xref: /openbmc/linux/drivers/net/ethernet/sfc/tc.c (revision faffb083)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /****************************************************************************
3  * Driver for Solarflare network controllers and boards
4  * Copyright 2019 Solarflare Communications Inc.
5  * Copyright 2020-2022 Xilinx Inc.
6  *
7  * This program is free software; you can redistribute it and/or modify it
8  * under the terms of the GNU General Public License version 2 as published
9  * by the Free Software Foundation, incorporated herein by reference.
10  */
11 
12 #include <net/pkt_cls.h>
13 #include "tc.h"
14 #include "tc_bindings.h"
15 #include "mae.h"
16 #include "ef100_rep.h"
17 #include "efx.h"
18 
19 #define EFX_EFV_PF	NULL
20 /* Look up the representor information (efv) for a device.
21  * May return NULL for the PF (us), or an error pointer for a device that
22  * isn't supported as a TC offload endpoint
23  */
24 static struct efx_rep *efx_tc_flower_lookup_efv(struct efx_nic *efx,
25 						struct net_device *dev)
26 {
27 	struct efx_rep *efv;
28 
29 	if (!dev)
30 		return ERR_PTR(-EOPNOTSUPP);
31 	/* Is it us (the PF)? */
32 	if (dev == efx->net_dev)
33 		return EFX_EFV_PF;
34 	/* Is it an efx vfrep at all? */
35 	if (dev->netdev_ops != &efx_ef100_rep_netdev_ops)
36 		return ERR_PTR(-EOPNOTSUPP);
37 	/* Is it ours?  We don't support TC rules that include another
38 	 * EF100's netdevices (not even on another port of the same NIC).
39 	 */
40 	efv = netdev_priv(dev);
41 	if (efv->parent != efx)
42 		return ERR_PTR(-EOPNOTSUPP);
43 	return efv;
44 }
45 
46 /* Convert a driver-internal vport ID into an external device (wire or VF) */
47 static s64 efx_tc_flower_external_mport(struct efx_nic *efx, struct efx_rep *efv)
48 {
49 	u32 mport;
50 
51 	if (IS_ERR(efv))
52 		return PTR_ERR(efv);
53 	if (!efv) /* device is PF (us) */
54 		efx_mae_mport_wire(efx, &mport);
55 	else /* device is repr */
56 		efx_mae_mport_mport(efx, efv->mport, &mport);
57 	return mport;
58 }
59 
60 static const struct rhashtable_params efx_tc_match_action_ht_params = {
61 	.key_len	= sizeof(unsigned long),
62 	.key_offset	= offsetof(struct efx_tc_flow_rule, cookie),
63 	.head_offset	= offsetof(struct efx_tc_flow_rule, linkage),
64 };
65 
66 static void efx_tc_free_action_set(struct efx_nic *efx,
67 				   struct efx_tc_action_set *act, bool in_hw)
68 {
69 	/* Failure paths calling this on the 'running action' set in_hw=false,
70 	 * because if the alloc had succeeded we'd've put it in acts.list and
71 	 * not still have it in act.
72 	 */
73 	if (in_hw) {
74 		efx_mae_free_action_set(efx, act->fw_id);
75 		/* in_hw is true iff we are on an acts.list; make sure to
76 		 * remove ourselves from that list before we are freed.
77 		 */
78 		list_del(&act->list);
79 	}
80 	if (act->count)
81 		efx_tc_flower_put_counter_index(efx, act->count);
82 	kfree(act);
83 }
84 
85 static void efx_tc_free_action_set_list(struct efx_nic *efx,
86 					struct efx_tc_action_set_list *acts,
87 					bool in_hw)
88 {
89 	struct efx_tc_action_set *act, *next;
90 
91 	/* Failure paths set in_hw=false, because usually the acts didn't get
92 	 * to efx_mae_alloc_action_set_list(); if they did, the failure tree
93 	 * has a separate efx_mae_free_action_set_list() before calling us.
