CAP_ESW(esw->dev, merged_eswitch))
127 MLX5_SET_TO_ONES(fte_match_set_misc, misc,
128 source_eswitch_owner_vhca_id);
130 spec->match_criteria_enable |= MLX5_MATCH_MISC_PARAMETERS;
133 if (MLX5_CAP_ESW_FLOWTABLE(esw->dev, flow_source) &&
134 attr->in_rep->vport == MLX5_VPORT_UPLINK)
135 spec->flow_context.flow_source = MLX5_FLOW_CONTEXT_FLOW_SOURCE_UPLINK;
138 struct mlx5_flow_handle *
139 mlx5_eswitch_add_offloaded_rule(struct mlx5_eswitch *esw,
140 struct mlx5_flow_spec *spec,
141 struct mlx5_esw_flow_attr *attr)
143 struct mlx5_flow_destination dest[MLX5_MAX_FLOW_FWD_VPORTS + 1] = {};
144 struct mlx5_flow_act flow_act = { .flags = FLOW_ACT_NO_APPEND, };
145 bool split = !!(attr->split_count);
146 struct mlx5_flow_handle *rule;
147 struct mlx5_flow_table *fdb;
150 if (esw->mode != MLX5_ESWITCH_OFFLOADS)
151 return ERR_PTR(-EOPNOTSUPP);
153 flow_act.action = attr->action;
154 /* if per flow vlan pop/push is emulated, don't set that into the firmware */
155 if (!mlx5_eswitch_vlan_actions_supported(esw->dev, 1))
156 flow_act.action &= ~(MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
157 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP);
158 else if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH) {
159 flow_act.vlan[0].ethtype = ntohs(attr->vlan_proto[0]);
160 flow_act.vlan[0].vid = attr->vlan_vid[0];
161 flow_act.vlan[0].prio = attr->vlan_prio[0];
162 if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2) {
163 flow_act.vlan[1].ethtype = ntohs(attr->vlan_proto[1]);
164 flow_act.vlan[1].vid = attr->vlan_vid[1];
165 flow_act.vlan[1].prio = attr->vlan_prio[1];
169 if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
170 if (attr->dest_chain) {
171 struct mlx5_flow_table *ft;
173 ft = esw_get_prio_table(esw, attr->dest_chain, 1, 0);
176 goto err_create_goto_table;
179 dest[i].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
183 for (j = attr->split_count; j < attr->out_count; j++) {
184 dest[i].type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
185 dest[i].vport.num = attr->dests[j].rep->vport;
186 dest[i].vport.vhca_id =
187 MLX5_CAP_GEN(attr->dests[j].mdev, vhca_id);
188 if (MLX5_CAP_ESW(esw->dev, merged_eswitch))
189 dest[i].vport.flags |=
190 MLX5_FLOW_DEST_VPORT_VHCA_ID;
191 if (attr->dests[j].flags & MLX5_ESW_DEST_ENCAP) {
192 flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT;
193 flow_act.reformat_id = attr->dests[j].encap_id;
194 dest[i].vport.flags |= MLX5_FLOW_DEST_VPORT_REFORMAT_ID;
195 dest[i].vport.reformat_id =
196 attr->dests[j].encap_id;
202 if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
203 dest[i].type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
204 dest[i].counter_id = mlx5_fc_id(attr->counter);
208 mlx5_eswitch_set_rule_source_port(esw, spec, attr);
210 if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_DECAP) {
211 if (attr->tunnel_match_level != MLX5_MATCH_NONE)
212 spec->match_criteria_enable |= MLX5_MATCH_OUTER_HEADERS;
213 if (attr->match_level != MLX5_MATCH_NONE)
214 spec->match_criteria_enable |= MLX5_MATCH_INNER_HEADERS;
215 } else if (attr->match_level != MLX5_MATCH_NONE) {
216 spec->match_criteria_enable |= MLX5_MATCH_OUTER_HEADERS;
219 if (flow_act.action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
220 flow_act.modify_id = attr->mod_hdr_id;
222 fdb = esw_get_prio_table(esw, attr->chain, attr->prio, !!split);
224 rule = ERR_CAST(fdb);
228 if (mlx5_eswitch_termtbl_required(esw, &flow_act, spec))
229 rule = mlx5_eswitch_add_termtbl_rule(esw, fdb, spec, attr,
232 rule = mlx5_add_flow_rules(fdb, spec, &flow_act, dest, i);
236 atomic64_inc(&esw->offloads.num_flows);
241 esw_put_prio_table(esw, attr->chain, attr->prio, !!split);
243 if (attr->dest_chain)
244 esw_put_prio_table(esw, attr->dest_chain, 1, 0);
245 err_create_goto_table:
249 struct mlx5_flow_handle *
250 mlx5_eswitch_add_fwd_rule(struct mlx5_eswitch *esw,
251 struct mlx5_flow_spec *spec,
252 struct mlx5_esw_flow_attr *attr)
254 struct mlx5_flow_destination dest[MLX5_MAX_FLOW_FWD_VPORTS + 1] = {};
255 struct mlx5_flow_act flow_act = { .flags = FLOW_ACT_NO_APPEND, };
256 struct mlx5_flow_table *fast_fdb;
257 struct mlx5_flow_table *fwd_fdb;
258 struct mlx5_flow_handle *rule;
261 fast_fdb = esw_get_prio_table(esw, attr->chain, attr->prio, 0);
262 if (IS_ERR(fast_fdb)) {
263 rule = ERR_CAST(fast_fdb);
267 fwd_fdb = esw_get_prio_table(esw, attr->chain, attr->prio, 1);
268 if (IS_ERR(fwd_fdb)) {
269 rule = ERR_CAST(fwd_fdb);
273 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
274 for (i = 0; i < attr->split_count; i++) {
275 dest[i].type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
276 dest[i].vport.num = attr->dests[i].rep->vport;
277 dest[i].vport.vhca_id =
278 MLX5_CAP_GEN(attr->dests[i].mdev, vhca_id);
279 if (MLX5_CAP_ESW(esw->dev, merged_eswitch))
280 dest[i].vport.flags |= MLX5_FLOW_DEST_VPORT_VHCA_ID;
281 if (attr->dests[i].flags & MLX5_ESW_DEST_ENCAP) {
282 dest[i].vport.flags |= MLX5_FLOW_DEST_VPORT_REFORMAT_ID;
283 dest[i].vport.reformat_id = attr->dests[i].encap_id;
286 dest[i].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
287 dest[i].ft = fwd_fdb,
290 mlx5_eswitch_set_rule_source_port(esw, spec, attr);
292 spec->match_criteria_enable |= MLX5_MATCH_MISC_PARAMETERS;
293 if (attr->match_level != MLX5_MATCH_NONE)
294 spec->match_criteria_enable |= MLX5_MATCH_OUTER_HEADERS;
296 rule = mlx5_add_flow_rules(fast_fdb, spec, &flow_act, dest, i);
301 atomic64_inc(&esw->offloads.num_flows);
305 esw_put_prio_table(esw, attr->chain, attr->prio, 1);
307 esw_put_prio_table(esw, attr->chain, attr->prio, 0);
313 __mlx5_eswitch_del_rule(struct mlx5_eswitch *esw,
314 struct mlx5_flow_handle *rule,
315 struct mlx5_esw_flow_attr *attr,
318 bool split = (attr->split_count > 0);
321 mlx5_del_flow_rules(rule);
323 /* unref the term table */
324 for (i = 0; i < MLX5_MAX_FLOW_FWD_VPORTS; i++) {
325 if (attr->dests[i].termtbl)
326 mlx5_eswitch_termtbl_put(esw, attr->dests[i].termtbl);
329 atomic64_dec(&esw->offloads.num_flows);
332 esw_put_prio_table(esw, attr->chain, attr->prio, 1);
333 esw_put_prio_table(esw, attr->chain, attr->prio, 0);
335 esw_put_prio_table(esw, attr->chain, attr->prio, !!split);
336 if (attr->dest_chain)
337 esw_put_prio_table(esw, attr->dest_chain, 1, 0);
342 mlx5_eswitch_del_offloaded_rule(struct mlx5_eswitch *esw,
343 struct mlx5_flow_handle *rule,
344 struct mlx5_esw_flow_attr *attr)
346 __mlx5_eswitch_del_rule(esw, rule, attr, false);
350 mlx5_eswitch_del_fwd_rule(struct mlx5_eswitch *esw,
351 struct mlx5_flow_handle *rule,
352 struct mlx5_esw_flow_attr *attr)
354 __mlx5_eswitch_del_rule(esw, rule, attr, true);
357 static int esw_set_global_vlan_pop(struct mlx5_eswitch *esw, u8 val)
359 struct mlx5_eswitch_rep *rep;
362 esw_debug(esw->dev, "%s applying global %s policy\n", __func__, val ? "pop" : "none");
363 mlx5_esw_for_each_host_func_rep(esw, i, rep, esw->esw_funcs.num_vfs) {
364 if (atomic_read(&rep->rep_data[REP_ETH].state) != REP_LOADED)
367 err = __mlx5_eswitch_set_vport_vlan(esw, rep->vport, 0, 0, val);
376 static struct mlx5_eswitch_rep *
377 esw_vlan_action_get_vport(struct mlx5_esw_flow_attr *attr, bool push, bool pop)
379 struct mlx5_eswitch_rep *in_rep, *out_rep, *vport = NULL;
381 in_rep = attr->in_rep;
382 out_rep = attr->dests[0].rep;
394 static int esw_add_vlan_action_check(struct mlx5_esw_flow_attr *attr,
395 bool push, bool pop, bool fwd)
397 struct mlx5_eswitch_rep *in_rep, *out_rep;
399 if ((push || pop) && !fwd)
402 in_rep = attr->in_rep;
403 out_rep = attr->dests[0].rep;
405 if (push && in_rep->vport == MLX5_VPORT_UPLINK)
408 if (pop && out_rep->vport == MLX5_VPORT_UPLINK)
411 /* vport has vlan push configured, can't offload VF --> wire rules w.o it */
412 if (!push && !pop && fwd)
413 if (in_rep->vlan && out_rep->vport == MLX5_VPORT_UPLINK)
416 /* protects against (1) setting rules with different vlans to push and
417 * (2) setting rules w.o vlans (attr->vlan = 0) && w. vlans to push (!= 0)
419 if (push && in_rep->vlan_refcount && (in_rep->vlan != attr->vlan_vid[0]))
428 int mlx5_eswitch_add_vlan_action(struct mlx5_eswitch *esw,
429 struct mlx5_esw_flow_attr *attr)
431 struct offloads_fdb *offloads = &esw->fdb_table.offloads;
