1 /*
2 * Copyright (c) 2015, Mellanox Technologies. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32
33 #include <linux/etherdevice.h>
34 #include <linux/mlx5/driver.h>
35 #include <linux/mlx5/mlx5_ifc.h>
36 #include <linux/mlx5/vport.h>
37 #include <linux/mlx5/fs.h>
38 #include "esw/acl/lgcy.h"
39 #include "mlx5_core.h"
40 #include "lib/eq.h"
41 #include "eswitch.h"
42 #include "fs_core.h"
43 #include "devlink.h"
44 #include "ecpf.h"
45 #include "en/mod_hdr.h"
46
47 enum {
48 MLX5_ACTION_NONE = 0,
49 MLX5_ACTION_ADD = 1,
50 MLX5_ACTION_DEL = 2,
51 };
52
53 /* Vport UC/MC hash node */
54 struct vport_addr {
55 struct l2addr_node node;
56 u8 action;
57 u16 vport;
58 struct mlx5_flow_handle *flow_rule;
59 bool mpfs; /* UC MAC was added to MPFs */
60 /* A flag indicating that mac was added due to mc promiscuous vport */
61 bool mc_promisc;
62 };
63
64 static void esw_destroy_legacy_fdb_table(struct mlx5_eswitch *esw);
65 static void esw_cleanup_vepa_rules(struct mlx5_eswitch *esw);
66
mlx5_eswitch_check(const struct mlx5_core_dev * dev)67 static int mlx5_eswitch_check(const struct mlx5_core_dev *dev)
68 {
69 if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
70 return -EOPNOTSUPP;
71
72 if (!MLX5_ESWITCH_MANAGER(dev))
73 return -EOPNOTSUPP;
74
75 return 0;
76 }
77
mlx5_devlink_eswitch_get(struct devlink * devlink)78 struct mlx5_eswitch *mlx5_devlink_eswitch_get(struct devlink *devlink)
79 {
80 struct mlx5_core_dev *dev = devlink_priv(devlink);
81 int err;
82
83 err = mlx5_eswitch_check(dev);
84 if (err)
85 return ERR_PTR(err);
86
87 return dev->priv.eswitch;
88 }
89
90 struct mlx5_vport *__must_check
mlx5_eswitch_get_vport(struct mlx5_eswitch * esw,u16 vport_num)91 mlx5_eswitch_get_vport(struct mlx5_eswitch *esw, u16 vport_num)
92 {
93 u16 idx;
94
95 if (!esw || !MLX5_CAP_GEN(esw->dev, vport_group_manager))
96 return ERR_PTR(-EPERM);
97
98 idx = mlx5_eswitch_vport_num_to_index(esw, vport_num);
99
100 if (idx > esw->total_vports - 1) {
101 esw_debug(esw->dev, "vport out of range: num(0x%x), idx(0x%x)\n",
102 vport_num, idx);
103 return ERR_PTR(-EINVAL);
104 }
105
106 return &esw->vports[idx];
107 }
108
arm_vport_context_events_cmd(struct mlx5_core_dev * dev,u16 vport,u32 events_mask)109 static int arm_vport_context_events_cmd(struct mlx5_core_dev *dev, u16 vport,
110 u32 events_mask)
111 {
112 u32 in[MLX5_ST_SZ_DW(modify_nic_vport_context_in)] = {};
113 void *nic_vport_ctx;
114
115 MLX5_SET(modify_nic_vport_context_in, in,
116 opcode, MLX5_CMD_OP_MODIFY_NIC_VPORT_CONTEXT);
117 MLX5_SET(modify_nic_vport_context_in, in, field_select.change_event, 1);
118 MLX5_SET(modify_nic_vport_context_in, in, vport_number, vport);
119 MLX5_SET(modify_nic_vport_context_in, in, other_vport, 1);
120 nic_vport_ctx = MLX5_ADDR_OF(modify_nic_vport_context_in,
121 in, nic_vport_context);
122
123 MLX5_SET(nic_vport_context, nic_vport_ctx, arm_change_event, 1);
124
125 if (events_mask & MLX5_VPORT_UC_ADDR_CHANGE)
126 MLX5_SET(nic_vport_context, nic_vport_ctx,
127 event_on_uc_address_change, 1);
128 if (events_mask & MLX5_VPORT_MC_ADDR_CHANGE)
129 MLX5_SET(nic_vport_context, nic_vport_ctx,
130 event_on_mc_address_change, 1);
131 if (events_mask & MLX5_VPORT_PROMISC_CHANGE)
132 MLX5_SET(nic_vport_context, nic_vport_ctx,
133 event_on_promisc_change, 1);
134
135 return mlx5_cmd_exec_in(dev, modify_nic_vport_context, in);
136 }
137
138 /* E-Switch vport context HW commands */
mlx5_eswitch_modify_esw_vport_context(struct mlx5_core_dev * dev,u16 vport,bool other_vport,void * in)139 int mlx5_eswitch_modify_esw_vport_context(struct mlx5_core_dev *dev, u16 vport,
140 bool other_vport, void *in)
141 {
142 MLX5_SET(modify_esw_vport_context_in, in, opcode,
143 MLX5_CMD_OP_MODIFY_ESW_VPORT_CONTEXT);
144 MLX5_SET(modify_esw_vport_context_in, in, vport_number, vport);
145 MLX5_SET(modify_esw_vport_context_in, in, other_vport, other_vport);
146 return mlx5_cmd_exec_in(dev, modify_esw_vport_context, in);
147 }
148
modify_esw_vport_cvlan(struct mlx5_core_dev * dev,u16 vport,u16 vlan,u8 qos,u8 set_flags)149 static int modify_esw_vport_cvlan(struct mlx5_core_dev *dev, u16 vport,
150 u16 vlan, u8 qos, u8 set_flags)
151 {
152 u32 in[MLX5_ST_SZ_DW(modify_esw_vport_context_in)] = {};
153
154 if (!MLX5_CAP_ESW(dev, vport_cvlan_strip) ||
155 !MLX5_CAP_ESW(dev, vport_cvlan_insert_if_not_exist))
156 return -EOPNOTSUPP;
157
158 esw_debug(dev, "Set Vport[%d] VLAN %d qos %d set=%x\n",
159 vport, vlan, qos, set_flags);
160
161 if (set_flags & SET_VLAN_STRIP)
162 MLX5_SET(modify_esw_vport_context_in, in,
163 esw_vport_context.vport_cvlan_strip, 1);
164
165 if (set_flags & SET_VLAN_INSERT) {
166 /* insert only if no vlan in packet */
167 MLX5_SET(modify_esw_vport_context_in, in,
168 esw_vport_context.vport_cvlan_insert, 1);
169
170 MLX5_SET(modify_esw_vport_context_in, in,
171 esw_vport_context.cvlan_pcp, qos);
172 MLX5_SET(modify_esw_vport_context_in, in,
173 esw_vport_context.cvlan_id, vlan);
174 }
175
176 MLX5_SET(modify_esw_vport_context_in, in,
177 field_select.vport_cvlan_strip, 1);
178 MLX5_SET(modify_esw_vport_context_in, in,
179 field_select.vport_cvlan_insert, 1);
180
181 return mlx5_eswitch_modify_esw_vport_context(dev, vport, true, in);
182 }
183
184 /* E-Switch FDB */
185 static struct mlx5_flow_handle *
__esw_fdb_set_vport_rule(struct mlx5_eswitch * esw,u16 vport,bool rx_rule,u8 mac_c[ETH_ALEN],u8 mac_v[ETH_ALEN])186 __esw_fdb_set_vport_rule(struct mlx5_eswitch *esw, u16 vport, bool rx_rule,
187 u8 mac_c[ETH_ALEN], u8 mac_v[ETH_ALEN])
188 {
189 int match_header = (is_zero_ether_addr(mac_c) ? 0 :
190 MLX5_MATCH_OUTER_HEADERS);
191 struct mlx5_flow_handle *flow_rule = NULL;
192 struct mlx5_flow_act flow_act = {0};
193 struct mlx5_flow_destination dest = {};
194 struct mlx5_flow_spec *spec;
195 void *mv_misc = NULL;
196 void *mc_misc = NULL;
197 u8 *dmac_v = NULL;
198 u8 *dmac_c = NULL;
199
200 if (rx_rule)
201 match_header |= MLX5_MATCH_MISC_PARAMETERS;
202
203 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
204 if (!spec)
205 return NULL;
206
207 dmac_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
208 outer_headers.dmac_47_16);
209 dmac_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
210 outer_headers.dmac_47_16);
211
212 if (match_header & MLX5_MATCH_OUTER_HEADERS) {
213 ether_addr_copy(dmac_v, mac_v);
214 ether_addr_copy(dmac_c, mac_c);
215 }
216
217 if (match_header & MLX5_MATCH_MISC_PARAMETERS) {
218 mv_misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
219 misc_parameters);
220 mc_misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
221 misc_parameters);
222 MLX5_SET(fte_match_set_misc, mv_misc, source_port, MLX5_VPORT_UPLINK);
223 MLX5_SET_TO_ONES(fte_match_set_misc, mc_misc, source_port);
224 }
225
226 dest.type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
227 dest.vport.num = vport;
228
229 esw_debug(esw->dev,
230 "\tFDB add rule dmac_v(%pM) dmac_c(%pM) -> vport(%d)\n",
231 dmac_v, dmac_c, vport);
232 spec->match_criteria_enable = match_header;
233 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
234 flow_rule =
235 mlx5_add_flow_rules(esw->fdb_table.legacy.fdb, spec,
236 &flow_act, &dest, 1);
237 if (IS_ERR(flow_rule)) {
238 esw_warn(esw->dev,
239 "FDB: Failed to add flow rule: dmac_v(%pM) dmac_c(%pM) -> vport(%d), err(%ld)\n",
240 dmac_v, dmac_c, vport, PTR_ERR(flow_rule));
241 flow_rule = NULL;
242 }
243
244 kvfree(spec);
245 return flow_rule;
246 }
247
248 static struct mlx5_flow_handle *
esw_fdb_set_vport_rule(struct mlx5_eswitch * esw,u8 mac[ETH_ALEN],u16 vport)249 esw_fdb_set_vport_rule(struct mlx5_eswitch *esw, u8 mac[ETH_ALEN], u16 vport)
250 {
251 u8 mac_c[ETH_ALEN];
252
253 eth_broadcast_addr(mac_c);
254 return __esw_fdb_set_vport_rule(esw, vport, false, mac_c, mac);
255 }
256
257 static struct mlx5_flow_handle *
esw_fdb_set_vport_allmulti_rule(struct mlx5_eswitch * esw,u16 vport)258 esw_fdb_set_vport_allmulti_rule(struct mlx5_eswitch *esw, u16 vport)
259 {
260 u8 mac_c[ETH_ALEN];
261 u8 mac_v[ETH_ALEN];
262
263 eth_zero_addr(mac_c);
264 eth_zero_addr(mac_v);
265 mac_c[0] = 0x01;
266 mac_v[0] = 0x01;
267 return __esw_fdb_set_vport_rule(esw, vport, false, mac_c, mac_v);
268 }
269
270 static struct mlx5_flow_handle *
esw_fdb_set_vport_promisc_rule(struct mlx5_eswitch * esw,u16 vport)271 esw_fdb_set_vport_promisc_rule(struct mlx5_eswitch *esw, u16 vport)
272 {
273 u8 mac_c[ETH_ALEN];
274 u8 mac_v[ETH_ALEN];
275
276 eth_zero_addr(mac_c);
277 eth_zero_addr(mac_v);
278 return __esw_fdb_set_vport_rule(esw, vport, true, mac_c, mac_v);
279 }
280
281 enum {
282 LEGACY_VEPA_PRIO = 0,
283 LEGACY_FDB_PRIO,
284 };
285
esw_create_legacy_vepa_table(struct mlx5_eswitch * esw)286 static int esw_create_legacy_vepa_table(struct mlx5_eswitch *esw)
287 {
288 struct mlx5_flow_table_attr ft_attr = {};
289 struct mlx5_core_dev *dev = esw->dev;
290 struct mlx5_flow_namespace *root_ns;
291 struct mlx5_flow_table *fdb;
292 int err;
293
294 root_ns = mlx5_get_fdb_sub_ns(dev, 0);
295 if (!root_ns) {
296 esw_warn(dev, "Failed to get FDB flow namespace\n");
297 return -EOPNOTSUPP;
298 }
299
300 /* num FTE 2, num FG 2 */
301 ft_attr.prio = LEGACY_VEPA_PRIO;
302 ft_attr.max_fte = 2;
303 ft_attr.autogroup.max_num_groups = 2;
304 fdb = mlx5_create_auto_grouped_flow_table(root_ns, &ft_attr);
305 if (IS_ERR(fdb)) {
306 err = PTR_ERR(fdb);
307 esw_warn(dev, "Failed to create VEPA FDB err %d\n", err);
308 return err;
309 }
310 esw->fdb_table.legacy.vepa_fdb = fdb;
311
312 return 0;
313 }
314
esw_create_legacy_fdb_table(struct mlx5_eswitch * esw)315 static int esw_create_legacy_fdb_table(struct mlx5_eswitch *esw)
316 {
317 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
318 struct mlx5_flow_table_attr ft_attr = {};
319 struct mlx5_core_dev *dev = esw->dev;
320 struct mlx5_flow_namespace *root_ns;
321 struct mlx5_flow_table *fdb;
322 struct mlx5_flow_group *g;
323 void *match_criteria;
324 int table_size;
325 u32 *flow_group_in;
326 u8 *dmac;
327 int err = 0;
328
329 esw_debug(dev, "Create FDB log_max_size(%d)\n",
330 MLX5_CAP_ESW_FLOWTABLE_FDB(dev, log_max_ft_size));
331
332 root_ns = mlx5_get_fdb_sub_ns(dev, 0);
333 if (!root_ns) {
334 esw_warn(dev, "Failed to get FDB flow namespace\n");
335 return -EOPNOTSUPP;
336 }
337
338 flow_group_in = kvzalloc(inlen, GFP_KERNEL);
339 if (!flow_group_in)
340 return -ENOMEM;
341
342 table_size = BIT(MLX5_CAP_ESW_FLOWTABLE_FDB(dev, log_max_ft_size));
343 ft_attr.max_fte = table_size;
344 ft_attr.prio = LEGACY_FDB_PRIO;
345 fdb = mlx5_create_flow_table(root_ns, &ft_attr);
346 if (IS_ERR(fdb)) {
347 err = PTR_ERR(fdb);
348 esw_warn(dev, "Failed to create FDB Table err %d\n", err);