94 	 */
95 	if (in_hw)
96 		efx_mae_free_action_set_list(efx, acts);
97 	/* Any act that's on the list will be in_hw even if the list isn't */
98 	list_for_each_entry_safe(act, next, &acts->list, list)
99 		efx_tc_free_action_set(efx, act, true);
100 	/* Don't kfree, as acts is embedded inside a struct efx_tc_flow_rule */
101 }
102 
103 static void efx_tc_delete_rule(struct efx_nic *efx, struct efx_tc_flow_rule *rule)
104 {
105 	efx_mae_delete_rule(efx, rule->fw_id);
106 
107 	/* Release entries in subsidiary tables */
108 	efx_tc_free_action_set_list(efx, &rule->acts, true);
109 	rule->fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
110 }
111 
112 static void efx_tc_flow_free(void *ptr, void *arg)
113 {
114 	struct efx_tc_flow_rule *rule = ptr;
115 	struct efx_nic *efx = arg;
116 
117 	netif_err(efx, drv, efx->net_dev,
118 		  "tc rule %lx still present at teardown, removing\n",
119 		  rule->cookie);
120 
121 	efx_mae_delete_rule(efx, rule->fw_id);
122 
123 	/* Release entries in subsidiary tables */
124 	efx_tc_free_action_set_list(efx, &rule->acts, true);
125 
126 	kfree(rule);
127 }
128 
129 /* Boilerplate for the simple 'copy a field' cases */
130 #define _MAP_KEY_AND_MASK(_name, _type, _tcget, _tcfield, _field)	\
131 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_##_name)) {		\
132 	struct flow_match_##_type fm;					\
133 									\
134 	flow_rule_match_##_tcget(rule, &fm);				\
135 	match->value._field = fm.key->_tcfield;				\
136 	match->mask._field = fm.mask->_tcfield;				\
137 }
138 #define MAP_KEY_AND_MASK(_name, _type, _tcfield, _field)	\
139 	_MAP_KEY_AND_MASK(_name, _type, _type, _tcfield, _field)
140 #define MAP_ENC_KEY_AND_MASK(_name, _type, _tcget, _tcfield, _field)	\
141 	_MAP_KEY_AND_MASK(ENC_##_name, _type, _tcget, _tcfield, _field)
142 
143 static int efx_tc_flower_parse_match(struct efx_nic *efx,
144 				     struct flow_rule *rule,
145 				     struct efx_tc_match *match,
146 				     struct netlink_ext_ack *extack)
147 {
148 	struct flow_dissector *dissector = rule->match.dissector;
149 	unsigned char ipv = 0;
150 
151 	/* Owing to internal TC infelicities, the IPV6_ADDRS key might be set
152 	 * even on IPv4 filters; so rather than relying on dissector->used_keys
153 	 * we check the addr_type in the CONTROL key.  If we don't find it (or
154 	 * it's masked, which should never happen), we treat both IPV4_ADDRS
155 	 * and IPV6_ADDRS as absent.
156 	 */
157 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
158 		struct flow_match_control fm;
159 
160 		flow_rule_match_control(rule, &fm);
161 		if (IS_ALL_ONES(fm.mask->addr_type))
162 			switch (fm.key->addr_type) {
163 			case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
164 				ipv = 4;
165 				break;
166 			case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
167 				ipv = 6;
168 				break;
169 			default:
170 				break;
171 			}
172 
173 		if (fm.mask->flags & FLOW_DIS_IS_FRAGMENT) {
174 			match->value.ip_frag = fm.key->flags & FLOW_DIS_IS_FRAGMENT;
175 			match->mask.ip_frag = true;
176 		}
177 		if (fm.mask->flags & FLOW_DIS_FIRST_FRAG) {
178 			match->value.ip_firstfrag = fm.key->flags & FLOW_DIS_FIRST_FRAG;
179 			match->mask.ip_firstfrag = true;
180 		}
181 		if (fm.mask->flags & ~(FLOW_DIS_IS_FRAGMENT | FLOW_DIS_FIRST_FRAG)) {
182 			NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported match on control.flags %#x",
183 					       fm.mask->flags);
184 			return -EOPNOTSUPP;
185 		}
186 	}
187 	if (dissector->used_keys &
188 	    ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
189 	      BIT(FLOW_DISSECTOR_KEY_BASIC) |
190 	      BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
191 	      BIT(FLOW_DISSECTOR_KEY_VLAN) |
192 	      BIT(FLOW_DISSECTOR_KEY_CVLAN) |
193 	      BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
194 	      BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
195 	      BIT(FLOW_DISSECTOR_KEY_PORTS) |
196 	      BIT(FLOW_DISSECTOR_KEY_TCP) |
197 	      BIT(FLOW_DISSECTOR_KEY_IP))) {
198 		NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported flower keys %#x",
199 				       dissector->used_keys);
200 		return -EOPNOTSUPP;
201 	}
202 
203 	MAP_KEY_AND_MASK(BASIC, basic, n_proto, eth_proto);
204 	/* Make sure we're IP if any L3/L4 keys used. */
205 	if (!IS_ALL_ONES(match->mask.eth_proto) ||
206 	    !(match->value.eth_proto == htons(ETH_P_IP) ||
207 	      match->value.eth_proto == htons(ETH_P_IPV6)))
208 		if (dissector->used_keys &
209 		    (BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