432 struct mlx5_eswitch_rep *vport = NULL;
436 /* nop if we're on the vlan push/pop non emulation mode */
437 if (mlx5_eswitch_vlan_actions_supported(esw->dev, 1))
440 push = !!(attr->action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH);
441 pop = !!(attr->action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP);
442 fwd = !!((attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
445 mutex_lock(&esw->state_lock);
447 err = esw_add_vlan_action_check(attr, push, pop, fwd);
451 attr->vlan_handled = false;
453 vport = esw_vlan_action_get_vport(attr, push, pop);
455 if (!push && !pop && fwd) {
456 /* tracks VF --> wire rules without vlan push action */
457 if (attr->dests[0].rep->vport == MLX5_VPORT_UPLINK) {
458 vport->vlan_refcount++;
459 attr->vlan_handled = true;
468 if (!(offloads->vlan_push_pop_refcount)) {
469 /* it's the 1st vlan rule, apply global vlan pop policy */
470 err = esw_set_global_vlan_pop(esw, SET_VLAN_STRIP);
474 offloads->vlan_push_pop_refcount++;
477 if (vport->vlan_refcount)
480 err = __mlx5_eswitch_set_vport_vlan(esw, vport->vport, attr->vlan_vid[0], 0,
481 SET_VLAN_INSERT | SET_VLAN_STRIP);
484 vport->vlan = attr->vlan_vid[0];
486 vport->vlan_refcount++;
490 attr->vlan_handled = true;
492 mutex_unlock(&esw->state_lock);
496 int mlx5_eswitch_del_vlan_action(struct mlx5_eswitch *esw,
497 struct mlx5_esw_flow_attr *attr)
499 struct offloads_fdb *offloads = &esw->fdb_table.offloads;
500 struct mlx5_eswitch_rep *vport = NULL;
504 /* nop if we're on the vlan push/pop non emulation mode */
505 if (mlx5_eswitch_vlan_actions_supported(esw->dev, 1))
508 if (!attr->vlan_handled)
511 push = !!(attr->action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH);
512 pop = !!(attr->action & MLX5_FLOW_CONTEXT_ACTION_VLAN_POP);
513 fwd = !!(attr->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST);
515 mutex_lock(&esw->state_lock);
517 vport = esw_vlan_action_get_vport(attr, push, pop);
519 if (!push && !pop && fwd) {
520 /* tracks VF --> wire rules without vlan push action */
521 if (attr->dests[0].rep->vport == MLX5_VPORT_UPLINK)
522 vport->vlan_refcount--;
528 vport->vlan_refcount--;
529 if (vport->vlan_refcount)
530 goto skip_unset_push;
533 err = __mlx5_eswitch_set_vport_vlan(esw, vport->vport,
534 0, 0, SET_VLAN_STRIP);
540 offloads->vlan_push_pop_refcount--;
541 if (offloads->vlan_push_pop_refcount)
544 /* no more vlan rules, stop global vlan pop policy */
545 err = esw_set_global_vlan_pop(esw, 0);
548 mutex_unlock(&esw->state_lock);
552 struct mlx5_flow_handle *
553 mlx5_eswitch_add_send_to_vport_rule(struct mlx5_eswitch *esw, u16 vport,
556 struct mlx5_flow_act flow_act = {0};
557 struct mlx5_flow_destination dest = {};
558 struct mlx5_flow_handle *flow_rule;
559 struct mlx5_flow_spec *spec;
562 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
564 flow_rule = ERR_PTR(-ENOMEM);
568 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value, misc_parameters);
569 MLX5_SET(fte_match_set_misc, misc, source_sqn, sqn);
570 /* source vport is the esw manager */
571 MLX5_SET(fte_match_set_misc, misc, source_port, esw->manager_vport);
573 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, misc_parameters);
574 MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_sqn);
575 MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
577 spec->match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS;
578 dest.type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
579 dest.vport.num = vport;
580 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
582 flow_rule = mlx5_add_flow_rules(esw->fdb_table.offloads.slow_fdb, spec,
583 &flow_act, &dest, 1);
584 if (IS_ERR(flow_rule))
585 esw_warn(esw->dev, "FDB: Failed to add send to vport rule err %ld\n", PTR_ERR(flow_rule));
590 EXPORT_SYMBOL(mlx5_eswitch_add_send_to_vport_rule);
592 void mlx5_eswitch_del_send_to_vport_rule(struct mlx5_flow_handle *rule)
594 mlx5_del_flow_rules(rule);
597 static int esw_set_passing_vport_metadata(struct mlx5_eswitch *esw, bool enable)
599 u32 out[MLX5_ST_SZ_DW(query_esw_vport_context_out)] = {};
600 u32 in[MLX5_ST_SZ_DW(modify_esw_vport_context_in)] = {};
601 u8 fdb_to_vport_reg_c_id;
604 if (!mlx5_eswitch_vport_match_metadata_enabled(esw))
607 err = mlx5_eswitch_query_esw_vport_context(esw, esw->manager_vport,
612 fdb_to_vport_reg_c_id = MLX5_GET(query_esw_vport_context_out, out,
613 esw_vport_context.fdb_to_vport_reg_c_id);
616 fdb_to_vport_reg_c_id |= MLX5_FDB_TO_VPORT_REG_C_0;
618 fdb_to_vport_reg_c_id &= ~MLX5_FDB_TO_VPORT_REG_C_0;
620 MLX5_SET(modify_esw_vport_context_in, in,
621 esw_vport_context.fdb_to_vport_reg_c_id, fdb_to_vport_reg_c_id);
623 MLX5_SET(modify_esw_vport_context_in, in,
624 field_select.fdb_to_vport_reg_c_id, 1);
626 return mlx5_eswitch_modify_esw_vport_context(esw, esw->manager_vport,
630 static void peer_miss_rules_setup(struct mlx5_eswitch *esw,
631 struct mlx5_core_dev *peer_dev,
632 struct mlx5_flow_spec *spec,
633 struct mlx5_flow_destination *dest)
637 if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
638 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
640 MLX5_SET_TO_ONES(fte_match_set_misc2, misc, metadata_reg_c_0);
642 spec->match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS_2;
644 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
647 MLX5_SET(fte_match_set_misc, misc, source_eswitch_owner_vhca_id,
648 MLX5_CAP_GEN(peer_dev, vhca_id));
650 spec->match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS;
652 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
654 MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
655 MLX5_SET_TO_ONES(fte_match_set_misc, misc,
656 source_eswitch_owner_vhca_id);
659 dest->type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
660 dest->vport.num = peer_dev->priv.eswitch->manager_vport;
661 dest->vport.vhca_id = MLX5_CAP_GEN(peer_dev, vhca_id);
662 dest->vport.flags |= MLX5_FLOW_DEST_VPORT_VHCA_ID;
665 static void esw_set_peer_miss_rule_source_port(struct mlx5_eswitch *esw,
666 struct mlx5_eswitch *peer_esw,
667 struct mlx5_flow_spec *spec,
672 if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
673 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
675 MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
676 mlx5_eswitch_get_vport_metadata_for_match(peer_esw,
679 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
681 MLX5_SET(fte_match_set_misc, misc, source_port, vport);
685 static int esw_add_fdb_peer_miss_rules(struct mlx5_eswitch *esw,
686 struct mlx5_core_dev *peer_dev)
688 struct mlx5_flow_destination dest = {};
689 struct mlx5_flow_act flow_act = {0};
690 struct mlx5_flow_handle **flows;
691 struct mlx5_flow_handle *flow;
692 struct mlx5_flow_spec *spec;
693 /* total vports is the same for both e-switches */
694 int nvports = esw->total_vports;
698 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
702 peer_miss_rules_setup(esw, peer_dev, spec, &dest);
704 flows = kvzalloc(nvports * sizeof(*flows), GFP_KERNEL);
707 goto alloc_flows_err;
710 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
711 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
714 if (mlx5_core_is_ecpf_esw_manager(esw->dev)) {
715 esw_set_peer_miss_rule_source_port(esw, peer_dev->priv.eswitch,
716 spec, MLX5_VPORT_PF);
718 flow = mlx5_add_flow_rules(esw->fdb_table.offloads.slow_fdb,
719 spec, &flow_act, &dest, 1);
722 goto add_pf_flow_err;
724 flows[MLX5_VPORT_PF] = flow;
727 if (mlx5_ecpf_vport_exists(esw->dev)) {
728 MLX5_SET(fte_match_set_misc, misc, source_port, MLX5_VPORT_ECPF);
729 flow = mlx5_add_flow_rules(esw->fdb_table.offloads.slow_fdb,
730 spec, &flow_act, &dest, 1);
733 goto add_ecpf_flow_err;
735 flows[mlx5_eswitch_ecpf_idx(esw)] = flow;
738 mlx5_esw_for_each_vf_vport_num(esw, i, mlx5_core_max_vfs(esw->dev)) {
739 esw_set_peer_miss_rule_source_port(esw,
740 peer_dev->priv.eswitch,
743 flow = mlx5_add_flow_rules(esw->fdb_table.offloads.slow_fdb,
744 spec, &flow_act, &dest, 1);
747 goto add_vf_flow_err;
752 esw->fdb_table.offloads.peer_miss_rules = flows;
759 mlx5_esw_for_each_vf_vport_num_reverse(esw, i, nvports)
760 mlx5_del_flow_rules(flows[i]);
762 if (mlx5_ecpf_vport_exists(esw->dev))
763 mlx5_del_flow_rules(flows[mlx5_eswitch_ecpf_idx(esw)]);
