xref: /linux/drivers/net/ethernet/sfc/tc.c (revision 4ae68b26c3ab5a82aa271e6e9fc9b1a06e1d6b40)
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 <net/vxlan.h>
14 #include <net/geneve.h>
15 #include "tc.h"
16 #include "tc_bindings.h"
17 #include "tc_encap_actions.h"
18 #include "mae.h"
19 #include "ef100_rep.h"
20 #include "efx.h"
21 
22 enum efx_encap_type efx_tc_indr_netdev_type(struct net_device *net_dev)
23 {
24 	if (netif_is_vxlan(net_dev))
25 		return EFX_ENCAP_TYPE_VXLAN;
26 	if (netif_is_geneve(net_dev))
27 		return EFX_ENCAP_TYPE_GENEVE;
28 
29 	return EFX_ENCAP_TYPE_NONE;
30 }
31 
32 #define EFX_EFV_PF	NULL
33 /* Look up the representor information (efv) for a device.
34  * May return NULL for the PF (us), or an error pointer for a device that
35  * isn't supported as a TC offload endpoint
36  */
37 struct efx_rep *efx_tc_flower_lookup_efv(struct efx_nic *efx,
38 					 struct net_device *dev)
39 {
40 	struct efx_rep *efv;
41 
42 	if (!dev)
43 		return ERR_PTR(-EOPNOTSUPP);
44 	/* Is it us (the PF)? */
45 	if (dev == efx->net_dev)
46 		return EFX_EFV_PF;
47 	/* Is it an efx vfrep at all? */
48 	if (dev->netdev_ops != &efx_ef100_rep_netdev_ops)
49 		return ERR_PTR(-EOPNOTSUPP);
50 	/* Is it ours?  We don't support TC rules that include another
51 	 * EF100's netdevices (not even on another port of the same NIC).
52 	 */
53 	efv = netdev_priv(dev);
54 	if (efv->parent != efx)
55 		return ERR_PTR(-EOPNOTSUPP);
56 	return efv;
57 }
58 
59 /* Convert a driver-internal vport ID into an internal device (PF or VF) */
60 static s64 efx_tc_flower_internal_mport(struct efx_nic *efx, struct efx_rep *efv)
61 {
62 	u32 mport;
63 
64 	if (IS_ERR(efv))
65 		return PTR_ERR(efv);
66 	if (!efv) /* device is PF (us) */
67 		efx_mae_mport_uplink(efx, &mport);
68 	else /* device is repr */
69 		efx_mae_mport_mport(efx, efv->mport, &mport);
70 	return mport;
71 }
72 
73 /* Convert a driver-internal vport ID into an external device (wire or VF) */
74 s64 efx_tc_flower_external_mport(struct efx_nic *efx, struct efx_rep *efv)
75 {
76 	u32 mport;
77 
78 	if (IS_ERR(efv))
79 		return PTR_ERR(efv);
80 	if (!efv) /* device is PF (us) */
81 		efx_mae_mport_wire(efx, &mport);
82 	else /* device is repr */
83 		efx_mae_mport_mport(efx, efv->mport, &mport);
84 	return mport;
85 }
86 
87 static const struct rhashtable_params efx_tc_encap_match_ht_params = {
88 	.key_len	= offsetof(struct efx_tc_encap_match, linkage),
89 	.key_offset	= 0,
90 	.head_offset	= offsetof(struct efx_tc_encap_match, linkage),
91 };
92 
93 static const struct rhashtable_params efx_tc_match_action_ht_params = {
94 	.key_len	= sizeof(unsigned long),
95 	.key_offset	= offsetof(struct efx_tc_flow_rule, cookie),
96 	.head_offset	= offsetof(struct efx_tc_flow_rule, linkage),
97 };
98 
99 static void efx_tc_free_action_set(struct efx_nic *efx,
100 				   struct efx_tc_action_set *act, bool in_hw)
101 {
102 	/* Failure paths calling this on the 'cursor' action set in_hw=false,
103 	 * because if the alloc had succeeded we'd've put it in acts.list and
104 	 * not still have it in act.
105 	 */
106 	if (in_hw) {
107 		efx_mae_free_action_set(efx, act->fw_id);
108 		/* in_hw is true iff we are on an acts.list; make sure to
109 		 * remove ourselves from that list before we are freed.
110 		 */
111 		list_del(&act->list);
112 	}
113 	if (act->count) {
114 		spin_lock_bh(&act->count->cnt->lock);
115 		if (!list_empty(&act->count_user))
116 			list_del(&act->count_user);
117 		spin_unlock_bh(&act->count->cnt->lock);
118 		efx_tc_flower_put_counter_index(efx, act->count);
119 	}
120 	if (act->encap_md) {
121 		list_del(&act->encap_user);
122 		efx_tc_flower_release_encap_md(efx, act->encap_md);
123 	}
124 	kfree(act);
125 }
126 
127 static void efx_tc_free_action_set_list(struct efx_nic *efx,
128 					struct efx_tc_action_set_list *acts,
129 					bool in_hw)
130 {
131 	struct efx_tc_action_set *act, *next;
132 
133 	/* Failure paths set in_hw=false, because usually the acts didn't get
134 	 * to efx_mae_alloc_action_set_list(); if they did, the failure tree
135 	 * has a separate efx_mae_free_action_set_list() before calling us.
