xref: /linux/drivers/net/ethernet/mellanox/mlx5/core/eswitch.c (revision bf4afc53b77aeaa48b5409da5c8da6bb4eff7f43)
1 /*
2  * Copyright (c) 2015, Mellanox Technologies. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #include <linux/etherdevice.h>
34 #include <linux/debugfs.h>
35 #include <linux/mlx5/driver.h>
36 #include <linux/mlx5/mlx5_ifc.h>
37 #include <linux/mlx5/vport.h>
38 #include <linux/mlx5/fs.h>
39 #include <linux/mlx5/mpfs.h>
40 #include "esw/acl/lgcy.h"
41 #include "esw/legacy.h"
42 #include "esw/qos.h"
43 #include "mlx5_core.h"
44 #include "lib/eq.h"
45 #include "lag/lag.h"
46 #include "eswitch.h"
47 #include "fs_core.h"
48 #include "devlink.h"
49 #include "ecpf.h"
50 #include "en/mod_hdr.h"
51 #include "en_accel/ipsec.h"
52 
53 enum {
54 	MLX5_ACTION_NONE = 0,
55 	MLX5_ACTION_ADD  = 1,
56 	MLX5_ACTION_DEL  = 2,
57 };
58 
59 /* Vport UC/MC hash node */
60 struct vport_addr {
61 	struct l2addr_node     node;
62 	u8                     action;
63 	u16                    vport;
64 	struct mlx5_flow_handle *flow_rule;
65 	bool mpfs; /* UC MAC was added to MPFs */
66 	/* A flag indicating that mac was added due to mc promiscuous vport */
67 	bool mc_promisc;
68 };
69 
mlx5_eswitch_check(const struct mlx5_core_dev * dev)70 static int mlx5_eswitch_check(const struct mlx5_core_dev *dev)
71 {
72 	if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
73 		return -EOPNOTSUPP;
74 
75 	if (!MLX5_ESWITCH_MANAGER(dev))
76 		return -EOPNOTSUPP;
77 
78 	return 0;
79 }
80 
__mlx5_devlink_eswitch_get(struct devlink * devlink,bool check)81 static struct mlx5_eswitch *__mlx5_devlink_eswitch_get(struct devlink *devlink, bool check)
82 {
83 	struct mlx5_core_dev *dev = devlink_priv(devlink);
84 	int err;
85 
86 	if (check) {
87 		err = mlx5_eswitch_check(dev);
88 		if (err)
89 			return ERR_PTR(err);
90 	}
91 
92 	return dev->priv.eswitch;
93 }
94 
95 struct mlx5_eswitch *__must_check
mlx5_devlink_eswitch_get(struct devlink * devlink)96 mlx5_devlink_eswitch_get(struct devlink *devlink)
97 {
98 	return __mlx5_devlink_eswitch_get(devlink, true);
99 }
100 
mlx5_devlink_eswitch_nocheck_get(struct devlink * devlink)101 struct mlx5_eswitch *mlx5_devlink_eswitch_nocheck_get(struct devlink *devlink)
102 {
103 	return __mlx5_devlink_eswitch_get(devlink, false);
104 }
105 
106 struct mlx5_vport *__must_check
mlx5_eswitch_get_vport(struct mlx5_eswitch * esw,u16 vport_num)107 mlx5_eswitch_get_vport(struct mlx5_eswitch *esw, u16 vport_num)
108 {
109 	struct mlx5_vport *vport;
110 
111 	if (!esw)
112 		return ERR_PTR(-EPERM);
113 
114 	vport = xa_load(&esw->vports, vport_num);
115 	if (!vport) {
116 		esw_debug(esw->dev, "vport out of range: num(0x%x)\n", vport_num);
117 		return ERR_PTR(-EINVAL);
118 	}
119 	return vport;
120 }
121 
arm_vport_context_events_cmd(struct mlx5_core_dev * dev,u16 vport,u32 events_mask)122 static int arm_vport_context_events_cmd(struct mlx5_core_dev *dev, u16 vport,
123 					u32 events_mask)
124 {
125 	u32 in[MLX5_ST_SZ_DW(modify_nic_vport_context_in)] = {};
126 	void *nic_vport_ctx;
127 
128 	MLX5_SET(modify_nic_vport_context_in, in,
129 		 opcode, MLX5_CMD_OP_MODIFY_NIC_VPORT_CONTEXT);
130 	MLX5_SET(modify_nic_vport_context_in, in, field_select.change_event, 1);
131 	MLX5_SET(modify_nic_vport_context_in, in, vport_number, vport);
132 	if (vport || mlx5_core_is_ecpf(dev))
133 		MLX5_SET(modify_nic_vport_context_in, in, other_vport, 1);
134 	nic_vport_ctx = MLX5_ADDR_OF(modify_nic_vport_context_in,
135 				     in, nic_vport_context);
136 
137 	MLX5_SET(nic_vport_context, nic_vport_ctx, arm_change_event, 1);
138 
139 	if (events_mask & MLX5_VPORT_UC_ADDR_CHANGE)
140 		MLX5_SET(nic_vport_context, nic_vport_ctx,
141 			 event_on_uc_address_change, 1);
142 	if (events_mask & MLX5_VPORT_MC_ADDR_CHANGE)
143 		MLX5_SET(nic_vport_context, nic_vport_ctx,
144 			 event_on_mc_address_change, 1);
145 	if (events_mask & MLX5_VPORT_PROMISC_CHANGE)
146 		MLX5_SET(nic_vport_context, nic_vport_ctx,
147 			 event_on_promisc_change, 1);
148 
149 	return mlx5_cmd_exec_in(dev, modify_nic_vport_context, in);
150 }
151 
152 /* E-Switch vport context HW commands */
mlx5_eswitch_modify_esw_vport_context(struct mlx5_core_dev * dev,u16 vport,bool other_vport,void * in)153 int mlx5_eswitch_modify_esw_vport_context(struct mlx5_core_dev *dev, u16 vport,
154 					  bool other_vport, void *in)
155 {
156 	MLX5_SET(modify_esw_vport_context_in, in, opcode,
157 		 MLX5_CMD_OP_MODIFY_ESW_VPORT_CONTEXT);
158 	MLX5_SET(modify_esw_vport_context_in, in, vport_number, vport);
159 	MLX5_SET(modify_esw_vport_context_in, in, other_vport, other_vport);
160 	return mlx5_cmd_exec_in(dev, modify_esw_vport_context, in);
161 }
162 
modify_esw_vport_cvlan(struct mlx5_core_dev * dev,u16 vport,u16 vlan,u8 qos,u8 set_flags)163 static int modify_esw_vport_cvlan(struct mlx5_core_dev *dev, u16 vport,
164 				  u16 vlan, u8 qos, u8 set_flags)
165 {
166 	u32 in[MLX5_ST_SZ_DW(modify_esw_vport_context_in)] = {};
167 
168 	if (!MLX5_CAP_ESW(dev, vport_cvlan_strip) ||
169 	    !MLX5_CAP_ESW(dev, vport_cvlan_insert_if_not_exist))
170 		return -EOPNOTSUPP;
171 
172 	esw_debug(dev, "Set Vport[%d] VLAN %d qos %d set=%x\n",
173 		  vport, vlan, qos, set_flags);
174 
175 	if (set_flags & SET_VLAN_STRIP)
176 		MLX5_SET(modify_esw_vport_context_in, in,
177 			 esw_vport_context.vport_cvlan_strip, 1);
178 
179 	if (set_flags & SET_VLAN_INSERT) {
180 		if (MLX5_CAP_ESW(dev, vport_cvlan_insert_always)) {
181 			/* insert either if vlan exist in packet or not */
182 			MLX5_SET(modify_esw_vport_context_in, in,
183 				 esw_vport_context.vport_cvlan_insert,
184 				 MLX5_VPORT_CVLAN_INSERT_ALWAYS);
185 		} else {
186 			/* insert only if no vlan in packet */
187 			MLX5_SET(modify_esw_vport_context_in, in,
188 				 esw_vport_context.vport_cvlan_insert,
189 				 MLX5_VPORT_CVLAN_INSERT_WHEN_NO_CVLAN);
190 		}
191 		MLX5_SET(modify_esw_vport_context_in, in,
192 			 esw_vport_context.cvlan_pcp, qos);
193 		MLX5_SET(modify_esw_vport_context_in, in,
194 			 esw_vport_context.cvlan_id, vlan);
195 	}
196 
197 	MLX5_SET(modify_esw_vport_context_in, in,
198 		 field_select.vport_cvlan_strip, 1);
199 	MLX5_SET(modify_esw_vport_context_in, in,
200 		 field_select.vport_cvlan_insert, 1);
201 
202 	return mlx5_eswitch_modify_esw_vport_context(dev, vport, true, in);
203 }
204 
205 /* E-Switch FDB */
206 static struct mlx5_flow_handle *
__esw_fdb_set_vport_rule(struct mlx5_eswitch * esw,u16 vport,bool rx_rule,u8 mac_c[ETH_ALEN],u8 mac_v[ETH_ALEN])207 __esw_fdb_set_vport_rule(struct mlx5_eswitch *esw, u16 vport, bool rx_rule,
208 			 u8 mac_c[ETH_ALEN], u8 mac_v[ETH_ALEN])
209 {
210 	int match_header = (is_zero_ether_addr(mac_c) ? 0 :
211 			    MLX5_MATCH_OUTER_HEADERS);
212 	struct mlx5_flow_handle *flow_rule = NULL;
213 	struct mlx5_flow_act flow_act = {0};
214 	struct mlx5_flow_destination dest = {};
215 	struct mlx5_flow_spec *spec;
216 	void *mv_misc = NULL;
217 	void *mc_misc = NULL;
218 	u8 *dmac_v = NULL;
219 	u8 *dmac_c = NULL;
220 
221 	if (rx_rule)
222 		match_header |= MLX5_MATCH_MISC_PARAMETERS;
223 
224 	spec = kvzalloc_obj(*spec);
225 	if (!spec)
226 		return NULL;
227 
228 	dmac_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
229 			      outer_headers.dmac_47_16);
230 	dmac_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
231 			      outer_headers.dmac_47_16);
232 
233 	if (match_header & MLX5_MATCH_OUTER_HEADERS) {
234 		ether_addr_copy(dmac_v, mac_v);
235 		ether_addr_copy(dmac_c, mac_c);
236 	}
237 
238 	if (match_header & MLX5_MATCH_MISC_PARAMETERS) {
239 		mv_misc  = MLX5_ADDR_OF(fte_match_param, spec->match_value,
240 					misc_parameters);
241 		mc_misc  = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
242 					misc_parameters);
243 		MLX5_SET(fte_match_set_misc, mv_misc, source_port, MLX5_VPORT_UPLINK);
244 		MLX5_SET_TO_ONES(fte_match_set_misc, mc_misc, source_port);
245 	}
246 
247 	dest.type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
248 	dest.vport.num = vport;
249 
250 	esw_debug(esw->dev,
251 		  "\tFDB add rule dmac_v(%pM) dmac_c(%pM) -> vport(%d)\n",
252 		  dmac_v, dmac_c, vport);
253 	spec->match_criteria_enable = match_header;
254 	flow_act.action =  MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
255 	flow_rule =
256 		mlx5_add_flow_rules(esw->fdb_table.legacy.fdb, spec,
257 				    &flow_act, &dest, 1);
258 	if (IS_ERR(flow_rule)) {
259 		esw_warn(esw->dev,
260 			 "FDB: Failed to add flow rule: dmac_v(%pM) dmac_c(%pM) -> vport(%d), err(%pe)\n",
261 			 dmac_v, dmac_c, vport, flow_rule);
262 		flow_rule = NULL;
263 	}
264 
265 	kvfree(spec);
266 	return flow_rule;
267 }
268 
269 static struct mlx5_flow_handle *
esw_fdb_set_vport_rule(struct mlx5_eswitch * esw,u8 mac[ETH_ALEN],u16 vport)270 esw_fdb_set_vport_rule(struct mlx5_eswitch *esw, u8 mac[ETH_ALEN], u16 vport)
271 {
272 	u8 mac_c[ETH_ALEN];
273 
274 	eth_broadcast_addr(mac_c);
275 	return __esw_fdb_set_vport_rule(esw, vport, false, mac_c, mac);
276 }
277 
278 static struct mlx5_flow_handle *
esw_fdb_set_vport_allmulti_rule(struct mlx5_eswitch * esw,u16 vport)279 esw_fdb_set_vport_allmulti_rule(struct mlx5_eswitch *esw, u16 vport)
280 {
281 	u8 mac_c[ETH_ALEN];
282 	u8 mac_v[ETH_ALEN];
283 
284 	eth_zero_addr(mac_c);
285 	eth_zero_addr(mac_v);
286 	mac_c[0] = 0x01;
287 	mac_v[0] = 0x01;
288 	return __esw_fdb_set_vport_rule(esw, vport, false, mac_c, mac_v);
289 }
290 
291 static struct mlx5_flow_handle *