765 if (mlx5_core_is_ecpf_esw_manager(esw->dev))
766 mlx5_del_flow_rules(flows[MLX5_VPORT_PF]);
768 esw_warn(esw->dev, "FDB: Failed to add peer miss flow rule err %d\n", err);
775 static void esw_del_fdb_peer_miss_rules(struct mlx5_eswitch *esw)
777 struct mlx5_flow_handle **flows;
780 flows = esw->fdb_table.offloads.peer_miss_rules;
782 mlx5_esw_for_each_vf_vport_num_reverse(esw, i,
783 mlx5_core_max_vfs(esw->dev))
784 mlx5_del_flow_rules(flows[i]);
786 if (mlx5_ecpf_vport_exists(esw->dev))
787 mlx5_del_flow_rules(flows[mlx5_eswitch_ecpf_idx(esw)]);
789 if (mlx5_core_is_ecpf_esw_manager(esw->dev))
790 mlx5_del_flow_rules(flows[MLX5_VPORT_PF]);
795 static int esw_add_fdb_miss_rule(struct mlx5_eswitch *esw)
797 struct mlx5_flow_act flow_act = {0};
798 struct mlx5_flow_destination dest = {};
799 struct mlx5_flow_handle *flow_rule = NULL;
800 struct mlx5_flow_spec *spec;
807 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
813 spec->match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
814 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
816 dmac_c = MLX5_ADDR_OF(fte_match_param, headers_c,
817 outer_headers.dmac_47_16);
820 dest.type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
821 dest.vport.num = esw->manager_vport;
822 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
824 flow_rule = mlx5_add_flow_rules(esw->fdb_table.offloads.slow_fdb, spec,
825 &flow_act, &dest, 1);
826 if (IS_ERR(flow_rule)) {
827 err = PTR_ERR(flow_rule);
828 esw_warn(esw->dev, "FDB: Failed to add unicast miss flow rule err %d\n", err);
832 esw->fdb_table.offloads.miss_rule_uni = flow_rule;
834 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
836 dmac_v = MLX5_ADDR_OF(fte_match_param, headers_v,
837 outer_headers.dmac_47_16);
839 flow_rule = mlx5_add_flow_rules(esw->fdb_table.offloads.slow_fdb, spec,
840 &flow_act, &dest, 1);
841 if (IS_ERR(flow_rule)) {
842 err = PTR_ERR(flow_rule);
843 esw_warn(esw->dev, "FDB: Failed to add multicast miss flow rule err %d\n", err);
844 mlx5_del_flow_rules(esw->fdb_table.offloads.miss_rule_uni);
848 esw->fdb_table.offloads.miss_rule_multi = flow_rule;
855 #define ESW_OFFLOADS_NUM_GROUPS 4
857 /* Firmware currently has 4 pool of 4 sizes that it supports (ESW_POOLS),
858 * and a virtual memory region of 16M (ESW_SIZE), this region is duplicated
859 * for each flow table pool. We can allocate up to 16M of each pool,
860 * and we keep track of how much we used via put/get_sz_to_pool.
861 * Firmware doesn't report any of this for now.
862 * ESW_POOL is expected to be sorted from large to small
864 #define ESW_SIZE (16 * 1024 * 1024)
865 const unsigned int ESW_POOLS[4] = { 4 * 1024 * 1024, 1 * 1024 * 1024,
866 64 * 1024, 4 * 1024 };
869 get_sz_from_pool(struct mlx5_eswitch *esw)
873 for (i = 0; i < ARRAY_SIZE(ESW_POOLS); i++) {
874 if (esw->fdb_table.offloads.fdb_left[i]) {
875 --esw->fdb_table.offloads.fdb_left[i];
885 put_sz_to_pool(struct mlx5_eswitch *esw, int sz)
889 for (i = 0; i < ARRAY_SIZE(ESW_POOLS); i++) {
890 if (sz >= ESW_POOLS[i]) {
891 ++esw->fdb_table.offloads.fdb_left[i];
897 static struct mlx5_flow_table *
898 create_next_size_table(struct mlx5_eswitch *esw,
899 struct mlx5_flow_namespace *ns,
904 struct mlx5_flow_table *fdb;
907 sz = get_sz_from_pool(esw);
909 return ERR_PTR(-ENOSPC);
911 fdb = mlx5_create_auto_grouped_flow_table(ns,
914 ESW_OFFLOADS_NUM_GROUPS,
918 esw_warn(esw->dev, "Failed to create FDB Table err %d (table prio: %d, level: %d, size: %d)\n",
919 (int)PTR_ERR(fdb), table_prio, level, sz);
920 put_sz_to_pool(esw, sz);
926 static struct mlx5_flow_table *
927 esw_get_prio_table(struct mlx5_eswitch *esw, u32 chain, u16 prio, int level)
929 struct mlx5_core_dev *dev = esw->dev;
930 struct mlx5_flow_table *fdb = NULL;
931 struct mlx5_flow_namespace *ns;
932 int table_prio, l = 0;
935 if (chain == FDB_SLOW_PATH_CHAIN)
936 return esw->fdb_table.offloads.slow_fdb;
938 mutex_lock(&esw->fdb_table.offloads.fdb_prio_lock);
940 fdb = fdb_prio_table(esw, chain, prio, level).fdb;
942 /* take ref on earlier levels as well */
944 fdb_prio_table(esw, chain, prio, level--).num_rules++;
945 mutex_unlock(&esw->fdb_table.offloads.fdb_prio_lock);
949 ns = mlx5_get_fdb_sub_ns(dev, chain);
951 esw_warn(dev, "Failed to get FDB sub namespace\n");
952 mutex_unlock(&esw->fdb_table.offloads.fdb_prio_lock);
953 return ERR_PTR(-EOPNOTSUPP);
956 if (esw->offloads.encap != DEVLINK_ESWITCH_ENCAP_MODE_NONE)
957 flags |= (MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT |
958 MLX5_FLOW_TABLE_TUNNEL_EN_DECAP);
960 table_prio = (chain * FDB_MAX_PRIO) + prio - 1;
962 /* create earlier levels for correct fs_core lookup when
965 for (l = 0; l <= level; l++) {
966 if (fdb_prio_table(esw, chain, prio, l).fdb) {
967 fdb_prio_table(esw, chain, prio, l).num_rules++;
971 fdb = create_next_size_table(esw, ns, table_prio, l, flags);
977 fdb_prio_table(esw, chain, prio, l).fdb = fdb;
978 fdb_prio_table(esw, chain, prio, l).num_rules = 1;
981 mutex_unlock(&esw->fdb_table.offloads.fdb_prio_lock);
985 mutex_unlock(&esw->fdb_table.offloads.fdb_prio_lock);
987 esw_put_prio_table(esw, chain, prio, l);
993 esw_put_prio_table(struct mlx5_eswitch *esw, u32 chain, u16 prio, int level)
997 if (chain == FDB_SLOW_PATH_CHAIN)
1000 mutex_lock(&esw->fdb_table.offloads.fdb_prio_lock);
1002 for (l = level; l >= 0; l--) {
1003 if (--(fdb_prio_table(esw, chain, prio, l).num_rules) > 0)
1006 put_sz_to_pool(esw, fdb_prio_table(esw, chain, prio, l).fdb->max_fte);
1007 mlx5_destroy_flow_table(fdb_prio_table(esw, chain, prio, l).fdb);
1008 fdb_prio_table(esw, chain, prio, l).fdb = NULL;
1011 mutex_unlock(&esw->fdb_table.offloads.fdb_prio_lock);
1014 static void esw_destroy_offloads_fast_fdb_tables(struct mlx5_eswitch *esw)
1016 /* If lazy creation isn't supported, deref the fast path tables */
1017 if (!(esw->fdb_table.flags & ESW_FDB_CHAINS_AND_PRIOS_SUPPORTED)) {
1018 esw_put_prio_table(esw, 0, 1, 1);
1019 esw_put_prio_table(esw, 0, 1, 0);
1023 #define MAX_PF_SQ 256
1024 #define MAX_SQ_NVPORTS 32
1026 static void esw_set_flow_group_source_port(struct mlx5_eswitch *esw,
1029 void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1033 if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
1034 MLX5_SET(create_flow_group_in, flow_group_in,
1035 match_criteria_enable,
1036 MLX5_MATCH_MISC_PARAMETERS_2);
1038 MLX5_SET_TO_ONES(fte_match_param, match_criteria,
1039 misc_parameters_2.metadata_reg_c_0);
1041 MLX5_SET(create_flow_group_in, flow_group_in,
1042 match_criteria_enable,
1043 MLX5_MATCH_MISC_PARAMETERS);
1045 MLX5_SET_TO_ONES(fte_match_param, match_criteria,
1046 misc_parameters.source_port);
1050 static int esw_create_offloads_fdb_tables(struct mlx5_eswitch *esw, int nvports)
1052 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1053 struct mlx5_flow_table_attr ft_attr = {};
1054 struct mlx5_core_dev *dev = esw->dev;
1055 u32 *flow_group_in, max_flow_counter;
1056 struct mlx5_flow_namespace *root_ns;
1057 struct mlx5_flow_table *fdb = NULL;
1058 int table_size, ix, err = 0, i;
1059 struct mlx5_flow_group *g;
1060 u32 flags = 0, fdb_max;
1061 void *match_criteria;
1064 esw_debug(esw->dev, "Create offloads FDB Tables\n");
1065 flow_group_in = kvzalloc(inlen, GFP_KERNEL);
1069 root_ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_FDB);
1071 esw_warn(dev, "Failed to get FDB flow namespace\n");
1076 max_flow_counter = (MLX5_CAP_GEN(dev, max_flow_counter_31_16) << 16) |
1077 MLX5_CAP_GEN(dev, max_flow_counter_15_0);
1078 fdb_max = 1 << MLX5_CAP_ESW_FLOWTABLE_FDB(dev, log_max_ft_size);
1080 esw_debug(dev, "Create offloads FDB table, min (max esw size(2^%d), max counters(%d), groups(%d), max flow table size(2^%d))\n",
1081 MLX5_CAP_ESW_FLOWTABLE_FDB(dev, log_max_ft_size),
1082 max_flow_counter, ESW_OFFLOADS_NUM_GROUPS,
1085 for (i = 0; i < ARRAY_SIZE(ESW_POOLS); i++)
1086 esw->fdb_table.offloads.fdb_left[i] =
1087 ESW_POOLS[i] <= fdb_max ? ESW_SIZE / ESW_POOLS[i] : 0;
1089 table_size = nvports * MAX_SQ_NVPORTS + MAX_PF_SQ +
1090 MLX5_ESW_MISS_FLOWS + esw->total_vports;
1092 /* create the slow path fdb with encap set, so further table instances
1093 * can be created at run time while VFs are probed if the FW allows that.