136 	 */
137 	if (in_hw)
138 		efx_mae_free_action_set_list(efx, acts);
139 	/* Any act that's on the list will be in_hw even if the list isn't */
140 	list_for_each_entry_safe(act, next, &acts->list, list)
141 		efx_tc_free_action_set(efx, act, true);
142 	/* Don't kfree, as acts is embedded inside a struct efx_tc_flow_rule */
143 }
144 
145 /* Boilerplate for the simple 'copy a field' cases */
146 #define _MAP_KEY_AND_MASK(_name, _type, _tcget, _tcfield, _field)	\
147 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_##_name)) {		\
148 	struct flow_match_##_type fm;					\
149 									\
150 	flow_rule_match_##_tcget(rule, &fm);				\
151 	match->value._field = fm.key->_tcfield;				\
152 	match->mask._field = fm.mask->_tcfield;				\
153 }
154 #define MAP_KEY_AND_MASK(_name, _type, _tcfield, _field)	\
155 	_MAP_KEY_AND_MASK(_name, _type, _type, _tcfield, _field)
156 #define MAP_ENC_KEY_AND_MASK(_name, _type, _tcget, _tcfield, _field)	\
157 	_MAP_KEY_AND_MASK(ENC_##_name, _type, _tcget, _tcfield, _field)
158 
159 static int efx_tc_flower_parse_match(struct efx_nic *efx,
160 				     struct flow_rule *rule,
161 				     struct efx_tc_match *match,
162 				     struct netlink_ext_ack *extack)
163 {
164 	struct flow_dissector *dissector = rule->match.dissector;
165 	unsigned char ipv = 0;
166 
167 	/* Owing to internal TC infelicities, the IPV6_ADDRS key might be set
168 	 * even on IPv4 filters; so rather than relying on dissector->used_keys
169 	 * we check the addr_type in the CONTROL key.  If we don't find it (or
170 	 * it's masked, which should never happen), we treat both IPV4_ADDRS
171 	 * and IPV6_ADDRS as absent.
172 	 */
173 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
174 		struct flow_match_control fm;
175 
176 		flow_rule_match_control(rule, &fm);
177 		if (IS_ALL_ONES(fm.mask->addr_type))
178 			switch (fm.key->addr_type) {
179 			case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
180 				ipv = 4;
181 				break;
182 			case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
183 				ipv = 6;
184 				break;
185 			default:
186 				break;
187 			}
188 
189 		if (fm.mask->flags & FLOW_DIS_IS_FRAGMENT) {
190 			match->value.ip_frag = fm.key->flags & FLOW_DIS_IS_FRAGMENT;
191 			match->mask.ip_frag = true;
192 		}
193 		if (fm.mask->flags & FLOW_DIS_FIRST_FRAG) {
194 			match->value.ip_firstfrag = fm.key->flags & FLOW_DIS_FIRST_FRAG;
195 			match->mask.ip_firstfrag = true;
196 		}
197 		if (fm.mask->flags & ~(FLOW_DIS_IS_FRAGMENT | FLOW_DIS_FIRST_FRAG)) {
198 			NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported match on control.flags %#x",
199 					       fm.mask->flags);
200 			return -EOPNOTSUPP;
201 		}
202 	}
203 	if (dissector->used_keys &
204 	    ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
205 	      BIT(FLOW_DISSECTOR_KEY_BASIC) |
206 	      BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
207 	      BIT(FLOW_DISSECTOR_KEY_VLAN) |
208 	      BIT(FLOW_DISSECTOR_KEY_CVLAN) |
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_ENC_KEYID) |
213 	      BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
214 	      BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
215 	      BIT(FLOW_DISSECTOR_KEY_ENC_IP) |
216 	      BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
217 	      BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL) |
218 	      BIT(FLOW_DISSECTOR_KEY_TCP) |
219 	      BIT(FLOW_DISSECTOR_KEY_IP))) {
220 		NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported flower keys %#x",
221 				       dissector->used_keys);
222 		return -EOPNOTSUPP;
223 	}
224 
225 	MAP_KEY_AND_MASK(BASIC, basic, n_proto, eth_proto);
226 	/* Make sure we're IP if any L3/L4 keys used. */
227 	if (!IS_ALL_ONES(match->mask.eth_proto) ||
228 	    !(match->value.eth_proto == htons(ETH_P_IP) ||
229 	      match->value.eth_proto == htons(ETH_P_IPV6)))
230 		if (dissector->used_keys &
231 		    (BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
232 		     BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
233 		     BIT(FLOW_DISSECTOR_KEY_PORTS) |
234 		     BIT(FLOW_DISSECTOR_KEY_IP) |
235 		     BIT(FLOW_DISSECTOR_KEY_TCP))) {
236 			NL_SET_ERR_MSG_FMT_MOD(extack, "L3/L4 flower keys %#x require protocol ipv[46]",
237 					       dissector->used_keys);
238 			return -EINVAL;
239 		}
240 
241 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
242 		struct flow_match_vlan fm;
243 
244 		flow_rule_match_vlan(rule, &fm);
245 		if (fm.mask->vlan_id || fm.mask->vlan_priority || fm.mask->vlan_tpid) {
246 			match->value.vlan_proto[0] = fm.key->vlan_tpid;
247 			match->mask.vlan_proto[0] = fm.mask->vlan_tpid;
248 			match->value.vlan_tci[0] = cpu_to_be16(fm.key->vlan_priority << 13 |
249 							       fm.key->vlan_id);
250 			match->mask.vlan_tci[0] = cpu_to_be16(fm.mask->vlan_priority << 13 |
251 							      fm.mask->vlan_id);
252 		}
253 	}
254 
255 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CVLAN)) {
256 		struct flow_match_vlan fm;
257 
258 		flow_rule_match_cvlan(rule, &fm);
259 		if (fm.mask->vlan_id || fm.mask->vlan_priority || fm.mask->vlan_tpid) {
260 			match->value.vlan_proto[1] = fm.key->vlan_tpid;
261 			match->mask.vlan_proto[1] = fm.mask->vlan_tpid;
262 			match->value.vlan_tci[1] = cpu_to_be16(fm.key->vlan_priority << 13 |
263 							       fm.key->vlan_id);
264 			match->mask.vlan_tci[1] = cpu_to_be16(fm.mask->vlan_priority << 13 |