349 goto out;
350 }
351 esw->fdb_table.legacy.fdb = fdb;
352
353 /* Addresses group : Full match unicast/multicast addresses */
354 MLX5_SET(create_flow_group_in, flow_group_in, match_criteria_enable,
355 MLX5_MATCH_OUTER_HEADERS);
356 match_criteria = MLX5_ADDR_OF(create_flow_group_in, flow_group_in, match_criteria);
357 dmac = MLX5_ADDR_OF(fte_match_param, match_criteria, outer_headers.dmac_47_16);
358 MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, 0);
359 /* Preserve 2 entries for allmulti and promisc rules*/
360 MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index, table_size - 3);
361 eth_broadcast_addr(dmac);
362 g = mlx5_create_flow_group(fdb, flow_group_in);
363 if (IS_ERR(g)) {
364 err = PTR_ERR(g);
365 esw_warn(dev, "Failed to create flow group err(%d)\n", err);
366 goto out;
367 }
368 esw->fdb_table.legacy.addr_grp = g;
369
370 /* Allmulti group : One rule that forwards any mcast traffic */
371 MLX5_SET(create_flow_group_in, flow_group_in, match_criteria_enable,
372 MLX5_MATCH_OUTER_HEADERS);
373 MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, table_size - 2);
374 MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index, table_size - 2);
375 eth_zero_addr(dmac);
376 dmac[0] = 0x01;
377 g = mlx5_create_flow_group(fdb, flow_group_in);
378 if (IS_ERR(g)) {
379 err = PTR_ERR(g);
380 esw_warn(dev, "Failed to create allmulti flow group err(%d)\n", err);
381 goto out;
382 }
383 esw->fdb_table.legacy.allmulti_grp = g;
384
385 /* Promiscuous group :
386 * One rule that forward all unmatched traffic from previous groups
387 */
388 eth_zero_addr(dmac);
389 MLX5_SET(create_flow_group_in, flow_group_in, match_criteria_enable,
390 MLX5_MATCH_MISC_PARAMETERS);
391 MLX5_SET_TO_ONES(fte_match_param, match_criteria, misc_parameters.source_port);
392 MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, table_size - 1);
393 MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index, table_size - 1);
394 g = mlx5_create_flow_group(fdb, flow_group_in);
395 if (IS_ERR(g)) {
396 err = PTR_ERR(g);
397 esw_warn(dev, "Failed to create promisc flow group err(%d)\n", err);
398 goto out;
399 }
400 esw->fdb_table.legacy.promisc_grp = g;
401
402 out:
403 if (err)
404 esw_destroy_legacy_fdb_table(esw);
405
406 kvfree(flow_group_in);
407 return err;
408 }
409
esw_destroy_legacy_vepa_table(struct mlx5_eswitch * esw)410 static void esw_destroy_legacy_vepa_table(struct mlx5_eswitch *esw)
411 {
412 esw_debug(esw->dev, "Destroy VEPA Table\n");
413 if (!esw->fdb_table.legacy.vepa_fdb)
414 return;
415
416 mlx5_destroy_flow_table(esw->fdb_table.legacy.vepa_fdb);
417 esw->fdb_table.legacy.vepa_fdb = NULL;
418 }
419
esw_destroy_legacy_fdb_table(struct mlx5_eswitch * esw)420 static void esw_destroy_legacy_fdb_table(struct mlx5_eswitch *esw)
421 {
422 esw_debug(esw->dev, "Destroy FDB Table\n");
423 if (!esw->fdb_table.legacy.fdb)
424 return;
425
426 if (esw->fdb_table.legacy.promisc_grp)
427 mlx5_destroy_flow_group(esw->fdb_table.legacy.promisc_grp);
428 if (esw->fdb_table.legacy.allmulti_grp)
429 mlx5_destroy_flow_group(esw->fdb_table.legacy.allmulti_grp);
430 if (esw->fdb_table.legacy.addr_grp)
431 mlx5_destroy_flow_group(esw->fdb_table.legacy.addr_grp);
432 mlx5_destroy_flow_table(esw->fdb_table.legacy.fdb);
433
434 esw->fdb_table.legacy.fdb = NULL;
435 esw->fdb_table.legacy.addr_grp = NULL;
436 esw->fdb_table.legacy.allmulti_grp = NULL;
437 esw->fdb_table.legacy.promisc_grp = NULL;
438 }
439
esw_create_legacy_table(struct mlx5_eswitch * esw)440 static int esw_create_legacy_table(struct mlx5_eswitch *esw)
441 {
442 int err;
443
444 memset(&esw->fdb_table.legacy, 0, sizeof(struct legacy_fdb));
445
446 err = esw_create_legacy_vepa_table(esw);
447 if (err)
448 return err;
449
450 err = esw_create_legacy_fdb_table(esw);
451 if (err)
452 esw_destroy_legacy_vepa_table(esw);
453
454 return err;
455 }
456
esw_destroy_legacy_table(struct mlx5_eswitch * esw)457 static void esw_destroy_legacy_table(struct mlx5_eswitch *esw)
458 {
459 esw_cleanup_vepa_rules(esw);
460 esw_destroy_legacy_fdb_table(esw);
461 esw_destroy_legacy_vepa_table(esw);
462 }
463
464 #define MLX5_LEGACY_SRIOV_VPORT_EVENTS (MLX5_VPORT_UC_ADDR_CHANGE | \
465 MLX5_VPORT_MC_ADDR_CHANGE | \
466 MLX5_VPORT_PROMISC_CHANGE)
467
esw_legacy_enable(struct mlx5_eswitch * esw)468 static int esw_legacy_enable(struct mlx5_eswitch *esw)
469 {
470 struct mlx5_vport *vport;
471 int ret, i;
472
473 ret = esw_create_legacy_table(esw);
474 if (ret)
475 return ret;
476
477 mlx5_esw_for_each_vf_vport(esw, i, vport, esw->esw_funcs.num_vfs)
478 vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
479
480 ret = mlx5_eswitch_enable_pf_vf_vports(esw, MLX5_LEGACY_SRIOV_VPORT_EVENTS);
481 if (ret)
482 esw_destroy_legacy_table(esw);
483 return ret;
484 }
485
esw_legacy_disable(struct mlx5_eswitch * esw)486 static void esw_legacy_disable(struct mlx5_eswitch *esw)
487 {
488 struct esw_mc_addr *mc_promisc;
489
490 mlx5_eswitch_disable_pf_vf_vports(esw);
491
492 mc_promisc = &esw->mc_promisc;
493 if (mc_promisc->uplink_rule)
494 mlx5_del_flow_rules(mc_promisc->uplink_rule);
495
496 esw_destroy_legacy_table(esw);
497 }
498
499 /* E-Switch vport UC/MC lists management */
500 typedef int (*vport_addr_action)(struct mlx5_eswitch *esw,
501 struct vport_addr *vaddr);
502
esw_add_uc_addr(struct mlx5_eswitch * esw,struct vport_addr * vaddr)503 static int esw_add_uc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
504 {
505 u8 *mac = vaddr->node.addr;
506 u16 vport = vaddr->vport;
507 int err;
508
509 /* Skip mlx5_mpfs_add_mac for eswitch_managers,
510 * it is already done by its netdev in mlx5e_execute_l2_action
511 */
512 if (mlx5_esw_is_manager_vport(esw, vport))
513 goto fdb_add;
514
515 err = mlx5_mpfs_add_mac(esw->dev, mac);
516 if (err) {
517 esw_warn(esw->dev,
518 "Failed to add L2 table mac(%pM) for vport(0x%x), err(%d)\n",
519 mac, vport, err);
520 return err;
521 }
522 vaddr->mpfs = true;
523
524 fdb_add:
525 /* SRIOV is enabled: Forward UC MAC to vport */
526 if (esw->fdb_table.legacy.fdb && esw->mode == MLX5_ESWITCH_LEGACY)
527 vaddr->flow_rule = esw_fdb_set_vport_rule(esw, mac, vport);
528
529 esw_debug(esw->dev, "\tADDED UC MAC: vport[%d] %pM fr(%p)\n",
530 vport, mac, vaddr->flow_rule);
531
532 return 0;
533 }
534
esw_del_uc_addr(struct mlx5_eswitch * esw,struct vport_addr * vaddr)535 static int esw_del_uc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
536 {
537 u8 *mac = vaddr->node.addr;
538 u16 vport = vaddr->vport;
539 int err = 0;
540
541 /* Skip mlx5_mpfs_del_mac for eswitch managers,
542 * it is already done by its netdev in mlx5e_execute_l2_action
543 */
544 if (!vaddr->mpfs || mlx5_esw_is_manager_vport(esw, vport))
545 goto fdb_del;
546
547 err = mlx5_mpfs_del_mac(esw->dev, mac);
548 if (err)
549 esw_warn(esw->dev,
550 "Failed to del L2 table mac(%pM) for vport(%d), err(%d)\n",
551 mac, vport, err);
552 vaddr->mpfs = false;
553
554 fdb_del:
555 if (vaddr->flow_rule)
556 mlx5_del_flow_rules(vaddr->flow_rule);
557 vaddr->flow_rule = NULL;
558
559 return 0;
560 }
561
update_allmulti_vports(struct mlx5_eswitch * esw,struct vport_addr * vaddr,struct esw_mc_addr * esw_mc)562 static void update_allmulti_vports(struct mlx5_eswitch *esw,
563 struct vport_addr *vaddr,
564 struct esw_mc_addr *esw_mc)
565 {
566 u8 *mac = vaddr->node.addr;
567 struct mlx5_vport *vport;
568 u16 i, vport_num;
569
570 mlx5_esw_for_all_vports(esw, i, vport) {
571 struct hlist_head *vport_hash = vport->mc_list;
572 struct vport_addr *iter_vaddr =
573 l2addr_hash_find(vport_hash,
574 mac,
575 struct vport_addr);
576 vport_num = vport->vport;
577 if (IS_ERR_OR_NULL(vport->allmulti_rule) ||
578 vaddr->vport == vport_num)
579 continue;
580 switch (vaddr->action) {
581 case MLX5_ACTION_ADD:
582 if (iter_vaddr)
583 continue;
584 iter_vaddr = l2addr_hash_add(vport_hash, mac,
585 struct vport_addr,
586 GFP_KERNEL);
587 if (!iter_vaddr) {
588 esw_warn(esw->dev,
589 "ALL-MULTI: Failed to add MAC(%pM) to vport[%d] DB\n",
590 mac, vport_num);
591 continue;
592 }
593 iter_vaddr->vport = vport_num;
594 iter_vaddr->flow_rule =
595 esw_fdb_set_vport_rule(esw,
596 mac,
597 vport_num);
598 iter_vaddr->mc_promisc = true;
599 break;
600 case MLX5_ACTION_DEL:
601 if (!iter_vaddr)
602 continue;
603 mlx5_del_flow_rules(iter_vaddr->flow_rule);
604 l2addr_hash_del(iter_vaddr);
605 break;
606 }
607 }
608 }
609
esw_add_mc_addr(struct mlx5_eswitch * esw,struct vport_addr * vaddr)610 static int esw_add_mc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
611 {
612 struct hlist_head *hash = esw->mc_table;
613 struct esw_mc_addr *esw_mc;
614 u8 *mac = vaddr->node.addr;
615 u16 vport = vaddr->vport;
616
617 if (!esw->fdb_table.legacy.fdb)
618 return 0;
619
620 esw_mc = l2addr_hash_find(hash, mac, struct esw_mc_addr);
621 if (esw_mc)
622 goto add;
623
624 esw_mc = l2addr_hash_add(hash, mac, struct esw_mc_addr, GFP_KERNEL);
625 if (!esw_mc)
626 return -ENOMEM;
627
628 esw_mc->uplink_rule = /* Forward MC MAC to Uplink */
629 esw_fdb_set_vport_rule(esw, mac, MLX5_VPORT_UPLINK);
630
631 /* Add this multicast mac to all the mc promiscuous vports */
632 update_allmulti_vports(esw, vaddr, esw_mc);
633
634 add:
635 /* If the multicast mac is added as a result of mc promiscuous vport,
636 * don't increment the multicast ref count
637 */
638 if (!vaddr->mc_promisc)
639 esw_mc->refcnt++;
640
641 /* Forward MC MAC to vport */
642 vaddr->flow_rule = esw_fdb_set_vport_rule(esw, mac, vport);
643 esw_debug(esw->dev,
644 "\tADDED MC MAC: vport[%d] %pM fr(%p) refcnt(%d) uplinkfr(%p)\n",
645 vport, mac, vaddr->flow_rule,
646 esw_mc->refcnt, esw_mc->uplink_rule);
647 return 0;
648 }
649
esw_del_mc_addr(struct mlx5_eswitch * esw,struct vport_addr * vaddr)650 static int esw_del_mc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
651 {
652 struct hlist_head *hash = esw->mc_table;
653 struct esw_mc_addr *esw_mc;
654 u8 *mac = vaddr->node.addr;
655 u16 vport = vaddr->vport;
656
657 if (!esw->fdb_table.legacy.fdb)
658 return 0;
659
660 esw_mc = l2addr_hash_find(hash, mac, struct esw_mc_addr);
661 if (!esw_mc) {
662 esw_warn(esw->dev,
663 "Failed to find eswitch MC addr for MAC(%pM) vport(%d)",
664 mac, vport);
665 return -EINVAL;
666 }
667 esw_debug(esw->dev,
668 "\tDELETE MC MAC: vport[%d] %pM fr(%p) refcnt(%d) uplinkfr(%p)\n",
669 vport, mac, vaddr->flow_rule, esw_mc->refcnt,
670 esw_mc->uplink_rule);
671
672 if (vaddr->flow_rule)
673 mlx5_del_flow_rules(vaddr->flow_rule);
674 vaddr->flow_rule = NULL;
675
676 /* If the multicast mac is added as a result of mc promiscuous vport,
677 * don't decrement the multicast ref count.