210 		     BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
211 		     BIT(FLOW_DISSECTOR_KEY_PORTS) |
212 		     BIT(FLOW_DISSECTOR_KEY_IP) |
213 		     BIT(FLOW_DISSECTOR_KEY_TCP))) {
214 			NL_SET_ERR_MSG_FMT_MOD(extack, "L3/L4 flower keys %#x require protocol ipv[46]",
215 					       dissector->used_keys);
216 			return -EINVAL;
217 		}
218 
219 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
220 		struct flow_match_vlan fm;
221 
222 		flow_rule_match_vlan(rule, &fm);
223 		if (fm.mask->vlan_id || fm.mask->vlan_priority || fm.mask->vlan_tpid) {
224 			match->value.vlan_proto[0] = fm.key->vlan_tpid;
225 			match->mask.vlan_proto[0] = fm.mask->vlan_tpid;
226 			match->value.vlan_tci[0] = cpu_to_be16(fm.key->vlan_priority << 13 |
227 							       fm.key->vlan_id);
228 			match->mask.vlan_tci[0] = cpu_to_be16(fm.mask->vlan_priority << 13 |
229 							      fm.mask->vlan_id);
230 		}
231 	}
232 
233 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CVLAN)) {
234 		struct flow_match_vlan fm;
235 
236 		flow_rule_match_cvlan(rule, &fm);
237 		if (fm.mask->vlan_id || fm.mask->vlan_priority || fm.mask->vlan_tpid) {
238 			match->value.vlan_proto[1] = fm.key->vlan_tpid;
239 			match->mask.vlan_proto[1] = fm.mask->vlan_tpid;
240 			match->value.vlan_tci[1] = cpu_to_be16(fm.key->vlan_priority << 13 |
241 							       fm.key->vlan_id);
242 			match->mask.vlan_tci[1] = cpu_to_be16(fm.mask->vlan_priority << 13 |
243 							      fm.mask->vlan_id);
244 		}
245 	}
246 
247 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
248 		struct flow_match_eth_addrs fm;
249 
250 		flow_rule_match_eth_addrs(rule, &fm);
251 		ether_addr_copy(match->value.eth_saddr, fm.key->src);
252 		ether_addr_copy(match->value.eth_daddr, fm.key->dst);
253 		ether_addr_copy(match->mask.eth_saddr, fm.mask->src);
254 		ether_addr_copy(match->mask.eth_daddr, fm.mask->dst);
255 	}
256 
257 	MAP_KEY_AND_MASK(BASIC, basic, ip_proto, ip_proto);
258 	/* Make sure we're TCP/UDP if any L4 keys used. */
259 	if ((match->value.ip_proto != IPPROTO_UDP &&
260 	     match->value.ip_proto != IPPROTO_TCP) || !IS_ALL_ONES(match->mask.ip_proto))
261 		if (dissector->used_keys &
262 		    (BIT(FLOW_DISSECTOR_KEY_PORTS) |
263 		     BIT(FLOW_DISSECTOR_KEY_TCP))) {
264 			NL_SET_ERR_MSG_FMT_MOD(extack, "L4 flower keys %#x require ipproto udp or tcp",
265 					       dissector->used_keys);
266 			return -EINVAL;
267 		}
268 	MAP_KEY_AND_MASK(IP, ip, tos, ip_tos);
269 	MAP_KEY_AND_MASK(IP, ip, ttl, ip_ttl);
270 	if (ipv == 4) {
271 		MAP_KEY_AND_MASK(IPV4_ADDRS, ipv4_addrs, src, src_ip);
272 		MAP_KEY_AND_MASK(IPV4_ADDRS, ipv4_addrs, dst, dst_ip);
273 	}
274 #ifdef CONFIG_IPV6
275 	else if (ipv == 6) {
276 		MAP_KEY_AND_MASK(IPV6_ADDRS, ipv6_addrs, src, src_ip6);
277 		MAP_KEY_AND_MASK(IPV6_ADDRS, ipv6_addrs, dst, dst_ip6);
278 	}
279 #endif
280 	MAP_KEY_AND_MASK(PORTS, ports, src, l4_sport);
281 	MAP_KEY_AND_MASK(PORTS, ports, dst, l4_dport);
282 	MAP_KEY_AND_MASK(TCP, tcp, flags, tcp_flags);
283 
284 	return 0;
285 }
286 
287 /* For details of action order constraints refer to SF-123102-TC-1§12.6.1 */
288 enum efx_tc_action_order {
289 	EFX_TC_AO_COUNT,
290 	EFX_TC_AO_DELIVER
291 };
292 /* Determine whether we can add @new action without violating order */
293 static bool efx_tc_flower_action_order_ok(const struct efx_tc_action_set *act,
294 					  enum efx_tc_action_order new)
295 {
296 	switch (new) {
297 	case EFX_TC_AO_COUNT:
298 		if (act->count)
299 			return false;
300 		fallthrough;
301 	case EFX_TC_AO_DELIVER:
302 		return !act->deliver;
303 	default:
304 		/* Bad caller.  Whatever they wanted to do, say they can't. */
305 		WARN_ON_ONCE(1);
306 		return false;
307 	}
308 }
309 
310 static int efx_tc_flower_replace(struct efx_nic *efx,
311 				 struct net_device *net_dev,
312 				 struct flow_cls_offload *tc,
313 				 struct efx_rep *efv)
314 {
315 	struct flow_rule *fr = flow_cls_offload_flow_rule(tc);
316 	struct netlink_ext_ack *extack = tc->common.extack;
317 	struct efx_tc_flow_rule *rule = NULL, *old;
318 	struct efx_tc_action_set *act = NULL;
319 	const struct flow_action_entry *fa;
320 	struct efx_rep *from_efv, *to_efv;
321 	struct efx_tc_match match;
322 	s64 rc;
323 	int i;
324 
325 	if (!tc_can_offload_extack(efx->net_dev, extack))
326 		return -EOPNOTSUPP;
327 	if (WARN_ON(!efx->tc))
328 		return -ENETDOWN;
329 	if (WARN_ON(!efx->tc->up))
330 		return -ENETDOWN;
331 
332 	from_efv = efx_tc_flower_lookup_efv(efx, net_dev);
333 	if (IS_ERR(from_efv)) {
334 		/* Might be a tunnel decap rule from an indirect block.