265 							      fm.mask->vlan_id);
266 		}
267 	}
268 
269 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
270 		struct flow_match_eth_addrs fm;
271 
272 		flow_rule_match_eth_addrs(rule, &fm);
273 		ether_addr_copy(match->value.eth_saddr, fm.key->src);
274 		ether_addr_copy(match->value.eth_daddr, fm.key->dst);
275 		ether_addr_copy(match->mask.eth_saddr, fm.mask->src);
276 		ether_addr_copy(match->mask.eth_daddr, fm.mask->dst);
277 	}
278 
279 	MAP_KEY_AND_MASK(BASIC, basic, ip_proto, ip_proto);
280 	/* Make sure we're TCP/UDP if any L4 keys used. */
281 	if ((match->value.ip_proto != IPPROTO_UDP &&
282 	     match->value.ip_proto != IPPROTO_TCP) || !IS_ALL_ONES(match->mask.ip_proto))
283 		if (dissector->used_keys &
284 		    (BIT(FLOW_DISSECTOR_KEY_PORTS) |
285 		     BIT(FLOW_DISSECTOR_KEY_TCP))) {
286 			NL_SET_ERR_MSG_FMT_MOD(extack, "L4 flower keys %#x require ipproto udp or tcp",
287 					       dissector->used_keys);
288 			return -EINVAL;
289 		}
290 	MAP_KEY_AND_MASK(IP, ip, tos, ip_tos);
291 	MAP_KEY_AND_MASK(IP, ip, ttl, ip_ttl);
292 	if (ipv == 4) {
293 		MAP_KEY_AND_MASK(IPV4_ADDRS, ipv4_addrs, src, src_ip);
294 		MAP_KEY_AND_MASK(IPV4_ADDRS, ipv4_addrs, dst, dst_ip);
295 	}
296 #ifdef CONFIG_IPV6
297 	else if (ipv == 6) {
298 		MAP_KEY_AND_MASK(IPV6_ADDRS, ipv6_addrs, src, src_ip6);
299 		MAP_KEY_AND_MASK(IPV6_ADDRS, ipv6_addrs, dst, dst_ip6);
300 	}
301 #endif
302 	MAP_KEY_AND_MASK(PORTS, ports, src, l4_sport);
303 	MAP_KEY_AND_MASK(PORTS, ports, dst, l4_dport);
304 	MAP_KEY_AND_MASK(TCP, tcp, flags, tcp_flags);
305 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
306 		struct flow_match_control fm;
307 
308 		flow_rule_match_enc_control(rule, &fm);
309 		if (fm.mask->flags) {
310 			NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported match on enc_control.flags %#x",
311 					       fm.mask->flags);
312 			return -EOPNOTSUPP;
313 		}
314 		if (!IS_ALL_ONES(fm.mask->addr_type)) {
315 			NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported enc addr_type mask %u (key %u)",
316 					       fm.mask->addr_type,
317 					       fm.key->addr_type);
318 			return -EOPNOTSUPP;
319 		}
320 		switch (fm.key->addr_type) {
321 		case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
322 			MAP_ENC_KEY_AND_MASK(IPV4_ADDRS, ipv4_addrs, enc_ipv4_addrs,
323 					     src, enc_src_ip);
324 			MAP_ENC_KEY_AND_MASK(IPV4_ADDRS, ipv4_addrs, enc_ipv4_addrs,
325 					     dst, enc_dst_ip);
326 			break;
327 #ifdef CONFIG_IPV6
328 		case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
329 			MAP_ENC_KEY_AND_MASK(IPV6_ADDRS, ipv6_addrs, enc_ipv6_addrs,
330 					     src, enc_src_ip6);
331 			MAP_ENC_KEY_AND_MASK(IPV6_ADDRS, ipv6_addrs, enc_ipv6_addrs,
332 					     dst, enc_dst_ip6);
333 			break;
334 #endif
335 		default:
336 			NL_SET_ERR_MSG_FMT_MOD(extack,
337 					       "Unsupported enc addr_type %u (supported are IPv4, IPv6)",
338 					       fm.key->addr_type);
339 			return -EOPNOTSUPP;
340 		}
341 		MAP_ENC_KEY_AND_MASK(IP, ip, enc_ip, tos, enc_ip_tos);
342 		MAP_ENC_KEY_AND_MASK(IP, ip, enc_ip, ttl, enc_ip_ttl);
343 		MAP_ENC_KEY_AND_MASK(PORTS, ports, enc_ports, src, enc_sport);
344 		MAP_ENC_KEY_AND_MASK(PORTS, ports, enc_ports, dst, enc_dport);
345 		MAP_ENC_KEY_AND_MASK(KEYID, enc_keyid, enc_keyid, keyid, enc_keyid);
346 	} else if (dissector->used_keys &
347 		   (BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
348 		    BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
349 		    BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
350 		    BIT(FLOW_DISSECTOR_KEY_ENC_IP) |
351 		    BIT(FLOW_DISSECTOR_KEY_ENC_PORTS))) {
352 		NL_SET_ERR_MSG_FMT_MOD(extack, "Flower enc keys require enc_control (keys: %#x)",
353 				       dissector->used_keys);
354 		return -EOPNOTSUPP;
355 	}
356 
357 	return 0;
358 }
359 
360 static void efx_tc_flower_release_encap_match(struct efx_nic *efx,
361 					      struct efx_tc_encap_match *encap)
362 {
363 	int rc;
364 
365 	if (!refcount_dec_and_test(&encap->ref))
366 		return; /* still in use */
367 
368 	if (encap->type == EFX_TC_EM_DIRECT) {
369 		rc = efx_mae_unregister_encap_match(efx, encap);
370 		if (rc)
371 			/* Display message but carry on and remove entry from our
372 			 * SW tables, because there's not much we can do about it.
373 			 */
374 			netif_err(efx, drv, efx->net_dev,
375 				  "Failed to release encap match %#x, rc %d\n",
376 				  encap->fw_id, rc);
377 	}
378 	rhashtable_remove_fast(&efx->tc->encap_match_ht, &encap->linkage,
379 			       efx_tc_encap_match_ht_params);
380 	if (encap->pseudo)
381 		efx_tc_flower_release_encap_match(efx, encap->pseudo);
382 	kfree(encap);
383 }
384 
385 static int efx_tc_flower_record_encap_match(struct efx_nic *efx,
386 					    struct efx_tc_match *match,
387 					    enum efx_encap_type type,
388 					    enum efx_tc_em_pseudo_type em_type,
389 					    u8 child_ip_tos_mask,
390 					    __be16 child_udp_sport_mask,
391 					    struct netlink_ext_ack *extack)
392 {
393 	struct efx_tc_encap_match *encap, *old, *pseudo = NULL;
394 	bool ipv6 = false;
395 	int rc;
396 
397 	/* We require that the socket-defining fields (IP addrs and UDP dest
398 	 * port) are present and exact-match.  Other fields may only be used
399 	 * if the field-set (and any masks) are the same for all encap
400 	 * matches on the same <sip,dip,dport> tuple; this is enforced by
401 	 * pseudo encap matches.