esw_fdb_set_vport_promisc_rule(struct mlx5_eswitch * esw,u16 vport)292 esw_fdb_set_vport_promisc_rule(struct mlx5_eswitch *esw, u16 vport)
293 {
294 	u8 mac_c[ETH_ALEN];
295 	u8 mac_v[ETH_ALEN];
296 
297 	eth_zero_addr(mac_c);
298 	eth_zero_addr(mac_v);
299 	return __esw_fdb_set_vport_rule(esw, vport, true, mac_c, mac_v);
300 }
301 
302 /* E-Switch vport UC/MC lists management */
303 typedef int (*vport_addr_action)(struct mlx5_eswitch *esw,
304 				 struct vport_addr *vaddr);
305 
esw_add_uc_addr(struct mlx5_eswitch * esw,struct vport_addr * vaddr)306 static int esw_add_uc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
307 {
308 	u8 *mac = vaddr->node.addr;
309 	u16 vport = vaddr->vport;
310 	int err;
311 
312 	/* Skip mlx5_mpfs_add_mac for eswitch_managers,
313 	 * it is already done by its netdev in mlx5e_execute_l2_action
314 	 */
315 	if (mlx5_esw_is_manager_vport(esw, vport))
316 		goto fdb_add;
317 
318 	err = mlx5_mpfs_add_mac(esw->dev, mac);
319 	if (err) {
320 		esw_warn(esw->dev,
321 			 "Failed to add L2 table mac(%pM) for vport(0x%x), err(%d)\n",
322 			 mac, vport, err);
323 		return err;
324 	}
325 	vaddr->mpfs = true;
326 
327 fdb_add:
328 	/* SRIOV is enabled: Forward UC MAC to vport */
329 	if (esw->fdb_table.legacy.fdb && esw->mode == MLX5_ESWITCH_LEGACY) {
330 		vaddr->flow_rule = esw_fdb_set_vport_rule(esw, mac, vport);
331 
332 		esw_debug(esw->dev, "\tADDED UC MAC: vport[%d] %pM fr(%p)\n",
333 			  vport, mac, vaddr->flow_rule);
334 	}
335 
336 	return 0;
337 }
338 
esw_del_uc_addr(struct mlx5_eswitch * esw,struct vport_addr * vaddr)339 static int esw_del_uc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
340 {
341 	u8 *mac = vaddr->node.addr;
342 	u16 vport = vaddr->vport;
343 	int err = 0;
344 
345 	/* Skip mlx5_mpfs_del_mac for eswitch managers,
346 	 * it is already done by its netdev in mlx5e_execute_l2_action
347 	 */
348 	if (!vaddr->mpfs || mlx5_esw_is_manager_vport(esw, vport))
349 		goto fdb_del;
350 
351 	err = mlx5_mpfs_del_mac(esw->dev, mac);
352 	if (err)
353 		esw_warn(esw->dev,
354 			 "Failed to del L2 table mac(%pM) for vport(%d), err(%d)\n",
355 			 mac, vport, err);
356 	vaddr->mpfs = false;
357 
358 fdb_del:
359 	if (vaddr->flow_rule)
360 		mlx5_del_flow_rules(vaddr->flow_rule);
361 	vaddr->flow_rule = NULL;
362 
363 	return 0;
364 }
365 
update_allmulti_vports(struct mlx5_eswitch * esw,struct vport_addr * vaddr,struct esw_mc_addr * esw_mc)366 static void update_allmulti_vports(struct mlx5_eswitch *esw,
367 				   struct vport_addr *vaddr,
368 				   struct esw_mc_addr *esw_mc)
369 {
370 	u8 *mac = vaddr->node.addr;
371 	struct mlx5_vport *vport;
372 	unsigned long i;
373 	u16 vport_num;
374 
375 	mlx5_esw_for_each_vport(esw, i, vport) {
376 		struct hlist_head *vport_hash = vport->mc_list;
377 		struct vport_addr *iter_vaddr =
378 					l2addr_hash_find(vport_hash,
379 							 mac,
380 							 struct vport_addr);
381 		vport_num = vport->vport;
382 		if (IS_ERR_OR_NULL(vport->allmulti_rule) ||
383 		    vaddr->vport == vport_num)
384 			continue;
385 		switch (vaddr->action) {
386 		case MLX5_ACTION_ADD:
387 			if (iter_vaddr)
388 				continue;
389 			iter_vaddr = l2addr_hash_add(vport_hash, mac,
390 						     struct vport_addr,
391 						     GFP_KERNEL);
392 			if (!iter_vaddr) {
393 				esw_warn(esw->dev,
394 					 "ALL-MULTI: Failed to add MAC(%pM) to vport[%d] DB\n",
395 					 mac, vport_num);
396 				continue;
397 			}
398 			iter_vaddr->vport = vport_num;
399 			iter_vaddr->flow_rule =
400 					esw_fdb_set_vport_rule(esw,
401 							       mac,
402 							       vport_num);
403 			iter_vaddr->mc_promisc = true;
404 			break;
405 		case MLX5_ACTION_DEL:
406 			if (!iter_vaddr)
407 				continue;
408 			mlx5_del_flow_rules(iter_vaddr->flow_rule);
409 			l2addr_hash_del(iter_vaddr);
410 			break;
411 		}
412 	}
413 }
414 
esw_add_mc_addr(struct mlx5_eswitch * esw,struct vport_addr * vaddr)415 static int esw_add_mc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
416 {
417 	struct hlist_head *hash = esw->mc_table;
418 	struct esw_mc_addr *esw_mc;
419 	u8 *mac = vaddr->node.addr;
420 	u16 vport = vaddr->vport;
421 
422 	if (!esw->fdb_table.legacy.fdb)
423 		return 0;
424 
425 	esw_mc = l2addr_hash_find(hash, mac, struct esw_mc_addr);
426 	if (esw_mc)
427 		goto add;
428 
429 	esw_mc = l2addr_hash_add(hash, mac, struct esw_mc_addr, GFP_KERNEL);
430 	if (!esw_mc)
431 		return -ENOMEM;
432 
433 	esw_mc->uplink_rule = /* Forward MC MAC to Uplink */
434 		esw_fdb_set_vport_rule(esw, mac, MLX5_VPORT_UPLINK);
435 
436 	/* Add this multicast mac to all the mc promiscuous vports */
437 	update_allmulti_vports(esw, vaddr, esw_mc);
438 
439 add:
440 	/* If the multicast mac is added as a result of mc promiscuous vport,
441 	 * don't increment the multicast ref count
442 	 */
443 	if (!vaddr->mc_promisc)
444 		esw_mc->refcnt++;
445 
446 	/* Forward MC MAC to vport */
447 	vaddr->flow_rule = esw_fdb_set_vport_rule(esw, mac, vport);
448 	esw_debug(esw->dev,
449 		  "\tADDED MC MAC: vport[%d] %pM fr(%p) refcnt(%d) uplinkfr(%p)\n",
450 		  vport, mac, vaddr->flow_rule,
451 		  esw_mc->refcnt, esw_mc->uplink_rule);
452 	return 0;
453 }
454 
esw_del_mc_addr(struct mlx5_eswitch * esw,struct vport_addr * vaddr)455 static int esw_del_mc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
456 {
457 	struct hlist_head *hash = esw->mc_table;
458 	struct esw_mc_addr *esw_mc;
459 	u8 *mac = vaddr->node.addr;
460 	u16 vport = vaddr->vport;
461 
462 	if (!esw->fdb_table.legacy.fdb)
463 		return 0;
464 
465 	esw_mc = l2addr_hash_find(hash, mac, struct esw_mc_addr);
466 	if (!esw_mc) {
467 		esw_warn(esw->dev,
468 			 "Failed to find eswitch MC addr for MAC(%pM) vport(%d)",
469 			 mac, vport);
470 		return -EINVAL;
471 	}
472 	esw_debug(esw->dev,
473 		  "\tDELETE MC MAC: vport[%d] %pM fr(%p) refcnt(%d) uplinkfr(%p)\n",
474 		  vport, mac, vaddr->flow_rule, esw_mc->refcnt,
475 		  esw_mc->uplink_rule);
476 
477 	if (vaddr->flow_rule)
478 		mlx5_del_flow_rules(vaddr->flow_rule);
479 	vaddr->flow_rule = NULL;
480 
481 	/* If the multicast mac is added as a result of mc promiscuous vport,
482 	 * don't decrement the multicast ref count.
483 	 */
484 	if (vaddr->mc_promisc || (--esw_mc->refcnt > 0))
485 		return 0;
486 
487 	/* Remove this multicast mac from all the mc promiscuous vports */
488 	update_allmulti_vports(esw, vaddr, esw_mc);
489 
490 	if (esw_mc->uplink_rule)
491 		mlx5_del_flow_rules(esw_mc->uplink_rule);
492 
493 	l2addr_hash_del(esw_mc);
494 	return 0;
495 }
496 
497 /* Apply vport UC/MC list to HW l2 table and FDB table */
esw_apply_vport_addr_list(struct mlx5_eswitch * esw,struct mlx5_vport * vport,int list_type)498 static void esw_apply_vport_addr_list(struct mlx5_eswitch *esw,
499 				      struct mlx5_vport *vport, int list_type)
500 {
501 	bool is_uc = list_type == MLX5_NVPRT_LIST_TYPE_UC;
502 	vport_addr_action vport_addr_add;
503 	vport_addr_action vport_addr_del;
504 	struct vport_addr *addr;
505 	struct l2addr_node *node;
506 	struct hlist_head *hash;
507 	struct hlist_node *tmp;
508 	int hi;
509 
510 	vport_addr_add = is_uc ? esw_add_uc_addr :
511 				 esw_add_mc_addr;
512 	vport_addr_del = is_uc ? esw_del_uc_addr :
513 				 esw_del_mc_addr;
514 
515 	hash = is_uc ? vport->uc_list : vport->mc_list;
516 	for_each_l2hash_node(node, tmp, hash, hi) {
517 		addr = container_of(node, struct vport_addr, node);
518 		switch (addr->action) {
519 		case MLX5_ACTION_ADD:
520 			vport_addr_add(esw, addr);
521 			addr->action = MLX5_ACTION_NONE;
522 			break;
523 		case MLX5_ACTION_DEL:
524 			vport_addr_del(esw, addr);
525 			l2addr_hash_del(addr);
526 			break;
527 		}
528 	}
529 }
530 
531 /* Sync vport UC/MC list from vport context */
esw_update_vport_addr_list(struct mlx5_eswitch * esw,struct mlx5_vport * vport,int list_type)532 static void esw_update_vport_addr_list(struct mlx5_eswitch *esw,
533 				       struct mlx5_vport *vport, int list_type)
534 {
535 	bool is_uc = list_type == MLX5_NVPRT_LIST_TYPE_UC;
536 	u8 (*mac_list)[ETH_ALEN];
537 	struct l2addr_node *node;
538 	struct vport_addr *addr;
539 	struct hlist_head *hash;
540 	struct hlist_node *tmp;
541 	int size;
542 	int err;
543 	int hi;
544 	int i;
545 
546 	size = is_uc ? MLX5_MAX_UC_PER_VPORT(esw->dev) :
547 		       MLX5_MAX_MC_PER_VPORT(esw->dev);
548 
549 	mac_list = kcalloc(size, ETH_ALEN, GFP_KERNEL);
550 	if (!mac_list)
551 		return;
552 
553 	hash = is_uc ? vport->uc_list : vport->mc_list;
554 
555 	for_each_l2hash_node(node, tmp, hash, hi) {
556 		addr = container_of(node, struct vport_addr, node);
557 		addr->action = MLX5_ACTION_DEL;
558 	}
559 
560 	if (!vport->enabled)
561 		goto out;
562 
563 	err = mlx5_query_nic_vport_mac_list(esw->dev, vport->vport, list_type,
564 					    mac_list, &size);
565 	if (err)
566 		goto out;
567 	esw_debug(esw->dev, "vport[%d] context update %s list size (%d)\n",
568 		  vport->vport, is_uc ? "UC" : "MC", size);
569 
570 	for (i = 0; i < size; i++) {
571 		if (is_uc && !is_valid_ether_addr(mac_list[i]))
572 			continue;
573 
574 		if (!is_uc && !is_multicast_ether_addr(mac_list[i]))
575 			continue;
576 
577 		addr = l2addr_hash_find(hash, mac_list[i], struct vport_addr);
578 		if (addr) {
579 			addr->action = MLX5_ACTION_NONE;
580 			/* If this mac was previously added because of allmulti
581 			 * promiscuous rx mode, its now converted to be original
582 			 * vport mac.