678 */
679 if (vaddr->mc_promisc || (--esw_mc->refcnt > 0))
680 return 0;
681
682 /* Remove this multicast mac from all the mc promiscuous vports */
683 update_allmulti_vports(esw, vaddr, esw_mc);
684
685 if (esw_mc->uplink_rule)
686 mlx5_del_flow_rules(esw_mc->uplink_rule);
687
688 l2addr_hash_del(esw_mc);
689 return 0;
690 }
691
692 /* Apply vport UC/MC list to HW l2 table and FDB table */
esw_apply_vport_addr_list(struct mlx5_eswitch * esw,struct mlx5_vport * vport,int list_type)693 static void esw_apply_vport_addr_list(struct mlx5_eswitch *esw,
694 struct mlx5_vport *vport, int list_type)
695 {
696 bool is_uc = list_type == MLX5_NVPRT_LIST_TYPE_UC;
697 vport_addr_action vport_addr_add;
698 vport_addr_action vport_addr_del;
699 struct vport_addr *addr;
700 struct l2addr_node *node;
701 struct hlist_head *hash;
702 struct hlist_node *tmp;
703 int hi;
704
705 vport_addr_add = is_uc ? esw_add_uc_addr :
706 esw_add_mc_addr;
707 vport_addr_del = is_uc ? esw_del_uc_addr :
708 esw_del_mc_addr;
709
710 hash = is_uc ? vport->uc_list : vport->mc_list;
711 for_each_l2hash_node(node, tmp, hash, hi) {
712 addr = container_of(node, struct vport_addr, node);
713 switch (addr->action) {
714 case MLX5_ACTION_ADD:
715 vport_addr_add(esw, addr);
716 addr->action = MLX5_ACTION_NONE;
717 break;
718 case MLX5_ACTION_DEL:
719 vport_addr_del(esw, addr);
720 l2addr_hash_del(addr);
721 break;
722 }
723 }
724 }
725
726 /* Sync vport UC/MC list from vport context */
esw_update_vport_addr_list(struct mlx5_eswitch * esw,struct mlx5_vport * vport,int list_type)727 static void esw_update_vport_addr_list(struct mlx5_eswitch *esw,
728 struct mlx5_vport *vport, int list_type)
729 {
730 bool is_uc = list_type == MLX5_NVPRT_LIST_TYPE_UC;
731 u8 (*mac_list)[ETH_ALEN];
732 struct l2addr_node *node;
733 struct vport_addr *addr;
734 struct hlist_head *hash;
735 struct hlist_node *tmp;
736 int size;
737 int err;
738 int hi;
739 int i;
740
741 size = is_uc ? MLX5_MAX_UC_PER_VPORT(esw->dev) :
742 MLX5_MAX_MC_PER_VPORT(esw->dev);
743
744 mac_list = kcalloc(size, ETH_ALEN, GFP_KERNEL);
745 if (!mac_list)
746 return;
747
748 hash = is_uc ? vport->uc_list : vport->mc_list;
749
750 for_each_l2hash_node(node, tmp, hash, hi) {
751 addr = container_of(node, struct vport_addr, node);
752 addr->action = MLX5_ACTION_DEL;
753 }
754
755 if (!vport->enabled)
756 goto out;
757
758 err = mlx5_query_nic_vport_mac_list(esw->dev, vport->vport, list_type,
759 mac_list, &size);
760 if (err)
761 goto out;
762 esw_debug(esw->dev, "vport[%d] context update %s list size (%d)\n",
763 vport->vport, is_uc ? "UC" : "MC", size);
764
765 for (i = 0; i < size; i++) {
766 if (is_uc && !is_valid_ether_addr(mac_list[i]))
767 continue;
768
769 if (!is_uc && !is_multicast_ether_addr(mac_list[i]))
770 continue;
771
772 addr = l2addr_hash_find(hash, mac_list[i], struct vport_addr);
773 if (addr) {
774 addr->action = MLX5_ACTION_NONE;
775 /* If this mac was previously added because of allmulti
776 * promiscuous rx mode, its now converted to be original
777 * vport mac.
778 */
779 if (addr->mc_promisc) {
780 struct esw_mc_addr *esw_mc =
781 l2addr_hash_find(esw->mc_table,
782 mac_list[i],
783 struct esw_mc_addr);
784 if (!esw_mc) {
785 esw_warn(esw->dev,
786 "Failed to MAC(%pM) in mcast DB\n",
787 mac_list[i]);
788 continue;
789 }
790 esw_mc->refcnt++;
791 addr->mc_promisc = false;
792 }
793 continue;
794 }
795
796 addr = l2addr_hash_add(hash, mac_list[i], struct vport_addr,
797 GFP_KERNEL);
798 if (!addr) {
799 esw_warn(esw->dev,
800 "Failed to add MAC(%pM) to vport[%d] DB\n",
801 mac_list[i], vport->vport);
802 continue;
803 }
804 addr->vport = vport->vport;
805 addr->action = MLX5_ACTION_ADD;
806 }
807 out:
808 kfree(mac_list);
809 }
810
811 /* Sync vport UC/MC list from vport context
812 * Must be called after esw_update_vport_addr_list
813 */
esw_update_vport_mc_promisc(struct mlx5_eswitch * esw,struct mlx5_vport * vport)814 static void esw_update_vport_mc_promisc(struct mlx5_eswitch *esw,
815 struct mlx5_vport *vport)
816 {
817 struct l2addr_node *node;
818 struct vport_addr *addr;
819 struct hlist_head *hash;
820 struct hlist_node *tmp;
821 int hi;
822
823 hash = vport->mc_list;
824
825 for_each_l2hash_node(node, tmp, esw->mc_table, hi) {
826 u8 *mac = node->addr;
827
828 addr = l2addr_hash_find(hash, mac, struct vport_addr);
829 if (addr) {
830 if (addr->action == MLX5_ACTION_DEL)
831 addr->action = MLX5_ACTION_NONE;
832 continue;
833 }
834 addr = l2addr_hash_add(hash, mac, struct vport_addr,
835 GFP_KERNEL);
836 if (!addr) {
837 esw_warn(esw->dev,
838 "Failed to add allmulti MAC(%pM) to vport[%d] DB\n",
839 mac, vport->vport);
840 continue;
841 }
842 addr->vport = vport->vport;
843 addr->action = MLX5_ACTION_ADD;
844 addr->mc_promisc = true;
845 }
846 }
847
848 /* Apply vport rx mode to HW FDB table */
esw_apply_vport_rx_mode(struct mlx5_eswitch * esw,struct mlx5_vport * vport,bool promisc,bool mc_promisc)849 static void esw_apply_vport_rx_mode(struct mlx5_eswitch *esw,
850 struct mlx5_vport *vport,
851 bool promisc, bool mc_promisc)
852 {
853 struct esw_mc_addr *allmulti_addr = &esw->mc_promisc;
854
855 if (IS_ERR_OR_NULL(vport->allmulti_rule) != mc_promisc)
856 goto promisc;
857
858 if (mc_promisc) {
859 vport->allmulti_rule =
860 esw_fdb_set_vport_allmulti_rule(esw, vport->vport);
861 if (!allmulti_addr->uplink_rule)
862 allmulti_addr->uplink_rule =
863 esw_fdb_set_vport_allmulti_rule(esw,
864 MLX5_VPORT_UPLINK);
865 allmulti_addr->refcnt++;
866 } else if (vport->allmulti_rule) {
867 mlx5_del_flow_rules(vport->allmulti_rule);
868 vport->allmulti_rule = NULL;
869
870 if (--allmulti_addr->refcnt > 0)
871 goto promisc;
872
873 if (allmulti_addr->uplink_rule)
874 mlx5_del_flow_rules(allmulti_addr->uplink_rule);
875 allmulti_addr->uplink_rule = NULL;
876 }
877
878 promisc:
879 if (IS_ERR_OR_NULL(vport->promisc_rule) != promisc)
880 return;
881
882 if (promisc) {
883 vport->promisc_rule =
884 esw_fdb_set_vport_promisc_rule(esw, vport->vport);
885 } else if (vport->promisc_rule) {
886 mlx5_del_flow_rules(vport->promisc_rule);
887 vport->promisc_rule = NULL;
888 }
889 }
890
891 /* Sync vport rx mode from vport context */
esw_update_vport_rx_mode(struct mlx5_eswitch * esw,struct mlx5_vport * vport)892 static void esw_update_vport_rx_mode(struct mlx5_eswitch *esw,
893 struct mlx5_vport *vport)
894 {
895 int promisc_all = 0;
896 int promisc_uc = 0;
897 int promisc_mc = 0;
898 int err;
899
900 err = mlx5_query_nic_vport_promisc(esw->dev,
901 vport->vport,
902 &promisc_uc,
903 &promisc_mc,
904 &promisc_all);
905 if (err)
906 return;
907 esw_debug(esw->dev, "vport[%d] context update rx mode promisc_all=%d, all_multi=%d\n",
908 vport->vport, promisc_all, promisc_mc);
909
910 if (!vport->info.trusted || !vport->enabled) {
911 promisc_uc = 0;
912 promisc_mc = 0;
913 promisc_all = 0;
914 }
915
916 esw_apply_vport_rx_mode(esw, vport, promisc_all,
917 (promisc_all || promisc_mc));
918 }
919
esw_vport_change_handle_locked(struct mlx5_vport * vport)920 static void esw_vport_change_handle_locked(struct mlx5_vport *vport)
921 {
922 struct mlx5_core_dev *dev = vport->dev;
923 struct mlx5_eswitch *esw = dev->priv.eswitch;
924 u8 mac[ETH_ALEN];
925
926 mlx5_query_nic_vport_mac_address(dev, vport->vport, true, mac);
927 esw_debug(dev, "vport[%d] Context Changed: perm mac: %pM\n",
928 vport->vport, mac);
929
930 if (vport->enabled_events & MLX5_VPORT_UC_ADDR_CHANGE) {
931 esw_update_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_UC);
932 esw_apply_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_UC);
933 }
934
935 if (vport->enabled_events & MLX5_VPORT_MC_ADDR_CHANGE)
936 esw_update_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_MC);
937
938 if (vport->enabled_events & MLX5_VPORT_PROMISC_CHANGE) {
939 esw_update_vport_rx_mode(esw, vport);
940 if (!IS_ERR_OR_NULL(vport->allmulti_rule))
941 esw_update_vport_mc_promisc(esw, vport);
942 }
943
944 if (vport->enabled_events & (MLX5_VPORT_PROMISC_CHANGE | MLX5_VPORT_MC_ADDR_CHANGE))
945 esw_apply_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_MC);
946
947 esw_debug(esw->dev, "vport[%d] Context Changed: Done\n", vport->vport);
948 if (vport->enabled)
949 arm_vport_context_events_cmd(dev, vport->vport,
950 vport->enabled_events);
951 }
952
esw_vport_change_handler(struct work_struct * work)953 static void esw_vport_change_handler(struct work_struct *work)
954 {
955 struct mlx5_vport *vport =
956 container_of(work, struct mlx5_vport, vport_change_handler);
957 struct mlx5_eswitch *esw = vport->dev->priv.eswitch;
958
959 mutex_lock(&esw->state_lock);
960 esw_vport_change_handle_locked(vport);
961 mutex_unlock(&esw->state_lock);
962 }
963
element_type_supported(struct mlx5_eswitch * esw,int type)964 static bool element_type_supported(struct mlx5_eswitch *esw, int type)
965 {
966 const struct mlx5_core_dev *dev = esw->dev;
967
968 switch (type) {
969 case SCHEDULING_CONTEXT_ELEMENT_TYPE_TSAR:
970 return MLX5_CAP_QOS(dev, esw_element_type) &
971 ELEMENT_TYPE_CAP_MASK_TASR;
972 case SCHEDULING_CONTEXT_ELEMENT_TYPE_VPORT:
973 return MLX5_CAP_QOS(dev, esw_element_type) &
974 ELEMENT_TYPE_CAP_MASK_VPORT;