335 		 * Support for those not implemented yet.
336 		 */
337 		return -EOPNOTSUPP;
338 	}
339 
340 	if (efv != from_efv) {
341 		/* can't happen */
342 		NL_SET_ERR_MSG_FMT_MOD(extack, "for %s efv is %snull but from_efv is %snull (can't happen)",
343 				       netdev_name(net_dev), efv ? "non-" : "",
344 				       from_efv ? "non-" : "");
345 		return -EINVAL;
346 	}
347 
348 	/* Parse match */
349 	memset(&match, 0, sizeof(match));
350 	rc = efx_tc_flower_external_mport(efx, from_efv);
351 	if (rc < 0) {
352 		NL_SET_ERR_MSG_MOD(extack, "Failed to identify ingress m-port");
353 		return rc;
354 	}
355 	match.value.ingress_port = rc;
356 	match.mask.ingress_port = ~0;
357 	rc = efx_tc_flower_parse_match(efx, fr, &match, extack);
358 	if (rc)
359 		return rc;
360 
361 	if (tc->common.chain_index) {
362 		NL_SET_ERR_MSG_MOD(extack, "No support for nonzero chain_index");
363 		return -EOPNOTSUPP;
364 	}
365 	match.mask.recirc_id = 0xff;
366 
367 	rc = efx_mae_match_check_caps(efx, &match.mask, extack);
368 	if (rc)
369 		return rc;
370 
371 	rule = kzalloc(sizeof(*rule), GFP_USER);
372 	if (!rule)
373 		return -ENOMEM;
374 	INIT_LIST_HEAD(&rule->acts.list);
375 	rule->cookie = tc->cookie;
376 	old = rhashtable_lookup_get_insert_fast(&efx->tc->match_action_ht,
377 						&rule->linkage,
378 						efx_tc_match_action_ht_params);
379 	if (old) {
380 		netif_dbg(efx, drv, efx->net_dev,
381 			  "Already offloaded rule (cookie %lx)\n", tc->cookie);
382 		rc = -EEXIST;
383 		NL_SET_ERR_MSG_MOD(extack, "Rule already offloaded");
384 		goto release;
385 	}
386 
387 	/* Parse actions */
388 	act = kzalloc(sizeof(*act), GFP_USER);
389 	if (!act) {
390 		rc = -ENOMEM;
391 		goto release;
392 	}
393 
394 	flow_action_for_each(i, fa, &fr->action) {
395 		struct efx_tc_action_set save;
396 
397 		if (!act) {
398 			/* more actions after a non-pipe action */
399 			NL_SET_ERR_MSG_MOD(extack, "Action follows non-pipe action");
400 			rc = -EINVAL;
401 			goto release;
402 		}
403 
404 		if ((fa->id == FLOW_ACTION_REDIRECT ||
405 		     fa->id == FLOW_ACTION_MIRRED ||
406 		     fa->id == FLOW_ACTION_DROP) && fa->hw_stats) {
407 			struct efx_tc_counter_index *ctr;
408 
409 			/* Currently the only actions that want stats are
410 			 * mirred and gact (ok, shot, trap, goto-chain), which
411 			 * means we want stats just before delivery.  Also,
412 			 * note that tunnel_key set shouldn't change the length
413 			 * — it's only the subsequent mirred that does that,
414 			 * and the stats are taken _before_ the mirred action
415 			 * happens.
416 			 */
417 			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_COUNT)) {
418 				/* All supported actions that count either steal
419 				 * (gact shot, mirred redirect) or clone act
420 				 * (mirred mirror), so we should never get two
421 				 * count actions on one action_set.