583 			 */
584 			if (addr->mc_promisc) {
585 				struct esw_mc_addr *esw_mc =
586 					l2addr_hash_find(esw->mc_table,
587 							 mac_list[i],
588 							 struct esw_mc_addr);
589 				if (!esw_mc) {
590 					esw_warn(esw->dev,
591 						 "Failed to MAC(%pM) in mcast DB\n",
592 						 mac_list[i]);
593 					continue;
594 				}
595 				esw_mc->refcnt++;
596 				addr->mc_promisc = false;
597 			}
598 			continue;
599 		}
600 
601 		addr = l2addr_hash_add(hash, mac_list[i], struct vport_addr,
602 				       GFP_KERNEL);
603 		if (!addr) {
604 			esw_warn(esw->dev,
605 				 "Failed to add MAC(%pM) to vport[%d] DB\n",
606 				 mac_list[i], vport->vport);
607 			continue;
608 		}
609 		addr->vport = vport->vport;
610 		addr->action = MLX5_ACTION_ADD;
611 	}
612 out:
613 	kfree(mac_list);
614 }
615 
616 /* Sync vport UC/MC list from vport context
617  * Must be called after esw_update_vport_addr_list
618  */
esw_update_vport_mc_promisc(struct mlx5_eswitch * esw,struct mlx5_vport * vport)619 static void esw_update_vport_mc_promisc(struct mlx5_eswitch *esw,
620 					struct mlx5_vport *vport)
621 {
622 	struct l2addr_node *node;
623 	struct vport_addr *addr;
624 	struct hlist_head *hash;
625 	struct hlist_node *tmp;
626 	int hi;
627 
628 	hash = vport->mc_list;
629 
630 	for_each_l2hash_node(node, tmp, esw->mc_table, hi) {
631 		u8 *mac = node->addr;
632 
633 		addr = l2addr_hash_find(hash, mac, struct vport_addr);
634 		if (addr) {
635 			if (addr->action == MLX5_ACTION_DEL)
636 				addr->action = MLX5_ACTION_NONE;
637 			continue;
638 		}
639 		addr = l2addr_hash_add(hash, mac, struct vport_addr,
640 				       GFP_KERNEL);
641 		if (!addr) {
642 			esw_warn(esw->dev,
643 				 "Failed to add allmulti MAC(%pM) to vport[%d] DB\n",
644 				 mac, vport->vport);
645 			continue;
646 		}
647 		addr->vport = vport->vport;
648 		addr->action = MLX5_ACTION_ADD;
649 		addr->mc_promisc = true;
650 	}
651 }
652 
653 /* 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)654 static void esw_apply_vport_rx_mode(struct mlx5_eswitch *esw,
655 				    struct mlx5_vport *vport,
656 				    bool promisc, bool mc_promisc)
657 {
658 	struct esw_mc_addr *allmulti_addr = &esw->mc_promisc;
659 
660 	if (IS_ERR_OR_NULL(vport->allmulti_rule) != mc_promisc)
661 		goto promisc;
662 
663 	if (mc_promisc) {
664 		vport->allmulti_rule =
665 			esw_fdb_set_vport_allmulti_rule(esw, vport->vport);
666 		if (!allmulti_addr->uplink_rule)
667 			allmulti_addr->uplink_rule =
668 				esw_fdb_set_vport_allmulti_rule(esw,
669 								MLX5_VPORT_UPLINK);
670 		allmulti_addr->refcnt++;
671 	} else if (vport->allmulti_rule) {
672 		mlx5_del_flow_rules(vport->allmulti_rule);
673 		vport->allmulti_rule = NULL;
674 
675 		if (--allmulti_addr->refcnt > 0)
676 			goto promisc;
677 
678 		if (allmulti_addr->uplink_rule)
679 			mlx5_del_flow_rules(allmulti_addr->uplink_rule);
680 		allmulti_addr->uplink_rule = NULL;
681 	}
682 
683 promisc:
684 	if (IS_ERR_OR_NULL(vport->promisc_rule) != promisc)
685 		return;
686 
687 	if (promisc) {
688 		vport->promisc_rule =
689 			esw_fdb_set_vport_promisc_rule(esw, vport->vport);
690 	} else if (vport->promisc_rule) {
691 		mlx5_del_flow_rules(vport->promisc_rule);
692 		vport->promisc_rule = NULL;
693 	}
694 }
695 
696 /* Sync vport rx mode from vport context */
esw_update_vport_rx_mode(struct mlx5_eswitch * esw,struct mlx5_vport * vport)697 static void esw_update_vport_rx_mode(struct mlx5_eswitch *esw,
698 				     struct mlx5_vport *vport)
699 {
700 	int promisc_all = 0;
701 	int promisc_uc = 0;
702 	int promisc_mc = 0;
703 	int err;
704 
705 	err = mlx5_query_nic_vport_promisc(esw->dev,
706 					   vport->vport,
707 					   &promisc_uc,
708 					   &promisc_mc,
709 					   &promisc_all);
710 	if (err)
711 		return;
712 	esw_debug(esw->dev, "vport[%d] context update rx mode promisc_all=%d, all_multi=%d\n",
713 		  vport->vport, promisc_all, promisc_mc);
714 
715 	if (!vport->info.trusted || !vport->enabled) {
716 		promisc_uc = 0;
717 		promisc_mc = 0;
718 		promisc_all = 0;
719 	}
720 
721 	esw_apply_vport_rx_mode(esw, vport, promisc_all,
722 				(promisc_all || promisc_mc));
723 }
724 
esw_vport_change_handle_locked(struct mlx5_vport * vport)725 void esw_vport_change_handle_locked(struct mlx5_vport *vport)
726 {
727 	struct mlx5_core_dev *dev = vport->dev;
728 	struct mlx5_eswitch *esw = dev->priv.eswitch;
729 	u8 mac[ETH_ALEN];
730 
731 	if (!MLX5_CAP_GEN(dev, log_max_l2_table))
732 		return;
733 
734 	mlx5_query_nic_vport_mac_address(dev, vport->vport, true, mac);
735 	esw_debug(dev, "vport[%d] Context Changed: perm mac: %pM\n",
736 		  vport->vport, mac);
737 
738 	if (vport->enabled_events & MLX5_VPORT_UC_ADDR_CHANGE) {
739 		esw_update_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_UC);
740 		esw_apply_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_UC);
741 	}
742 
743 	if (vport->enabled_events & MLX5_VPORT_MC_ADDR_CHANGE)
744 		esw_update_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_MC);
745 
746 	if (vport->enabled_events & MLX5_VPORT_PROMISC_CHANGE) {
747 		esw_update_vport_rx_mode(esw, vport);
748 		if (!IS_ERR_OR_NULL(vport->allmulti_rule))
749 			esw_update_vport_mc_promisc(esw, vport);
750 	}
751 
752 	if (vport->enabled_events & (MLX5_VPORT_PROMISC_CHANGE | MLX5_VPORT_MC_ADDR_CHANGE))
753 		esw_apply_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_MC);
754 
755 	esw_debug(esw->dev, "vport[%d] Context Changed: Done\n", vport->vport);
756 	if (vport->enabled)
757 		arm_vport_context_events_cmd(dev, vport->vport,
758 					     vport->enabled_events);
759 }
760 
esw_vport_change_handler(struct work_struct * work)761 static void esw_vport_change_handler(struct work_struct *work)
762 {
763 	struct mlx5_vport *vport =
764 		container_of(work, struct mlx5_vport, vport_change_handler);
765 	struct mlx5_eswitch *esw = vport->dev->priv.eswitch;
766 
767 	mutex_lock(&esw->state_lock);
768 	esw_vport_change_handle_locked(vport);
769 	mutex_unlock(&esw->state_lock);
770 }
771 
node_guid_gen_from_mac(u64 * node_guid,const u8 * mac)772 static void node_guid_gen_from_mac(u64 *node_guid, const u8 *mac)
773 {
774 	((u8 *)node_guid)[7] = mac[0];
775 	((u8 *)node_guid)[6] = mac[1];
776 	((u8 *)node_guid)[5] = mac[2];
777 	((u8 *)node_guid)[4] = 0xff;
778 	((u8 *)node_guid)[3] = 0xfe;
779 	((u8 *)node_guid)[2] = mac[3];
780 	((u8 *)node_guid)[1] = mac[4];
781 	((u8 *)node_guid)[0] = mac[5];
782 }
783 
esw_vport_setup_acl(struct mlx5_eswitch * esw,struct mlx5_vport * vport)784 static int esw_vport_setup_acl(struct mlx5_eswitch *esw,
785 			       struct mlx5_vport *vport)
786 {
787 	if (esw->mode == MLX5_ESWITCH_LEGACY)
788 		return esw_legacy_vport_acl_setup(esw, vport);
789 	else
790 		return esw_vport_create_offloads_acl_tables(esw, vport);
791 }
792 
esw_vport_cleanup_acl(struct mlx5_eswitch * esw,struct mlx5_vport * vport)793 static void esw_vport_cleanup_acl(struct mlx5_eswitch *esw,
794 				  struct mlx5_vport *vport)
795 {
796 	if (esw->mode == MLX5_ESWITCH_LEGACY)
797 		esw_legacy_vport_acl_cleanup(esw, vport);
798 	else
799 		esw_vport_destroy_offloads_acl_tables(esw, vport);
800 }
801 
mlx5_esw_vport_caps_get(struct mlx5_eswitch * esw,struct mlx5_vport * vport)802 static int mlx5_esw_vport_caps_get(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
803 {
804 	int query_out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
805 	void *query_ctx;
806 	void *hca_caps;
807 	int err;
808 
809 	if (!MLX5_CAP_GEN(esw->dev, vhca_resource_manager))
810 		return 0;
811 
812 	query_ctx = kzalloc(query_out_sz, GFP_KERNEL);
813 	if (!query_ctx)
814 		return -ENOMEM;
815 
816 	err = mlx5_vport_get_other_func_cap(esw->dev, vport->vport, query_ctx,
817 					    MLX5_CAP_GENERAL);
818 	if (err)
819 		goto out_free;
820 
821 	hca_caps = MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability);
822 	vport->info.roce_enabled = MLX5_GET(cmd_hca_cap, hca_caps, roce);
823 	vport->vhca_id = MLX5_GET(cmd_hca_cap, hca_caps, vhca_id);
824 
825 	if (!MLX5_CAP_GEN_MAX(esw->dev, hca_cap_2))
826 		goto out_free;
827 
828 	memset(query_ctx, 0, query_out_sz);
829 	err = mlx5_vport_get_other_func_cap(esw->dev, vport->vport, query_ctx,
830 					    MLX5_CAP_GENERAL_2);
831 	if (err)
832 		goto out_free;
833 
834 	hca_caps = MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability);
835 	vport->info.mig_enabled = MLX5_GET(cmd_hca_cap_2, hca_caps, migratable);
836 
837 	err = mlx5_esw_ipsec_vf_offload_get(esw->dev, vport);
838 out_free:
839 	kfree(query_ctx);
840 	return err;
841 }
842 
mlx5_esw_vport_vhca_id(struct mlx5_eswitch * esw,u16 vportn,u16 * vhca_id)843 bool mlx5_esw_vport_vhca_id(struct mlx5_eswitch *esw, u16 vportn, u16 *vhca_id)
844 {
845 	struct mlx5_vport *vport;
846 
847 	vport = mlx5_eswitch_get_vport(esw, vportn);
848 	if (IS_ERR(vport) || MLX5_VPORT_INVAL_VHCA_ID(vport))
849 		return false;
850 
851 	*vhca_id = vport->vhca_id;
852 	return true;
853 }
854 
esw_vport_setup(struct mlx5_eswitch * esw,struct mlx5_vport * vport)855 static int esw_vport_setup(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
856 {
857 	bool vst_mode_steering = esw_vst_mode_is_steering(esw);
858 	u16 vport_num = vport->vport;
859 	int flags;
860 	int err;
861 
862 	err = esw_vport_setup_acl(esw, vport);
863 	if (err)
864 		return err;
865 
866 	if (mlx5_esw_is_manager_vport(esw, vport_num))
867 		return 0;
868 
869 	err = mlx5_esw_vport_caps_get(esw, vport);
870 	if (err)
871 		goto err_caps;
872 
873 	mlx5_modify_vport_admin_state(esw->dev,
874 				      MLX5_VPORT_STATE_OP_MOD_ESW_VPORT,
875 				      vport_num, 1,
876 				      vport->info.link_state);
877 
878 	mlx5_query_nic_vport_mac_address(esw->dev, vport_num, true,
879 					 vport->info.mac);
880 	mlx5_query_nic_vport_node_guid(esw->dev, vport_num, true,
881 				       &vport->info.node_guid);
882 
883 	flags = (vport->info.vlan || vport->info.qos) ?
884 		SET_VLAN_STRIP | SET_VLAN_INSERT : 0;
885 	if (esw->mode == MLX5_ESWITCH_OFFLOADS || !vst_mode_steering)
886 		modify_esw_vport_cvlan(esw->dev, vport_num, vport->info.vlan,
887 				       vport->info.qos, flags);
888 
889 	return 0;
890 
891 err_caps:
892 	esw_vport_cleanup_acl(esw, vport);
893 	return err;
894 }
895 
896 /* Don't cleanup vport->info, it's needed to restore vport configuration */
esw_vport_cleanup(struct mlx5_eswitch * esw,struct mlx5_vport * vport)897 static void esw_vport_cleanup(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
898 {
899 	u16 vport_num = vport->vport;
900 
901 	if (!mlx5_esw_is_manager_vport(esw, vport_num))
902 		mlx5_modify_vport_admin_state(esw->dev,
903 					      MLX5_VPORT_STATE_OP_MOD_ESW_VPORT,
904 					      vport_num, 1,
905 					      MLX5_VPORT_ADMIN_STATE_DOWN);
906 
907 	mlx5_esw_qos_vport_disable(vport);
908 	esw_vport_cleanup_acl(esw, vport);
909 }
910 
mlx5_esw_vport_enable(struct mlx5_eswitch * esw,struct mlx5_vport * vport,enum mlx5_eswitch_vport_event enabled_events)911 int mlx5_esw_vport_enable(struct mlx5_eswitch *esw, struct mlx5_vport *vport,
912 			  enum mlx5_eswitch_vport_event enabled_events)
913 {
914 	u16 vport_num = vport->vport;
915 	int ret;
916 
917 	mutex_lock(&esw->state_lock);
918 	WARN_ON(vport->enabled);
919 
920 	esw_debug(esw->dev, "Enabling VPORT(%d)\n", vport_num);
921 
922 	ret = esw_vport_setup(esw, vport);
923 	if (ret)
924 		goto done;
925 
926 	/* Sync with current vport context */
927 	vport->enabled_events = enabled_events;
928 	vport->enabled = true;
929 	if (vport->vport != MLX5_VPORT_PF &&
930 	    (vport->info.ipsec_crypto_enabled || vport->info.ipsec_packet_enabled))
931 		esw->enabled_ipsec_vf_count++;
932 
933 	/* Esw manager is trusted by default. Host PF (vport 0) is trusted as well
934 	 * in smartNIC as it's a vport group manager.
935 	 */
936 	if (mlx5_esw_is_manager_vport(esw, vport_num) ||
937 	    (!vport_num && mlx5_core_is_ecpf(esw->dev)))
938 		vport->info.trusted = true;
939 
940 	if (!mlx5_esw_is_manager_vport(esw, vport_num) &&
941 	    MLX5_CAP_GEN(esw->dev, vhca_resource_manager)) {
942 		ret = mlx5_esw_vport_vhca_id_map(esw, vport);
943 		if (ret)
944 			goto err_vhca_mapping;
945 	}
946 
947 	esw_vport_change_handle_locked(vport);
948 
949 	esw->enabled_vports++;
950 	esw_debug(esw->dev, "Enabled VPORT(%d)\n", vport_num);
951 done:
952 	mutex_unlock(&esw->state_lock);
953 	return ret;
954 
955 err_vhca_mapping:
956 	esw_vport_cleanup(esw, vport);
957 	mutex_unlock(&esw->state_lock);
958 	return ret;
959 }
960 
mlx5_esw_vport_disable(struct mlx5_eswitch * esw,struct mlx5_vport * vport)961 void mlx5_esw_vport_disable(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
962 {
963 	u16 vport_num = vport->vport;
964 
965 	mutex_lock(&esw->state_lock);
966 
967 	if (!vport->enabled)
968 		goto done;
969 
970 	esw_debug(esw->dev, "Disabling vport(%d)\n", vport_num);
971 	/* Mark this vport as disabled to discard new events */
972 	vport->enabled = false;
973 
974 	/* Disable events from this vport */
975 	if (MLX5_CAP_GEN(esw->dev, log_max_l2_table))
976 		arm_vport_context_events_cmd(esw->dev, vport_num, 0);
977 
978 	if (!mlx5_esw_is_manager_vport(esw, vport_num) &&
979 	    MLX5_CAP_GEN(esw->dev, vhca_resource_manager))
980 		mlx5_esw_vport_vhca_id_unmap(esw, vport);
981 
982 	if (vport->vport != MLX5_VPORT_PF &&
983 	    (vport->info.ipsec_crypto_enabled || vport->info.ipsec_packet_enabled))
984 		esw->enabled_ipsec_vf_count--;
985 
986 	/* We don't assume VFs will cleanup after themselves.
987 	 * Calling vport change handler while vport is disabled will cleanup
988 	 * the vport resources.
989 	 */
990 	esw_vport_change_handle_locked(vport);
991 	vport->enabled_events = 0;
992 	esw_apply_vport_rx_mode(esw, vport, false, false);
993 	esw_vport_cleanup(esw, vport);
994 	esw->enabled_vports--;
995 
996 done:
997 	mutex_unlock(&esw->state_lock);
998 }
999 
eswitch_vport_event(struct notifier_block * nb,unsigned long type,void * data)1000 static int eswitch_vport_event(struct notifier_block *nb,
1001 			       unsigned long type, void *data)
1002 {
1003 	struct mlx5_eswitch *esw = mlx5_nb_cof(nb, struct mlx5_eswitch, nb);
1004 	struct mlx5_eqe *eqe = data;
1005 	struct mlx5_vport *vport;
1006 	u16 vport_num;
1007 
1008 	vport_num = be16_to_cpu(eqe->data.vport_change.vport_num);
1009 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1010 	if (!IS_ERR(vport))
1011 		queue_work(esw->work_queue, &vport->vport_change_handler);
1012 	return NOTIFY_OK;
1013 }
1014 
1015 /**
1016  * mlx5_esw_query_functions - Returns raw output about functions state
1017  * @dev:	Pointer to device to query
1018  *
1019  * mlx5_esw_query_functions() allocates and returns functions changed
1020  * raw output memory pointer from device on success. Otherwise returns ERR_PTR.