975 case SCHEDULING_CONTEXT_ELEMENT_TYPE_VPORT_TC:
976 return MLX5_CAP_QOS(dev, esw_element_type) &
977 ELEMENT_TYPE_CAP_MASK_VPORT_TC;
978 case SCHEDULING_CONTEXT_ELEMENT_TYPE_PARA_VPORT_TC:
979 return MLX5_CAP_QOS(dev, esw_element_type) &
980 ELEMENT_TYPE_CAP_MASK_PARA_VPORT_TC;
981 }
982 return false;
983 }
984
985 /* Vport QoS management */
esw_create_tsar(struct mlx5_eswitch * esw)986 static void esw_create_tsar(struct mlx5_eswitch *esw)
987 {
988 u32 tsar_ctx[MLX5_ST_SZ_DW(scheduling_context)] = {0};
989 struct mlx5_core_dev *dev = esw->dev;
990 __be32 *attr;
991 int err;
992
993 if (!MLX5_CAP_GEN(dev, qos) || !MLX5_CAP_QOS(dev, esw_scheduling))
994 return;
995
996 if (!element_type_supported(esw, SCHEDULING_CONTEXT_ELEMENT_TYPE_TSAR))
997 return;
998
999 if (esw->qos.enabled)
1000 return;
1001
1002 MLX5_SET(scheduling_context, tsar_ctx, element_type,
1003 SCHEDULING_CONTEXT_ELEMENT_TYPE_TSAR);
1004
1005 attr = MLX5_ADDR_OF(scheduling_context, tsar_ctx, element_attributes);
1006 *attr = cpu_to_be32(TSAR_ELEMENT_TSAR_TYPE_DWRR << 16);
1007
1008 err = mlx5_create_scheduling_element_cmd(dev,
1009 SCHEDULING_HIERARCHY_E_SWITCH,
1010 tsar_ctx,
1011 &esw->qos.root_tsar_id);
1012 if (err) {
1013 esw_warn(esw->dev, "E-Switch create TSAR failed (%d)\n", err);
1014 return;
1015 }
1016
1017 esw->qos.enabled = true;
1018 }
1019
esw_destroy_tsar(struct mlx5_eswitch * esw)1020 static void esw_destroy_tsar(struct mlx5_eswitch *esw)
1021 {
1022 int err;
1023
1024 if (!esw->qos.enabled)
1025 return;
1026
1027 err = mlx5_destroy_scheduling_element_cmd(esw->dev,
1028 SCHEDULING_HIERARCHY_E_SWITCH,
1029 esw->qos.root_tsar_id);
1030 if (err)
1031 esw_warn(esw->dev, "E-Switch destroy TSAR failed (%d)\n", err);
1032
1033 esw->qos.enabled = false;
1034 }
1035
esw_vport_enable_qos(struct mlx5_eswitch * esw,struct mlx5_vport * vport,u32 initial_max_rate,u32 initial_bw_share)1036 static int esw_vport_enable_qos(struct mlx5_eswitch *esw,
1037 struct mlx5_vport *vport,
1038 u32 initial_max_rate, u32 initial_bw_share)
1039 {
1040 u32 sched_ctx[MLX5_ST_SZ_DW(scheduling_context)] = {0};
1041 struct mlx5_core_dev *dev = esw->dev;
1042 void *vport_elem;
1043 int err = 0;
1044
1045 if (!esw->qos.enabled || !MLX5_CAP_GEN(dev, qos) ||
1046 !MLX5_CAP_QOS(dev, esw_scheduling))
1047 return 0;
1048
1049 if (vport->qos.enabled)
1050 return -EEXIST;
1051
1052 MLX5_SET(scheduling_context, sched_ctx, element_type,
1053 SCHEDULING_CONTEXT_ELEMENT_TYPE_VPORT);
1054 vport_elem = MLX5_ADDR_OF(scheduling_context, sched_ctx,
1055 element_attributes);
1056 MLX5_SET(vport_element, vport_elem, vport_number, vport->vport);
1057 MLX5_SET(scheduling_context, sched_ctx, parent_element_id,
1058 esw->qos.root_tsar_id);
1059 MLX5_SET(scheduling_context, sched_ctx, max_average_bw,
1060 initial_max_rate);
1061 MLX5_SET(scheduling_context, sched_ctx, bw_share, initial_bw_share);
1062
1063 err = mlx5_create_scheduling_element_cmd(dev,
1064 SCHEDULING_HIERARCHY_E_SWITCH,
1065 sched_ctx,
1066 &vport->qos.esw_tsar_ix);
1067 if (err) {
1068 esw_warn(esw->dev, "E-Switch create TSAR vport element failed (vport=%d,err=%d)\n",
1069 vport->vport, err);
1070 return err;
1071 }
1072
1073 vport->qos.enabled = true;
1074 return 0;
1075 }
1076
esw_vport_disable_qos(struct mlx5_eswitch * esw,struct mlx5_vport * vport)1077 static void esw_vport_disable_qos(struct mlx5_eswitch *esw,
1078 struct mlx5_vport *vport)
1079 {
1080 int err;
1081
1082 if (!vport->qos.enabled)
1083 return;
1084
1085 err = mlx5_destroy_scheduling_element_cmd(esw->dev,
1086 SCHEDULING_HIERARCHY_E_SWITCH,
1087 vport->qos.esw_tsar_ix);
1088 if (err)
1089 esw_warn(esw->dev, "E-Switch destroy TSAR vport element failed (vport=%d,err=%d)\n",
1090 vport->vport, err);
1091
1092 vport->qos.enabled = false;
1093 }
1094
esw_vport_qos_config(struct mlx5_eswitch * esw,struct mlx5_vport * vport,u32 max_rate,u32 bw_share)1095 static int esw_vport_qos_config(struct mlx5_eswitch *esw,
1096 struct mlx5_vport *vport,
1097 u32 max_rate, u32 bw_share)
1098 {
1099 u32 sched_ctx[MLX5_ST_SZ_DW(scheduling_context)] = {0};
1100 struct mlx5_core_dev *dev = esw->dev;
1101 void *vport_elem;
1102 u32 bitmask = 0;
1103 int err = 0;
1104
1105 if (!MLX5_CAP_GEN(dev, qos) || !MLX5_CAP_QOS(dev, esw_scheduling))
1106 return -EOPNOTSUPP;
1107
1108 if (!vport->qos.enabled)
1109 return -EIO;
1110
1111 MLX5_SET(scheduling_context, sched_ctx, element_type,
1112 SCHEDULING_CONTEXT_ELEMENT_TYPE_VPORT);
1113 vport_elem = MLX5_ADDR_OF(scheduling_context, sched_ctx,
1114 element_attributes);
1115 MLX5_SET(vport_element, vport_elem, vport_number, vport->vport);
1116 MLX5_SET(scheduling_context, sched_ctx, parent_element_id,
1117 esw->qos.root_tsar_id);
1118 MLX5_SET(scheduling_context, sched_ctx, max_average_bw,
1119 max_rate);
1120 MLX5_SET(scheduling_context, sched_ctx, bw_share, bw_share);
1121 bitmask |= MODIFY_SCHEDULING_ELEMENT_IN_MODIFY_BITMASK_MAX_AVERAGE_BW;
1122 bitmask |= MODIFY_SCHEDULING_ELEMENT_IN_MODIFY_BITMASK_BW_SHARE;
1123
1124 err = mlx5_modify_scheduling_element_cmd(dev,
1125 SCHEDULING_HIERARCHY_E_SWITCH,
1126 sched_ctx,
1127 vport->qos.esw_tsar_ix,
1128 bitmask);
1129 if (err) {
1130 esw_warn(esw->dev, "E-Switch modify TSAR vport element failed (vport=%d,err=%d)\n",
1131 vport->vport, err);
1132 return err;
1133 }
1134
1135 return 0;
1136 }
1137
mlx5_esw_modify_vport_rate(struct mlx5_eswitch * esw,u16 vport_num,u32 rate_mbps)1138 int mlx5_esw_modify_vport_rate(struct mlx5_eswitch *esw, u16 vport_num,
1139 u32 rate_mbps)
1140 {
1141 u32 ctx[MLX5_ST_SZ_DW(scheduling_context)] = {};
1142 struct mlx5_vport *vport;
1143
1144 vport = mlx5_eswitch_get_vport(esw, vport_num);
1145
1146 if (!vport->qos.enabled)
1147 return -EOPNOTSUPP;
1148
1149 MLX5_SET(scheduling_context, ctx, max_average_bw, rate_mbps);
1150
1151 return mlx5_modify_scheduling_element_cmd(esw->dev,
1152 SCHEDULING_HIERARCHY_E_SWITCH,
1153 ctx,
1154 vport->qos.esw_tsar_ix,
1155 MODIFY_SCHEDULING_ELEMENT_IN_MODIFY_BITMASK_MAX_AVERAGE_BW);
1156 }
1157
node_guid_gen_from_mac(u64 * node_guid,const u8 * mac)1158 static void node_guid_gen_from_mac(u64 *node_guid, const u8 *mac)
1159 {
1160 ((u8 *)node_guid)[7] = mac[0];
1161 ((u8 *)node_guid)[6] = mac[1];
1162 ((u8 *)node_guid)[5] = mac[2];
1163 ((u8 *)node_guid)[4] = 0xff;
1164 ((u8 *)node_guid)[3] = 0xfe;
1165 ((u8 *)node_guid)[2] = mac[3];
1166 ((u8 *)node_guid)[1] = mac[4];
1167 ((u8 *)node_guid)[0] = mac[5];
1168 }
1169
esw_vport_create_legacy_acl_tables(struct mlx5_eswitch * esw,struct mlx5_vport * vport)1170 static int esw_vport_create_legacy_acl_tables(struct mlx5_eswitch *esw,
1171 struct mlx5_vport *vport)
1172 {
1173 int ret;
1174
1175 /* Only non manager vports need ACL in legacy mode */
1176 if (mlx5_esw_is_manager_vport(esw, vport->vport))
1177 return 0;
1178
1179 ret = esw_acl_ingress_lgcy_setup(esw, vport);
1180 if (ret)
1181 goto ingress_err;
1182
1183 ret = esw_acl_egress_lgcy_setup(esw, vport);
1184 if (ret)
1185 goto egress_err;
1186
1187 return 0;
1188
1189 egress_err:
1190 esw_acl_ingress_lgcy_cleanup(esw, vport);
1191 ingress_err:
1192 return ret;
1193 }
1194
esw_vport_setup_acl(struct mlx5_eswitch * esw,struct mlx5_vport * vport)1195 static int esw_vport_setup_acl(struct mlx5_eswitch *esw,
1196 struct mlx5_vport *vport)
1197 {
1198 if (esw->mode == MLX5_ESWITCH_LEGACY)
1199 return esw_vport_create_legacy_acl_tables(esw, vport);
1200 else
1201 return esw_vport_create_offloads_acl_tables(esw, vport);
1202 }
1203
esw_vport_destroy_legacy_acl_tables(struct mlx5_eswitch * esw,struct mlx5_vport * vport)1204 static void esw_vport_destroy_legacy_acl_tables(struct mlx5_eswitch *esw,
1205 struct mlx5_vport *vport)
1206
1207 {
1208 if (mlx5_esw_is_manager_vport(esw, vport->vport))
1209 return;
1210
1211 esw_acl_egress_lgcy_cleanup(esw, vport);
1212 esw_acl_ingress_lgcy_cleanup(esw, vport);
1213 }
1214
esw_vport_cleanup_acl(struct mlx5_eswitch * esw,struct mlx5_vport * vport)1215 static void esw_vport_cleanup_acl(struct mlx5_eswitch *esw,
1216 struct mlx5_vport *vport)
1217 {
1218 if (esw->mode == MLX5_ESWITCH_LEGACY)
1219 esw_vport_destroy_legacy_acl_tables(esw, vport);
1220 else
1221 esw_vport_destroy_offloads_acl_tables(esw, vport);
1222 }
1223
esw_vport_setup(struct mlx5_eswitch * esw,struct mlx5_vport * vport)1224 static int esw_vport_setup(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
1225 {
1226 u16 vport_num = vport->vport;
1227 int flags;
1228 int err;
1229
1230 err = esw_vport_setup_acl(esw, vport);
1231 if (err)
1232 return err;
1233
1234 /* Attach vport to the eswitch rate limiter */
1235 esw_vport_enable_qos(esw, vport, vport->info.max_rate, vport->qos.bw_share);
1236
1237 if (mlx5_esw_is_manager_vport(esw, vport_num))
1238 return 0;
1239
1240 mlx5_modify_vport_admin_state(esw->dev,
1241 MLX5_VPORT_STATE_OP_MOD_ESW_VPORT,
1242 vport_num, 1,
1243 vport->info.link_state);
1244
1245 /* Host PF has its own mac/guid. */
1246 if (vport_num) {
1247 mlx5_modify_nic_vport_mac_address(esw->dev, vport_num,
1248 vport->info.mac);
1249 mlx5_modify_nic_vport_node_guid(esw->dev, vport_num,
1250 vport->info.node_guid);
1251 }
1252
1253 flags = (vport->info.vlan || vport->info.qos) ?