422 				 */
423 				NL_SET_ERR_MSG_MOD(extack, "Count-action conflict (can't happen)");
424 				rc = -EOPNOTSUPP;
425 				goto release;
426 			}
427 
428 			if (!(fa->hw_stats & FLOW_ACTION_HW_STATS_DELAYED)) {
429 				NL_SET_ERR_MSG_FMT_MOD(extack, "hw_stats_type %u not supported (only 'delayed')",
430 						       fa->hw_stats);
431 				rc = -EOPNOTSUPP;
432 				goto release;
433 			}
434 
435 			ctr = efx_tc_flower_get_counter_index(efx, tc->cookie,
436 							      EFX_TC_COUNTER_TYPE_AR);
437 			if (IS_ERR(ctr)) {
438 				rc = PTR_ERR(ctr);
439 				NL_SET_ERR_MSG_MOD(extack, "Failed to obtain a counter");
440 				goto release;
441 			}
442 			act->count = ctr;
443 		}
444 
445 		switch (fa->id) {
446 		case FLOW_ACTION_DROP:
447 			rc = efx_mae_alloc_action_set(efx, act);
448 			if (rc) {
449 				NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (drop)");
450 				goto release;
451 			}
452 			list_add_tail(&act->list, &rule->acts.list);
453 			act = NULL; /* end of the line */
454 			break;
455 		case FLOW_ACTION_REDIRECT:
456 		case FLOW_ACTION_MIRRED:
457 			save = *act;
458 
459 			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_DELIVER)) {
460 				/* can't happen */
461 				rc = -EOPNOTSUPP;
462 				NL_SET_ERR_MSG_MOD(extack, "Deliver action violates action order (can't happen)");
463 				goto release;
464 			}
465 
466 			to_efv = efx_tc_flower_lookup_efv(efx, fa->dev);
467 			if (IS_ERR(to_efv)) {
468 				NL_SET_ERR_MSG_MOD(extack, "Mirred egress device not on switch");
469 				rc = PTR_ERR(to_efv);
470 				goto release;
471 			}
472 			rc = efx_tc_flower_external_mport(efx, to_efv);
473 			if (rc < 0) {
474 				NL_SET_ERR_MSG_MOD(extack, "Failed to identify egress m-port");
475 				goto release;
476 			}
477 			act->dest_mport = rc;
478 			act->deliver = 1;
479 			rc = efx_mae_alloc_action_set(efx, act);
480 			if (rc) {
481 				NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (mirred)");
482 				goto release;
483 			}
484 			list_add_tail(&act->list, &rule->acts.list);
485 			act = NULL;
486 			if (fa->id == FLOW_ACTION_REDIRECT)
487 				break; /* end of the line */
488 			/* Mirror, so continue on with saved act */
489 			save.count = NULL;
490 			act = kzalloc(sizeof(*act), GFP_USER);
491 			if (!act) {
492 				rc = -ENOMEM;
493 				goto release;
494 			}
495 			*act = save;
496 			break;
497 		default:
498 			NL_SET_ERR_MSG_FMT_MOD(extack, "Unhandled action %u",
499 					       fa->id);
500 			rc = -EOPNOTSUPP;
501 			goto release;
502 		}
503 	}
504 
505 	if (act) {
506 		/* Not shot/redirected, so deliver to default dest */
507 		if (from_efv == EFX_EFV_PF)
508 			/* Rule applies to traffic from the wire,
509 			 * and default dest is thus the PF
510 			 */
511 			efx_mae_mport_uplink(efx, &act->dest_mport);
512 		else
513 			/* Representor, so rule applies to traffic from
514 			 * representee, and default dest is thus the rep.
515 			 * All reps use the same mport for delivery
516 			 */
517 			efx_mae_mport_mport(efx, efx->tc->reps_mport_id,
518 					    &act->dest_mport);
519 		act->deliver = 1;
520 		rc = efx_mae_alloc_action_set(efx, act);
521 		if (rc) {
522 			NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (deliver)");
523 			goto release;
524 		}
525 		list_add_tail(&act->list, &rule->acts.list);
526 		act = NULL; /* Prevent double-free in error path */
527 	}
528 
529 	netif_dbg(efx, drv, efx->net_dev,
530 		  "Successfully parsed filter (cookie %lx)\n",
531 		  tc->cookie);
532 
533 	rule->match = match;
534 
535 	rc = efx_mae_alloc_action_set_list(efx, &rule->acts);
536 	if (rc) {
537 		NL_SET_ERR_MSG_MOD(extack, "Failed to write action set list to hw");
538 		goto release;
539 	}
540 	rc = efx_mae_insert_rule(efx, &rule->match, EFX_TC_PRIO_TC,
541 				 rule->acts.fw_id, &rule->fw_id);
542 	if (rc) {
543 		NL_SET_ERR_MSG_MOD(extack, "Failed to insert rule in hw");
544 		goto release_acts;
545 	}
546 	return 0;
547 
548 release_acts:
549 	efx_mae_free_action_set_list(efx, &rule->acts);
550 release:
551 	/* We failed to insert the rule, so free up any entries we created in
552 	 * subsidiary tables.
553 	 */
554 	if (act)
555 		efx_tc_free_action_set(efx, act, false);
556 	if (rule) {
557 		rhashtable_remove_fast(&efx->tc->match_action_ht,
558 				       &rule->linkage,
559 				       efx_tc_match_action_ht_params);
560 		efx_tc_free_action_set_list(efx, &rule->acts, false);
561 	}
562 	kfree(rule);
563 	return rc;
564 }
565 
566 static int efx_tc_flower_destroy(struct efx_nic *efx,
567 				 struct net_device *net_dev,
568 				 struct flow_cls_offload *tc)
569 {
570 	struct netlink_ext_ack *extack = tc->common.extack;
571 	struct efx_tc_flow_rule *rule;
572 
573 	rule = rhashtable_lookup_fast(&efx->tc->match_action_ht, &tc->cookie,
574 				      efx_tc_match_action_ht_params);
575 	if (!rule) {
576 		/* Only log a message if we're the ingress device.  Otherwise
577 		 * it's a foreign filter and we might just not have been
578 		 * interested (e.g. we might not have been the egress device
579 		 * either).