1095 if (esw->offloads.encap != DEVLINK_ESWITCH_ENCAP_MODE_NONE)
1096 flags |= (MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT |
1097 MLX5_FLOW_TABLE_TUNNEL_EN_DECAP);
1099 ft_attr.flags = flags;
1100 ft_attr.max_fte = table_size;
1101 ft_attr.prio = FDB_SLOW_PATH;
1103 fdb = mlx5_create_flow_table(root_ns, &ft_attr);
1106 esw_warn(dev, "Failed to create slow path FDB Table err %d\n", err);
1109 esw->fdb_table.offloads.slow_fdb = fdb;
1111 /* If lazy creation isn't supported, open the fast path tables now */
1112 if (!MLX5_CAP_ESW_FLOWTABLE(esw->dev, multi_fdb_encap) &&
1113 esw->offloads.encap != DEVLINK_ESWITCH_ENCAP_MODE_NONE) {
1114 esw->fdb_table.flags &= ~ESW_FDB_CHAINS_AND_PRIOS_SUPPORTED;
1115 esw_warn(dev, "Lazy creation of flow tables isn't supported, ignoring priorities\n");
1116 esw_get_prio_table(esw, 0, 1, 0);
1117 esw_get_prio_table(esw, 0, 1, 1);
1119 esw_debug(dev, "Lazy creation of flow tables supported, deferring table opening\n");
1120 esw->fdb_table.flags |= ESW_FDB_CHAINS_AND_PRIOS_SUPPORTED;
1123 /* create send-to-vport group */
1124 MLX5_SET(create_flow_group_in, flow_group_in, match_criteria_enable,
1125 MLX5_MATCH_MISC_PARAMETERS);
1127 match_criteria = MLX5_ADDR_OF(create_flow_group_in, flow_group_in, match_criteria);
1129 MLX5_SET_TO_ONES(fte_match_param, match_criteria, misc_parameters.source_sqn);
1130 MLX5_SET_TO_ONES(fte_match_param, match_criteria, misc_parameters.source_port);
1132 ix = nvports * MAX_SQ_NVPORTS + MAX_PF_SQ;
1133 MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, 0);
1134 MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index, ix - 1);
1136 g = mlx5_create_flow_group(fdb, flow_group_in);
1139 esw_warn(dev, "Failed to create send-to-vport flow group err(%d)\n", err);
1140 goto send_vport_err;
1142 esw->fdb_table.offloads.send_to_vport_grp = g;
1144 /* create peer esw miss group */
1145 memset(flow_group_in, 0, inlen);
1147 esw_set_flow_group_source_port(esw, flow_group_in);
1149 if (!mlx5_eswitch_vport_match_metadata_enabled(esw)) {
1150 match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1154 MLX5_SET_TO_ONES(fte_match_param, match_criteria,
1155 misc_parameters.source_eswitch_owner_vhca_id);
1157 MLX5_SET(create_flow_group_in, flow_group_in,
1158 source_eswitch_owner_vhca_id_valid, 1);
1161 MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, ix);
1162 MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index,
1163 ix + esw->total_vports - 1);
1164 ix += esw->total_vports;
1166 g = mlx5_create_flow_group(fdb, flow_group_in);
1169 esw_warn(dev, "Failed to create peer miss flow group err(%d)\n", err);
1172 esw->fdb_table.offloads.peer_miss_grp = g;
1174 /* create miss group */
1175 memset(flow_group_in, 0, inlen);
1176 MLX5_SET(create_flow_group_in, flow_group_in, match_criteria_enable,
1177 MLX5_MATCH_OUTER_HEADERS);
1178 match_criteria = MLX5_ADDR_OF(create_flow_group_in, flow_group_in,
1180 dmac = MLX5_ADDR_OF(fte_match_param, match_criteria,
1181 outer_headers.dmac_47_16);
1184 MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, ix);
1185 MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index,
1186 ix + MLX5_ESW_MISS_FLOWS);
1188 g = mlx5_create_flow_group(fdb, flow_group_in);
1191 esw_warn(dev, "Failed to create miss flow group err(%d)\n", err);
1194 esw->fdb_table.offloads.miss_grp = g;
1196 err = esw_add_fdb_miss_rule(esw);
1200 esw->nvports = nvports;
1201 kvfree(flow_group_in);
1205 mlx5_destroy_flow_group(esw->fdb_table.offloads.miss_grp);
1207 mlx5_destroy_flow_group(esw->fdb_table.offloads.peer_miss_grp);
1209 mlx5_destroy_flow_group(esw->fdb_table.offloads.send_to_vport_grp);
1211 esw_destroy_offloads_fast_fdb_tables(esw);
1212 mlx5_destroy_flow_table(esw->fdb_table.offloads.slow_fdb);
1215 kvfree(flow_group_in);
1219 static void esw_destroy_offloads_fdb_tables(struct mlx5_eswitch *esw)
1221 if (!esw->fdb_table.offloads.slow_fdb)
1224 esw_debug(esw->dev, "Destroy offloads FDB Tables\n");
1225 mlx5_del_flow_rules(esw->fdb_table.offloads.miss_rule_multi);
1226 mlx5_del_flow_rules(esw->fdb_table.offloads.miss_rule_uni);
1227 mlx5_destroy_flow_group(esw->fdb_table.offloads.send_to_vport_grp);
1228 mlx5_destroy_flow_group(esw->fdb_table.offloads.peer_miss_grp);
1229 mlx5_destroy_flow_group(esw->fdb_table.offloads.miss_grp);
1231 mlx5_destroy_flow_table(esw->fdb_table.offloads.slow_fdb);
1232 esw_destroy_offloads_fast_fdb_tables(esw);
1235 static int esw_create_offloads_table(struct mlx5_eswitch *esw, int nvports)
1237 struct mlx5_flow_table_attr ft_attr = {};
1238 struct mlx5_core_dev *dev = esw->dev;
1239 struct mlx5_flow_table *ft_offloads;
1240 struct mlx5_flow_namespace *ns;
1243 ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_OFFLOADS);
1245 esw_warn(esw->dev, "Failed to get offloads flow namespace\n");
1249 ft_attr.max_fte = nvports + MLX5_ESW_MISS_FLOWS;
1251 ft_offloads = mlx5_create_flow_table(ns, &ft_attr);
1252 if (IS_ERR(ft_offloads)) {
1253 err = PTR_ERR(ft_offloads);
1254 esw_warn(esw->dev, "Failed to create offloads table, err %d\n", err);
1258 esw->offloads.ft_offloads = ft_offloads;
1262 static void esw_destroy_offloads_table(struct mlx5_eswitch *esw)
1264 struct mlx5_esw_offload *offloads = &esw->offloads;
1266 mlx5_destroy_flow_table(offloads->ft_offloads);
1269 static int esw_create_vport_rx_group(struct mlx5_eswitch *esw, int nvports)
1271 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1272 struct mlx5_flow_group *g;
1276 nvports = nvports + MLX5_ESW_MISS_FLOWS;
1277 flow_group_in = kvzalloc(inlen, GFP_KERNEL);
1281 /* create vport rx group */
1282 esw_set_flow_group_source_port(esw, flow_group_in);
1284 MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, 0);
1285 MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index, nvports - 1);
1287 g = mlx5_create_flow_group(esw->offloads.ft_offloads, flow_group_in);
1291 mlx5_core_warn(esw->dev, "Failed to create vport rx group err %d\n", err);
1295 esw->offloads.vport_rx_group = g;
1297 kvfree(flow_group_in);
1301 static void esw_destroy_vport_rx_group(struct mlx5_eswitch *esw)
1303 mlx5_destroy_flow_group(esw->offloads.vport_rx_group);
1306 struct mlx5_flow_handle *
1307 mlx5_eswitch_create_vport_rx_rule(struct mlx5_eswitch *esw, u16 vport,
1308 struct mlx5_flow_destination *dest)
1310 struct mlx5_flow_act flow_act = {0};
1311 struct mlx5_flow_handle *flow_rule;
1312 struct mlx5_flow_spec *spec;
1315 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
1317 flow_rule = ERR_PTR(-ENOMEM);
1321 if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
1322 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value, misc_parameters_2);
1323 MLX5_SET(fte_match_set_misc2, misc, metadata_reg_c_0,
1324 mlx5_eswitch_get_vport_metadata_for_match(esw, vport));
1326 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, misc_parameters_2);
1327 MLX5_SET_TO_ONES(fte_match_set_misc2, misc, metadata_reg_c_0);
1329 spec->match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS_2;
1331 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value, misc_parameters);