1021  * Caller must free the memory using kvfree() when valid pointer is returned.
1022  */
mlx5_esw_query_functions(struct mlx5_core_dev * dev)1023 const u32 *mlx5_esw_query_functions(struct mlx5_core_dev *dev)
1024 {
1025 	int outlen = MLX5_ST_SZ_BYTES(query_esw_functions_out);
1026 	u32 in[MLX5_ST_SZ_DW(query_esw_functions_in)] = {};
1027 	u32 *out;
1028 	int err;
1029 
1030 	out = kvzalloc(outlen, GFP_KERNEL);
1031 	if (!out)
1032 		return ERR_PTR(-ENOMEM);
1033 
1034 	MLX5_SET(query_esw_functions_in, in, opcode,
1035 		 MLX5_CMD_OP_QUERY_ESW_FUNCTIONS);
1036 
1037 	err = mlx5_cmd_exec(dev, in, sizeof(in), out, outlen);
1038 	if (!err)
1039 		return out;
1040 
1041 	kvfree(out);
1042 	return ERR_PTR(err);
1043 }
1044 
mlx5_esw_host_functions_enabled_query(struct mlx5_eswitch * esw)1045 static int mlx5_esw_host_functions_enabled_query(struct mlx5_eswitch *esw)
1046 {
1047 	const u32 *query_host_out;
1048 
1049 	if (!mlx5_core_is_ecpf_esw_manager(esw->dev))
1050 		return 0;
1051 
1052 	query_host_out = mlx5_esw_query_functions(esw->dev);
1053 	if (IS_ERR(query_host_out))
1054 		return PTR_ERR(query_host_out);
1055 
1056 	esw->esw_funcs.host_funcs_disabled =
1057 		MLX5_GET(query_esw_functions_out, query_host_out,
1058 			 host_params_context.host_pf_not_exist);
1059 
1060 	kvfree(query_host_out);
1061 	return 0;
1062 }
1063 
mlx5_eswitch_event_handler_register(struct mlx5_eswitch * esw)1064 static void mlx5_eswitch_event_handler_register(struct mlx5_eswitch *esw)
1065 {
1066 	if (esw->mode == MLX5_ESWITCH_OFFLOADS && mlx5_eswitch_is_funcs_handler(esw->dev)) {
1067 		MLX5_NB_INIT(&esw->esw_funcs.nb, mlx5_esw_funcs_changed_handler,
1068 			     ESW_FUNCTIONS_CHANGED);
1069 		mlx5_eq_notifier_register(esw->dev, &esw->esw_funcs.nb);
1070 	}
1071 }
1072 
mlx5_eswitch_event_handler_unregister(struct mlx5_eswitch * esw)1073 static void mlx5_eswitch_event_handler_unregister(struct mlx5_eswitch *esw)
1074 {
1075 	if (esw->mode == MLX5_ESWITCH_OFFLOADS && mlx5_eswitch_is_funcs_handler(esw->dev))
1076 		mlx5_eq_notifier_unregister(esw->dev, &esw->esw_funcs.nb);
1077 
1078 	flush_workqueue(esw->work_queue);
1079 }
1080 
mlx5_eswitch_clear_vf_vports_info(struct mlx5_eswitch * esw)1081 static void mlx5_eswitch_clear_vf_vports_info(struct mlx5_eswitch *esw)
1082 {
1083 	struct mlx5_vport *vport;
1084 	unsigned long i;
1085 
1086 	mlx5_esw_for_each_vf_vport(esw, i, vport, esw->esw_funcs.num_vfs) {
1087 		mlx5_esw_qos_vport_qos_free(vport);
1088 		memset(&vport->info, 0, sizeof(vport->info));
1089 		vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
1090 	}
1091 }
1092 
mlx5_eswitch_clear_ec_vf_vports_info(struct mlx5_eswitch * esw)1093 static void mlx5_eswitch_clear_ec_vf_vports_info(struct mlx5_eswitch *esw)
1094 {
1095 	struct mlx5_vport *vport;
1096 	unsigned long i;
1097 
1098 	mlx5_esw_for_each_ec_vf_vport(esw, i, vport, esw->esw_funcs.num_ec_vfs) {
1099 		mlx5_esw_qos_vport_qos_free(vport);
1100 		memset(&vport->info, 0, sizeof(vport->info));
1101 		vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
1102 	}
1103 }
1104 
mlx5_eswitch_load_vport(struct mlx5_eswitch * esw,struct mlx5_vport * vport,enum mlx5_eswitch_vport_event enabled_events)1105 static int mlx5_eswitch_load_vport(struct mlx5_eswitch *esw, struct mlx5_vport *vport,
1106 				   enum mlx5_eswitch_vport_event enabled_events)
1107 {
1108 	int err;
1109 
1110 	err = mlx5_esw_vport_enable(esw, vport, enabled_events);
1111 	if (err)
1112 		return err;
1113 
1114 	err = mlx5_esw_offloads_load_rep(esw, vport);
1115 	if (err)
1116 		goto err_rep;
1117 
1118 	return err;
1119 
1120 err_rep:
1121 	mlx5_esw_vport_disable(esw, vport);
1122 	return err;
1123 }
1124 
mlx5_eswitch_unload_vport(struct mlx5_eswitch * esw,struct mlx5_vport * vport)1125 static void mlx5_eswitch_unload_vport(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
1126 {
1127 	mlx5_esw_offloads_unload_rep(esw, vport);
1128 	mlx5_esw_vport_disable(esw, vport);
1129 }
1130 
mlx5_eswitch_load_pf_vf_vport(struct mlx5_eswitch * esw,u16 vport_num,enum mlx5_eswitch_vport_event enabled_events)1131 static int mlx5_eswitch_load_pf_vf_vport(struct mlx5_eswitch *esw, u16 vport_num,
1132 					 enum mlx5_eswitch_vport_event enabled_events)
1133 {
1134 	struct mlx5_vport *vport;
1135 	int err;
1136 
1137 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1138 	if (IS_ERR(vport))
1139 		return PTR_ERR(vport);
1140 
1141 	err = mlx5_esw_offloads_init_pf_vf_rep(esw, vport);
1142 	if (err)
1143 		return err;
1144 
1145 	err = mlx5_eswitch_load_vport(esw, vport, enabled_events);
1146 	if (err)
1147 		goto err_load;
1148 	return 0;
1149 
1150 err_load:
1151 	mlx5_esw_offloads_cleanup_pf_vf_rep(esw, vport);
1152 	return err;
1153 }
1154 
mlx5_eswitch_unload_pf_vf_vport(struct mlx5_eswitch * esw,u16 vport_num)1155 static void mlx5_eswitch_unload_pf_vf_vport(struct mlx5_eswitch *esw, u16 vport_num)
1156 {
1157 	struct mlx5_vport *vport;
1158 
1159 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1160 	if (IS_ERR(vport))
1161 		return;
1162 
1163 	mlx5_eswitch_unload_vport(esw, vport);
1164 	mlx5_esw_offloads_cleanup_pf_vf_rep(esw, vport);
1165 }
1166 
mlx5_eswitch_load_sf_vport(struct mlx5_eswitch * esw,u16 vport_num,enum mlx5_eswitch_vport_event enabled_events,struct mlx5_devlink_port * dl_port,u32 controller,u32 sfnum)1167 int mlx5_eswitch_load_sf_vport(struct mlx5_eswitch *esw, u16 vport_num,
1168 			       enum mlx5_eswitch_vport_event enabled_events,
1169 			       struct mlx5_devlink_port *dl_port, u32 controller, u32 sfnum)
1170 {
1171 	struct mlx5_vport *vport;
1172 	int err;
1173 
1174 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1175 	if (IS_ERR(vport))
1176 		return PTR_ERR(vport);
1177 
1178 	err = mlx5_esw_offloads_init_sf_rep(esw, vport, dl_port, controller, sfnum);
1179 	if (err)
1180 		return err;
1181 
1182 	err = mlx5_eswitch_load_vport(esw, vport, enabled_events);
1183 	if (err)
1184 		goto err_load;
1185 
1186 	return 0;
1187 
1188 err_load:
1189 	mlx5_esw_offloads_cleanup_sf_rep(esw, vport);
1190 	return err;
1191 }
1192 
mlx5_eswitch_unload_sf_vport(struct mlx5_eswitch * esw,u16 vport_num)1193 void mlx5_eswitch_unload_sf_vport(struct mlx5_eswitch *esw, u16 vport_num)
1194 {
1195 	struct mlx5_vport *vport;
1196 
1197 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1198 	if (IS_ERR(vport))
1199 		return;
1200 
1201 	mlx5_eswitch_unload_vport(esw, vport);
1202 	mlx5_esw_offloads_cleanup_sf_rep(esw, vport);
1203 }
1204 
mlx5_eswitch_unload_vf_vports(struct mlx5_eswitch * esw,u16 num_vfs)1205 void mlx5_eswitch_unload_vf_vports(struct mlx5_eswitch *esw, u16 num_vfs)
1206 {
1207 	struct mlx5_vport *vport;
1208 	unsigned long i;
1209 
1210 	mlx5_esw_for_each_vf_vport(esw, i, vport, num_vfs) {
1211 		/* Adjacent VFs are unloaded separately */
1212 		if (!vport->enabled || vport->adjacent)
1213 			continue;
1214 		mlx5_eswitch_unload_pf_vf_vport(esw, vport->vport);
1215 	}
1216 }
1217 
mlx5_eswitch_unload_ec_vf_vports(struct mlx5_eswitch * esw,u16 num_ec_vfs)1218 static void mlx5_eswitch_unload_ec_vf_vports(struct mlx5_eswitch *esw,
1219 					     u16 num_ec_vfs)
1220 {
1221 	struct mlx5_vport *vport;
1222 	unsigned long i;
1223 
1224 	mlx5_esw_for_each_ec_vf_vport(esw, i, vport, num_ec_vfs) {
1225 		if (!vport->enabled)
1226 			continue;
1227 		mlx5_eswitch_unload_pf_vf_vport(esw, vport->vport);
1228 	}
1229 }
1230 
mlx5_eswitch_unload_adj_vf_vports(struct mlx5_eswitch * esw)1231 static void mlx5_eswitch_unload_adj_vf_vports(struct mlx5_eswitch *esw)
1232 {
1233 	struct mlx5_vport *vport;
1234 	unsigned long i;
1235 
1236 	mlx5_esw_for_each_vf_vport(esw, i, vport, U16_MAX) {
1237 		if (!vport->enabled || !vport->adjacent)
1238 			continue;
1239 		mlx5_eswitch_unload_pf_vf_vport(esw, vport->vport);
1240 	}
1241 }
1242 
1243 static int
mlx5_eswitch_load_adj_vf_vports(struct mlx5_eswitch * esw,enum mlx5_eswitch_vport_event enabled_events)1244 mlx5_eswitch_load_adj_vf_vports(struct mlx5_eswitch *esw,
1245 				enum mlx5_eswitch_vport_event enabled_events)
1246 {
1247 	struct mlx5_vport *vport;
1248 	unsigned long i;
1249 	int err;
1250 
1251 	mlx5_esw_for_each_vf_vport(esw, i, vport, U16_MAX) {
1252 		if (!vport->adjacent)
1253 			continue;
1254 		err = mlx5_eswitch_load_pf_vf_vport(esw, vport->vport,
1255 						    enabled_events);
1256 		if (err)
1257 			goto unload_adj_vf_vport;
1258 	}
1259 
1260 	return 0;
1261 
1262 unload_adj_vf_vport:
1263 	mlx5_eswitch_unload_adj_vf_vports(esw);
1264 	return err;
1265 }
1266 
mlx5_eswitch_load_vf_vports(struct mlx5_eswitch * esw,u16 num_vfs,enum mlx5_eswitch_vport_event enabled_events)1267 int mlx5_eswitch_load_vf_vports(struct mlx5_eswitch *esw, u16 num_vfs,
1268 				enum mlx5_eswitch_vport_event enabled_events)
1269 {
1270 	struct mlx5_vport *vport;
1271 	unsigned long i;
1272 	int err;
1273 
1274 	mlx5_esw_for_each_vf_vport(esw, i, vport, num_vfs) {
1275 		err = mlx5_eswitch_load_pf_vf_vport(esw, vport->vport, enabled_events);
1276 		if (err)
1277 			goto vf_err;
1278 	}
1279 
1280 	return 0;
1281 
1282 vf_err:
1283 	mlx5_eswitch_unload_vf_vports(esw, num_vfs);
1284 	return err;
1285 }
1286 
mlx5_eswitch_load_ec_vf_vports(struct mlx5_eswitch * esw,u16 num_ec_vfs,enum mlx5_eswitch_vport_event enabled_events)1287 static int mlx5_eswitch_load_ec_vf_vports(struct mlx5_eswitch *esw, u16 num_ec_vfs,
1288 					  enum mlx5_eswitch_vport_event enabled_events)
1289 {
1290 	struct mlx5_vport *vport;
1291 	unsigned long i;
1292 	int err;
1293 
1294 	mlx5_esw_for_each_ec_vf_vport(esw, i, vport, num_ec_vfs) {
1295 		err = mlx5_eswitch_load_pf_vf_vport(esw, vport->vport, enabled_events);
1296 		if (err)
1297 			goto vf_err;
1298 	}
1299 
1300 	return 0;
1301 
1302 vf_err:
1303 	mlx5_eswitch_unload_ec_vf_vports(esw, num_ec_vfs);
1304 	return err;
1305 }
1306 
mlx5_esw_host_pf_enable_hca(struct mlx5_core_dev * dev)1307 int mlx5_esw_host_pf_enable_hca(struct mlx5_core_dev *dev)
1308 {
1309 	struct mlx5_eswitch *esw = dev->priv.eswitch;
1310 	struct mlx5_vport *vport;
1311 	int err;
1312 
1313 	if (!mlx5_core_is_ecpf(dev) || !mlx5_esw_allowed(esw))
1314 		return 0;
1315 
1316 	vport = mlx5_eswitch_get_vport(esw, MLX5_VPORT_PF);
1317 	if (IS_ERR(vport))
1318 		return PTR_ERR(vport);
1319 
1320 	/* Once vport and representor are ready, take out the external host PF
1321 	 * out of initializing state. Enabling HCA clears the iser->initializing
1322 	 * bit and host PF driver loading can progress.