580 		 */
581 		if (!IS_ERR(efx_tc_flower_lookup_efv(efx, net_dev)))
582 			netif_warn(efx, drv, efx->net_dev,
583 				   "Filter %lx not found to remove\n", tc->cookie);
584 		NL_SET_ERR_MSG_MOD(extack, "Flow cookie not found in offloaded rules");
585 		return -ENOENT;
586 	}
587 
588 	/* Remove it from HW */
589 	efx_tc_delete_rule(efx, rule);
590 	/* Delete it from SW */
591 	rhashtable_remove_fast(&efx->tc->match_action_ht, &rule->linkage,
592 			       efx_tc_match_action_ht_params);
593 	netif_dbg(efx, drv, efx->net_dev, "Removed filter %lx\n", rule->cookie);
594 	kfree(rule);
595 	return 0;
596 }
597 
598 static int efx_tc_flower_stats(struct efx_nic *efx, struct net_device *net_dev,
599 			       struct flow_cls_offload *tc)
600 {
601 	struct netlink_ext_ack *extack = tc->common.extack;
602 	struct efx_tc_counter_index *ctr;
603 	struct efx_tc_counter *cnt;
604 	u64 packets, bytes;
605 
606 	ctr = efx_tc_flower_find_counter_index(efx, tc->cookie);
607 	if (!ctr) {
608 		/* See comment in efx_tc_flower_destroy() */
609 		if (!IS_ERR(efx_tc_flower_lookup_efv(efx, net_dev)))
610 			if (net_ratelimit())
611 				netif_warn(efx, drv, efx->net_dev,
612 					   "Filter %lx not found for stats\n",
613 					   tc->cookie);
614 		NL_SET_ERR_MSG_MOD(extack, "Flow cookie not found in offloaded rules");
615 		return -ENOENT;
616 	}
617 	if (WARN_ON(!ctr->cnt)) /* can't happen */
618 		return -EIO;
619 	cnt = ctr->cnt;
620 
621 	spin_lock_bh(&cnt->lock);
622 	/* Report only new pkts/bytes since last time TC asked */
623 	packets = cnt->packets;
624 	bytes = cnt->bytes;
625 	flow_stats_update(&tc->stats, bytes - cnt->old_bytes,
626 			  packets - cnt->old_packets, 0, cnt->touched,
627 			  FLOW_ACTION_HW_STATS_DELAYED);
628 	cnt->old_packets = packets;
629 	cnt->old_bytes = bytes;
630 	spin_unlock_bh(&cnt->lock);
631 	return 0;
632 }
633 
634 int efx_tc_flower(struct efx_nic *efx, struct net_device *net_dev,
635 		  struct flow_cls_offload *tc, struct efx_rep *efv)
636 {
637 	int rc;
638 
639 	if (!efx->tc)
640 		return -EOPNOTSUPP;
641 
642 	mutex_lock(&efx->tc->mutex);
643 	switch (tc->command) {
644 	case FLOW_CLS_REPLACE:
645 		rc = efx_tc_flower_replace(efx, net_dev, tc, efv);
646 		break;
647 	case FLOW_CLS_DESTROY:
648 		rc = efx_tc_flower_destroy(efx, net_dev, tc);
649 		break;
650 	case FLOW_CLS_STATS:
651 		rc = efx_tc_flower_stats(efx, net_dev, tc);
652 		break;
653 	default:
654 		rc = -EOPNOTSUPP;
655 		break;
656 	}
657 	mutex_unlock(&efx->tc->mutex);
658 	return rc;
659 }
660 
661 static int efx_tc_configure_default_rule(struct efx_nic *efx, u32 ing_port,
662 					 u32 eg_port, struct efx_tc_flow_rule *rule)
663 {
664 	struct efx_tc_action_set_list *acts = &rule->acts;
665 	struct efx_tc_match *match = &rule->match;
666 	struct efx_tc_action_set *act;
667 	int rc;
668 
669 	match->value.ingress_port = ing_port;
670 	match->mask.ingress_port = ~0;
671 	act = kzalloc(sizeof(*act), GFP_KERNEL);
672 	if (!act)
673 		return -ENOMEM;
674 	act->deliver = 1;
675 	act->dest_mport = eg_port;
676 	rc = efx_mae_alloc_action_set(efx, act);
677 	if (rc)
678 		goto fail1;
679 	EFX_WARN_ON_PARANOID(!list_empty(&acts->list));
680 	list_add_tail(&act->list, &acts->list);
681 	rc = efx_mae_alloc_action_set_list(efx, acts);
682 	if (rc)
683 		goto fail2;
684 	rc = efx_mae_insert_rule(efx, match, EFX_TC_PRIO_DFLT,
685 				 acts->fw_id, &rule->fw_id);
686 	if (rc)
687 		goto fail3;
688 	return 0;
689 fail3:
690 	efx_mae_free_action_set_list(efx, acts);
691 fail2:
692 	list_del(&act->list);
693 	efx_mae_free_action_set(efx, act->fw_id);
694 fail1:
695 	kfree(act);
696 	return rc;
697 }
698 
699 static int efx_tc_configure_default_rule_pf(struct efx_nic *efx)
700 {