1332 MLX5_SET(fte_match_set_misc, misc, source_port, vport);
1334 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, misc_parameters);
1335 MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
1337 spec->match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS;
1340 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1341 flow_rule = mlx5_add_flow_rules(esw->offloads.ft_offloads, spec,
1342 &flow_act, dest, 1);
1343 if (IS_ERR(flow_rule)) {
1344 esw_warn(esw->dev, "fs offloads: Failed to add vport rx rule err %ld\n", PTR_ERR(flow_rule));
1353 static int esw_offloads_start(struct mlx5_eswitch *esw,
1354 struct netlink_ext_ack *extack)
1358 if (esw->mode != MLX5_ESWITCH_LEGACY &&
1359 !mlx5_core_is_ecpf_esw_manager(esw->dev)) {
1360 NL_SET_ERR_MSG_MOD(extack,
1361 "Can't set offloads mode, SRIOV legacy not enabled");
1365 mlx5_eswitch_disable(esw);
1366 mlx5_eswitch_update_num_of_vfs(esw, esw->dev->priv.sriov.num_vfs);
1367 err = mlx5_eswitch_enable(esw, MLX5_ESWITCH_OFFLOADS);
1369 NL_SET_ERR_MSG_MOD(extack,
1370 "Failed setting eswitch to offloads");
1371 err1 = mlx5_eswitch_enable(esw, MLX5_ESWITCH_LEGACY);
1373 NL_SET_ERR_MSG_MOD(extack,
1374 "Failed setting eswitch back to legacy");
1377 if (esw->offloads.inline_mode == MLX5_INLINE_MODE_NONE) {
1378 if (mlx5_eswitch_inline_mode_get(esw,
1379 &esw->offloads.inline_mode)) {
1380 esw->offloads.inline_mode = MLX5_INLINE_MODE_L2;
1381 NL_SET_ERR_MSG_MOD(extack,
1382 "Inline mode is different between vports");
1388 void esw_offloads_cleanup_reps(struct mlx5_eswitch *esw)
1390 kfree(esw->offloads.vport_reps);
1393 int esw_offloads_init_reps(struct mlx5_eswitch *esw)
1395 int total_vports = esw->total_vports;
1396 struct mlx5_eswitch_rep *rep;
1400 esw->offloads.vport_reps = kcalloc(total_vports,
1401 sizeof(struct mlx5_eswitch_rep),
1403 if (!esw->offloads.vport_reps)
1406 mlx5_esw_for_all_reps(esw, vport_index, rep) {
1407 rep->vport = mlx5_eswitch_index_to_vport_num(esw, vport_index);
1408 rep->vport_index = vport_index;
1410 for (rep_type = 0; rep_type < NUM_REP_TYPES; rep_type++)
1411 atomic_set(&rep->rep_data[rep_type].state,
1418 static void __esw_offloads_unload_rep(struct mlx5_eswitch *esw,
1419 struct mlx5_eswitch_rep *rep, u8 rep_type)
1421 if (atomic_cmpxchg(&rep->rep_data[rep_type].state,
1422 REP_LOADED, REP_REGISTERED) == REP_LOADED)
1423 esw->offloads.rep_ops[rep_type]->unload(rep);
1426 static void __unload_reps_special_vport(struct mlx5_eswitch *esw, u8 rep_type)
1428 struct mlx5_eswitch_rep *rep;
1430 if (mlx5_ecpf_vport_exists(esw->dev)) {
1431 rep = mlx5_eswitch_get_rep(esw, MLX5_VPORT_ECPF);
1432 __esw_offloads_unload_rep(esw, rep, rep_type);
1435 if (mlx5_core_is_ecpf_esw_manager(esw->dev)) {
1436 rep = mlx5_eswitch_get_rep(esw, MLX5_VPORT_PF);
1437 __esw_offloads_unload_rep(esw, rep, rep_type);
1440 rep = mlx5_eswitch_get_rep(esw, MLX5_VPORT_UPLINK);
1441 __esw_offloads_unload_rep(esw, rep, rep_type);
1444 static void __unload_reps_vf_vport(struct mlx5_eswitch *esw, int nvports,
1447 struct mlx5_eswitch_rep *rep;
1450 mlx5_esw_for_each_vf_rep_reverse(esw, i, rep, nvports)
1451 __esw_offloads_unload_rep(esw, rep, rep_type);
1454 static void esw_offloads_unload_vf_reps(struct mlx5_eswitch *esw, int nvports)
1456 u8 rep_type = NUM_REP_TYPES;
1458 while (rep_type-- > 0)
1459 __unload_reps_vf_vport(esw, nvports, rep_type);
1462 static void __unload_reps_all_vport(struct mlx5_eswitch *esw, u8 rep_type)
1464 __unload_reps_vf_vport(esw, esw->esw_funcs.num_vfs, rep_type);
1466 /* Special vports must be the last to unload. */
1467 __unload_reps_special_vport(esw, rep_type);
1470 static void esw_offloads_unload_all_reps(struct mlx5_eswitch *esw)
1472 u8 rep_type = NUM_REP_TYPES;
1474 while (rep_type-- > 0)
1475 __unload_reps_all_vport(esw, rep_type);
1478 static int __esw_offloads_load_rep(struct mlx5_eswitch *esw,
1479 struct mlx5_eswitch_rep *rep, u8 rep_type)
1483 if (atomic_cmpxchg(&rep->rep_data[rep_type].state,
1484 REP_REGISTERED, REP_LOADED) == REP_REGISTERED) {
1485 err = esw->offloads.rep_ops[rep_type]->load(esw->dev, rep);
1487 atomic_set(&rep->rep_data[rep_type].state,
1494 static int __load_reps_special_vport(struct mlx5_eswitch *esw, u8 rep_type)
1496 struct mlx5_eswitch_rep *rep;
1499 rep = mlx5_eswitch_get_rep(esw, MLX5_VPORT_UPLINK);
1500 err = __esw_offloads_load_rep(esw, rep, rep_type);
1504 if (mlx5_core_is_ecpf_esw_manager(esw->dev)) {
1505 rep = mlx5_eswitch_get_rep(esw, MLX5_VPORT_PF);
1506 err = __esw_offloads_load_rep(esw, rep, rep_type);
1511 if (mlx5_ecpf_vport_exists(esw->dev)) {
1512 rep = mlx5_eswitch_get_rep(esw, MLX5_VPORT_ECPF);
1513 err = __esw_offloads_load_rep(esw, rep, rep_type);
1521 if (mlx5_core_is_ecpf_esw_manager(esw->dev)) {
1522 rep = mlx5_eswitch_get_rep(esw, MLX5_VPORT_PF);
1523 __esw_offloads_unload_rep(esw, rep, rep_type);
1527 rep = mlx5_eswitch_get_rep(esw, MLX5_VPORT_UPLINK);
1528 __esw_offloads_unload_rep(esw, rep, rep_type);
1532 static int __load_reps_vf_vport(struct mlx5_eswitch *esw, int nvports,
1535 struct mlx5_eswitch_rep *rep;
1538 mlx5_esw_for_each_vf_rep(esw, i, rep, nvports) {
1539 err = __esw_offloads_load_rep(esw, rep, rep_type);
1547 __unload_reps_vf_vport(esw, --i, rep_type);
1551 static int __load_reps_all_vport(struct mlx5_eswitch *esw, u8 rep_type)
1555 /* Special vports must be loaded first, uplink rep creates mdev resource. */
1556 err = __load_reps_special_vport(esw, rep_type);
1560 err = __load_reps_vf_vport(esw, esw->esw_funcs.num_vfs, rep_type);
1567 __unload_reps_special_vport(esw, rep_type);
1571 static int esw_offloads_load_vf_reps(struct mlx5_eswitch *esw, int nvports)
1576 for (rep_type = 0; rep_type < NUM_REP_TYPES; rep_type++) {
1577 err = __load_reps_vf_vport(esw, nvports, rep_type);
1585 while (rep_type-- > 0)
1586 __unload_reps_vf_vport(esw, nvports, rep_type);
1590 static int esw_offloads_load_all_reps(struct mlx5_eswitch *esw)
1595 for (rep_type = 0; rep_type < NUM_REP_TYPES; rep_type++) {
1596 err = __load_reps_all_vport(esw, rep_type);
1604 while (rep_type-- > 0)
1605 __unload_reps_all_vport(esw, rep_type);
1609 #define ESW_OFFLOADS_DEVCOM_PAIR (0)
1610 #define ESW_OFFLOADS_DEVCOM_UNPAIR (1)
1612 static int mlx5_esw_offloads_pair(struct mlx5_eswitch *esw,
1613 struct mlx5_eswitch *peer_esw)
1617 err = esw_add_fdb_peer_miss_rules(esw, peer_esw->dev);
1624 static void mlx5_esw_offloads_unpair(struct mlx5_eswitch *esw)
1626 mlx5e_tc_clean_fdb_peer_flows(esw);
1627 esw_del_fdb_peer_miss_rules(esw);
1630 static int mlx5_esw_offloads_devcom_event(int event,
1634 struct mlx5_eswitch *esw = my_data;
1635 struct mlx5_eswitch *peer_esw = event_data;
1636 struct mlx5_devcom *devcom = esw->dev->priv.devcom;
1640 case ESW_OFFLOADS_DEVCOM_PAIR:
1641 if (mlx5_eswitch_vport_match_metadata_enabled(esw) !=
1642 mlx5_eswitch_vport_match_metadata_enabled(peer_esw))
1645 err = mlx5_esw_offloads_pair(esw, peer_esw);
1649 err = mlx5_esw_offloads_pair(peer_esw, esw);
1653 mlx5_devcom_set_paired(devcom, MLX5_DEVCOM_ESW_OFFLOADS, true);