402 	 */
403 	if (match->mask.enc_dst_ip | match->mask.enc_src_ip) {
404 		if (!IS_ALL_ONES(match->mask.enc_dst_ip)) {
405 			NL_SET_ERR_MSG_MOD(extack,
406 					   "Egress encap match is not exact on dst IP address");
407 			return -EOPNOTSUPP;
408 		}
409 		if (!IS_ALL_ONES(match->mask.enc_src_ip)) {
410 			NL_SET_ERR_MSG_MOD(extack,
411 					   "Egress encap match is not exact on src IP address");
412 			return -EOPNOTSUPP;
413 		}
414 #ifdef CONFIG_IPV6
415 		if (!ipv6_addr_any(&match->mask.enc_dst_ip6) ||
416 		    !ipv6_addr_any(&match->mask.enc_src_ip6)) {
417 			NL_SET_ERR_MSG_MOD(extack,
418 					   "Egress encap match on both IPv4 and IPv6, don't understand");
419 			return -EOPNOTSUPP;
420 		}
421 	} else {
422 		ipv6 = true;
423 		if (!efx_ipv6_addr_all_ones(&match->mask.enc_dst_ip6)) {
424 			NL_SET_ERR_MSG_MOD(extack,
425 					   "Egress encap match is not exact on dst IP address");
426 			return -EOPNOTSUPP;
427 		}
428 		if (!efx_ipv6_addr_all_ones(&match->mask.enc_src_ip6)) {
429 			NL_SET_ERR_MSG_MOD(extack,
430 					   "Egress encap match is not exact on src IP address");
431 			return -EOPNOTSUPP;
432 		}
433 #endif
434 	}
435 	if (!IS_ALL_ONES(match->mask.enc_dport)) {
436 		NL_SET_ERR_MSG_MOD(extack, "Egress encap match is not exact on dst UDP port");
437 		return -EOPNOTSUPP;
438 	}
439 	if (match->mask.enc_sport || match->mask.enc_ip_tos) {
440 		struct efx_tc_match pmatch = *match;
441 
442 		if (em_type == EFX_TC_EM_PSEUDO_MASK) { /* can't happen */
443 			NL_SET_ERR_MSG_MOD(extack, "Bad recursion in egress encap match handler");
444 			return -EOPNOTSUPP;
445 		}
446 		pmatch.value.enc_ip_tos = 0;
447 		pmatch.mask.enc_ip_tos = 0;
448 		pmatch.value.enc_sport = 0;
449 		pmatch.mask.enc_sport = 0;
450 		rc = efx_tc_flower_record_encap_match(efx, &pmatch, type,
451 						      EFX_TC_EM_PSEUDO_MASK,
452 						      match->mask.enc_ip_tos,
453 						      match->mask.enc_sport,
454 						      extack);
455 		if (rc)
456 			return rc;
457 		pseudo = pmatch.encap;
458 	}
459 	if (match->mask.enc_ip_ttl) {
460 		NL_SET_ERR_MSG_MOD(extack, "Egress encap match on IP TTL not supported");
461 		rc = -EOPNOTSUPP;
462 		goto fail_pseudo;
463 	}
464 
465 	rc = efx_mae_check_encap_match_caps(efx, ipv6, match->mask.enc_ip_tos,
466 					    match->mask.enc_sport, extack);
467 	if (rc)
468 		goto fail_pseudo;
469 
470 	encap = kzalloc(sizeof(*encap), GFP_USER);
471 	if (!encap) {
472 		rc = -ENOMEM;
473 		goto fail_pseudo;
474 	}
475 	encap->src_ip = match->value.enc_src_ip;
476 	encap->dst_ip = match->value.enc_dst_ip;
477 #ifdef CONFIG_IPV6
478 	encap->src_ip6 = match->value.enc_src_ip6;
479 	encap->dst_ip6 = match->value.enc_dst_ip6;
480 #endif
481 	encap->udp_dport = match->value.enc_dport;
482 	encap->tun_type = type;
483 	encap->ip_tos = match->value.enc_ip_tos;
484 	encap->ip_tos_mask = match->mask.enc_ip_tos;
485 	encap->child_ip_tos_mask = child_ip_tos_mask;
486 	encap->udp_sport = match->value.enc_sport;
487 	encap->udp_sport_mask = match->mask.enc_sport;
488 	encap->child_udp_sport_mask = child_udp_sport_mask;
489 	encap->type = em_type;
490 	encap->pseudo = pseudo;
491 	old = rhashtable_lookup_get_insert_fast(&efx->tc->encap_match_ht,
492 						&encap->linkage,
493 						efx_tc_encap_match_ht_params);
494 	if (old) {
495 		/* don't need our new entry */
496 		kfree(encap);
497 		if (pseudo) /* don't need our new pseudo either */
498 			efx_tc_flower_release_encap_match(efx, pseudo);
499 		/* check old and new em_types are compatible */
500 		switch (old->type) {
501 		case EFX_TC_EM_DIRECT:
502 			/* old EM is in hardware, so mustn't overlap with a
503 			 * pseudo, but may be shared with another direct EM
504 			 */
505 			if (em_type == EFX_TC_EM_DIRECT)
506 				break;
507 			NL_SET_ERR_MSG_MOD(extack, "Pseudo encap match conflicts with existing direct entry");
508 			return -EEXIST;
509 		case EFX_TC_EM_PSEUDO_MASK:
510 			/* old EM is protecting a ToS- or src port-qualified
511 			 * filter, so may only be shared with another pseudo
512 			 * for the same ToS and src port masks.