1323 	 */
1324 	err = mlx5_cmd_host_pf_enable_hca(dev);
1325 	if (err)
1326 		return err;
1327 
1328 	vport->pf_activated = true;
1329 
1330 	return 0;
1331 }
1332 
mlx5_esw_host_pf_disable_hca(struct mlx5_core_dev * dev)1333 int mlx5_esw_host_pf_disable_hca(struct mlx5_core_dev *dev)
1334 {
1335 	struct mlx5_eswitch *esw = dev->priv.eswitch;
1336 	struct mlx5_vport *vport;
1337 	int err;
1338 
1339 	if (!mlx5_core_is_ecpf(dev) || !mlx5_esw_allowed(esw))
1340 		return 0;
1341 
1342 	vport = mlx5_eswitch_get_vport(esw, MLX5_VPORT_PF);
1343 	if (IS_ERR(vport))
1344 		return PTR_ERR(vport);
1345 
1346 	err = mlx5_cmd_host_pf_disable_hca(dev);
1347 	if (err)
1348 		return err;
1349 
1350 	vport->pf_activated = false;
1351 
1352 	return 0;
1353 }
1354 
1355 /* mlx5_eswitch_enable_pf_vf_vports() enables vports of PF, ECPF and VFs
1356  * whichever are present on the eswitch.
1357  */
1358 int
mlx5_eswitch_enable_pf_vf_vports(struct mlx5_eswitch * esw,enum mlx5_eswitch_vport_event enabled_events)1359 mlx5_eswitch_enable_pf_vf_vports(struct mlx5_eswitch *esw,
1360 				 enum mlx5_eswitch_vport_event enabled_events)
1361 {
1362 	bool pf_needed;
1363 	int ret;
1364 
1365 	pf_needed = mlx5_core_is_ecpf_esw_manager(esw->dev) ||
1366 		    esw->mode == MLX5_ESWITCH_LEGACY;
1367 
1368 	/* Enable PF vport */
1369 	if (pf_needed && mlx5_esw_host_functions_enabled(esw->dev)) {
1370 		ret = mlx5_eswitch_load_pf_vf_vport(esw, MLX5_VPORT_PF,
1371 						    enabled_events);
1372 		if (ret)
1373 			return ret;
1374 	}
1375 
1376 	if (mlx5_esw_host_functions_enabled(esw->dev)) {
1377 		/* Enable external host PF HCA */
1378 		ret = mlx5_esw_host_pf_enable_hca(esw->dev);
1379 		if (ret)
1380 			goto pf_hca_err;
1381 	}
1382 
1383 	/* Enable ECPF vport */
1384 	if (mlx5_ecpf_vport_exists(esw->dev)) {
1385 		ret = mlx5_eswitch_load_pf_vf_vport(esw, MLX5_VPORT_ECPF, enabled_events);
1386 		if (ret)
1387 			goto ecpf_err;
1388 	}
1389 
1390 	/* Enable ECVF vports */
1391 	if (mlx5_core_ec_sriov_enabled(esw->dev)) {
1392 		ret = mlx5_eswitch_load_ec_vf_vports(esw,
1393 						     esw->esw_funcs.num_ec_vfs,
1394 						     enabled_events);
1395 		if (ret)
1396 			goto ec_vf_err;
1397 	}
1398 
1399 	/* Enable VF vports */
1400 	ret = mlx5_eswitch_load_vf_vports(esw, esw->esw_funcs.num_vfs,
1401 					  enabled_events);
1402 	if (ret)
1403 		goto vf_err;
1404 
1405 	/* Enable adjacent VF vports */
1406 	ret = mlx5_eswitch_load_adj_vf_vports(esw, enabled_events);
1407 	if (ret)
1408 		goto unload_vf_vports;
1409 
1410 	return 0;
1411 
1412 unload_vf_vports:
1413 	mlx5_eswitch_unload_vf_vports(esw, esw->esw_funcs.num_vfs);
1414 vf_err:
1415 	if (mlx5_core_ec_sriov_enabled(esw->dev))
1416 		mlx5_eswitch_unload_ec_vf_vports(esw, esw->esw_funcs.num_ec_vfs);
1417 ec_vf_err:
1418 	if (mlx5_ecpf_vport_exists(esw->dev))
1419 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_ECPF);
1420 ecpf_err:
1421 	if (mlx5_esw_host_functions_enabled(esw->dev))
1422 		mlx5_esw_host_pf_disable_hca(esw->dev);
1423 pf_hca_err:
1424 	if (pf_needed && mlx5_esw_host_functions_enabled(esw->dev))
1425 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_PF);
1426 	return ret;
1427 }
1428 
1429 /* mlx5_eswitch_disable_pf_vf_vports() disables vports of PF, ECPF and VFs
1430  * whichever are previously enabled on the eswitch.
1431  */
mlx5_eswitch_disable_pf_vf_vports(struct mlx5_eswitch * esw)1432 void mlx5_eswitch_disable_pf_vf_vports(struct mlx5_eswitch *esw)
1433 {
1434 	mlx5_eswitch_unload_adj_vf_vports(esw);
1435 
1436 	mlx5_eswitch_unload_vf_vports(esw, esw->esw_funcs.num_vfs);
1437 
1438 	if (mlx5_core_ec_sriov_enabled(esw->dev))
1439 		mlx5_eswitch_unload_ec_vf_vports(esw,
1440 						 esw->esw_funcs.num_ec_vfs);
1441 
1442 	if (mlx5_ecpf_vport_exists(esw->dev)) {
1443 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_ECPF);
1444 	}
1445 
1446 	if (mlx5_esw_host_functions_enabled(esw->dev))
1447 		mlx5_esw_host_pf_disable_hca(esw->dev);
1448 
1449 	if ((mlx5_core_is_ecpf_esw_manager(esw->dev) ||
1450 	     esw->mode == MLX5_ESWITCH_LEGACY) &&
1451 	    mlx5_esw_host_functions_enabled(esw->dev))
1452 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_PF);
1453 }
1454 
mlx5_eswitch_get_devlink_param(struct mlx5_eswitch * esw)1455 static void mlx5_eswitch_get_devlink_param(struct mlx5_eswitch *esw)
1456 {
1457 	struct devlink *devlink = priv_to_devlink(esw->dev);
1458 	union devlink_param_value val;
1459 	int err;
1460 
1461 	err = devl_param_driverinit_value_get(devlink,
1462 					      MLX5_DEVLINK_PARAM_ID_ESW_LARGE_GROUP_NUM,
1463 					      &val);
1464 	if (!err) {
1465 		esw->params.large_group_num = val.vu32;
1466 	} else {
1467 		esw_warn(esw->dev,
1468 			 "Devlink can't get param fdb_large_groups, uses default (%d).\n",
1469 			 ESW_OFFLOADS_DEFAULT_NUM_GROUPS);
1470 		esw->params.large_group_num = ESW_OFFLOADS_DEFAULT_NUM_GROUPS;
1471 	}
1472 }
1473 
1474 static void
mlx5_eswitch_update_num_of_vfs(struct mlx5_eswitch * esw,int num_vfs)1475 mlx5_eswitch_update_num_of_vfs(struct mlx5_eswitch *esw, int num_vfs)
1476 {
1477 	const u32 *out;
1478 
1479 	if (num_vfs < 0)
1480 		return;
1481 
1482 	if (!mlx5_core_is_ecpf_esw_manager(esw->dev)) {
1483 		esw->esw_funcs.num_vfs = num_vfs;
1484 		return;
1485 	}
1486 
1487 	out = mlx5_esw_query_functions(esw->dev);
1488 	if (IS_ERR(out))
1489 		return;
1490 
1491 	esw->esw_funcs.num_vfs = MLX5_GET(query_esw_functions_out, out,
1492 					  host_params_context.host_num_of_vfs);
1493 	if (mlx5_core_ec_sriov_enabled(esw->dev))
1494 		esw->esw_funcs.num_ec_vfs = num_vfs;
1495 
1496 	kvfree(out);
1497 }
1498 
mlx5_esw_mode_change_notify(struct mlx5_eswitch * esw,u16 mode)1499 static void mlx5_esw_mode_change_notify(struct mlx5_eswitch *esw, u16 mode)
1500 {
1501 	struct mlx5_esw_event_info info = {};
1502 
1503 	info.new_mode = mode;
1504 
1505 	blocking_notifier_call_chain(&esw->dev->priv.esw_n_head, 0, &info);
1506 }
1507 
mlx5_esw_egress_acls_init(struct mlx5_core_dev * dev)1508 static int mlx5_esw_egress_acls_init(struct mlx5_core_dev *dev)
1509 {
1510 	struct mlx5_flow_steering *steering = dev->priv.steering;
1511 	int total_vports = mlx5_eswitch_get_total_vports(dev);
1512 	int err;
1513 	int i;
1514 
1515 	for (i = 0; i < total_vports; i++) {
1516 		err = mlx5_fs_vport_egress_acl_ns_add(steering, i);
1517 		if (err)
1518 			goto acl_ns_remove;
1519 	}
1520 	return 0;
1521 
1522 acl_ns_remove:
1523 	while (i--)
1524 		mlx5_fs_vport_egress_acl_ns_remove(steering, i);
1525 	return err;
1526 }
1527 
mlx5_esw_egress_acls_cleanup(struct mlx5_core_dev * dev)1528 static void mlx5_esw_egress_acls_cleanup(struct mlx5_core_dev *dev)
1529 {
1530 	struct mlx5_flow_steering *steering = dev->priv.steering;
1531 	int total_vports = mlx5_eswitch_get_total_vports(dev);
1532 	int i;
1533 
1534 	for (i = total_vports - 1; i >= 0; i--)
1535 		mlx5_fs_vport_egress_acl_ns_remove(steering, i);
1536 }
1537 
mlx5_esw_ingress_acls_init(struct mlx5_core_dev * dev)1538 static int mlx5_esw_ingress_acls_init(struct mlx5_core_dev *dev)
1539 {
1540 	struct mlx5_flow_steering *steering = dev->priv.steering;
1541 	int total_vports = mlx5_eswitch_get_total_vports(dev);
1542 	int err;
1543 	int i;
1544 
1545 	for (i = 0; i < total_vports; i++) {
1546 		err = mlx5_fs_vport_ingress_acl_ns_add(steering, i);
1547 		if (err)
1548 			goto acl_ns_remove;
1549 	}
1550 	return 0;
1551 
1552 acl_ns_remove:
1553 	while (i--)
1554 		mlx5_fs_vport_ingress_acl_ns_remove(steering, i);
1555 	return err;
1556 }
1557 
mlx5_esw_ingress_acls_cleanup(struct mlx5_core_dev * dev)1558 static void mlx5_esw_ingress_acls_cleanup(struct mlx5_core_dev *dev)
1559 {
1560 	struct mlx5_flow_steering *steering = dev->priv.steering;
1561 	int total_vports = mlx5_eswitch_get_total_vports(dev);
1562 	int i;
1563 
1564 	for (i = total_vports - 1; i >= 0; i--)
1565 		mlx5_fs_vport_ingress_acl_ns_remove(steering, i);
1566 }
1567 
mlx5_esw_acls_ns_init(struct mlx5_eswitch * esw)1568 static int mlx5_esw_acls_ns_init(struct mlx5_eswitch *esw)
1569 {
1570 	struct mlx5_core_dev *dev = esw->dev;
1571 	int err;
1572 
1573 	if (esw->flags & MLX5_ESWITCH_VPORT_ACL_NS_CREATED)
1574 		return 0;
1575 
1576 	if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
1577 		err = mlx5_esw_egress_acls_init(dev);
1578 		if (err)
1579 			return err;
1580 	} else {
1581 		esw_warn(dev, "egress ACL is not supported by FW\n");
1582 	}
1583 
1584 	if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
1585 		err = mlx5_esw_ingress_acls_init(dev);
1586 		if (err)
1587 			goto err;
1588 	} else {
1589 		esw_warn(dev, "ingress ACL is not supported by FW\n");
1590 	}
1591 	esw->flags |= MLX5_ESWITCH_VPORT_ACL_NS_CREATED;
1592 	return 0;
1593 
1594 err:
1595 	if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support))
1596 		mlx5_esw_egress_acls_cleanup(dev);
1597 	return err;
1598 }
1599 
mlx5_esw_acls_ns_cleanup(struct mlx5_eswitch * esw)1600 static void mlx5_esw_acls_ns_cleanup(struct mlx5_eswitch *esw)
1601 {
1602 	struct mlx5_core_dev *dev = esw->dev;
1603 
1604 	esw->flags &= ~MLX5_ESWITCH_VPORT_ACL_NS_CREATED;
1605 	if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support))
1606 		mlx5_esw_ingress_acls_cleanup(dev);
1607 	if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support))
1608 		mlx5_esw_egress_acls_cleanup(dev);
1609 }
1610 
1611 /**
1612  * mlx5_eswitch_enable_locked - Enable eswitch
1613  * @esw:	Pointer to eswitch
1614  * @num_vfs:	Enable eswitch for given number of VFs. This is optional.
1615  *		Valid value are 0, > 0 and MLX5_ESWITCH_IGNORE_NUM_VFS.
1616  *		Caller should pass num_vfs > 0 when enabling eswitch for
1617  *		vf vports. Caller should pass num_vfs = 0, when eswitch
1618  *		is enabled without sriov VFs or when caller
1619  *		is unaware of the sriov state of the host PF on ECPF based
1620  *		eswitch. Caller should pass < 0 when num_vfs should be
1621  *		completely ignored. This is typically the case when eswitch
1622  *		is enabled without sriov regardless of PF/ECPF system.
1623  * mlx5_eswitch_enable_locked() Enables eswitch in either legacy or offloads
1624  * mode. If num_vfs >=0 is provided, it setup VF related eswitch vports.
1625  * It returns 0 on success or error code on failure.
1626  */
mlx5_eswitch_enable_locked(struct mlx5_eswitch * esw,int num_vfs)1627 int mlx5_eswitch_enable_locked(struct mlx5_eswitch *esw, int num_vfs)
1628 {
1629 	int err;
1630 
1631 	devl_assert_locked(priv_to_devlink(esw->dev));
1632 
1633 	if (!MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, ft_support)) {
1634 		esw_warn(esw->dev, "FDB is not supported, aborting ...\n");
1635 		return -EOPNOTSUPP;
1636 	}
1637 
1638 	mlx5_eswitch_get_devlink_param(esw);
1639 
1640 	err = mlx5_esw_acls_ns_init(esw);
1641 	if (err)
1642 		return err;
1643 
1644 	mlx5_eswitch_update_num_of_vfs(esw, num_vfs);
1645 
1646 	MLX5_NB_INIT(&esw->nb, eswitch_vport_event, NIC_VPORT_CHANGE);
1647 	mlx5_eq_notifier_register(esw->dev, &esw->nb);
1648 
1649 	err = mlx5_esw_qos_init(esw);
1650 	if (err)
1651 		goto err_esw_init;
1652 
1653 	if (esw->mode == MLX5_ESWITCH_LEGACY) {
1654 		err = esw_legacy_enable(esw);
1655 	} else {
1656 		err = esw_offloads_enable(esw);
1657 	}
1658 
1659 	if (err)
1660 		goto err_esw_init;
1661 
1662 	esw->fdb_table.flags |= MLX5_ESW_FDB_CREATED;
1663 
1664 	mlx5_eswitch_event_handler_register(esw);
1665 
1666 	esw_info(esw->dev, "Enable: mode(%s), nvfs(%d), necvfs(%d), active vports(%d)\n",
1667 		 esw->mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
1668 		 esw->esw_funcs.num_vfs, esw->esw_funcs.num_ec_vfs, esw->enabled_vports);
1669 
1670 	mlx5_esw_mode_change_notify(esw, esw->mode);
1671 
1672 	return 0;
1673 
1674 err_esw_init:
1675 	mlx5_eq_notifier_unregister(esw->dev, &esw->nb);
1676 	mlx5_esw_acls_ns_cleanup(esw);
1677 	return err;
1678 }
1679 
1680 /**
1681  * mlx5_eswitch_enable - Enable eswitch
1682  * @esw:	Pointer to eswitch
1683  * @num_vfs:	Enable eswitch switch for given number of VFs.