1656 case ESW_OFFLOADS_DEVCOM_UNPAIR:
1657 if (!mlx5_devcom_is_paired(devcom, MLX5_DEVCOM_ESW_OFFLOADS))
1660 mlx5_devcom_set_paired(devcom, MLX5_DEVCOM_ESW_OFFLOADS, false);
1661 mlx5_esw_offloads_unpair(peer_esw);
1662 mlx5_esw_offloads_unpair(esw);
1669 mlx5_esw_offloads_unpair(esw);
1672 mlx5_core_err(esw->dev, "esw offloads devcom event failure, event %u err %d",
1677 static void esw_offloads_devcom_init(struct mlx5_eswitch *esw)
1679 struct mlx5_devcom *devcom = esw->dev->priv.devcom;
1681 INIT_LIST_HEAD(&esw->offloads.peer_flows);
1682 mutex_init(&esw->offloads.peer_mutex);
1684 if (!MLX5_CAP_ESW(esw->dev, merged_eswitch))
1687 mlx5_devcom_register_component(devcom,
1688 MLX5_DEVCOM_ESW_OFFLOADS,
1689 mlx5_esw_offloads_devcom_event,
1692 mlx5_devcom_send_event(devcom,
1693 MLX5_DEVCOM_ESW_OFFLOADS,
1694 ESW_OFFLOADS_DEVCOM_PAIR, esw);
1697 static void esw_offloads_devcom_cleanup(struct mlx5_eswitch *esw)
1699 struct mlx5_devcom *devcom = esw->dev->priv.devcom;
1701 if (!MLX5_CAP_ESW(esw->dev, merged_eswitch))
1704 mlx5_devcom_send_event(devcom, MLX5_DEVCOM_ESW_OFFLOADS,
1705 ESW_OFFLOADS_DEVCOM_UNPAIR, esw);
1707 mlx5_devcom_unregister_component(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
1710 static int esw_vport_ingress_prio_tag_config(struct mlx5_eswitch *esw,
1711 struct mlx5_vport *vport)
1713 struct mlx5_flow_act flow_act = {0};
1714 struct mlx5_flow_spec *spec;
1717 /* For prio tag mode, there is only 1 FTEs:
1718 * 1) Untagged packets - push prio tag VLAN and modify metadata if
1720 * Unmatched traffic is allowed by default
1723 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
1729 /* Untagged packets - push prio tag VLAN, allow */
1730 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria, outer_headers.cvlan_tag);
1731 MLX5_SET(fte_match_param, spec->match_value, outer_headers.cvlan_tag, 0);
1732 spec->match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
1733 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1734 MLX5_FLOW_CONTEXT_ACTION_ALLOW;
1735 flow_act.vlan[0].ethtype = ETH_P_8021Q;
1736 flow_act.vlan[0].vid = 0;
1737 flow_act.vlan[0].prio = 0;
1739 if (vport->ingress.modify_metadata_rule) {
1740 flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_MOD_HDR;
1741 flow_act.modify_id = vport->ingress.modify_metadata_id;
1744 vport->ingress.allow_rule =
1745 mlx5_add_flow_rules(vport->ingress.acl, spec,
1746 &flow_act, NULL, 0);
1747 if (IS_ERR(vport->ingress.allow_rule)) {
1748 err = PTR_ERR(vport->ingress.allow_rule);
1750 "vport[%d] configure ingress untagged allow rule, err(%d)\n",
1752 vport->ingress.allow_rule = NULL;
1760 esw_vport_cleanup_ingress_rules(esw, vport);
1764 static int esw_vport_add_ingress_acl_modify_metadata(struct mlx5_eswitch *esw,
1765 struct mlx5_vport *vport)
1767 u8 action[MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto)] = {};
1768 static const struct mlx5_flow_spec spec = {};
1769 struct mlx5_flow_act flow_act = {};
1772 MLX5_SET(set_action_in, action, action_type, MLX5_ACTION_TYPE_SET);
1773 MLX5_SET(set_action_in, action, field, MLX5_ACTION_IN_FIELD_METADATA_REG_C_0);
1774 MLX5_SET(set_action_in, action, data,
1775 mlx5_eswitch_get_vport_metadata_for_match(esw, vport->vport));
1777 err = mlx5_modify_header_alloc(esw->dev, MLX5_FLOW_NAMESPACE_ESW_INGRESS,
1778 1, action, &vport->ingress.modify_metadata_id);
1781 "failed to alloc modify header for vport %d ingress acl (%d)\n",
1786 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_MOD_HDR | MLX5_FLOW_CONTEXT_ACTION_ALLOW;
1787 flow_act.modify_id = vport->ingress.modify_metadata_id;
1788 vport->ingress.modify_metadata_rule = mlx5_add_flow_rules(vport->ingress.acl,
1789 &spec, &flow_act, NULL, 0);
1790 if (IS_ERR(vport->ingress.modify_metadata_rule)) {
1791 err = PTR_ERR(vport->ingress.modify_metadata_rule);
1793 "failed to add setting metadata rule for vport %d ingress acl, err(%d)\n",
1795 vport->ingress.modify_metadata_rule = NULL;
1801 mlx5_modify_header_dealloc(esw->dev, vport->ingress.modify_metadata_id);
1805 void esw_vport_del_ingress_acl_modify_metadata(struct mlx5_eswitch *esw,
1806 struct mlx5_vport *vport)
1808 if (vport->ingress.modify_metadata_rule) {
1809 mlx5_del_flow_rules(vport->ingress.modify_metadata_rule);
1810 mlx5_modify_header_dealloc(esw->dev, vport->ingress.modify_metadata_id);
1812 vport->ingress.modify_metadata_rule = NULL;
1816 static int esw_vport_egress_prio_tag_config(struct mlx5_eswitch *esw,
1817 struct mlx5_vport *vport)
1819 struct mlx5_flow_act flow_act = {0};
1820 struct mlx5_flow_spec *spec;
1823 if (!MLX5_CAP_GEN(esw->dev, prio_tag_required))
1826 /* For prio tag mode, there is only 1 FTEs:
1827 * 1) prio tag packets - pop the prio tag VLAN, allow
1828 * Unmatched traffic is allowed by default
1831 esw_vport_cleanup_egress_rules(esw, vport);
1833 err = esw_vport_enable_egress_acl(esw, vport);
1835 mlx5_core_warn(esw->dev,
1836 "failed to enable egress acl (%d) on vport[%d]\n",
1842 "vport[%d] configure prio tag egress rules\n", vport->vport);
1844 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
1850 /* prio tag vlan rule - pop it so VF receives untagged packets */
1851 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria, outer_headers.cvlan_tag);
1852 MLX5_SET_TO_ONES(fte_match_param, spec->match_value, outer_headers.cvlan_tag);
1853 MLX5_SET_TO_ONES(fte_match_param, spec->match_criteria, outer_headers.first_vid);
1854 MLX5_SET(fte_match_param, spec->match_value, outer_headers.first_vid, 0);
1856 spec->match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
1857 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1858 MLX5_FLOW_CONTEXT_ACTION_ALLOW;
1859 vport->egress.allowed_vlan =
1860 mlx5_add_flow_rules(vport->egress.acl, spec,
1861 &flow_act, NULL, 0);
1862 if (IS_ERR(vport->egress.allowed_vlan)) {
1863 err = PTR_ERR(vport->egress.allowed_vlan);
1865 "vport[%d] configure egress pop prio tag vlan rule failed, err(%d)\n",
1867 vport->egress.allowed_vlan = NULL;
1875 esw_vport_cleanup_egress_rules(esw, vport);
1879 static int esw_vport_ingress_common_config(struct mlx5_eswitch *esw,
1880 struct mlx5_vport *vport)
1884 if (!mlx5_eswitch_vport_match_metadata_enabled(esw) &&
1885 !MLX5_CAP_GEN(esw->dev, prio_tag_required))
1888 esw_vport_cleanup_ingress_rules(esw, vport);
1890 err = esw_vport_enable_ingress_acl(esw, vport);
1893 "failed to enable ingress acl (%d) on vport[%d]\n",
1899 "vport[%d] configure ingress rules\n", vport->vport);
1901 if (mlx5_eswitch_vport_match_metadata_enabled(esw)) {
1902 err = esw_vport_add_ingress_acl_modify_metadata(esw, vport);
1907 if (MLX5_CAP_GEN(esw->dev, prio_tag_required) &&
1908 mlx5_eswitch_is_vf_vport(esw, vport->vport)) {
1909 err = esw_vport_ingress_prio_tag_config(esw, vport);
1916 esw_vport_disable_ingress_acl(esw, vport);
1921 esw_check_vport_match_metadata_supported(const struct mlx5_eswitch *esw)
1923 if (!MLX5_CAP_ESW(esw->dev, esw_uplink_ingress_acl))
1926 if (!(MLX5_CAP_ESW_FLOWTABLE(esw->dev, fdb_to_vport_reg_c_id) &