513 			 */
514 			if (em_type != EFX_TC_EM_PSEUDO_MASK) {
515 				NL_SET_ERR_MSG_FMT_MOD(extack,
516 						       "%s encap match conflicts with existing pseudo(MASK) entry",
517 						       em_type ? "Pseudo" : "Direct");
518 				return -EEXIST;
519 			}
520 			if (child_ip_tos_mask != old->child_ip_tos_mask) {
521 				NL_SET_ERR_MSG_FMT_MOD(extack,
522 						       "Pseudo encap match for TOS mask %#04x conflicts with existing pseudo(MASK) entry for TOS mask %#04x",
523 						       child_ip_tos_mask,
524 						       old->child_ip_tos_mask);
525 				return -EEXIST;
526 			}
527 			if (child_udp_sport_mask != old->child_udp_sport_mask) {
528 				NL_SET_ERR_MSG_FMT_MOD(extack,
529 						       "Pseudo encap match for UDP src port mask %#x conflicts with existing pseudo(MASK) entry for mask %#x",
530 						       child_udp_sport_mask,
531 						       old->child_udp_sport_mask);
532 				return -EEXIST;
533 			}
534 			break;
535 		default: /* Unrecognised pseudo-type.  Just say no */
536 			NL_SET_ERR_MSG_FMT_MOD(extack,
537 					       "%s encap match conflicts with existing pseudo(%d) entry",
538 					       em_type ? "Pseudo" : "Direct",
539 					       old->type);
540 			return -EEXIST;
541 		}
542 		/* check old and new tun_types are compatible */
543 		if (old->tun_type != type) {
544 			NL_SET_ERR_MSG_FMT_MOD(extack,
545 					       "Egress encap match with conflicting tun_type %u != %u",
546 					       old->tun_type, type);
547 			return -EEXIST;
548 		}
549 		if (!refcount_inc_not_zero(&old->ref))
550 			return -EAGAIN;
551 		/* existing entry found */
552 		encap = old;
553 	} else {
554 		if (em_type == EFX_TC_EM_DIRECT) {
555 			rc = efx_mae_register_encap_match(efx, encap);
556 			if (rc) {
557 				NL_SET_ERR_MSG_MOD(extack, "Failed to record egress encap match in HW");
558 				goto fail;
559 			}
560 		}
561 		refcount_set(&encap->ref, 1);
562 	}
563 	match->encap = encap;
564 	return 0;
565 fail:
566 	rhashtable_remove_fast(&efx->tc->encap_match_ht, &encap->linkage,
567 			       efx_tc_encap_match_ht_params);
568 	kfree(encap);
569 fail_pseudo:
570 	if (pseudo)
571 		efx_tc_flower_release_encap_match(efx, pseudo);
572 	return rc;
573 }
574 
575 static void efx_tc_delete_rule(struct efx_nic *efx, struct efx_tc_flow_rule *rule)
576 {
577 	efx_mae_delete_rule(efx, rule->fw_id);
578 
579 	/* Release entries in subsidiary tables */
580 	efx_tc_free_action_set_list(efx, &rule->acts, true);
581 	if (rule->match.encap)
582 		efx_tc_flower_release_encap_match(efx, rule->match.encap);
583 	rule->fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
584 }
585 
586 static const char *efx_tc_encap_type_name(enum efx_encap_type typ)
587 {
588 	switch (typ) {
589 	case EFX_ENCAP_TYPE_NONE:
590 		return "none";
591 	case EFX_ENCAP_TYPE_VXLAN:
592 		return "vxlan";
593 	case EFX_ENCAP_TYPE_GENEVE:
594 		return "geneve";
595 	default:
596 		pr_warn_once("Unknown efx_encap_type %d encountered\n", typ);
597 		return "unknown";
598 	}
599 }
600 
601 /* For details of action order constraints refer to SF-123102-TC-1§12.6.1 */
602 enum efx_tc_action_order {
603 	EFX_TC_AO_DECAP,
604 	EFX_TC_AO_VLAN_POP,
605 	EFX_TC_AO_VLAN_PUSH,
606 	EFX_TC_AO_COUNT,
607 	EFX_TC_AO_ENCAP,
608 	EFX_TC_AO_DELIVER
609 };
610 /* Determine whether we can add @new action without violating order */
611 static bool efx_tc_flower_action_order_ok(const struct efx_tc_action_set *act,
612 					  enum efx_tc_action_order new)
613 {
614 	switch (new) {
615 	case EFX_TC_AO_DECAP:
616 		if (act->decap)
617 			return false;
618 		fallthrough;
619 	case EFX_TC_AO_VLAN_POP:
620 		if (act->vlan_pop >= 2)
621 			return false;
622 		/* If we've already pushed a VLAN, we can't then pop it;
623 		 * the hardware would instead try to pop an existing VLAN
624 		 * before pushing the new one.
625 		 */
626 		if (act->vlan_push)
627 			return false;
628 		fallthrough;
629 	case EFX_TC_AO_VLAN_PUSH:
630 		if (act->vlan_push >= 2)
631 			return false;
632 		fallthrough;
633 	case EFX_TC_AO_COUNT:
634 		if (act->count)
635 			return false;
636 		fallthrough;
637 	case EFX_TC_AO_ENCAP:
638 		if (act->encap_md)
639 			return false;
640 		fallthrough;
641 	case EFX_TC_AO_DELIVER:
642 		return !act->deliver;
643 	default:
644 		/* Bad caller.  Whatever they wanted to do, say they can't. */
645 		WARN_ON_ONCE(1);
646 		return false;
647 	}
648 }
649 
650 static int efx_tc_flower_replace_foreign(struct efx_nic *efx,
651 					 struct net_device *net_dev,
652 					 struct flow_cls_offload *tc)
653 {
654 	struct flow_rule *fr = flow_cls_offload_flow_rule(tc);
655 	struct netlink_ext_ack *extack = tc->common.extack;
656 	struct efx_tc_flow_rule *rule = NULL, *old = NULL;
657 	struct efx_tc_action_set *act = NULL;
658 	bool found = false, uplinked = false;
659 	const struct flow_action_entry *fa;
660 	struct efx_tc_match match;
661 	struct efx_rep *to_efv;
662 	s64 rc;
663 	int i;
664 
665 	/* Parse match */
666 	memset(&match, 0, sizeof(match));
667 	rc = efx_tc_flower_parse_match(efx, fr, &match, NULL);
668 	if (rc)
669 		return rc;
670 	/* The rule as given to us doesn't specify a source netdevice.