1684  *		Caller must pass num_vfs > 0 when enabling eswitch for
1685  *		vf vports.
1686  * mlx5_eswitch_enable() returns 0 on success or error code on failure.
1687  */
mlx5_eswitch_enable(struct mlx5_eswitch * esw,int num_vfs)1688 int mlx5_eswitch_enable(struct mlx5_eswitch *esw, int num_vfs)
1689 {
1690 	bool toggle_lag;
1691 	int ret = 0;
1692 
1693 	if (!mlx5_esw_allowed(esw))
1694 		return 0;
1695 
1696 	devl_assert_locked(priv_to_devlink(esw->dev));
1697 
1698 	toggle_lag = !mlx5_esw_is_fdb_created(esw);
1699 
1700 	if (toggle_lag)
1701 		mlx5_lag_disable_change(esw->dev);
1702 
1703 	if (!mlx5_esw_is_fdb_created(esw)) {
1704 		ret = mlx5_eswitch_enable_locked(esw, num_vfs);
1705 	} else {
1706 		enum mlx5_eswitch_vport_event vport_events;
1707 
1708 		vport_events = (esw->mode == MLX5_ESWITCH_LEGACY) ?
1709 					MLX5_LEGACY_SRIOV_VPORT_EVENTS : MLX5_VPORT_UC_ADDR_CHANGE;
1710 		/* If this is the ECPF the number of host VFs is managed via the
1711 		 * eswitch function change event handler, and any num_vfs provided
1712 		 * here are intended to be EC VFs.
1713 		 */
1714 		if (!mlx5_core_is_ecpf(esw->dev)) {
1715 			ret = mlx5_eswitch_load_vf_vports(esw, num_vfs, vport_events);
1716 			if (!ret)
1717 				esw->esw_funcs.num_vfs = num_vfs;
1718 		} else if (mlx5_core_ec_sriov_enabled(esw->dev)) {
1719 			ret = mlx5_eswitch_load_ec_vf_vports(esw, num_vfs, vport_events);
1720 			if (!ret)
1721 				esw->esw_funcs.num_ec_vfs = num_vfs;
1722 		}
1723 	}
1724 
1725 	if (toggle_lag)
1726 		mlx5_lag_enable_change(esw->dev);
1727 
1728 	return ret;
1729 }
1730 
1731 /* When disabling sriov, free driver level resources. */
mlx5_eswitch_disable_sriov(struct mlx5_eswitch * esw,bool clear_vf)1732 void mlx5_eswitch_disable_sriov(struct mlx5_eswitch *esw, bool clear_vf)
1733 {
1734 	if (!mlx5_esw_allowed(esw))
1735 		return;
1736 
1737 	devl_assert_locked(priv_to_devlink(esw->dev));
1738 	/* If driver is unloaded, this function is called twice by remove_one()
1739 	 * and mlx5_unload(). Prevent the second call.
1740 	 */
1741 	if (!esw->esw_funcs.num_vfs && !esw->esw_funcs.num_ec_vfs && !clear_vf)
1742 		return;
1743 
1744 	esw_info(esw->dev, "Unload vfs: mode(%s), nvfs(%d), necvfs(%d), active vports(%d)\n",
1745 		 esw->mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
1746 		 esw->esw_funcs.num_vfs, esw->esw_funcs.num_ec_vfs, esw->enabled_vports);
1747 
1748 	if (!mlx5_core_is_ecpf(esw->dev)) {
1749 		mlx5_eswitch_unload_vf_vports(esw, esw->esw_funcs.num_vfs);
1750 		if (clear_vf)
1751 			mlx5_eswitch_clear_vf_vports_info(esw);
1752 	} else if (mlx5_core_ec_sriov_enabled(esw->dev)) {
1753 		mlx5_eswitch_unload_ec_vf_vports(esw, esw->esw_funcs.num_ec_vfs);
1754 		if (clear_vf)
1755 			mlx5_eswitch_clear_ec_vf_vports_info(esw);
1756 	}
1757 
1758 	if (esw->mode == MLX5_ESWITCH_OFFLOADS) {
1759 		struct devlink *devlink = priv_to_devlink(esw->dev);
1760 
1761 		devl_rate_nodes_destroy(devlink);
1762 	}
1763 	/* Destroy legacy fdb when disabling sriov in legacy mode. */
1764 	if (esw->mode == MLX5_ESWITCH_LEGACY)
1765 		mlx5_eswitch_disable_locked(esw);
1766 
1767 	if (!mlx5_core_is_ecpf(esw->dev))
1768 		esw->esw_funcs.num_vfs = 0;
1769 	else
1770 		esw->esw_funcs.num_ec_vfs = 0;
1771 }
1772 
1773 /* Free resources for corresponding eswitch mode. It is called by devlink
1774  * when changing eswitch mode or modprobe when unloading driver.
1775  */
mlx5_eswitch_disable_locked(struct mlx5_eswitch * esw)1776 void mlx5_eswitch_disable_locked(struct mlx5_eswitch *esw)
1777 {
1778 	struct devlink *devlink = priv_to_devlink(esw->dev);
1779 
1780 	/* Notify eswitch users that it is exiting from current mode.
1781 	 * So that it can do necessary cleanup before the eswitch is disabled.
1782 	 */
1783 	mlx5_esw_mode_change_notify(esw, MLX5_ESWITCH_LEGACY);
1784 
1785 	mlx5_eq_notifier_unregister(esw->dev, &esw->nb);
1786 	mlx5_eswitch_event_handler_unregister(esw);
1787 
1788 	esw_info(esw->dev, "Disable: mode(%s), nvfs(%d), necvfs(%d), active vports(%d)\n",
1789 		 esw->mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
1790 		 esw->esw_funcs.num_vfs, esw->esw_funcs.num_ec_vfs, esw->enabled_vports);
1791 
1792 	if (esw->fdb_table.flags & MLX5_ESW_FDB_CREATED) {
1793 		esw->fdb_table.flags &= ~MLX5_ESW_FDB_CREATED;
1794 		if (esw->mode == MLX5_ESWITCH_OFFLOADS)
1795 			esw_offloads_disable(esw);
1796 		else if (esw->mode == MLX5_ESWITCH_LEGACY)
1797 			esw_legacy_disable(esw);
1798 		mlx5_esw_acls_ns_cleanup(esw);
1799 	}
1800 
1801 	if (esw->mode == MLX5_ESWITCH_OFFLOADS)
1802 		devl_rate_nodes_destroy(devlink);
1803 }
1804 
mlx5_eswitch_disable(struct mlx5_eswitch * esw)1805 void mlx5_eswitch_disable(struct mlx5_eswitch *esw)
1806 {
1807 	if (!mlx5_esw_allowed(esw))
1808 		return;
1809 
1810 	devl_assert_locked(priv_to_devlink(esw->dev));
1811 	mlx5_lag_disable_change(esw->dev);
1812 	mlx5_eswitch_disable_locked(esw);
1813 	esw->mode = MLX5_ESWITCH_LEGACY;
1814 	mlx5_lag_enable_change(esw->dev);
1815 }
1816 
mlx5_query_hca_cap_host_pf(struct mlx5_core_dev * dev,void * out)1817 static int mlx5_query_hca_cap_host_pf(struct mlx5_core_dev *dev, void *out)
1818 {
1819 	u16 opmod = (MLX5_CAP_GENERAL << 1) | (HCA_CAP_OPMOD_GET_MAX & 0x01);
1820 	u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)] = {};
1821 
1822 	MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
1823 	MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
1824 	MLX5_SET(query_hca_cap_in, in, function_id, MLX5_VPORT_PF);
1825 	MLX5_SET(query_hca_cap_in, in, other_function, true);
1826 	return mlx5_cmd_exec_inout(dev, query_hca_cap, in, out);
1827 }
1828 
mlx5_esw_sf_max_hpf_functions(struct mlx5_core_dev * dev,u16 * max_sfs,u16 * sf_base_id)1829 int mlx5_esw_sf_max_hpf_functions(struct mlx5_core_dev *dev, u16 *max_sfs, u16 *sf_base_id)
1830 
1831 {
1832 	int query_out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
1833 	void *query_ctx;
1834 	void *hca_caps;
1835 	int err;
1836 
1837 	if (!mlx5_core_is_ecpf(dev) ||
1838 	    !mlx5_esw_host_functions_enabled(dev)) {
1839 		*max_sfs = 0;
1840 		return 0;
1841 	}
1842 
1843 	query_ctx = kzalloc(query_out_sz, GFP_KERNEL);
1844 	if (!query_ctx)
1845 		return -ENOMEM;
1846 
1847 	err = mlx5_query_hca_cap_host_pf(dev, query_ctx);
1848 	if (err)
1849 		goto out_free;
1850 
1851 	hca_caps = MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability);
1852 	*max_sfs = MLX5_GET(cmd_hca_cap, hca_caps, max_num_sf);
1853 	*sf_base_id = MLX5_GET(cmd_hca_cap, hca_caps, sf_base_id);
1854 
1855 out_free:
1856 	kfree(query_ctx);
1857 	return err;
1858 }
1859 
mlx5_esw_vport_alloc(struct mlx5_eswitch * esw,int index,u16 vport_num)1860 int mlx5_esw_vport_alloc(struct mlx5_eswitch *esw, int index, u16 vport_num)
1861 {
1862 	struct mlx5_vport *vport;
1863 	int err;
1864 
1865 	vport = kzalloc_obj(*vport);
1866 	if (!vport)
1867 		return -ENOMEM;
1868 
1869 	vport->dev = esw->dev;
1870 	vport->vport = vport_num;
1871 	vport->index = index;
1872 	vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
1873 	vport->vhca_id = MLX5_VHCA_ID_INVALID;
1874 	INIT_WORK(&vport->vport_change_handler, esw_vport_change_handler);
1875 	err = xa_insert(&esw->vports, vport_num, vport, GFP_KERNEL);
1876 	if (err)
1877 		goto insert_err;
1878 
1879 	esw->total_vports++;
1880 	return 0;
1881 
1882 insert_err:
1883 	kfree(vport);
1884 	return err;
1885 }
1886 
mlx5_esw_vport_free(struct mlx5_eswitch * esw,struct mlx5_vport * vport)1887 void mlx5_esw_vport_free(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
1888 {
1889 	esw->total_vports--;
1890 	xa_erase(&esw->vports, vport->vport);
1891 	kfree(vport);
1892 }
1893 
mlx5_esw_vports_cleanup(struct mlx5_eswitch * esw)1894 static void mlx5_esw_vports_cleanup(struct mlx5_eswitch *esw)
1895 {
1896 	struct mlx5_vport *vport;
1897 	unsigned long i;
1898 
1899 	mlx5_esw_for_each_vport(esw, i, vport)
1900 		mlx5_esw_vport_free(esw, vport);
1901 	xa_destroy(&esw->vports);
1902 }
1903 
mlx5_esw_vports_init(struct mlx5_eswitch * esw)1904 static int mlx5_esw_vports_init(struct mlx5_eswitch *esw)
1905 {
1906 	struct mlx5_core_dev *dev = esw->dev;
1907 	u16 max_host_pf_sfs;
1908 	u16 base_sf_num;
1909 	int idx = 0;
1910 	int err;
1911 	int i;
1912 
1913 	xa_init(&esw->vports);
1914 
1915 	if (mlx5_esw_host_functions_enabled(dev)) {
1916 		err = mlx5_esw_vport_alloc(esw, idx, MLX5_VPORT_PF);
1917 		if (err)
1918 			goto err;
1919 		if (esw->first_host_vport == MLX5_VPORT_PF)
1920 			xa_set_mark(&esw->vports, idx, MLX5_ESW_VPT_HOST_FN);
1921 		idx++;
1922 		for (i = 0; i < mlx5_core_max_vfs(dev); i++) {
1923 			err = mlx5_esw_vport_alloc(esw, idx, idx);
1924 			if (err)
1925 				goto err;
1926 			xa_set_mark(&esw->vports, idx, MLX5_ESW_VPT_VF);
1927 			xa_set_mark(&esw->vports, idx, MLX5_ESW_VPT_HOST_FN);
1928 			idx++;
1929 		}
1930 	}
1931 
1932 	base_sf_num = mlx5_sf_start_function_id(dev);
1933 	for (i = 0; i < mlx5_sf_max_functions(dev); i++) {
1934 		err = mlx5_esw_vport_alloc(esw, idx, base_sf_num + i);
1935 		if (err)
1936 			goto err;
1937 		xa_set_mark(&esw->vports, base_sf_num + i, MLX5_ESW_VPT_SF);
1938 		idx++;
1939 	}
1940 
1941 	err = mlx5_esw_sf_max_hpf_functions(dev, &max_host_pf_sfs, &base_sf_num);
1942 	if (err)
1943 		goto err;
1944 	for (i = 0; i < max_host_pf_sfs; i++) {
1945 		err = mlx5_esw_vport_alloc(esw, idx, base_sf_num + i);
1946 		if (err)
1947 			goto err;
1948 		xa_set_mark(&esw->vports, base_sf_num + i, MLX5_ESW_VPT_SF);
1949 		idx++;
1950 	}
1951 
1952 	if (mlx5_core_ec_sriov_enabled(esw->dev)) {
1953 		int ec_vf_base_num = mlx5_core_ec_vf_vport_base(dev);
1954 
1955 		for (i = 0; i < mlx5_core_max_ec_vfs(esw->dev); i++) {
1956 			err = mlx5_esw_vport_alloc(esw, idx, ec_vf_base_num + i);
1957 			if (err)
1958 				goto err;
1959 			idx++;
1960 		}
1961 	}
1962 
1963 	if (mlx5_ecpf_vport_exists(dev) ||
1964 	    mlx5_core_is_ecpf_esw_manager(dev)) {
1965 		err = mlx5_esw_vport_alloc(esw, idx, MLX5_VPORT_ECPF);
1966 		if (err)
1967 			goto err;
1968 		idx++;
1969 	}
1970 	err = mlx5_esw_vport_alloc(esw, idx, MLX5_VPORT_UPLINK);
1971 	if (err)
1972 		goto err;
1973 
1974 	/* Adjacent vports or other dynamically create vports will use this */
1975 	esw->last_vport_idx = ++idx;
1976 	return 0;
1977 
1978 err:
1979 	mlx5_esw_vports_cleanup(esw);
1980 	return err;
1981 }
1982 
mlx5_devlink_esw_multiport_set(struct devlink * devlink,u32 id,struct devlink_param_gset_ctx * ctx,struct netlink_ext_ack * extack)1983 static int mlx5_devlink_esw_multiport_set(struct devlink *devlink, u32 id,