1927 MLX5_FDB_TO_VPORT_REG_C_0))
1930 if (!MLX5_CAP_ESW_FLOWTABLE(esw->dev, flow_source))
1933 if (mlx5_core_is_ecpf_esw_manager(esw->dev) ||
1934 mlx5_ecpf_vport_exists(esw->dev))
1940 static int esw_create_offloads_acl_tables(struct mlx5_eswitch *esw)
1942 struct mlx5_vport *vport;
1946 if (esw_check_vport_match_metadata_supported(esw))
1947 esw->flags |= MLX5_ESWITCH_VPORT_MATCH_METADATA;
1949 mlx5_esw_for_all_vports(esw, i, vport) {
1950 err = esw_vport_ingress_common_config(esw, vport);
1954 if (mlx5_eswitch_is_vf_vport(esw, vport->vport)) {
1955 err = esw_vport_egress_prio_tag_config(esw, vport);
1961 if (mlx5_eswitch_vport_match_metadata_enabled(esw))
1962 esw_info(esw->dev, "Use metadata reg_c as source vport to match\n");
1967 esw_vport_disable_ingress_acl(esw, vport);
1969 for (j = MLX5_VPORT_PF; j < i; j++) {
1970 vport = &esw->vports[j];
1971 esw_vport_disable_egress_acl(esw, vport);
1972 esw_vport_disable_ingress_acl(esw, vport);
1978 static void esw_destroy_offloads_acl_tables(struct mlx5_eswitch *esw)
1980 struct mlx5_vport *vport;
1983 mlx5_esw_for_all_vports(esw, i, vport) {
1984 esw_vport_disable_egress_acl(esw, vport);
1985 esw_vport_disable_ingress_acl(esw, vport);
1988 esw->flags &= ~MLX5_ESWITCH_VPORT_MATCH_METADATA;
1991 static int esw_offloads_steering_init(struct mlx5_eswitch *esw)
1993 int num_vfs = esw->esw_funcs.num_vfs;
1997 if (mlx5_core_is_ecpf_esw_manager(esw->dev))
1998 total_vports = esw->total_vports;
2000 total_vports = num_vfs + MLX5_SPECIAL_VPORTS(esw->dev);
2002 memset(&esw->fdb_table.offloads, 0, sizeof(struct offloads_fdb));
2003 mutex_init(&esw->fdb_table.offloads.fdb_prio_lock);
2005 err = esw_create_offloads_acl_tables(esw);
2009 err = esw_create_offloads_fdb_tables(esw, total_vports);
2011 goto create_fdb_err;
2013 err = esw_create_offloads_table(esw, total_vports);
2017 err = esw_create_vport_rx_group(esw, total_vports);
2024 esw_destroy_offloads_table(esw);
2027 esw_destroy_offloads_fdb_tables(esw);
2030 esw_destroy_offloads_acl_tables(esw);
2035 static void esw_offloads_steering_cleanup(struct mlx5_eswitch *esw)
2037 esw_destroy_vport_rx_group(esw);
2038 esw_destroy_offloads_table(esw);
2039 esw_destroy_offloads_fdb_tables(esw);
2040 esw_destroy_offloads_acl_tables(esw);
2044 esw_vfs_changed_event_handler(struct mlx5_eswitch *esw, const u32 *out)
2046 bool host_pf_disabled;
2049 new_num_vfs = MLX5_GET(query_esw_functions_out, out,
2050 host_params_context.host_num_of_vfs);
2051 host_pf_disabled = MLX5_GET(query_esw_functions_out, out,
2052 host_params_context.host_pf_disabled);
2054 if (new_num_vfs == esw->esw_funcs.num_vfs || host_pf_disabled)
2057 /* Number of VFs can only change from "0 to x" or "x to 0". */
2058 if (esw->esw_funcs.num_vfs > 0) {
2059 esw_offloads_unload_vf_reps(esw, esw->esw_funcs.num_vfs);
2063 err = esw_offloads_load_vf_reps(esw, new_num_vfs);
2067 esw->esw_funcs.num_vfs = new_num_vfs;
2070 static void esw_functions_changed_event_handler(struct work_struct *work)
2072 struct mlx5_host_work *host_work;
2073 struct mlx5_eswitch *esw;
2076 host_work = container_of(work, struct mlx5_host_work, work);
2077 esw = host_work->esw;
2079 out = mlx5_esw_query_functions(esw->dev);
2083 esw_vfs_changed_event_handler(esw, out);
2089 int mlx5_esw_funcs_changed_handler(struct notifier_block *nb, unsigned long type, void *data)
2091 struct mlx5_esw_functions *esw_funcs;
2092 struct mlx5_host_work *host_work;
2093 struct mlx5_eswitch *esw;
2095 host_work = kzalloc(sizeof(*host_work), GFP_ATOMIC);
2099 esw_funcs = mlx5_nb_cof(nb, struct mlx5_esw_functions, nb);
2100 esw = container_of(esw_funcs, struct mlx5_eswitch, esw_funcs);
2102 host_work->esw = esw;
2104 INIT_WORK(&host_work->work, esw_functions_changed_event_handler);
2105 queue_work(esw->work_queue, &host_work->work);
2110 int esw_offloads_enable(struct mlx5_eswitch *esw)
2114 if (MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, reformat) &&
2115 MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, decap))
2116 esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_BASIC;
2118 esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_NONE;
2120 err = esw_offloads_steering_init(esw);
2124 err = esw_set_passing_vport_metadata(esw, true);
2126 goto err_vport_metadata;
2128 mlx5_eswitch_enable_pf_vf_vports(esw, MLX5_VPORT_UC_ADDR_CHANGE);
2130 err = esw_offloads_load_all_reps(esw);
2134 esw_offloads_devcom_init(esw);
2135 mutex_init(&esw->offloads.termtbl_mutex);
2137 mlx5_rdma_enable_roce(esw->dev);
2142 mlx5_eswitch_disable_pf_vf_vports(esw);
2143 esw_set_passing_vport_metadata(esw, false);
2145 esw_offloads_steering_cleanup(esw);
2149 static int esw_offloads_stop(struct mlx5_eswitch *esw,
2150 struct netlink_ext_ack *extack)
2154 mlx5_eswitch_disable(esw);
2155 err = mlx5_eswitch_enable(esw, MLX5_ESWITCH_LEGACY);
2157 NL_SET_ERR_MSG_MOD(extack, "Failed setting eswitch to legacy");
2158 err1 = mlx5_eswitch_enable(esw, MLX5_ESWITCH_OFFLOADS);
2160 NL_SET_ERR_MSG_MOD(extack,
2161 "Failed setting eswitch back to offloads");
2168 void esw_offloads_disable(struct mlx5_eswitch *esw)
2170 mlx5_rdma_disable_roce(esw->dev);
2171 esw_offloads_devcom_cleanup(esw);
2172 esw_offloads_unload_all_reps(esw);
2173 mlx5_eswitch_disable_pf_vf_vports(esw);
2174 esw_set_passing_vport_metadata(esw, false);
2175 esw_offloads_steering_cleanup(esw);
2176 esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_NONE;
2179 static int esw_mode_from_devlink(u16 mode, u16 *mlx5_mode)
2182 case DEVLINK_ESWITCH_MODE_LEGACY:
2183 *mlx5_mode = MLX5_ESWITCH_LEGACY;
2185 case DEVLINK_ESWITCH_MODE_SWITCHDEV:
2186 *mlx5_mode = MLX5_ESWITCH_OFFLOADS;
2195 static int esw_mode_to_devlink(u16 mlx5_mode, u16 *mode)
2197 switch (mlx5_mode) {
2198 case MLX5_ESWITCH_LEGACY:
2199 *mode = DEVLINK_ESWITCH_MODE_LEGACY;
2201 case MLX5_ESWITCH_OFFLOADS:
2202 *mode = DEVLINK_ESWITCH_MODE_SWITCHDEV;
2211 static int esw_inline_mode_from_devlink(u8 mode, u8 *mlx5_mode)
2214 case DEVLINK_ESWITCH_INLINE_MODE_NONE:
2215 *mlx5_mode = MLX5_INLINE_MODE_NONE;
2217 case DEVLINK_ESWITCH_INLINE_MODE_LINK:
2218 *mlx5_mode = MLX5_INLINE_MODE_L2;
2220 case DEVLINK_ESWITCH_INLINE_MODE_NETWORK:
2221 *mlx5_mode = MLX5_INLINE_MODE_IP;
2223 case DEVLINK_ESWITCH_INLINE_MODE_TRANSPORT:
2224 *mlx5_mode = MLX5_INLINE_MODE_TCP_UDP;
2233 static int esw_inline_mode_to_devlink(u8 mlx5_mode, u8 *mode)
2235 switch (mlx5_mode) {
2236 case MLX5_INLINE_MODE_NONE:
2237 *mode = DEVLINK_ESWITCH_INLINE_MODE_NONE;
2239 case MLX5_INLINE_MODE_L2:
2240 *mode = DEVLINK_ESWITCH_INLINE_MODE_LINK;
2242 case MLX5_INLINE_MODE_IP:
2243 *mode = DEVLINK_ESWITCH_INLINE_MODE_NETWORK;
2245 case MLX5_INLINE_MODE_TCP_UDP:
2246 *mode = DEVLINK_ESWITCH_INLINE_MODE_TRANSPORT;
2255 static int mlx5_devlink_eswitch_check(struct devlink *devlink)
2257 struct mlx5_core_dev *dev = devlink_priv(devlink);
2259 if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
2262 if(!MLX5_ESWITCH_MANAGER(dev))
2265 if (dev->priv.eswitch->mode == MLX5_ESWITCH_NONE &&
2266 !mlx5_core_is_ecpf_esw_manager(dev))