1254 SET_VLAN_STRIP | SET_VLAN_INSERT : 0;
1255 modify_esw_vport_cvlan(esw->dev, vport_num, vport->info.vlan,
1256 vport->info.qos, flags);
1257
1258 return 0;
1259 }
1260
1261 /* Don't cleanup vport->info, it's needed to restore vport configuration */
esw_vport_cleanup(struct mlx5_eswitch * esw,struct mlx5_vport * vport)1262 static void esw_vport_cleanup(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
1263 {
1264 u16 vport_num = vport->vport;
1265
1266 if (!mlx5_esw_is_manager_vport(esw, vport_num))
1267 mlx5_modify_vport_admin_state(esw->dev,
1268 MLX5_VPORT_STATE_OP_MOD_ESW_VPORT,
1269 vport_num, 1,
1270 MLX5_VPORT_ADMIN_STATE_DOWN);
1271
1272 esw_vport_disable_qos(esw, vport);
1273 esw_vport_cleanup_acl(esw, vport);
1274 }
1275
esw_enable_vport(struct mlx5_eswitch * esw,u16 vport_num,enum mlx5_eswitch_vport_event enabled_events)1276 static int esw_enable_vport(struct mlx5_eswitch *esw, u16 vport_num,
1277 enum mlx5_eswitch_vport_event enabled_events)
1278 {
1279 struct mlx5_vport *vport;
1280 int ret;
1281
1282 vport = mlx5_eswitch_get_vport(esw, vport_num);
1283
1284 mutex_lock(&esw->state_lock);
1285 WARN_ON(vport->enabled);
1286
1287 esw_debug(esw->dev, "Enabling VPORT(%d)\n", vport_num);
1288
1289 ret = esw_vport_setup(esw, vport);
1290 if (ret)
1291 goto done;
1292
1293 /* Sync with current vport context */
1294 vport->enabled_events = enabled_events;
1295 vport->enabled = true;
1296
1297 /* Esw manager is trusted by default. Host PF (vport 0) is trusted as well
1298 * in smartNIC as it's a vport group manager.
1299 */
1300 if (mlx5_esw_is_manager_vport(esw, vport_num) ||
1301 (!vport_num && mlx5_core_is_ecpf(esw->dev)))
1302 vport->info.trusted = true;
1303
1304 esw_vport_change_handle_locked(vport);
1305
1306 esw->enabled_vports++;
1307 esw_debug(esw->dev, "Enabled VPORT(%d)\n", vport_num);
1308 done:
1309 mutex_unlock(&esw->state_lock);
1310 return ret;
1311 }
1312
esw_disable_vport(struct mlx5_eswitch * esw,u16 vport_num)1313 static void esw_disable_vport(struct mlx5_eswitch *esw, u16 vport_num)
1314 {
1315 struct mlx5_vport *vport;
1316
1317 vport = mlx5_eswitch_get_vport(esw, vport_num);
1318
1319 mutex_lock(&esw->state_lock);
1320 if (!vport->enabled)
1321 goto done;
1322
1323 esw_debug(esw->dev, "Disabling vport(%d)\n", vport_num);
1324 /* Mark this vport as disabled to discard new events */
1325 vport->enabled = false;
1326
1327 /* Disable events from this vport */
1328 arm_vport_context_events_cmd(esw->dev, vport->vport, 0);
1329 /* We don't assume VFs will cleanup after themselves.
1330 * Calling vport change handler while vport is disabled will cleanup
1331 * the vport resources.
1332 */
1333 esw_vport_change_handle_locked(vport);
1334 vport->enabled_events = 0;
1335 esw_vport_cleanup(esw, vport);
1336 esw->enabled_vports--;
1337
1338 done:
1339 mutex_unlock(&esw->state_lock);
1340 }
1341
eswitch_vport_event(struct notifier_block * nb,unsigned long type,void * data)1342 static int eswitch_vport_event(struct notifier_block *nb,
1343 unsigned long type, void *data)
1344 {
1345 struct mlx5_eswitch *esw = mlx5_nb_cof(nb, struct mlx5_eswitch, nb);
1346 struct mlx5_eqe *eqe = data;
1347 struct mlx5_vport *vport;
1348 u16 vport_num;
1349
1350 vport_num = be16_to_cpu(eqe->data.vport_change.vport_num);
1351 vport = mlx5_eswitch_get_vport(esw, vport_num);
1352 if (!IS_ERR(vport))
1353 queue_work(esw->work_queue, &vport->vport_change_handler);
1354 return NOTIFY_OK;
1355 }
1356
1357 /**
1358 * mlx5_esw_query_functions - Returns raw output about functions state
1359 * @dev: Pointer to device to query
1360 *
1361 * mlx5_esw_query_functions() allocates and returns functions changed
1362 * raw output memory pointer from device on success. Otherwise returns ERR_PTR.
1363 * Caller must free the memory using kvfree() when valid pointer is returned.
1364 */
mlx5_esw_query_functions(struct mlx5_core_dev * dev)1365 const u32 *mlx5_esw_query_functions(struct mlx5_core_dev *dev)
1366 {
1367 int outlen = MLX5_ST_SZ_BYTES(query_esw_functions_out);
1368 u32 in[MLX5_ST_SZ_DW(query_esw_functions_in)] = {};
1369 u32 *out;
1370 int err;
1371
1372 out = kvzalloc(outlen, GFP_KERNEL);
1373 if (!out)
1374 return ERR_PTR(-ENOMEM);
1375
1376 MLX5_SET(query_esw_functions_in, in, opcode,
1377 MLX5_CMD_OP_QUERY_ESW_FUNCTIONS);
1378
1379 err = mlx5_cmd_exec_inout(dev, query_esw_functions, in, out);
1380 if (!err)
1381 return out;
1382
1383 kvfree(out);
1384 return ERR_PTR(err);
1385 }
1386
mlx5_eswitch_event_handlers_register(struct mlx5_eswitch * esw)1387 static void mlx5_eswitch_event_handlers_register(struct mlx5_eswitch *esw)
1388 {
1389 MLX5_NB_INIT(&esw->nb, eswitch_vport_event, NIC_VPORT_CHANGE);
1390 mlx5_eq_notifier_register(esw->dev, &esw->nb);
1391
1392 if (esw->mode == MLX5_ESWITCH_OFFLOADS && mlx5_eswitch_is_funcs_handler(esw->dev)) {
1393 MLX5_NB_INIT(&esw->esw_funcs.nb, mlx5_esw_funcs_changed_handler,
1394 ESW_FUNCTIONS_CHANGED);
1395 mlx5_eq_notifier_register(esw->dev, &esw->esw_funcs.nb);
1396 }
1397 }
1398
mlx5_eswitch_event_handlers_unregister(struct mlx5_eswitch * esw)1399 static void mlx5_eswitch_event_handlers_unregister(struct mlx5_eswitch *esw)
1400 {
1401 if (esw->mode == MLX5_ESWITCH_OFFLOADS && mlx5_eswitch_is_funcs_handler(esw->dev))
1402 mlx5_eq_notifier_unregister(esw->dev, &esw->esw_funcs.nb);
1403
1404 mlx5_eq_notifier_unregister(esw->dev, &esw->nb);
1405
1406 flush_workqueue(esw->work_queue);
1407 }
1408
mlx5_eswitch_clear_vf_vports_info(struct mlx5_eswitch * esw)1409 static void mlx5_eswitch_clear_vf_vports_info(struct mlx5_eswitch *esw)
1410 {
1411 struct mlx5_vport *vport;
1412 int i;
1413
1414 mlx5_esw_for_each_vf_vport(esw, i, vport, esw->esw_funcs.num_vfs) {
1415 memset(&vport->qos, 0, sizeof(vport->qos));
1416 memset(&vport->info, 0, sizeof(vport->info));
1417 vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
1418 }
1419 }
1420
1421 /* Public E-Switch API */
1422 #define ESW_ALLOWED(esw) ((esw) && MLX5_ESWITCH_MANAGER((esw)->dev))
1423
mlx5_eswitch_load_vport(struct mlx5_eswitch * esw,u16 vport_num,enum mlx5_eswitch_vport_event enabled_events)1424 int mlx5_eswitch_load_vport(struct mlx5_eswitch *esw, u16 vport_num,
1425 enum mlx5_eswitch_vport_event enabled_events)
1426 {
1427 int err;
1428
1429 err = esw_enable_vport(esw, vport_num, enabled_events);
1430 if (err)
1431 return err;
1432
1433 err = esw_offloads_load_rep(esw, vport_num);
1434 if (err)
1435 goto err_rep;
1436
1437 return err;
1438
1439 err_rep:
1440 esw_disable_vport(esw, vport_num);
1441 return err;
1442 }
1443
mlx5_eswitch_unload_vport(struct mlx5_eswitch * esw,u16 vport_num)1444 void mlx5_eswitch_unload_vport(struct mlx5_eswitch *esw, u16 vport_num)
1445 {
1446 esw_offloads_unload_rep(esw, vport_num);
1447 esw_disable_vport(esw, vport_num);
1448 }
1449
mlx5_eswitch_unload_vf_vports(struct mlx5_eswitch * esw,u16 num_vfs)1450 void mlx5_eswitch_unload_vf_vports(struct mlx5_eswitch *esw, u16 num_vfs)
1451 {
1452 int i;
1453
1454 mlx5_esw_for_each_vf_vport_num_reverse(esw, i, num_vfs)
1455 mlx5_eswitch_unload_vport(esw, i);
1456 }
1457
mlx5_eswitch_load_vf_vports(struct mlx5_eswitch * esw,u16 num_vfs,enum mlx5_eswitch_vport_event enabled_events)1458 int mlx5_eswitch_load_vf_vports(struct mlx5_eswitch *esw, u16 num_vfs,
1459 enum mlx5_eswitch_vport_event enabled_events)
1460 {
1461 int err;
1462 int i;
1463
1464 mlx5_esw_for_each_vf_vport_num(esw, i, num_vfs) {
1465 err = mlx5_eswitch_load_vport(esw, i, enabled_events);
1466 if (err)
1467 goto vf_err;
1468 }
1469
1470 return 0;
1471
1472 vf_err:
1473 mlx5_eswitch_unload_vf_vports(esw, i - 1);
1474 return err;
1475 }
1476
1477 /* mlx5_eswitch_enable_pf_vf_vports() enables vports of PF, ECPF and VFs
1478 * whichever are present on the eswitch.
1479 */
1480 int
mlx5_eswitch_enable_pf_vf_vports(struct mlx5_eswitch * esw,enum mlx5_eswitch_vport_event enabled_events)1481 mlx5_eswitch_enable_pf_vf_vports(struct mlx5_eswitch *esw,
1482 enum mlx5_eswitch_vport_event enabled_events)
1483 {
1484 int ret;
1485
1486 /* Enable PF vport */
1487 ret = mlx5_eswitch_load_vport(esw, MLX5_VPORT_PF, enabled_events);
1488 if (ret)
1489 return ret;
1490
1491 /* Enable ECPF vport */
1492 if (mlx5_ecpf_vport_exists(esw->dev)) {
1493 ret = mlx5_eswitch_load_vport(esw, MLX5_VPORT_ECPF, enabled_events);
1494 if (ret)
1495 goto ecpf_err;
1496 }
1497
1498 /* Enable VF vports */
1499 ret = mlx5_eswitch_load_vf_vports(esw, esw->esw_funcs.num_vfs,
1500 enabled_events);
1501 if (ret)
1502 goto vf_err;
1503 return 0;
1504
1505 vf_err:
1506 if (mlx5_ecpf_vport_exists(esw->dev))
1507 mlx5_eswitch_unload_vport(esw, MLX5_VPORT_ECPF);
1508
1509 ecpf_err:
1510 mlx5_eswitch_unload_vport(esw, MLX5_VPORT_PF);
1511 return ret;
1512 }
1513
1514 /* mlx5_eswitch_disable_pf_vf_vports() disables vports of PF, ECPF and VFs
1515 * whichever are previously enabled on the eswitch.
1516 */
mlx5_eswitch_disable_pf_vf_vports(struct mlx5_eswitch * esw)1517 void mlx5_eswitch_disable_pf_vf_vports(struct mlx5_eswitch *esw)
1518 {
1519 mlx5_eswitch_unload_vf_vports(esw, esw->esw_funcs.num_vfs);
1520
1521 if (mlx5_ecpf_vport_exists(esw->dev))
1522 mlx5_eswitch_unload_vport(esw, MLX5_VPORT_ECPF);
1523
1524 mlx5_eswitch_unload_vport(esw, MLX5_VPORT_PF);
1525 }
1526
mlx5_eswitch_get_devlink_param(struct mlx5_eswitch * esw)1527 static void mlx5_eswitch_get_devlink_param(struct mlx5_eswitch *esw)
1528 {
1529 struct devlink *devlink = priv_to_devlink(esw->dev);
1530 union devlink_param_value val;
1531 int err;
1532
1533 err = devlink_param_driverinit_value_get(devlink,
1534 MLX5_DEVLINK_PARAM_ID_ESW_LARGE_GROUP_NUM,
1535 &val);
1536 if (!err) {
1537 esw->params.large_group_num = val.vu32;
1538 } else {
1539 esw_warn(esw->dev,
1540 "Devlink can't get param fdb_large_groups, uses default (%d).\n",
1541 ESW_OFFLOADS_DEFAULT_NUM_GROUPS);
1542 esw->params.large_group_num = ESW_OFFLOADS_DEFAULT_NUM_GROUPS;
1543 }
1544 }
1545
1546 static void
mlx5_eswitch_update_num_of_vfs(struct mlx5_eswitch * esw,int num_vfs)1547 mlx5_eswitch_update_num_of_vfs(struct mlx5_eswitch *esw, int num_vfs)
1548 {
1549 const u32 *out;
1550
1551 WARN_ON_ONCE(esw->mode != MLX5_ESWITCH_NONE);
1552
1553 if (num_vfs < 0)
1554 return;
1555
1556 if (!mlx5_core_is_ecpf_esw_manager(esw->dev)) {
1557 esw->esw_funcs.num_vfs = num_vfs;
1558 return;
1559 }
1560
1561 out = mlx5_esw_query_functions(esw->dev);
1562 if (IS_ERR(out))
1563 return;
1564
1565 esw->esw_funcs.num_vfs = MLX5_GET(query_esw_functions_out, out,
1566 host_params_context.host_num_of_vfs);
1567 kvfree(out);
1568 }
1569
1570 /**
1571 * mlx5_eswitch_enable_locked - Enable eswitch
1572 * @esw: Pointer to eswitch
1573 * @mode: Eswitch mode to enable
1574 * @num_vfs: Enable eswitch for given number of VFs. This is optional.