1984 					  struct devlink_param_gset_ctx *ctx,
1985 					  struct netlink_ext_ack *extack)
1986 {
1987 	struct mlx5_core_dev *dev = devlink_priv(devlink);
1988 
1989 	if (!MLX5_ESWITCH_MANAGER(dev))
1990 		return -EOPNOTSUPP;
1991 
1992 	if (ctx->val.vbool)
1993 		return mlx5_lag_mpesw_enable(dev);
1994 
1995 	mlx5_lag_mpesw_disable(dev);
1996 	return 0;
1997 }
1998 
mlx5_devlink_esw_multiport_get(struct devlink * devlink,u32 id,struct devlink_param_gset_ctx * ctx,struct netlink_ext_ack * extack)1999 static int mlx5_devlink_esw_multiport_get(struct devlink *devlink, u32 id,
2000 					  struct devlink_param_gset_ctx *ctx,
2001 					  struct netlink_ext_ack *extack)
2002 {
2003 	struct mlx5_core_dev *dev = devlink_priv(devlink);
2004 
2005 	ctx->val.vbool = mlx5_lag_is_mpesw(dev);
2006 	return 0;
2007 }
2008 
2009 static const struct devlink_param mlx5_eswitch_params[] = {
2010 	DEVLINK_PARAM_DRIVER(MLX5_DEVLINK_PARAM_ID_ESW_MULTIPORT,
2011 			     "esw_multiport", DEVLINK_PARAM_TYPE_BOOL,
2012 			     BIT(DEVLINK_PARAM_CMODE_RUNTIME),
2013 			     mlx5_devlink_esw_multiport_get,
2014 			     mlx5_devlink_esw_multiport_set, NULL),
2015 };
2016 
mlx5_eswitch_init(struct mlx5_core_dev * dev)2017 int mlx5_eswitch_init(struct mlx5_core_dev *dev)
2018 {
2019 	struct mlx5_eswitch *esw;
2020 	int err;
2021 
2022 	if (!MLX5_VPORT_MANAGER(dev) && !MLX5_ESWITCH_MANAGER(dev))
2023 		return 0;
2024 
2025 	esw = kzalloc_obj(*esw);
2026 	if (!esw)
2027 		return -ENOMEM;
2028 
2029 	err = devl_params_register(priv_to_devlink(dev), mlx5_eswitch_params,
2030 				   ARRAY_SIZE(mlx5_eswitch_params));
2031 	if (err)
2032 		goto free_esw;
2033 
2034 	esw->dev = dev;
2035 	dev->priv.eswitch = esw;
2036 	esw->manager_vport = mlx5_eswitch_manager_vport(dev);
2037 	esw->first_host_vport = mlx5_eswitch_first_host_vport_num(dev);
2038 
2039 	esw->debugfs_root = debugfs_create_dir("esw", mlx5_debugfs_get_dev_root(dev));
2040 	esw->work_queue = create_singlethread_workqueue("mlx5_esw_wq");
2041 	if (!esw->work_queue) {
2042 		err = -ENOMEM;
2043 		goto abort;
2044 	}
2045 
2046 	err = mlx5_esw_host_functions_enabled_query(esw);
2047 	if (err)
2048 		goto abort;
2049 
2050 	err = mlx5_esw_vports_init(esw);
2051 	if (err)
2052 		goto abort;
2053 
2054 	err = esw_offloads_init(esw);
2055 	if (err)
2056 		goto reps_err;
2057 
2058 	esw->mode = MLX5_ESWITCH_LEGACY;
2059 	err = mlx5_esw_qos_init(esw);
2060 	if (err)
2061 		goto reps_err;
2062 
2063 	mutex_init(&esw->offloads.encap_tbl_lock);
2064 	hash_init(esw->offloads.encap_tbl);
2065 	mutex_init(&esw->offloads.decap_tbl_lock);
2066 	hash_init(esw->offloads.decap_tbl);
2067 	mlx5e_mod_hdr_tbl_init(&esw->offloads.mod_hdr);
2068 	atomic64_set(&esw->offloads.num_flows, 0);
2069 	ida_init(&esw->offloads.vport_metadata_ida);
2070 	xa_init_flags(&esw->offloads.vhca_map, XA_FLAGS_ALLOC);
2071 	mutex_init(&esw->state_lock);
2072 	init_rwsem(&esw->mode_lock);
2073 	refcount_set(&esw->qos.refcnt, 0);
2074 
2075 	esw->enabled_vports = 0;
2076 	esw->offloads.inline_mode = MLX5_INLINE_MODE_NONE;
2077 	if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, reformat) &&
2078 	    MLX5_CAP_ESW_FLOWTABLE_FDB(dev, decap))
2079 		esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_BASIC;
2080 	else
2081 		esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_NONE;
2082 
2083 	esw_info(dev,
2084 		 "Total vports %d, per vport: max uc(%d) max mc(%d)\n",
2085 		 esw->total_vports,
2086 		 MLX5_MAX_UC_PER_VPORT(dev),
2087 		 MLX5_MAX_MC_PER_VPORT(dev));
2088 	return 0;
2089 
2090 reps_err:
2091 	mlx5_esw_vports_cleanup(esw);
2092 	dev->priv.eswitch = NULL;
2093 abort:
2094 	if (esw->work_queue)
2095 		destroy_workqueue(esw->work_queue);
2096 	debugfs_remove_recursive(esw->debugfs_root);
2097 	devl_params_unregister(priv_to_devlink(dev), mlx5_eswitch_params,
2098 			       ARRAY_SIZE(mlx5_eswitch_params));
2099 free_esw:
2100 	kfree(esw);
2101 	return err;
2102 }
2103 
mlx5_eswitch_cleanup(struct mlx5_eswitch * esw)2104 void mlx5_eswitch_cleanup(struct mlx5_eswitch *esw)
2105 {
2106 	if (!esw)
2107 		return;
2108 
2109 	esw_info(esw->dev, "cleanup\n");
2110 
2111 	mlx5_esw_qos_cleanup(esw);
2112 	destroy_workqueue(esw->work_queue);
2113 	WARN_ON(refcount_read(&esw->qos.refcnt));
2114 	mutex_destroy(&esw->state_lock);
2115 	WARN_ON(!xa_empty(&esw->offloads.vhca_map));
2116 	xa_destroy(&esw->offloads.vhca_map);
2117 	ida_destroy(&esw->offloads.vport_metadata_ida);
2118 	mlx5e_mod_hdr_tbl_destroy(&esw->offloads.mod_hdr);
2119 	mutex_destroy(&esw->offloads.encap_tbl_lock);
2120 	mutex_destroy(&esw->offloads.decap_tbl_lock);
2121 	esw_offloads_cleanup(esw);
2122 	esw->dev->priv.eswitch = NULL;
2123 	mlx5_esw_vports_cleanup(esw);
2124 	debugfs_remove_recursive(esw->debugfs_root);
2125 	devl_params_unregister(priv_to_devlink(esw->dev), mlx5_eswitch_params,
2126 			       ARRAY_SIZE(mlx5_eswitch_params));
2127 	kfree(esw);
2128 }
2129 
2130 /* Vport Administration */
2131 static int
mlx5_esw_set_vport_mac_locked(struct mlx5_eswitch * esw,struct mlx5_vport * evport,const u8 * mac)2132 mlx5_esw_set_vport_mac_locked(struct mlx5_eswitch *esw,
2133 			      struct mlx5_vport *evport, const u8 *mac)
2134 {
2135 	u16 vport_num = evport->vport;
2136 	u64 node_guid;
2137 	int err = 0;
2138 
2139 	if (is_multicast_ether_addr(mac))
2140 		return -EINVAL;
2141 
2142 	if (evport->info.spoofchk && !is_valid_ether_addr(mac))
2143 		mlx5_core_warn(esw->dev,
2144 			       "Set invalid MAC while spoofchk is on, vport(%d)\n",
2145 			       vport_num);
2146 
2147 	err = mlx5_modify_nic_vport_mac_address(esw->dev, vport_num, mac);
2148 	if (err) {
2149 		mlx5_core_warn(esw->dev,
2150 			       "Failed to mlx5_modify_nic_vport_mac vport(%d) err=(%d)\n",
2151 			       vport_num, err);
2152 		return err;
2153 	}
2154 
2155 	node_guid_gen_from_mac(&node_guid, mac);
2156 	err = mlx5_modify_nic_vport_node_guid(esw->dev, vport_num, node_guid);
2157 	if (err)
2158 		mlx5_core_warn(esw->dev,
2159 			       "Failed to set vport %d node guid, err = %d. RDMA_CM will not function properly for this VF.\n",
2160 			       vport_num, err);
2161 
2162 	ether_addr_copy(evport->info.mac, mac);
2163 	evport->info.node_guid = node_guid;
2164 	if (evport->enabled && esw->mode == MLX5_ESWITCH_LEGACY)
2165 		err = esw_acl_ingress_lgcy_setup(esw, evport);
2166 
2167 	return err;
2168 }
2169 
mlx5_eswitch_set_vport_mac(struct mlx5_eswitch * esw,u16 vport,const u8 * mac)2170 int mlx5_eswitch_set_vport_mac(struct mlx5_eswitch *esw,
2171 			       u16 vport, const u8 *mac)
2172 {
2173 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2174 	int err = 0;
2175 
2176 	if (IS_ERR(evport))
2177 		return PTR_ERR(evport);
2178 
2179 	mutex_lock(&esw->state_lock);
2180 	err = mlx5_esw_set_vport_mac_locked(esw, evport, mac);
2181 	mutex_unlock(&esw->state_lock);
2182 	return err;
2183 }
2184 
mlx5_esw_check_port_type(struct mlx5_eswitch * esw,u16 vport_num,xa_mark_t mark)2185 static bool mlx5_esw_check_port_type(struct mlx5_eswitch *esw, u16 vport_num, xa_mark_t mark)
2186 {
2187 	return xa_get_mark(&esw->vports, vport_num, mark);
2188 }
2189 
mlx5_eswitch_is_vf_vport(struct mlx5_eswitch * esw,u16 vport_num)2190 bool mlx5_eswitch_is_vf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2191 {
2192 	return mlx5_esw_check_port_type(esw, vport_num, MLX5_ESW_VPT_VF);
2193 }
2194 
mlx5_eswitch_is_pf_vf_vport(struct mlx5_eswitch * esw,u16 vport_num)2195 bool mlx5_eswitch_is_pf_vf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2196 {
2197 	return vport_num == MLX5_VPORT_PF ||
2198 		mlx5_eswitch_is_vf_vport(esw, vport_num);
2199 }
2200 
mlx5_esw_is_sf_vport(struct mlx5_eswitch * esw,u16 vport_num)2201 bool mlx5_esw_is_sf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2202 {
2203 	return mlx5_esw_check_port_type(esw, vport_num, MLX5_ESW_VPT_SF);
2204 }
2205 
mlx5_eswitch_set_vport_state(struct mlx5_eswitch * esw,u16 vport,int link_state)2206 int mlx5_eswitch_set_vport_state(struct mlx5_eswitch *esw,
2207 				 u16 vport, int link_state)
2208 {
2209 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2210 	int opmod = MLX5_VPORT_STATE_OP_MOD_ESW_VPORT;
2211 	int other_vport = 1;
2212 	int err = 0;
2213 
2214 	if (!mlx5_esw_allowed(esw))
2215 		return -EPERM;
2216 	if (IS_ERR(evport))
2217 		return PTR_ERR(evport);
2218 
2219 	if (vport == MLX5_VPORT_UPLINK) {
2220 		opmod = MLX5_VPORT_STATE_OP_MOD_UPLINK;
2221 		other_vport = 0;
2222 		vport = 0;
2223 	}
2224 	mutex_lock(&esw->state_lock);
2225 	if (esw->mode != MLX5_ESWITCH_LEGACY) {
2226 		err = -EOPNOTSUPP;
2227 		goto unlock;
2228 	}
2229 
2230 	err = mlx5_modify_vport_admin_state(esw->dev, opmod, vport, other_vport, link_state);
2231 	if (err) {
2232 		mlx5_core_warn(esw->dev, "Failed to set vport %d link state, opmod = %d, err = %d",
2233 			       vport, opmod, err);
2234 		goto unlock;
2235 	}
2236 
2237 	evport->info.link_state = link_state;
2238 
2239 unlock:
2240 	mutex_unlock(&esw->state_lock);
2241 	return err;
2242 }
2243 
mlx5_eswitch_get_vport_config(struct mlx5_eswitch * esw,u16 vport,struct ifla_vf_info * ivi)2244 int mlx5_eswitch_get_vport_config(struct mlx5_eswitch *esw,
2245 				  u16 vport, struct ifla_vf_info *ivi)
2246 {
2247 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2248 	u32 max_rate, min_rate;
2249 
2250 	if (IS_ERR(evport))
2251 		return PTR_ERR(evport);
2252 
2253 	memset(ivi, 0, sizeof(*ivi));
2254 	ivi->vf = vport - 1;
2255 
2256 	mutex_lock(&esw->state_lock);
2257 
2258 	mlx5_query_nic_vport_mac_address(esw->dev, vport, true,
2259 					 evport->info.mac);
2260 	ether_addr_copy(ivi->mac, evport->info.mac);
2261 	ivi->linkstate = evport->info.link_state;
2262 	ivi->vlan = evport->info.vlan;
2263 	ivi->qos = evport->info.qos;
2264 	ivi->spoofchk = evport->info.spoofchk;
2265 	ivi->trusted = evport->info.trusted;
2266 
2267 	if (mlx5_esw_qos_get_vport_rate(evport, &max_rate, &min_rate)) {
2268 		ivi->max_tx_rate = max_rate;
2269 		ivi->min_tx_rate = min_rate;
2270 	}
2271 	mutex_unlock(&esw->state_lock);
2272 
2273 	return 0;
2274 }
2275 
__mlx5_eswitch_set_vport_vlan(struct mlx5_eswitch * esw,u16 vport,u16 vlan,u8 qos,u8 set_flags)2276 int __mlx5_eswitch_set_vport_vlan(struct mlx5_eswitch *esw,
2277 				  u16 vport, u16 vlan, u8 qos, u8 set_flags)
2278 {
2279 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2280 	bool vst_mode_steering = esw_vst_mode_is_steering(esw);
2281 	int err = 0;
2282 
2283 	if (IS_ERR(evport))
2284 		return PTR_ERR(evport);
2285 	if (vlan > 4095 || qos > 7)
2286 		return -EINVAL;
2287 