701 	struct efx_tc_flow_rule *rule = &efx->tc->dflt.pf;
702 	u32 ing_port, eg_port;
703 
704 	efx_mae_mport_uplink(efx, &ing_port);
705 	efx_mae_mport_wire(efx, &eg_port);
706 	return efx_tc_configure_default_rule(efx, ing_port, eg_port, rule);
707 }
708 
709 static int efx_tc_configure_default_rule_wire(struct efx_nic *efx)
710 {
711 	struct efx_tc_flow_rule *rule = &efx->tc->dflt.wire;
712 	u32 ing_port, eg_port;
713 
714 	efx_mae_mport_wire(efx, &ing_port);
715 	efx_mae_mport_uplink(efx, &eg_port);
716 	return efx_tc_configure_default_rule(efx, ing_port, eg_port, rule);
717 }
718 
719 int efx_tc_configure_default_rule_rep(struct efx_rep *efv)
720 {
721 	struct efx_tc_flow_rule *rule = &efv->dflt;
722 	struct efx_nic *efx = efv->parent;
723 	u32 ing_port, eg_port;
724 
725 	efx_mae_mport_mport(efx, efv->mport, &ing_port);
726 	efx_mae_mport_mport(efx, efx->tc->reps_mport_id, &eg_port);
727 	return efx_tc_configure_default_rule(efx, ing_port, eg_port, rule);
728 }
729 
730 void efx_tc_deconfigure_default_rule(struct efx_nic *efx,
731 				     struct efx_tc_flow_rule *rule)
732 {
733 	if (rule->fw_id != MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL)
734 		efx_tc_delete_rule(efx, rule);
735 	rule->fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
736 }
737 
738 static int efx_tc_configure_rep_mport(struct efx_nic *efx)
739 {
740 	u32 rep_mport_label;
741 	int rc;
742 
743 	rc = efx_mae_allocate_mport(efx, &efx->tc->reps_mport_id, &rep_mport_label);
744 	if (rc)
745 		return rc;
746 	pci_dbg(efx->pci_dev, "created rep mport 0x%08x (0x%04x)\n",
747 		efx->tc->reps_mport_id, rep_mport_label);
748 	/* Use mport *selector* as vport ID */
749 	efx_mae_mport_mport(efx, efx->tc->reps_mport_id,
750 			    &efx->tc->reps_mport_vport_id);
751 	return 0;
752 }
753 
754 static void efx_tc_deconfigure_rep_mport(struct efx_nic *efx)
755 {
756 	efx_mae_free_mport(efx, efx->tc->reps_mport_id);
757 	efx->tc->reps_mport_id = MAE_MPORT_SELECTOR_NULL;
758 }
759 
760 int efx_tc_insert_rep_filters(struct efx_nic *efx)
761 {
762 	struct efx_filter_spec promisc, allmulti;
763 	int rc;
764 
765 	if (efx->type->is_vf)
766 		return 0;
767 	if (!efx->tc)
768 		return 0;
769 	efx_filter_init_rx(&promisc, EFX_FILTER_PRI_REQUIRED, 0, 0);
770 	efx_filter_set_uc_def(&promisc);
771 	efx_filter_set_vport_id(&promisc, efx->tc->reps_mport_vport_id);
772 	rc = efx_filter_insert_filter(efx, &promisc, false);
773 	if (rc < 0)
774 		return rc;
775 	efx->tc->reps_filter_uc = rc;
776 	efx_filter_init_rx(&allmulti, EFX_FILTER_PRI_REQUIRED, 0, 0);
777 	efx_filter_set_mc_def(&allmulti);
778 	efx_filter_set_vport_id(&allmulti, efx->tc->reps_mport_vport_id);
779 	rc = efx_filter_insert_filter(efx, &allmulti, false);
780 	if (rc < 0)
781 		return rc;
782 	efx->tc->reps_filter_mc = rc;
783 	return 0;
784 }
785 
786 void efx_tc_remove_rep_filters(struct efx_nic *efx)
787 {
788 	if (efx->type->is_vf)
789 		return;
790 	if (!efx->tc)
791 		return;
792 	if (efx->tc->reps_filter_mc >= 0)
793 		efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED, efx->tc->reps_filter_mc);
794 	efx->tc->reps_filter_mc = -1;
795 	if (efx->tc->reps_filter_uc >= 0)
796 		efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED, efx->tc->reps_filter_uc);
797 	efx->tc->reps_filter_uc = -1;
798 }
799 
800 int efx_init_tc(struct efx_nic *efx)
801 {
802 	int rc;
803 
804 	rc = efx_mae_get_caps(efx, efx->tc->caps);
805 	if (rc)
806 		return rc;
807 	if (efx->tc->caps->match_field_count > MAE_NUM_FIELDS)
808 		/* Firmware supports some match fields the driver doesn't know
809 		 * about.  Not fatal, unless any of those fields are required
810 		 * (MAE_FIELD_SUPPORTED_MATCH_ALWAYS) but if so we don't know.