1575 * Valid value are 0, > 0 and MLX5_ESWITCH_IGNORE_NUM_VFS.
1576 * Caller should pass num_vfs > 0 when enabling eswitch for
1577 * vf vports. Caller should pass num_vfs = 0, when eswitch
1578 * is enabled without sriov VFs or when caller
1579 * is unaware of the sriov state of the host PF on ECPF based
1580 * eswitch. Caller should pass < 0 when num_vfs should be
1581 * completely ignored. This is typically the case when eswitch
1582 * is enabled without sriov regardless of PF/ECPF system.
1583 * mlx5_eswitch_enable_locked() Enables eswitch in either legacy or offloads
1584 * mode. If num_vfs >=0 is provided, it setup VF related eswitch vports.
1585 * It returns 0 on success or error code on failure.
1586 */
mlx5_eswitch_enable_locked(struct mlx5_eswitch * esw,int mode,int num_vfs)1587 int mlx5_eswitch_enable_locked(struct mlx5_eswitch *esw, int mode, int num_vfs)
1588 {
1589 int err;
1590
1591 lockdep_assert_held(&esw->mode_lock);
1592
1593 if (!MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, ft_support)) {
1594 esw_warn(esw->dev, "FDB is not supported, aborting ...\n");
1595 return -EOPNOTSUPP;
1596 }
1597
1598 if (!MLX5_CAP_ESW_INGRESS_ACL(esw->dev, ft_support))
1599 esw_warn(esw->dev, "ingress ACL is not supported by FW\n");
1600
1601 if (!MLX5_CAP_ESW_EGRESS_ACL(esw->dev, ft_support))
1602 esw_warn(esw->dev, "engress ACL is not supported by FW\n");
1603
1604 mlx5_eswitch_get_devlink_param(esw);
1605
1606 mlx5_eswitch_update_num_of_vfs(esw, num_vfs);
1607
1608 esw_create_tsar(esw);
1609
1610 esw->mode = mode;
1611
1612 mlx5_lag_update(esw->dev);
1613
1614 if (mode == MLX5_ESWITCH_LEGACY) {
1615 err = esw_legacy_enable(esw);
1616 } else {
1617 mlx5_reload_interface(esw->dev, MLX5_INTERFACE_PROTOCOL_ETH);
1618 mlx5_reload_interface(esw->dev, MLX5_INTERFACE_PROTOCOL_IB);
1619 err = esw_offloads_enable(esw);
1620 }
1621
1622 if (err)
1623 goto abort;
1624
1625 mlx5_eswitch_event_handlers_register(esw);
1626
1627 esw_info(esw->dev, "Enable: mode(%s), nvfs(%d), active vports(%d)\n",
1628 mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
1629 esw->esw_funcs.num_vfs, esw->enabled_vports);
1630
1631 return 0;
1632
1633 abort:
1634 esw->mode = MLX5_ESWITCH_NONE;
1635
1636 if (mode == MLX5_ESWITCH_OFFLOADS) {
1637 mlx5_reload_interface(esw->dev, MLX5_INTERFACE_PROTOCOL_IB);
1638 mlx5_reload_interface(esw->dev, MLX5_INTERFACE_PROTOCOL_ETH);
1639 }
1640 esw_destroy_tsar(esw);
1641 return err;
1642 }
1643
1644 /**
1645 * mlx5_eswitch_enable - Enable eswitch
1646 * @esw: Pointer to eswitch
1647 * @num_vfs: Enable eswitch swich for given number of VFs.
1648 * Caller must pass num_vfs > 0 when enabling eswitch for
1649 * vf vports.
1650 * mlx5_eswitch_enable() returns 0 on success or error code on failure.
1651 */
mlx5_eswitch_enable(struct mlx5_eswitch * esw,int num_vfs)1652 int mlx5_eswitch_enable(struct mlx5_eswitch *esw, int num_vfs)
1653 {
1654 int ret;
1655
1656 if (!ESW_ALLOWED(esw))
1657 return 0;
1658
1659 mutex_lock(&esw->mode_lock);
1660 if (esw->mode == MLX5_ESWITCH_NONE) {
1661 ret = mlx5_eswitch_enable_locked(esw, MLX5_ESWITCH_LEGACY, num_vfs);
1662 } else {
1663 enum mlx5_eswitch_vport_event vport_events;
1664
1665 vport_events = (esw->mode == MLX5_ESWITCH_LEGACY) ?
1666 MLX5_LEGACY_SRIOV_VPORT_EVENTS : MLX5_VPORT_UC_ADDR_CHANGE;
1667 ret = mlx5_eswitch_load_vf_vports(esw, num_vfs, vport_events);
1668 if (!ret)
1669 esw->esw_funcs.num_vfs = num_vfs;
1670 }
1671 mutex_unlock(&esw->mode_lock);
1672 return ret;
1673 }
1674
mlx5_eswitch_disable_locked(struct mlx5_eswitch * esw,bool clear_vf)1675 void mlx5_eswitch_disable_locked(struct mlx5_eswitch *esw, bool clear_vf)
1676 {
1677 int old_mode;
1678
1679 lockdep_assert_held_write(&esw->mode_lock);
1680
1681 if (esw->mode == MLX5_ESWITCH_NONE)
1682 return;
1683
1684 esw_info(esw->dev, "Disable: mode(%s), nvfs(%d), active vports(%d)\n",
1685 esw->mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
1686 esw->esw_funcs.num_vfs, esw->enabled_vports);
1687
1688 mlx5_eswitch_event_handlers_unregister(esw);
1689
1690 if (esw->mode == MLX5_ESWITCH_LEGACY)
1691 esw_legacy_disable(esw);
1692 else if (esw->mode == MLX5_ESWITCH_OFFLOADS)
1693 esw_offloads_disable(esw);
1694
1695 old_mode = esw->mode;
1696 esw->mode = MLX5_ESWITCH_NONE;
1697
1698 mlx5_lag_update(esw->dev);
1699
1700 if (old_mode == MLX5_ESWITCH_OFFLOADS) {
1701 mlx5_reload_interface(esw->dev, MLX5_INTERFACE_PROTOCOL_IB);
1702 mlx5_reload_interface(esw->dev, MLX5_INTERFACE_PROTOCOL_ETH);
1703 }
1704 esw_destroy_tsar(esw);
1705
1706 if (clear_vf)
1707 mlx5_eswitch_clear_vf_vports_info(esw);
1708 }
1709
mlx5_eswitch_disable(struct mlx5_eswitch * esw,bool clear_vf)1710 void mlx5_eswitch_disable(struct mlx5_eswitch *esw, bool clear_vf)
1711 {
1712 if (!ESW_ALLOWED(esw))
1713 return;
1714
1715 mutex_lock(&esw->mode_lock);
1716 mlx5_eswitch_disable_locked(esw, clear_vf);
1717 esw->esw_funcs.num_vfs = 0;
1718 mutex_unlock(&esw->mode_lock);
1719 }
1720
mlx5_eswitch_init(struct mlx5_core_dev * dev)1721 int mlx5_eswitch_init(struct mlx5_core_dev *dev)
1722 {
1723 struct mlx5_eswitch *esw;
1724 struct mlx5_vport *vport;
1725 int total_vports;
1726 int err, i;
1727
1728 if (!MLX5_VPORT_MANAGER(dev))
1729 return 0;
1730
1731 total_vports = mlx5_eswitch_get_total_vports(dev);
1732
1733 esw_info(dev,
1734 "Total vports %d, per vport: max uc(%d) max mc(%d)\n",
1735 total_vports,
1736 MLX5_MAX_UC_PER_VPORT(dev),
1737 MLX5_MAX_MC_PER_VPORT(dev));
1738
1739 esw = kzalloc(sizeof(*esw), GFP_KERNEL);
1740 if (!esw)
1741 return -ENOMEM;
1742
1743 esw->dev = dev;
1744 esw->manager_vport = mlx5_eswitch_manager_vport(dev);
1745 esw->first_host_vport = mlx5_eswitch_first_host_vport_num(dev);
1746
1747 esw->work_queue = create_singlethread_workqueue("mlx5_esw_wq");
1748 if (!esw->work_queue) {
1749 err = -ENOMEM;
1750 goto abort;
1751 }
1752
1753 esw->vports = kcalloc(total_vports, sizeof(struct mlx5_vport),
1754 GFP_KERNEL);
1755 if (!esw->vports) {
1756 err = -ENOMEM;
1757 goto abort;
1758 }
1759
1760 esw->total_vports = total_vports;
1761
1762 err = esw_offloads_init_reps(esw);
1763 if (err)
1764 goto abort;
1765
1766 mutex_init(&esw->offloads.encap_tbl_lock);
1767 hash_init(esw->offloads.encap_tbl);
1768 mutex_init(&esw->offloads.decap_tbl_lock);
1769 hash_init(esw->offloads.decap_tbl);
1770 mlx5e_mod_hdr_tbl_init(&esw->offloads.mod_hdr);
1771 atomic64_set(&esw->offloads.num_flows, 0);
1772 ida_init(&esw->offloads.vport_metadata_ida);
1773 mutex_init(&esw->state_lock);
1774 mutex_init(&esw->mode_lock);
1775
1776 mlx5_esw_for_all_vports(esw, i, vport) {
1777 vport->vport = mlx5_eswitch_index_to_vport_num(esw, i);
1778 vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
1779 vport->dev = dev;
1780 INIT_WORK(&vport->vport_change_handler,
1781 esw_vport_change_handler);
1782 }
1783
1784 esw->enabled_vports = 0;
1785 esw->mode = MLX5_ESWITCH_NONE;
1786 esw->offloads.inline_mode = MLX5_INLINE_MODE_NONE;
1787
1788 dev->priv.eswitch = esw;
1789 return 0;
1790 abort:
1791 if (esw->work_queue)
1792 destroy_workqueue(esw->work_queue);
1793 esw_offloads_cleanup_reps(esw);
1794 kfree(esw->vports);
1795 kfree(esw);
1796 return err;
1797 }
1798
mlx5_eswitch_cleanup(struct mlx5_eswitch * esw)1799 void mlx5_eswitch_cleanup(struct mlx5_eswitch *esw)
1800 {
1801 if (!esw || !MLX5_VPORT_MANAGER(esw->dev))
1802 return;
1803
1804 esw_info(esw->dev, "cleanup\n");
1805
1806 esw->dev->priv.eswitch = NULL;
1807 destroy_workqueue(esw->work_queue);
1808 esw_offloads_cleanup_reps(esw);
1809 mutex_destroy(&esw->mode_lock);
1810 mutex_destroy(&esw->state_lock);
1811 ida_destroy(&esw->offloads.vport_metadata_ida);
1812 mlx5e_mod_hdr_tbl_destroy(&esw->offloads.mod_hdr);
1813 mutex_destroy(&esw->offloads.encap_tbl_lock);
1814 mutex_destroy(&esw->offloads.decap_tbl_lock);
1815 kfree(esw->vports);
1816 kfree(esw);
1817 }
1818
1819 /* Vport Administration */
1820 static int
mlx5_esw_set_vport_mac_locked(struct mlx5_eswitch * esw,struct mlx5_vport * evport,const u8 * mac)1821 mlx5_esw_set_vport_mac_locked(struct mlx5_eswitch *esw,
1822 struct mlx5_vport *evport, const u8 *mac)
1823 {
1824 u16 vport_num = evport->vport;
1825 u64 node_guid;
1826 int err = 0;
1827
1828 if (is_multicast_ether_addr(mac))
1829 return -EINVAL;
1830
1831 if (evport->info.spoofchk && !is_valid_ether_addr(mac))
1832 mlx5_core_warn(esw->dev,
1833 "Set invalid MAC while spoofchk is on, vport(%d)\n",
1834 vport_num);
1835
1836 err = mlx5_modify_nic_vport_mac_address(esw->dev, vport_num, mac);
1837 if (err) {
1838 mlx5_core_warn(esw->dev,
1839 "Failed to mlx5_modify_nic_vport_mac vport(%d) err=(%d)\n",
1840 vport_num, err);
1841 return err;
1842 }
1843
1844 node_guid_gen_from_mac(&node_guid, mac);
1845 err = mlx5_modify_nic_vport_node_guid(esw->dev, vport_num, node_guid);
1846 if (err)
1847 mlx5_core_warn(esw->dev,
1848 "Failed to set vport %d node guid, err = %d. RDMA_CM will not function properly for this VF.\n",
1849 vport_num, err);
1850
1851 ether_addr_copy(evport->info.mac, mac);
1852 evport->info.node_guid = node_guid;
1853 if (evport->enabled && esw->mode == MLX5_ESWITCH_LEGACY)
1854 err = esw_acl_ingress_lgcy_setup(esw, evport);
1855
1856 return err;
1857 }
1858
mlx5_eswitch_set_vport_mac(struct mlx5_eswitch * esw,u16 vport,const u8 * mac)1859 int mlx5_eswitch_set_vport_mac(struct mlx5_eswitch *esw,
1860 u16 vport, const u8 *mac)
1861 {
1862 struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
1863 int err = 0;
1864
1865 if (IS_ERR(evport))
1866 return PTR_ERR(evport);
1867
1868 mutex_lock(&esw->state_lock);
1869 err = mlx5_esw_set_vport_mac_locked(esw, evport, mac);
1870 mutex_unlock(&esw->state_lock);
1871 return err;
1872 }
1873
1874 static bool
is_port_function_supported(const struct mlx5_eswitch * esw,u16 vport_num)1875 is_port_function_supported(const struct mlx5_eswitch *esw, u16 vport_num)