2272 int mlx5_devlink_eswitch_mode_set(struct devlink *devlink, u16 mode,
2273 struct netlink_ext_ack *extack)
2275 struct mlx5_core_dev *dev = devlink_priv(devlink);
2276 u16 cur_mlx5_mode, mlx5_mode = 0;
2279 err = mlx5_devlink_eswitch_check(devlink);
2283 cur_mlx5_mode = dev->priv.eswitch->mode;
2285 if (esw_mode_from_devlink(mode, &mlx5_mode))
2288 if (cur_mlx5_mode == mlx5_mode)
2291 if (mode == DEVLINK_ESWITCH_MODE_SWITCHDEV)
2292 return esw_offloads_start(dev->priv.eswitch, extack);
2293 else if (mode == DEVLINK_ESWITCH_MODE_LEGACY)
2294 return esw_offloads_stop(dev->priv.eswitch, extack);
2299 int mlx5_devlink_eswitch_mode_get(struct devlink *devlink, u16 *mode)
2301 struct mlx5_core_dev *dev = devlink_priv(devlink);
2304 err = mlx5_devlink_eswitch_check(devlink);
2308 return esw_mode_to_devlink(dev->priv.eswitch->mode, mode);
2311 int mlx5_devlink_eswitch_inline_mode_set(struct devlink *devlink, u8 mode,
2312 struct netlink_ext_ack *extack)
2314 struct mlx5_core_dev *dev = devlink_priv(devlink);
2315 struct mlx5_eswitch *esw = dev->priv.eswitch;
2316 int err, vport, num_vport;
2319 err = mlx5_devlink_eswitch_check(devlink);
2323 switch (MLX5_CAP_ETH(dev, wqe_inline_mode)) {
2324 case MLX5_CAP_INLINE_MODE_NOT_REQUIRED:
2325 if (mode == DEVLINK_ESWITCH_INLINE_MODE_NONE)
2328 case MLX5_CAP_INLINE_MODE_L2:
2329 NL_SET_ERR_MSG_MOD(extack, "Inline mode can't be set");
2331 case MLX5_CAP_INLINE_MODE_VPORT_CONTEXT:
2335 if (atomic64_read(&esw->offloads.num_flows) > 0) {
2336 NL_SET_ERR_MSG_MOD(extack,
2337 "Can't set inline mode when flows are configured");
2341 err = esw_inline_mode_from_devlink(mode, &mlx5_mode);
2345 mlx5_esw_for_each_host_func_vport(esw, vport, esw->esw_funcs.num_vfs) {
2346 err = mlx5_modify_nic_vport_min_inline(dev, vport, mlx5_mode);
2348 NL_SET_ERR_MSG_MOD(extack,
2349 "Failed to set min inline on vport");
2350 goto revert_inline_mode;
2354 esw->offloads.inline_mode = mlx5_mode;
2358 num_vport = --vport;
2359 mlx5_esw_for_each_host_func_vport_reverse(esw, vport, num_vport)
2360 mlx5_modify_nic_vport_min_inline(dev,
2362 esw->offloads.inline_mode);
2367 int mlx5_devlink_eswitch_inline_mode_get(struct devlink *devlink, u8 *mode)
2369 struct mlx5_core_dev *dev = devlink_priv(devlink);
2370 struct mlx5_eswitch *esw = dev->priv.eswitch;
2373 err = mlx5_devlink_eswitch_check(devlink);
2377 return esw_inline_mode_to_devlink(esw->offloads.inline_mode, mode);
2380 int mlx5_eswitch_inline_mode_get(struct mlx5_eswitch *esw, u8 *mode)
2382 u8 prev_mlx5_mode, mlx5_mode = MLX5_INLINE_MODE_L2;
2383 struct mlx5_core_dev *dev = esw->dev;
2386 if (!MLX5_CAP_GEN(dev, vport_group_manager))
2389 if (esw->mode == MLX5_ESWITCH_NONE)
2392 switch (MLX5_CAP_ETH(dev, wqe_inline_mode)) {
2393 case MLX5_CAP_INLINE_MODE_NOT_REQUIRED:
2394 mlx5_mode = MLX5_INLINE_MODE_NONE;
2396 case MLX5_CAP_INLINE_MODE_L2:
2397 mlx5_mode = MLX5_INLINE_MODE_L2;
2399 case MLX5_CAP_INLINE_MODE_VPORT_CONTEXT:
2404 mlx5_query_nic_vport_min_inline(dev, esw->first_host_vport, &prev_mlx5_mode);
2405 mlx5_esw_for_each_host_func_vport(esw, vport, esw->esw_funcs.num_vfs) {
2406 mlx5_query_nic_vport_min_inline(dev, vport, &mlx5_mode);
2407 if (prev_mlx5_mode != mlx5_mode)
2409 prev_mlx5_mode = mlx5_mode;
2417 int mlx5_devlink_eswitch_encap_mode_set(struct devlink *devlink,
2418 enum devlink_eswitch_encap_mode encap,
2419 struct netlink_ext_ack *extack)
2421 struct mlx5_core_dev *dev = devlink_priv(devlink);
2422 struct mlx5_eswitch *esw = dev->priv.eswitch;
2425 err = mlx5_devlink_eswitch_check(devlink);
2429 if (encap != DEVLINK_ESWITCH_ENCAP_MODE_NONE &&
2430 (!MLX5_CAP_ESW_FLOWTABLE_FDB(dev, reformat) ||
2431 !MLX5_CAP_ESW_FLOWTABLE_FDB(dev, decap)))
2434 if (encap && encap != DEVLINK_ESWITCH_ENCAP_MODE_BASIC)
2437 if (esw->mode == MLX5_ESWITCH_LEGACY) {
2438 esw->offloads.encap = encap;
2442 if (esw->offloads.encap == encap)
2445 if (atomic64_read(&esw->offloads.num_flows) > 0) {
2446 NL_SET_ERR_MSG_MOD(extack,
2447 "Can't set encapsulation when flows are configured");
2451 esw_destroy_offloads_fdb_tables(esw);
2453 esw->offloads.encap = encap;
2455 err = esw_create_offloads_fdb_tables(esw, esw->nvports);
2458 NL_SET_ERR_MSG_MOD(extack,
2459 "Failed re-creating fast FDB table");
2460 esw->offloads.encap = !encap;
2461 (void)esw_create_offloads_fdb_tables(esw, esw->nvports);
2467 int mlx5_devlink_eswitch_encap_mode_get(struct devlink *devlink,
2468 enum devlink_eswitch_encap_mode *encap)
2470 struct mlx5_core_dev *dev = devlink_priv(devlink);
2471 struct mlx5_eswitch *esw = dev->priv.eswitch;
2474 err = mlx5_devlink_eswitch_check(devlink);
2478 *encap = esw->offloads.encap;
2482 void mlx5_eswitch_register_vport_reps(struct mlx5_eswitch *esw,
2483 const struct mlx5_eswitch_rep_ops *ops,
2486 struct mlx5_eswitch_rep_data *rep_data;
2487 struct mlx5_eswitch_rep *rep;
2490 esw->offloads.rep_ops[rep_type] = ops;
2491 mlx5_esw_for_all_reps(esw, i, rep) {
2492 rep_data = &rep->rep_data[rep_type];
2493 atomic_set(&rep_data->state, REP_REGISTERED);
2496 EXPORT_SYMBOL(mlx5_eswitch_register_vport_reps);
2498 void mlx5_eswitch_unregister_vport_reps(struct mlx5_eswitch *esw, u8 rep_type)
2500 struct mlx5_eswitch_rep *rep;
2503 if (esw->mode == MLX5_ESWITCH_OFFLOADS)
2504 __unload_reps_all_vport(esw, rep_type);
2506 mlx5_esw_for_all_reps(esw, i, rep)
2507 atomic_set(&rep->rep_data[rep_type].state, REP_UNREGISTERED);
2509 EXPORT_SYMBOL(mlx5_eswitch_unregister_vport_reps);
2511 void *mlx5_eswitch_get_uplink_priv(struct mlx5_eswitch *esw, u8 rep_type)
2513 struct mlx5_eswitch_rep *rep;
2515 rep = mlx5_eswitch_get_rep(esw, MLX5_VPORT_UPLINK);
2516 return rep->rep_data[rep_type].priv;
2519 void *mlx5_eswitch_get_proto_dev(struct mlx5_eswitch *esw,
2523 struct mlx5_eswitch_rep *rep;
2525 rep = mlx5_eswitch_get_rep(esw, vport);
2527 if (atomic_read(&rep->rep_data[rep_type].state) == REP_LOADED &&
2528 esw->offloads.rep_ops[rep_type]->get_proto_dev)
2529 return esw->offloads.rep_ops[rep_type]->get_proto_dev(rep);
2532 EXPORT_SYMBOL(mlx5_eswitch_get_proto_dev);
2534 void *mlx5_eswitch_uplink_get_proto_dev(struct mlx5_eswitch *esw, u8 rep_type)
2536 return mlx5_eswitch_get_proto_dev(esw, MLX5_VPORT_UPLINK, rep_type);
2538 EXPORT_SYMBOL(mlx5_eswitch_uplink_get_proto_dev);
2540 struct mlx5_eswitch_rep *mlx5_eswitch_vport_rep(struct mlx5_eswitch *esw,
2543 return mlx5_eswitch_get_rep(esw, vport);
2545 EXPORT_SYMBOL(mlx5_eswitch_vport_rep);
2547 bool mlx5_eswitch_is_vf_vport(const struct mlx5_eswitch *esw, u16 vport_num)
2549 return vport_num >= MLX5_VPORT_FIRST_VF &&
2550 vport_num <= esw->dev->priv.sriov.max_vfs;
2553 bool mlx5_eswitch_vport_match_metadata_enabled(const struct mlx5_eswitch *esw)
2555 return !!(esw->flags & MLX5_ESWITCH_VPORT_MATCH_METADATA);
2557 EXPORT_SYMBOL(mlx5_eswitch_vport_match_metadata_enabled);
2559 u32 mlx5_eswitch_get_vport_metadata_for_match(const struct mlx5_eswitch *esw,
2562 return ((MLX5_CAP_GEN(esw->dev, vhca_id) & 0xffff) << 16) | vport_num;
2564 EXPORT_SYMBOL(mlx5_eswitch_get_vport_metadata_for_match);