2288 	if (esw->mode == MLX5_ESWITCH_OFFLOADS || !vst_mode_steering) {
2289 		err = modify_esw_vport_cvlan(esw->dev, vport, vlan, qos, set_flags);
2290 		if (err)
2291 			return err;
2292 	}
2293 
2294 	evport->info.vlan = vlan;
2295 	evport->info.qos = qos;
2296 	if (evport->enabled && esw->mode == MLX5_ESWITCH_LEGACY) {
2297 		err = esw_acl_ingress_lgcy_setup(esw, evport);
2298 		if (err)
2299 			return err;
2300 		err = esw_acl_egress_lgcy_setup(esw, evport);
2301 	}
2302 
2303 	return err;
2304 }
2305 
mlx5_eswitch_get_vport_stats(struct mlx5_eswitch * esw,u16 vport_num,struct ifla_vf_stats * vf_stats)2306 int mlx5_eswitch_get_vport_stats(struct mlx5_eswitch *esw,
2307 				 u16 vport_num,
2308 				 struct ifla_vf_stats *vf_stats)
2309 {
2310 	struct mlx5_vport *vport = mlx5_eswitch_get_vport(esw, vport_num);
2311 	int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
2312 	u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)] = {};
2313 	struct mlx5_vport_drop_stats stats = {};
2314 	int err = 0;
2315 	u32 *out;
2316 
2317 	if (IS_ERR(vport))
2318 		return PTR_ERR(vport);
2319 
2320 	out = kvzalloc(outlen, GFP_KERNEL);
2321 	if (!out)
2322 		return -ENOMEM;
2323 
2324 	MLX5_SET(query_vport_counter_in, in, opcode,
2325 		 MLX5_CMD_OP_QUERY_VPORT_COUNTER);
2326 	MLX5_SET(query_vport_counter_in, in, op_mod, 0);
2327 	MLX5_SET(query_vport_counter_in, in, vport_number, vport->vport);
2328 	MLX5_SET(query_vport_counter_in, in, other_vport, 1);
2329 
2330 	err = mlx5_cmd_exec_inout(esw->dev, query_vport_counter, in, out);
2331 	if (err)
2332 		goto free_out;
2333 
2334 	#define MLX5_GET_CTR(p, x) \
2335 		MLX5_GET64(query_vport_counter_out, p, x)
2336 
2337 	memset(vf_stats, 0, sizeof(*vf_stats));
2338 	vf_stats->rx_packets =
2339 		MLX5_GET_CTR(out, received_eth_unicast.packets) +
2340 		MLX5_GET_CTR(out, received_ib_unicast.packets) +
2341 		MLX5_GET_CTR(out, received_eth_multicast.packets) +
2342 		MLX5_GET_CTR(out, received_ib_multicast.packets) +
2343 		MLX5_GET_CTR(out, received_eth_broadcast.packets);
2344 
2345 	vf_stats->rx_bytes =
2346 		MLX5_GET_CTR(out, received_eth_unicast.octets) +
2347 		MLX5_GET_CTR(out, received_ib_unicast.octets) +
2348 		MLX5_GET_CTR(out, received_eth_multicast.octets) +
2349 		MLX5_GET_CTR(out, received_ib_multicast.octets) +
2350 		MLX5_GET_CTR(out, received_eth_broadcast.octets);
2351 
2352 	vf_stats->tx_packets =
2353 		MLX5_GET_CTR(out, transmitted_eth_unicast.packets) +
2354 		MLX5_GET_CTR(out, transmitted_ib_unicast.packets) +
2355 		MLX5_GET_CTR(out, transmitted_eth_multicast.packets) +
2356 		MLX5_GET_CTR(out, transmitted_ib_multicast.packets) +
2357 		MLX5_GET_CTR(out, transmitted_eth_broadcast.packets);
2358 
2359 	vf_stats->tx_bytes =
2360 		MLX5_GET_CTR(out, transmitted_eth_unicast.octets) +
2361 		MLX5_GET_CTR(out, transmitted_ib_unicast.octets) +
2362 		MLX5_GET_CTR(out, transmitted_eth_multicast.octets) +
2363 		MLX5_GET_CTR(out, transmitted_ib_multicast.octets) +
2364 		MLX5_GET_CTR(out, transmitted_eth_broadcast.octets);
2365 
2366 	vf_stats->multicast =
2367 		MLX5_GET_CTR(out, received_eth_multicast.packets) +
2368 		MLX5_GET_CTR(out, received_ib_multicast.packets);
2369 
2370 	vf_stats->broadcast =
2371 		MLX5_GET_CTR(out, received_eth_broadcast.packets);
2372 
2373 	err = mlx5_esw_query_vport_drop_stats(esw->dev, vport, &stats);
2374 	if (err)
2375 		goto free_out;
2376 	vf_stats->rx_dropped = stats.rx_dropped;
2377 	vf_stats->tx_dropped = stats.tx_dropped;
2378 
2379 free_out:
2380 	kvfree(out);
2381 	return err;
2382 }
2383 
mlx5_eswitch_mode(const struct mlx5_core_dev * dev)2384 u8 mlx5_eswitch_mode(const struct mlx5_core_dev *dev)
2385 {
2386 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2387 
2388 	return mlx5_esw_allowed(esw) ? esw->mode : MLX5_ESWITCH_LEGACY;
2389 }
2390 EXPORT_SYMBOL_GPL(mlx5_eswitch_mode);
2391 
2392 enum devlink_eswitch_encap_mode
mlx5_eswitch_get_encap_mode(const struct mlx5_core_dev * dev)2393 mlx5_eswitch_get_encap_mode(const struct mlx5_core_dev *dev)
2394 {
2395 	struct mlx5_eswitch *esw;
2396 
2397 	esw = dev->priv.eswitch;
2398 	return (mlx5_eswitch_mode(dev) == MLX5_ESWITCH_OFFLOADS)  ? esw->offloads.encap :
2399 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
2400 }
2401 EXPORT_SYMBOL(mlx5_eswitch_get_encap_mode);
2402 
mlx5_esw_multipath_prereq(struct mlx5_core_dev * dev0,struct mlx5_core_dev * dev1)2403 bool mlx5_esw_multipath_prereq(struct mlx5_core_dev *dev0,
2404 			       struct mlx5_core_dev *dev1)
2405 {
2406 	return (dev0->priv.eswitch->mode == MLX5_ESWITCH_OFFLOADS &&
2407 		dev1->priv.eswitch->mode == MLX5_ESWITCH_OFFLOADS);
2408 }
2409 
mlx5_esw_event_notifier_register(struct mlx5_core_dev * dev,struct notifier_block * nb)2410 int mlx5_esw_event_notifier_register(struct mlx5_core_dev *dev,
2411 				     struct notifier_block *nb)
2412 {
2413 	return blocking_notifier_chain_register(&dev->priv.esw_n_head, nb);
2414 }
2415 
mlx5_esw_event_notifier_unregister(struct mlx5_core_dev * dev,struct notifier_block * nb)2416 void mlx5_esw_event_notifier_unregister(struct mlx5_core_dev *dev,
2417 					struct notifier_block *nb)
2418 {
2419 	blocking_notifier_chain_unregister(&dev->priv.esw_n_head, nb);
2420 }
2421 
2422 /**
2423  * mlx5_esw_hold() - Try to take a read lock on esw mode lock.
2424  * @mdev: mlx5 core device.
2425  *
2426  * Should be called by esw resources callers.
2427  *
2428  * Return: true on success or false.
2429  */
mlx5_esw_hold(struct mlx5_core_dev * mdev)2430 bool mlx5_esw_hold(struct mlx5_core_dev *mdev)
2431 {
2432 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2433 
2434 	/* e.g. VF doesn't have eswitch so nothing to do */
2435 	if (!mlx5_esw_allowed(esw))
2436 		return true;
2437 
2438 	if (down_read_trylock(&esw->mode_lock) != 0) {
2439 		if (esw->eswitch_operation_in_progress) {
2440 			up_read(&esw->mode_lock);
2441 			return false;
2442 		}
2443 		return true;
2444 	}
2445 
2446 	return false;
2447 }
2448 
2449 /**
2450  * mlx5_esw_release() - Release a read lock on esw mode lock.
2451  * @mdev: mlx5 core device.
2452  */
mlx5_esw_release(struct mlx5_core_dev * mdev)2453 void mlx5_esw_release(struct mlx5_core_dev *mdev)
2454 {
2455 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2456 
2457 	if (mlx5_esw_allowed(esw))
2458 		up_read(&esw->mode_lock);
2459 }
2460 
2461 /**
2462  * mlx5_esw_get() - Increase esw user count.
2463  * @mdev: mlx5 core device.
2464  */
mlx5_esw_get(struct mlx5_core_dev * mdev)2465 void mlx5_esw_get(struct mlx5_core_dev *mdev)
2466 {
2467 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2468 
2469 	if (mlx5_esw_allowed(esw))
2470 		atomic64_inc(&esw->user_count);
2471 }
2472 
2473 /**
2474  * mlx5_esw_put() - Decrease esw user count.
2475  * @mdev: mlx5 core device.
2476  */
mlx5_esw_put(struct mlx5_core_dev * mdev)2477 void mlx5_esw_put(struct mlx5_core_dev *mdev)
2478 {
2479 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2480 
2481 	if (mlx5_esw_allowed(esw))
2482 		atomic64_dec_if_positive(&esw->user_count);
2483 }
2484 
2485 /**
2486  * mlx5_esw_try_lock() - Take a write lock on esw mode lock.
2487  * @esw: eswitch device.
2488  *
2489  * Should be called by esw mode change routine.
2490  *
2491  * Return:
2492  * * 0       - esw mode if successfully locked and refcount is 0.
2493  * * -EBUSY  - refcount is not 0.
2494  * * -EINVAL - In the middle of switching mode or lock is already held.
2495  */
mlx5_esw_try_lock(struct mlx5_eswitch * esw)2496 int mlx5_esw_try_lock(struct mlx5_eswitch *esw)
2497 {
2498 	if (down_write_trylock(&esw->mode_lock) == 0)
2499 		return -EINVAL;
2500 
2501 	if (esw->eswitch_operation_in_progress ||
2502 	    atomic64_read(&esw->user_count) > 0) {
2503 		up_write(&esw->mode_lock);
2504 		return -EBUSY;
2505 	}
2506 
2507 	return esw->mode;
2508 }
2509 
mlx5_esw_lock(struct mlx5_eswitch * esw)2510 int mlx5_esw_lock(struct mlx5_eswitch *esw)
2511 {
2512 	down_write(&esw->mode_lock);
2513 
2514 	if (esw->eswitch_operation_in_progress) {
2515 		up_write(&esw->mode_lock);
2516 		return -EBUSY;
2517 	}
2518 
2519 	return 0;
2520 }
2521 
2522 /**
2523  * mlx5_esw_unlock() - Release write lock on esw mode lock
2524  * @esw: eswitch device.
2525  */
mlx5_esw_unlock(struct mlx5_eswitch * esw)2526 void mlx5_esw_unlock(struct mlx5_eswitch *esw)
2527 {
2528 	up_write(&esw->mode_lock);
2529 }
2530 
2531 /**
2532  * mlx5_eswitch_get_total_vports - Get total vports of the eswitch
2533  *
2534  * @dev: Pointer to core device
2535  *
2536  * mlx5_eswitch_get_total_vports returns total number of eswitch vports.
2537  */
mlx5_eswitch_get_total_vports(const struct mlx5_core_dev * dev)2538 u16 mlx5_eswitch_get_total_vports(const struct mlx5_core_dev *dev)
2539 {
2540 	struct mlx5_eswitch *esw;
2541 
2542 	esw = dev->priv.eswitch;
2543 	return mlx5_esw_allowed(esw) ? esw->total_vports : 0;
2544 }
2545 EXPORT_SYMBOL_GPL(mlx5_eswitch_get_total_vports);
2546 
2547 /**
2548  * mlx5_eswitch_get_core_dev - Get the mdev device
2549  * @esw : eswitch device.
2550  *
2551  * Return the mellanox core device which manages the eswitch.
2552  */
mlx5_eswitch_get_core_dev(struct mlx5_eswitch * esw)2553 struct mlx5_core_dev *mlx5_eswitch_get_core_dev(struct mlx5_eswitch *esw)
2554 {
2555 	return mlx5_esw_allowed(esw) ? esw->dev : NULL;
2556 }
2557 EXPORT_SYMBOL(mlx5_eswitch_get_core_dev);
2558 
mlx5_eswitch_block_ipsec(struct mlx5_core_dev * dev)2559 bool mlx5_eswitch_block_ipsec(struct mlx5_core_dev *dev)
2560 {
2561 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2562 
2563 	if (!mlx5_esw_allowed(esw))
2564 		return true;
2565 
2566 	mutex_lock(&esw->state_lock);
2567 	if (esw->enabled_ipsec_vf_count) {
2568 		mutex_unlock(&esw->state_lock);
2569 		return false;
2570 	}
2571 
2572 	dev->num_ipsec_offloads++;
2573 	mutex_unlock(&esw->state_lock);
2574 	return true;
2575 }
2576 
mlx5_eswitch_unblock_ipsec(struct mlx5_core_dev * dev)2577 void mlx5_eswitch_unblock_ipsec(struct mlx5_core_dev *dev)
2578 {
2579 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2580 
2581 	if (!mlx5_esw_allowed(esw))
2582 		/* Failure means no eswitch => core dev is not a PF */
2583 		return;
2584 
2585 	mutex_lock(&esw->state_lock);
2586 	dev->num_ipsec_offloads--;
2587 	mutex_unlock(&esw->state_lock);
2588 }
2589 
mlx5_esw_host_functions_enabled(const struct mlx5_core_dev * dev)2590 bool mlx5_esw_host_functions_enabled(const struct mlx5_core_dev *dev)
2591 {
2592 	if (!dev->priv.eswitch)
2593 		return true;
2594 
2595 	return !dev->priv.eswitch->esw_funcs.host_funcs_disabled;
2596 }
2597