811 		 */
812 		netif_warn(efx, probe, efx->net_dev,
813 			   "FW reports additional match fields %u\n",
814 			   efx->tc->caps->match_field_count);
815 	if (efx->tc->caps->action_prios < EFX_TC_PRIO__NUM) {
816 		netif_err(efx, probe, efx->net_dev,
817 			  "Too few action prios supported (have %u, need %u)\n",
818 			  efx->tc->caps->action_prios, EFX_TC_PRIO__NUM);
819 		return -EIO;
820 	}
821 	rc = efx_tc_configure_default_rule_pf(efx);
822 	if (rc)
823 		return rc;
824 	rc = efx_tc_configure_default_rule_wire(efx);
825 	if (rc)
826 		return rc;
827 	rc = efx_tc_configure_rep_mport(efx);
828 	if (rc)
829 		return rc;
830 	efx->tc->up = true;
831 	rc = flow_indr_dev_register(efx_tc_indr_setup_cb, efx);
832 	if (rc)
833 		return rc;
834 	return 0;
835 }
836 
837 void efx_fini_tc(struct efx_nic *efx)
838 {
839 	/* We can get called even if efx_init_struct_tc() failed */
840 	if (!efx->tc)
841 		return;
842 	if (efx->tc->up)
843 		flow_indr_dev_unregister(efx_tc_indr_setup_cb, efx, efx_tc_block_unbind);
844 	efx_tc_deconfigure_rep_mport(efx);
845 	efx_tc_deconfigure_default_rule(efx, &efx->tc->dflt.pf);
846 	efx_tc_deconfigure_default_rule(efx, &efx->tc->dflt.wire);
847 	efx->tc->up = false;
848 }
849 
850 int efx_init_struct_tc(struct efx_nic *efx)
851 {
852 	int rc;
853 
854 	if (efx->type->is_vf)
855 		return 0;
856 
857 	efx->tc = kzalloc(sizeof(*efx->tc), GFP_KERNEL);
858 	if (!efx->tc)
859 		return -ENOMEM;
860 	efx->tc->caps = kzalloc(sizeof(struct mae_caps), GFP_KERNEL);
861 	if (!efx->tc->caps) {
862 		rc = -ENOMEM;
863 		goto fail_alloc_caps;
864 	}
865 	INIT_LIST_HEAD(&efx->tc->block_list);
866 
867 	mutex_init(&efx->tc->mutex);
868 	init_waitqueue_head(&efx->tc->flush_wq);
869 	rc = efx_tc_init_counters(efx);
870 	if (rc < 0)
871 		goto fail_counters;
872 	rc = rhashtable_init(&efx->tc->match_action_ht, &efx_tc_match_action_ht_params);
873 	if (rc < 0)
874 		goto fail_match_action_ht;
875 	efx->tc->reps_filter_uc = -1;
876 	efx->tc->reps_filter_mc = -1;
877 	INIT_LIST_HEAD(&efx->tc->dflt.pf.acts.list);
878 	efx->tc->dflt.pf.fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
879 	INIT_LIST_HEAD(&efx->tc->dflt.wire.acts.list);
880 	efx->tc->dflt.wire.fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
881 	efx->extra_channel_type[EFX_EXTRA_CHANNEL_TC] = &efx_tc_channel_type;
882 	return 0;
883 fail_match_action_ht:
884 	efx_tc_destroy_counters(efx);
885 fail_counters:
886 	mutex_destroy(&efx->tc->mutex);
887 	kfree(efx->tc->caps);
888 fail_alloc_caps:
889 	kfree(efx->tc);
890 	efx->tc = NULL;
891 	return rc;
892 }
893 
894 void efx_fini_struct_tc(struct efx_nic *efx)
895 {
896 	if (!efx->tc)
897 		return;
898 
899 	mutex_lock(&efx->tc->mutex);
900 	EFX_WARN_ON_PARANOID(efx->tc->dflt.pf.fw_id !=
901 			     MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL);
902 	EFX_WARN_ON_PARANOID(efx->tc->dflt.wire.fw_id !=
903 			     MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL);
904 	rhashtable_free_and_destroy(&efx->tc->match_action_ht, efx_tc_flow_free,
905 				    efx);
906 	efx_tc_fini_counters(efx);
907 	mutex_unlock(&efx->tc->mutex);
908 	mutex_destroy(&efx->tc->mutex);
909 	kfree(efx->tc->caps);
910 	kfree(efx->tc);
911 	efx->tc = NULL;
912 }
913