1876 {
1877 return vport_num == MLX5_VPORT_PF ||
1878 mlx5_eswitch_is_vf_vport(esw, vport_num);
1879 }
1880
mlx5_devlink_port_function_hw_addr_get(struct devlink * devlink,struct devlink_port * port,u8 * hw_addr,int * hw_addr_len,struct netlink_ext_ack * extack)1881 int mlx5_devlink_port_function_hw_addr_get(struct devlink *devlink,
1882 struct devlink_port *port,
1883 u8 *hw_addr, int *hw_addr_len,
1884 struct netlink_ext_ack *extack)
1885 {
1886 struct mlx5_eswitch *esw;
1887 struct mlx5_vport *vport;
1888 int err = -EOPNOTSUPP;
1889 u16 vport_num;
1890
1891 esw = mlx5_devlink_eswitch_get(devlink);
1892 if (IS_ERR(esw))
1893 return PTR_ERR(esw);
1894
1895 vport_num = mlx5_esw_devlink_port_index_to_vport_num(port->index);
1896 if (!is_port_function_supported(esw, vport_num))
1897 return -EOPNOTSUPP;
1898
1899 vport = mlx5_eswitch_get_vport(esw, vport_num);
1900 if (IS_ERR(vport)) {
1901 NL_SET_ERR_MSG_MOD(extack, "Invalid port");
1902 return PTR_ERR(vport);
1903 }
1904
1905 mutex_lock(&esw->state_lock);
1906 if (vport->enabled) {
1907 ether_addr_copy(hw_addr, vport->info.mac);
1908 *hw_addr_len = ETH_ALEN;
1909 err = 0;
1910 }
1911 mutex_unlock(&esw->state_lock);
1912 return err;
1913 }
1914
mlx5_devlink_port_function_hw_addr_set(struct devlink * devlink,struct devlink_port * port,const u8 * hw_addr,int hw_addr_len,struct netlink_ext_ack * extack)1915 int mlx5_devlink_port_function_hw_addr_set(struct devlink *devlink,
1916 struct devlink_port *port,
1917 const u8 *hw_addr, int hw_addr_len,
1918 struct netlink_ext_ack *extack)
1919 {
1920 struct mlx5_eswitch *esw;
1921 struct mlx5_vport *vport;
1922 int err = -EOPNOTSUPP;
1923 u16 vport_num;
1924
1925 esw = mlx5_devlink_eswitch_get(devlink);
1926 if (IS_ERR(esw)) {
1927 NL_SET_ERR_MSG_MOD(extack, "Eswitch doesn't support set hw_addr");
1928 return PTR_ERR(esw);
1929 }
1930
1931 vport_num = mlx5_esw_devlink_port_index_to_vport_num(port->index);
1932 if (!is_port_function_supported(esw, vport_num)) {
1933 NL_SET_ERR_MSG_MOD(extack, "Port doesn't support set hw_addr");
1934 return -EINVAL;
1935 }
1936 vport = mlx5_eswitch_get_vport(esw, vport_num);
1937 if (IS_ERR(vport)) {
1938 NL_SET_ERR_MSG_MOD(extack, "Invalid port");
1939 return PTR_ERR(vport);
1940 }
1941
1942 mutex_lock(&esw->state_lock);
1943 if (vport->enabled)
1944 err = mlx5_esw_set_vport_mac_locked(esw, vport, hw_addr);
1945 else
1946 NL_SET_ERR_MSG_MOD(extack, "Eswitch vport is disabled");
1947 mutex_unlock(&esw->state_lock);
1948 return err;
1949 }
1950
mlx5_eswitch_set_vport_state(struct mlx5_eswitch * esw,u16 vport,int link_state)1951 int mlx5_eswitch_set_vport_state(struct mlx5_eswitch *esw,
1952 u16 vport, int link_state)
1953 {
1954 struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
1955 int opmod = MLX5_VPORT_STATE_OP_MOD_ESW_VPORT;
1956 int other_vport = 1;
1957 int err = 0;
1958
1959 if (!ESW_ALLOWED(esw))
1960 return -EPERM;
1961 if (IS_ERR(evport))
1962 return PTR_ERR(evport);
1963
1964 if (vport == MLX5_VPORT_UPLINK) {
1965 opmod = MLX5_VPORT_STATE_OP_MOD_UPLINK;
1966 other_vport = 0;
1967 vport = 0;
1968 }
1969 mutex_lock(&esw->state_lock);
1970
1971 err = mlx5_modify_vport_admin_state(esw->dev, opmod, vport, other_vport, link_state);
1972 if (err) {
1973 mlx5_core_warn(esw->dev, "Failed to set vport %d link state, opmod = %d, err = %d",
1974 vport, opmod, err);
1975 goto unlock;
1976 }
1977
1978 evport->info.link_state = link_state;
1979
1980 unlock:
1981 mutex_unlock(&esw->state_lock);
1982 return err;
1983 }
1984
mlx5_eswitch_get_vport_config(struct mlx5_eswitch * esw,u16 vport,struct ifla_vf_info * ivi)1985 int mlx5_eswitch_get_vport_config(struct mlx5_eswitch *esw,
1986 u16 vport, struct ifla_vf_info *ivi)
1987 {
1988 struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
1989
1990 if (IS_ERR(evport))
1991 return PTR_ERR(evport);
1992
1993 memset(ivi, 0, sizeof(*ivi));
1994 ivi->vf = vport - 1;
1995
1996 mutex_lock(&esw->state_lock);
1997 ether_addr_copy(ivi->mac, evport->info.mac);
1998 ivi->linkstate = evport->info.link_state;
1999 ivi->vlan = evport->info.vlan;
2000 ivi->qos = evport->info.qos;
2001 ivi->spoofchk = evport->info.spoofchk;
2002 ivi->trusted = evport->info.trusted;
2003 ivi->min_tx_rate = evport->info.min_rate;
2004 ivi->max_tx_rate = evport->info.max_rate;
2005 mutex_unlock(&esw->state_lock);
2006
2007 return 0;
2008 }
2009
__mlx5_eswitch_set_vport_vlan(struct mlx5_eswitch * esw,u16 vport,u16 vlan,u8 qos,u8 set_flags)2010 int __mlx5_eswitch_set_vport_vlan(struct mlx5_eswitch *esw,
2011 u16 vport, u16 vlan, u8 qos, u8 set_flags)
2012 {
2013 struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2014 int err = 0;
2015
2016 if (IS_ERR(evport))
2017 return PTR_ERR(evport);
2018 if (vlan > 4095 || qos > 7)
2019 return -EINVAL;
2020
2021 err = modify_esw_vport_cvlan(esw->dev, vport, vlan, qos, set_flags);
2022 if (err)
2023 return err;
2024
2025 evport->info.vlan = vlan;
2026 evport->info.qos = qos;
2027 if (evport->enabled && esw->mode == MLX5_ESWITCH_LEGACY) {
2028 err = esw_acl_ingress_lgcy_setup(esw, evport);
2029 if (err)
2030 return err;
2031 err = esw_acl_egress_lgcy_setup(esw, evport);
2032 }
2033
2034 return err;
2035 }
2036
mlx5_eswitch_set_vport_vlan(struct mlx5_eswitch * esw,u16 vport,u16 vlan,u8 qos)2037 int mlx5_eswitch_set_vport_vlan(struct mlx5_eswitch *esw,
2038 u16 vport, u16 vlan, u8 qos)
2039 {
2040 u8 set_flags = 0;
2041 int err;
2042
2043 if (!ESW_ALLOWED(esw))
2044 return -EPERM;
2045
2046 if (vlan || qos)
2047 set_flags = SET_VLAN_STRIP | SET_VLAN_INSERT;
2048
2049 mutex_lock(&esw->state_lock);
2050 err = __mlx5_eswitch_set_vport_vlan(esw, vport, vlan, qos, set_flags);
2051 mutex_unlock(&esw->state_lock);
2052
2053 return err;
2054 }
2055
mlx5_eswitch_set_vport_spoofchk(struct mlx5_eswitch * esw,u16 vport,bool spoofchk)2056 int mlx5_eswitch_set_vport_spoofchk(struct mlx5_eswitch *esw,
2057 u16 vport, bool spoofchk)
2058 {
2059 struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2060 bool pschk;
2061 int err = 0;
2062
2063 if (!ESW_ALLOWED(esw))
2064 return -EPERM;
2065 if (IS_ERR(evport))
2066 return PTR_ERR(evport);
2067
2068 mutex_lock(&esw->state_lock);
2069 pschk = evport->info.spoofchk;
2070 evport->info.spoofchk = spoofchk;
2071 if (pschk && !is_valid_ether_addr(evport->info.mac))
2072 mlx5_core_warn(esw->dev,
2073 "Spoofchk in set while MAC is invalid, vport(%d)\n",
2074 evport->vport);
2075 if (evport->enabled && esw->mode == MLX5_ESWITCH_LEGACY)
2076 err = esw_acl_ingress_lgcy_setup(esw, evport);
2077 if (err)
2078 evport->info.spoofchk = pschk;
2079 mutex_unlock(&esw->state_lock);
2080
2081 return err;
2082 }
2083
esw_cleanup_vepa_rules(struct mlx5_eswitch * esw)2084 static void esw_cleanup_vepa_rules(struct mlx5_eswitch *esw)
2085 {
2086 if (esw->fdb_table.legacy.vepa_uplink_rule)
2087 mlx5_del_flow_rules(esw->fdb_table.legacy.vepa_uplink_rule);
2088
2089 if (esw->fdb_table.legacy.vepa_star_rule)
2090 mlx5_del_flow_rules(esw->fdb_table.legacy.vepa_star_rule);
2091
2092 esw->fdb_table.legacy.vepa_uplink_rule = NULL;
2093 esw->fdb_table.legacy.vepa_star_rule = NULL;
2094 }
2095
_mlx5_eswitch_set_vepa_locked(struct mlx5_eswitch * esw,u8 setting)2096 static int _mlx5_eswitch_set_vepa_locked(struct mlx5_eswitch *esw,
2097 u8 setting)
2098 {
2099 struct mlx5_flow_destination dest = {};
2100 struct mlx5_flow_act flow_act = {};
2101 struct mlx5_flow_handle *flow_rule;
2102 struct mlx5_flow_spec *spec;
2103 int err = 0;
2104 void *misc;
2105
2106 if (!setting) {
2107 esw_cleanup_vepa_rules(esw);
2108 return 0;
2109 }
2110
2111 if (esw->fdb_table.legacy.vepa_uplink_rule)
2112 return 0;
2113
2114 spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
2115 if (!spec)
2116 return -ENOMEM;
2117
2118 /* Uplink rule forward uplink traffic to FDB */
2119 misc = MLX5_ADDR_OF(fte_match_param, spec->match_value, misc_parameters);
2120 MLX5_SET(fte_match_set_misc, misc, source_port, MLX5_VPORT_UPLINK);
2121
2122 misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria, misc_parameters);
2123 MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
2124
2125 spec->match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS;
2126 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
2127 dest.ft = esw->fdb_table.legacy.fdb;
2128 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2129 flow_rule = mlx5_add_flow_rules(esw->fdb_table.legacy.vepa_fdb, spec,
2130 &flow_act, &dest, 1);
2131 if (IS_ERR(flow_rule)) {
2132 err = PTR_ERR(flow_rule);
2133 goto out;
2134 } else {
2135 esw->fdb_table.legacy.vepa_uplink_rule = flow_rule;
2136 }
2137
2138 /* Star rule to forward all traffic to uplink vport */
2139 memset(&dest, 0, sizeof(dest));
2140 dest.type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
2141 dest.vport.num = MLX5_VPORT_UPLINK;
2142 flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2143 flow_rule = mlx5_add_flow_rules(esw->fdb_table.legacy.vepa_fdb, NULL,
2144 &flow_act, &dest, 1);
2145 if (IS_ERR(flow_rule)) {
2146 err = PTR_ERR(flow_rule);
2147 goto out;
2148 } else {
2149 esw->fdb_table.legacy.vepa_star_rule = flow_rule;
2150 }
2151
2152 out:
2153 kvfree(spec);
2154 if (err)
2155 esw_cleanup_vepa_rules(esw);
2156 return err;
2157 }
2158
mlx5_eswitch_set_vepa(struct mlx5_eswitch * esw,u8 setting)2159 int mlx5_eswitch_set_vepa(struct mlx5_eswitch *esw, u8 setting)
2160 {
2161 int err = 0;
2162
2163 if (!esw)
2164 return -EOPNOTSUPP;
2165
2166 if (!ESW_ALLOWED(esw))
2167 return -EPERM;
2168
2169 mutex_lock(&esw->state_lock);
2170 if (esw->mode != MLX5_ESWITCH_LEGACY) {
2171 err = -EOPNOTSUPP;
2172 goto out;
2173 }
2174
2175 err = _mlx5_eswitch_set_vepa_locked(esw, setting);
2176
2177 out:
2178 mutex_unlock(&esw->state_lock);
2179 return err;
2180 }
2181