671 	 * But, determining whether packets from a VF should match it is
672 	 * complicated, so leave those to the software slowpath: qualify
673 	 * the filter with source m-port == wire.
674 	 */
675 	rc = efx_tc_flower_external_mport(efx, EFX_EFV_PF);
676 	if (rc < 0) {
677 		NL_SET_ERR_MSG_MOD(extack, "Failed to identify ingress m-port for foreign filter");
678 		return rc;
679 	}
680 	match.value.ingress_port = rc;
681 	match.mask.ingress_port = ~0;
682 
683 	if (tc->common.chain_index) {
684 		NL_SET_ERR_MSG_MOD(extack, "No support for nonzero chain_index");
685 		return -EOPNOTSUPP;
686 	}
687 	match.mask.recirc_id = 0xff;
688 
689 	flow_action_for_each(i, fa, &fr->action) {
690 		switch (fa->id) {
691 		case FLOW_ACTION_REDIRECT:
692 		case FLOW_ACTION_MIRRED: /* mirred means mirror here */
693 			to_efv = efx_tc_flower_lookup_efv(efx, fa->dev);
694 			if (IS_ERR(to_efv))
695 				continue;
696 			found = true;
697 			break;
698 		default:
699 			break;
700 		}
701 	}
702 	if (!found) { /* We don't care. */
703 		netif_dbg(efx, drv, efx->net_dev,
704 			  "Ignoring foreign filter that doesn't egdev us\n");
705 		return -EOPNOTSUPP;
706 	}
707 
708 	rc = efx_mae_match_check_caps(efx, &match.mask, NULL);
709 	if (rc)
710 		return rc;
711 
712 	if (efx_tc_match_is_encap(&match.mask)) {
713 		enum efx_encap_type type;
714 
715 		type = efx_tc_indr_netdev_type(net_dev);
716 		if (type == EFX_ENCAP_TYPE_NONE) {
717 			NL_SET_ERR_MSG_MOD(extack,
718 					   "Egress encap match on unsupported tunnel device");
719 			return -EOPNOTSUPP;
720 		}
721 
722 		rc = efx_mae_check_encap_type_supported(efx, type);
723 		if (rc) {
724 			NL_SET_ERR_MSG_FMT_MOD(extack,
725 					       "Firmware reports no support for %s encap match",
726 					       efx_tc_encap_type_name(type));
727 			return rc;
728 		}
729 
730 		rc = efx_tc_flower_record_encap_match(efx, &match, type,
731 						      EFX_TC_EM_DIRECT, 0, 0,
732 						      extack);
733 		if (rc)
734 			return rc;
735 	} else {
736 		/* This is not a tunnel decap rule, ignore it */
737 		netif_dbg(efx, drv, efx->net_dev,
738 			  "Ignoring foreign filter without encap match\n");
739 		return -EOPNOTSUPP;
740 	}
741 
742 	rule = kzalloc(sizeof(*rule), GFP_USER);
743 	if (!rule) {
744 		rc = -ENOMEM;
745 		goto out_free;
746 	}
747 	INIT_LIST_HEAD(&rule->acts.list);
748 	rule->cookie = tc->cookie;
749 	old = rhashtable_lookup_get_insert_fast(&efx->tc->match_action_ht,
750 						&rule->linkage,
751 						efx_tc_match_action_ht_params);
752 	if (old) {
753 		netif_dbg(efx, drv, efx->net_dev,
754 			  "Ignoring already-offloaded rule (cookie %lx)\n",
755 			  tc->cookie);
756 		rc = -EEXIST;
757 		goto out_free;
758 	}
759 
760 	act = kzalloc(sizeof(*act), GFP_USER);
761 	if (!act) {
762 		rc = -ENOMEM;
763 		goto release;
764 	}
765 
766 	/* Parse actions.  For foreign rules we only support decap & redirect.
767 	 * See corresponding code in efx_tc_flower_replace() for theory of
768 	 * operation & how 'act' cursor is used.
769 	 */
770 	flow_action_for_each(i, fa, &fr->action) {
771 		struct efx_tc_action_set save;
772 
773 		switch (fa->id) {
774 		case FLOW_ACTION_REDIRECT:
775 		case FLOW_ACTION_MIRRED:
776 			/* See corresponding code in efx_tc_flower_replace() for
777 			 * long explanations of what's going on here.
778 			 */
779 			save = *act;
780 			if (fa->hw_stats) {
781 				struct efx_tc_counter_index *ctr;
782 
783 				if (!(fa->hw_stats & FLOW_ACTION_HW_STATS_DELAYED)) {
784 					NL_SET_ERR_MSG_FMT_MOD(extack,
785 							       "hw_stats_type %u not supported (only 'delayed')",
786 							       fa->hw_stats);
787 					rc = -EOPNOTSUPP;
788 					goto release;
789 				}
790 				if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_COUNT)) {
791 					rc = -EOPNOTSUPP;
792 					goto release;
793 				}
794 
795 				ctr = efx_tc_flower_get_counter_index(efx,
796 								      tc->cookie,
797 								      EFX_TC_COUNTER_TYPE_AR);
798 				if (IS_ERR(ctr)) {
799 					rc = PTR_ERR(ctr);
800 					NL_SET_ERR_MSG_MOD(extack, "Failed to obtain a counter");
801 					goto release;
802 				}
803 				act->count = ctr;
804 				INIT_LIST_HEAD(&act->count_user);
805 			}
806 
807 			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_DELIVER)) {
808 				/* can't happen */
809 				rc = -EOPNOTSUPP;
810 				NL_SET_ERR_MSG_MOD(extack,
811 						   "Deliver action violates action order (can't happen)");
812 				goto release;
813 			}
814 			to_efv = efx_tc_flower_lookup_efv(efx, fa->dev);
815 			/* PF implies egdev is us, in which case we really
816 			 * want to deliver to the uplink (because this is an
817 			 * ingress filter).  If we don't recognise the egdev
818 			 * at all, then we'd better trap so SW can handle it.