mlx5_eswitch_get_vepa(struct mlx5_eswitch * esw,u8 * setting)2182 int mlx5_eswitch_get_vepa(struct mlx5_eswitch *esw, u8 *setting)
2183 {
2184 if (!esw)
2185 return -EOPNOTSUPP;
2186
2187 if (!ESW_ALLOWED(esw))
2188 return -EPERM;
2189
2190 if (esw->mode != MLX5_ESWITCH_LEGACY)
2191 return -EOPNOTSUPP;
2192
2193 *setting = esw->fdb_table.legacy.vepa_uplink_rule ? 1 : 0;
2194 return 0;
2195 }
2196
mlx5_eswitch_set_vport_trust(struct mlx5_eswitch * esw,u16 vport,bool setting)2197 int mlx5_eswitch_set_vport_trust(struct mlx5_eswitch *esw,
2198 u16 vport, bool setting)
2199 {
2200 struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2201
2202 if (!ESW_ALLOWED(esw))
2203 return -EPERM;
2204 if (IS_ERR(evport))
2205 return PTR_ERR(evport);
2206
2207 mutex_lock(&esw->state_lock);
2208 evport->info.trusted = setting;
2209 if (evport->enabled)
2210 esw_vport_change_handle_locked(evport);
2211 mutex_unlock(&esw->state_lock);
2212
2213 return 0;
2214 }
2215
calculate_vports_min_rate_divider(struct mlx5_eswitch * esw)2216 static u32 calculate_vports_min_rate_divider(struct mlx5_eswitch *esw)
2217 {
2218 u32 fw_max_bw_share = MLX5_CAP_QOS(esw->dev, max_tsar_bw_share);
2219 struct mlx5_vport *evport;
2220 u32 max_guarantee = 0;
2221 int i;
2222
2223 mlx5_esw_for_all_vports(esw, i, evport) {
2224 if (!evport->enabled || evport->info.min_rate < max_guarantee)
2225 continue;
2226 max_guarantee = evport->info.min_rate;
2227 }
2228
2229 if (max_guarantee)
2230 return max_t(u32, max_guarantee / fw_max_bw_share, 1);
2231 return 0;
2232 }
2233
normalize_vports_min_rate(struct mlx5_eswitch * esw)2234 static int normalize_vports_min_rate(struct mlx5_eswitch *esw)
2235 {
2236 u32 fw_max_bw_share = MLX5_CAP_QOS(esw->dev, max_tsar_bw_share);
2237 u32 divider = calculate_vports_min_rate_divider(esw);
2238 struct mlx5_vport *evport;
2239 u32 vport_max_rate;
2240 u32 vport_min_rate;
2241 u32 bw_share;
2242 int err;
2243 int i;
2244
2245 mlx5_esw_for_all_vports(esw, i, evport) {
2246 if (!evport->enabled)
2247 continue;
2248 vport_min_rate = evport->info.min_rate;
2249 vport_max_rate = evport->info.max_rate;
2250 bw_share = 0;
2251
2252 if (divider)
2253 bw_share = MLX5_RATE_TO_BW_SHARE(vport_min_rate,
2254 divider,
2255 fw_max_bw_share);
2256
2257 if (bw_share == evport->qos.bw_share)
2258 continue;
2259
2260 err = esw_vport_qos_config(esw, evport, vport_max_rate,
2261 bw_share);
2262 if (!err)
2263 evport->qos.bw_share = bw_share;
2264 else
2265 return err;
2266 }
2267
2268 return 0;
2269 }
2270
mlx5_eswitch_set_vport_rate(struct mlx5_eswitch * esw,u16 vport,u32 max_rate,u32 min_rate)2271 int mlx5_eswitch_set_vport_rate(struct mlx5_eswitch *esw, u16 vport,
2272 u32 max_rate, u32 min_rate)
2273 {
2274 struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2275 u32 fw_max_bw_share;
2276 u32 previous_min_rate;
2277 bool min_rate_supported;
2278 bool max_rate_supported;
2279 int err = 0;
2280
2281 if (!ESW_ALLOWED(esw))
2282 return -EPERM;
2283 if (IS_ERR(evport))
2284 return PTR_ERR(evport);
2285
2286 fw_max_bw_share = MLX5_CAP_QOS(esw->dev, max_tsar_bw_share);
2287 min_rate_supported = MLX5_CAP_QOS(esw->dev, esw_bw_share) &&
2288 fw_max_bw_share >= MLX5_MIN_BW_SHARE;
2289 max_rate_supported = MLX5_CAP_QOS(esw->dev, esw_rate_limit);
2290
2291 if ((min_rate && !min_rate_supported) || (max_rate && !max_rate_supported))
2292 return -EOPNOTSUPP;
2293
2294 mutex_lock(&esw->state_lock);
2295
2296 if (min_rate == evport->info.min_rate)
2297 goto set_max_rate;
2298
2299 previous_min_rate = evport->info.min_rate;
2300 evport->info.min_rate = min_rate;
2301 err = normalize_vports_min_rate(esw);
2302 if (err) {
2303 evport->info.min_rate = previous_min_rate;
2304 goto unlock;
2305 }
2306
2307 set_max_rate:
2308 if (max_rate == evport->info.max_rate)
2309 goto unlock;
2310
2311 err = esw_vport_qos_config(esw, evport, max_rate, evport->qos.bw_share);
2312 if (!err)
2313 evport->info.max_rate = max_rate;
2314
2315 unlock:
2316 mutex_unlock(&esw->state_lock);
2317 return err;
2318 }
2319
mlx5_eswitch_query_vport_drop_stats(struct mlx5_core_dev * dev,struct mlx5_vport * vport,struct mlx5_vport_drop_stats * stats)2320 static int mlx5_eswitch_query_vport_drop_stats(struct mlx5_core_dev *dev,
2321 struct mlx5_vport *vport,
2322 struct mlx5_vport_drop_stats *stats)
2323 {
2324 struct mlx5_eswitch *esw = dev->priv.eswitch;
2325 u64 rx_discard_vport_down, tx_discard_vport_down;
2326 u64 bytes = 0;
2327 int err = 0;
2328
2329 if (esw->mode != MLX5_ESWITCH_LEGACY)
2330 return 0;
2331
2332 mutex_lock(&esw->state_lock);
2333 if (!vport->enabled)
2334 goto unlock;
2335
2336 if (!IS_ERR_OR_NULL(vport->egress.legacy.drop_counter))
2337 mlx5_fc_query(dev, vport->egress.legacy.drop_counter,
2338 &stats->rx_dropped, &bytes);
2339
2340 if (vport->ingress.legacy.drop_counter)
2341 mlx5_fc_query(dev, vport->ingress.legacy.drop_counter,
2342 &stats->tx_dropped, &bytes);
2343
2344 if (!MLX5_CAP_GEN(dev, receive_discard_vport_down) &&
2345 !MLX5_CAP_GEN(dev, transmit_discard_vport_down))
2346 goto unlock;
2347
2348 err = mlx5_query_vport_down_stats(dev, vport->vport, 1,
2349 &rx_discard_vport_down,
2350 &tx_discard_vport_down);
2351 if (err)
2352 goto unlock;
2353
2354 if (MLX5_CAP_GEN(dev, receive_discard_vport_down))
2355 stats->rx_dropped += rx_discard_vport_down;
2356 if (MLX5_CAP_GEN(dev, transmit_discard_vport_down))
2357 stats->tx_dropped += tx_discard_vport_down;
2358
2359 unlock:
2360 mutex_unlock(&esw->state_lock);
2361 return err;
2362 }
2363
mlx5_eswitch_get_vport_stats(struct mlx5_eswitch * esw,u16 vport_num,struct ifla_vf_stats * vf_stats)2364 int mlx5_eswitch_get_vport_stats(struct mlx5_eswitch *esw,
2365 u16 vport_num,
2366 struct ifla_vf_stats *vf_stats)
2367 {
2368 struct mlx5_vport *vport = mlx5_eswitch_get_vport(esw, vport_num);
2369 int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
2370 u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)] = {};
2371 struct mlx5_vport_drop_stats stats = {};
2372 int err = 0;
2373 u32 *out;
2374
2375 if (IS_ERR(vport))
2376 return PTR_ERR(vport);
2377
2378 out = kvzalloc(outlen, GFP_KERNEL);
2379 if (!out)
2380 return -ENOMEM;
2381
2382 MLX5_SET(query_vport_counter_in, in, opcode,
2383 MLX5_CMD_OP_QUERY_VPORT_COUNTER);
2384 MLX5_SET(query_vport_counter_in, in, op_mod, 0);
2385 MLX5_SET(query_vport_counter_in, in, vport_number, vport->vport);
2386 MLX5_SET(query_vport_counter_in, in, other_vport, 1);
2387
2388 err = mlx5_cmd_exec_inout(esw->dev, query_vport_counter, in, out);
2389 if (err)
2390 goto free_out;
2391
2392 #define MLX5_GET_CTR(p, x) \
2393 MLX5_GET64(query_vport_counter_out, p, x)
2394
2395 memset(vf_stats, 0, sizeof(*vf_stats));
2396 vf_stats->rx_packets =
2397 MLX5_GET_CTR(out, received_eth_unicast.packets) +
2398 MLX5_GET_CTR(out, received_ib_unicast.packets) +
2399 MLX5_GET_CTR(out, received_eth_multicast.packets) +
2400 MLX5_GET_CTR(out, received_ib_multicast.packets) +
2401 MLX5_GET_CTR(out, received_eth_broadcast.packets);
2402
2403 vf_stats->rx_bytes =
2404 MLX5_GET_CTR(out, received_eth_unicast.octets) +
2405 MLX5_GET_CTR(out, received_ib_unicast.octets) +
2406 MLX5_GET_CTR(out, received_eth_multicast.octets) +
2407 MLX5_GET_CTR(out, received_ib_multicast.octets) +
2408 MLX5_GET_CTR(out, received_eth_broadcast.octets);
2409
2410 vf_stats->tx_packets =
2411 MLX5_GET_CTR(out, transmitted_eth_unicast.packets) +
2412 MLX5_GET_CTR(out, transmitted_ib_unicast.packets) +
2413 MLX5_GET_CTR(out, transmitted_eth_multicast.packets) +
2414 MLX5_GET_CTR(out, transmitted_ib_multicast.packets) +
2415 MLX5_GET_CTR(out, transmitted_eth_broadcast.packets);
2416
2417 vf_stats->tx_bytes =
2418 MLX5_GET_CTR(out, transmitted_eth_unicast.octets) +
2419 MLX5_GET_CTR(out, transmitted_ib_unicast.octets) +
2420 MLX5_GET_CTR(out, transmitted_eth_multicast.octets) +
2421 MLX5_GET_CTR(out, transmitted_ib_multicast.octets) +
2422 MLX5_GET_CTR(out, transmitted_eth_broadcast.octets);
2423
2424 vf_stats->multicast =
2425 MLX5_GET_CTR(out, received_eth_multicast.packets) +
2426 MLX5_GET_CTR(out, received_ib_multicast.packets);
2427
2428 vf_stats->broadcast =
2429 MLX5_GET_CTR(out, received_eth_broadcast.packets);
2430
2431 err = mlx5_eswitch_query_vport_drop_stats(esw->dev, vport, &stats);
2432 if (err)
2433 goto free_out;
2434 vf_stats->rx_dropped = stats.rx_dropped;
2435 vf_stats->tx_dropped = stats.tx_dropped;
2436
2437 free_out:
2438 kvfree(out);
2439 return err;
2440 }
2441
mlx5_eswitch_mode(struct mlx5_eswitch * esw)2442 u8 mlx5_eswitch_mode(struct mlx5_eswitch *esw)
2443 {
2444 return ESW_ALLOWED(esw) ? esw->mode : MLX5_ESWITCH_NONE;
2445 }
2446 EXPORT_SYMBOL_GPL(mlx5_eswitch_mode);
2447
2448 enum devlink_eswitch_encap_mode
mlx5_eswitch_get_encap_mode(const struct mlx5_core_dev * dev)2449 mlx5_eswitch_get_encap_mode(const struct mlx5_core_dev *dev)
2450 {
2451 struct mlx5_eswitch *esw;
2452
2453 esw = dev->priv.eswitch;
2454 return ESW_ALLOWED(esw) ? esw->offloads.encap :
2455 DEVLINK_ESWITCH_ENCAP_MODE_NONE;
2456 }
2457 EXPORT_SYMBOL(mlx5_eswitch_get_encap_mode);
2458
mlx5_esw_lag_prereq(struct mlx5_core_dev * dev0,struct mlx5_core_dev * dev1)2459 bool mlx5_esw_lag_prereq(struct mlx5_core_dev *dev0, struct mlx5_core_dev *dev1)
2460 {
2461 if ((dev0->priv.eswitch->mode == MLX5_ESWITCH_NONE &&
2462 dev1->priv.eswitch->mode == MLX5_ESWITCH_NONE) ||
2463 (dev0->priv.eswitch->mode == MLX5_ESWITCH_OFFLOADS &&
2464 dev1->priv.eswitch->mode == MLX5_ESWITCH_OFFLOADS))
2465 return true;
2466
2467 return false;
2468 }
2469
mlx5_esw_multipath_prereq(struct mlx5_core_dev * dev0,struct mlx5_core_dev * dev1)2470 bool mlx5_esw_multipath_prereq(struct mlx5_core_dev *dev0,
2471 struct mlx5_core_dev *dev1)
2472 {
2473 return (dev0->priv.eswitch->mode == MLX5_ESWITCH_OFFLOADS &&
2474 dev1->priv.eswitch->mode == MLX5_ESWITCH_OFFLOADS);
2475 }
2476
2477
2478