819 			 */
820 			if (IS_ERR(to_efv))
821 				to_efv = EFX_EFV_PF;
822 			if (to_efv == EFX_EFV_PF) {
823 				if (uplinked)
824 					break;
825 				uplinked = true;
826 			}
827 			rc = efx_tc_flower_internal_mport(efx, to_efv);
828 			if (rc < 0) {
829 				NL_SET_ERR_MSG_MOD(extack, "Failed to identify egress m-port");
830 				goto release;
831 			}
832 			act->dest_mport = rc;
833 			act->deliver = 1;
834 			rc = efx_mae_alloc_action_set(efx, act);
835 			if (rc) {
836 				NL_SET_ERR_MSG_MOD(extack,
837 						   "Failed to write action set to hw (mirred)");
838 				goto release;
839 			}
840 			list_add_tail(&act->list, &rule->acts.list);
841 			act = NULL;
842 			if (fa->id == FLOW_ACTION_REDIRECT)
843 				break; /* end of the line */
844 			/* Mirror, so continue on with saved act */
845 			act = kzalloc(sizeof(*act), GFP_USER);
846 			if (!act) {
847 				rc = -ENOMEM;
848 				goto release;
849 			}
850 			*act = save;
851 			break;
852 		case FLOW_ACTION_TUNNEL_DECAP:
853 			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_DECAP)) {
854 				rc = -EINVAL;
855 				NL_SET_ERR_MSG_MOD(extack, "Decap action violates action order");
856 				goto release;
857 			}
858 			act->decap = 1;
859 			/* If we previously delivered/trapped to uplink, now
860 			 * that we've decapped we'll want another copy if we
861 			 * try to deliver/trap to uplink again.
862 			 */
863 			uplinked = false;
864 			break;
865 		default:
866 			NL_SET_ERR_MSG_FMT_MOD(extack, "Unhandled action %u",
867 					       fa->id);
868 			rc = -EOPNOTSUPP;
869 			goto release;
870 		}
871 	}
872 
873 	if (act) {
874 		if (!uplinked) {
875 			/* Not shot/redirected, so deliver to default dest (which is
876 			 * the uplink, as this is an ingress filter)
877 			 */
878 			efx_mae_mport_uplink(efx, &act->dest_mport);
879 			act->deliver = 1;
880 		}
881 		rc = efx_mae_alloc_action_set(efx, act);
882 		if (rc) {
883 			NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (deliver)");
884 			goto release;
885 		}
886 		list_add_tail(&act->list, &rule->acts.list);
887 		act = NULL; /* Prevent double-free in error path */
888 	}
889 
890 	rule->match = match;
891 
892 	netif_dbg(efx, drv, efx->net_dev,
893 		  "Successfully parsed foreign filter (cookie %lx)\n",
894 		  tc->cookie);
895 
896 	rc = efx_mae_alloc_action_set_list(efx, &rule->acts);
897 	if (rc) {
898 		NL_SET_ERR_MSG_MOD(extack, "Failed to write action set list to hw");
899 		goto release;
900 	}
901 	rc = efx_mae_insert_rule(efx, &rule->match, EFX_TC_PRIO_TC,
902 				 rule->acts.fw_id, &rule->fw_id);
903 	if (rc) {
904 		NL_SET_ERR_MSG_MOD(extack, "Failed to insert rule in hw");
905 		goto release_acts;
906 	}
907 	return 0;
908 
909 release_acts:
910 	efx_mae_free_action_set_list(efx, &rule->acts);
911 release:
912 	/* We failed to insert the rule, so free up any entries we created in
913 	 * subsidiary tables.
914 	 */
915 	if (act)
916 		efx_tc_free_action_set(efx, act, false);
917 	if (rule) {
918 		rhashtable_remove_fast(&efx->tc->match_action_ht,
919 				       &rule->linkage,
920 				       efx_tc_match_action_ht_params);
921 		efx_tc_free_action_set_list(efx, &rule->acts, false);
922 	}
923 out_free:
924 	kfree(rule);
925 	if (match.encap)
926 		efx_tc_flower_release_encap_match(efx, match.encap);
927 	return rc;
928 }
929 
930 static int efx_tc_flower_replace(struct efx_nic *efx,
931 				 struct net_device *net_dev,
932 				 struct flow_cls_offload *tc,
933 				 struct efx_rep *efv)
934 {
935 	struct flow_rule *fr = flow_cls_offload_flow_rule(tc);
936 	struct netlink_ext_ack *extack = tc->common.extack;
937 	const struct ip_tunnel_info *encap_info = NULL;
938 	struct efx_tc_flow_rule *rule = NULL, *old;
939 	struct efx_tc_action_set *act = NULL;
940 	const struct flow_action_entry *fa;
941 	struct efx_rep *from_efv, *to_efv;
942 	struct efx_tc_match match;
943 	u32 acts_id;
944 	s64 rc;
945 	int i;
946 
947 	if (!tc_can_offload_extack(efx->net_dev, extack))
948 		return -EOPNOTSUPP;
949 	if (WARN_ON(!efx->tc))
950 		return -ENETDOWN;
951 	if (WARN_ON(!efx->tc->up))
952 		return -ENETDOWN;
953 
954 	from_efv = efx_tc_flower_lookup_efv(efx, net_dev);
955 	if (IS_ERR(from_efv)) {
956 		/* Not from our PF or representors, so probably a tunnel dev */
957 		return efx_tc_flower_replace_foreign(efx, net_dev, tc);
958 	}
959 
960 	if (efv != from_efv) {
961 		/* can't happen */
962 		NL_SET_ERR_MSG_FMT_MOD(extack, "for %s efv is %snull but from_efv is %snull (can't happen)",
963 				       netdev_name(net_dev), efv ? "non-" : "",
964 				       from_efv ? "non-" : "");
965 		return -EINVAL;
966 	}
967 
968 	/* Parse match */
969 	memset(&match, 0, sizeof(match));
970 	rc = efx_tc_flower_external_mport(efx, from_efv);
971 	if (rc < 0) {
972 		NL_SET_ERR_MSG_MOD(extack, "Failed to identify ingress m-port");
973 		return rc;
974 	}
975 	match.value.ingress_port = rc;
976 	match.mask.ingress_port = ~0;
977 	rc = efx_tc_flower_parse_match(efx, fr, &match, extack);
978 	if (rc)
979 		return rc;
980 	if (efx_tc_match_is_encap(&match.mask)) {
981 		NL_SET_ERR_MSG_MOD(extack, "Ingress enc_key matches not supported");
982 		return -EOPNOTSUPP;
983 	}
984 
985 	if (tc->common.chain_index) {
986 		NL_SET_ERR_MSG_MOD(extack, "No support for nonzero chain_index");
987 		return -EOPNOTSUPP;
988 	}
989 	match.mask.recirc_id = 0xff;
990 
991 	rc = efx_mae_match_check_caps(efx, &match.mask, extack);
992 	if (rc)
993 		return rc;
994 
995 	rule = kzalloc(sizeof(*rule), GFP_USER);
996 	if (!rule)
997 		return -ENOMEM;
998 	INIT_LIST_HEAD(&rule->acts.list);
999 	rule->cookie = tc->cookie;
1000 	old = rhashtable_lookup_get_insert_fast(&efx->tc->match_action_ht,
1001 						&rule->linkage,
1002 						efx_tc_match_action_ht_params);
1003 	if (old) {
1004 		netif_dbg(efx, drv, efx->net_dev,
1005 			  "Already offloaded rule (cookie %lx)\n", tc->cookie);
1006 		NL_SET_ERR_MSG_MOD(extack, "Rule already offloaded");
1007 		kfree(rule);
1008 		return -EEXIST;
1009 	}
1010 
1011 	/* Parse actions */
1012 	act = kzalloc(sizeof(*act), GFP_USER);
1013 	if (!act) {
1014 		rc = -ENOMEM;
1015 		goto release;
1016 	}
1017 
1018 	/**
1019 	 * DOC: TC action translation
1020 	 *
1021 	 * Actions in TC are sequential and cumulative, with delivery actions
1022 	 * potentially anywhere in the order.  The EF100 MAE, however, takes
1023 	 * an 'action set list' consisting of 'action sets', each of which is
1024 	 * applied to the _original_ packet, and consists of a set of optional
1025 	 * actions in a fixed order with delivery at the end.
1026 	 * To translate between these two models, we maintain a 'cursor', @act,
1027 	 * which describes the cumulative effect of all the packet-mutating
1028 	 * actions encountered so far; on handling a delivery (mirred or drop)
1029 	 * action, once the action-set has been inserted into hardware, we
1030 	 * append @act to the action-set list (@rule->acts); if this is a pipe
1031 	 * action (mirred mirror) we then allocate a new @act with a copy of
1032 	 * the cursor state _before_ the delivery action, otherwise we set @act
1033 	 * to %NULL.
1034 	 * This ensures that every allocated action-set is either attached to
1035 	 * @rule->acts or pointed to by @act (and never both), and that only
1036 	 * those action-sets in @rule->acts exist in hardware.  Consequently,
1037 	 * in the failure path, @act only needs to be freed in memory, whereas
1038 	 * for @rule->acts we remove each action-set from hardware before
1039 	 * freeing it (efx_tc_free_action_set_list()), even if the action-set
1040 	 * list itself is not in hardware.
1041 	 */
1042 	flow_action_for_each(i, fa, &fr->action) {
1043 		struct efx_tc_action_set save;
1044 		u16 tci;
1045 
1046 		if (!act) {
1047 			/* more actions after a non-pipe action */
1048 			NL_SET_ERR_MSG_MOD(extack, "Action follows non-pipe action");
1049 			rc = -EINVAL;
1050 			goto release;
1051 		}
1052 
1053 		if ((fa->id == FLOW_ACTION_REDIRECT ||
1054 		     fa->id == FLOW_ACTION_MIRRED ||
1055 		     fa->id == FLOW_ACTION_DROP) && fa->hw_stats) {
1056 			struct efx_tc_counter_index *ctr;
1057 
1058 			/* Currently the only actions that want stats are
1059 			 * mirred and gact (ok, shot, trap, goto-chain), which
1060 			 * means we want stats just before delivery.  Also,
1061 			 * note that tunnel_key set shouldn't change the length
1062 			 * — it's only the subsequent mirred that does that,
1063 			 * and the stats are taken _before_ the mirred action
1064 			 * happens.
1065 			 */
1066 			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_COUNT)) {
1067 				/* All supported actions that count either steal
1068 				 * (gact shot, mirred redirect) or clone act
1069 				 * (mirred mirror), so we should never get two
1070 				 * count actions on one action_set.
1071 				 */
1072 				NL_SET_ERR_MSG_MOD(extack, "Count-action conflict (can't happen)");
1073 				rc = -EOPNOTSUPP;
1074 				goto release;
1075 			}
1076 
1077 			if (!(fa->hw_stats & FLOW_ACTION_HW_STATS_DELAYED)) {
1078 				NL_SET_ERR_MSG_FMT_MOD(extack, "hw_stats_type %u not supported (only 'delayed')",
1079 						       fa->hw_stats);
1080 				rc = -EOPNOTSUPP;
1081 				goto release;
1082 			}
1083 
1084 			ctr = efx_tc_flower_get_counter_index(efx, tc->cookie,
1085 							      EFX_TC_COUNTER_TYPE_AR);
1086 			if (IS_ERR(ctr)) {
1087 				rc = PTR_ERR(ctr);
1088 				NL_SET_ERR_MSG_MOD(extack, "Failed to obtain a counter");
1089 				goto release;
1090 			}
1091 			act->count = ctr;
1092 			INIT_LIST_HEAD(&act->count_user);
1093 		}
1094 
1095 		switch (fa->id) {
1096 		case FLOW_ACTION_DROP:
1097 			rc = efx_mae_alloc_action_set(efx, act);
1098 			if (rc) {
1099 				NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (drop)");
1100 				goto release;
1101 			}
1102 			list_add_tail(&act->list, &rule->acts.list);
1103 			act = NULL; /* end of the line */
1104 			break;
1105 		case FLOW_ACTION_REDIRECT:
1106 		case FLOW_ACTION_MIRRED:
1107 			save = *act;
1108 
1109 			if (encap_info) {
1110 				struct efx_tc_encap_action *encap;
1111 
1112 				if (!efx_tc_flower_action_order_ok(act,
1113 								   EFX_TC_AO_ENCAP)) {
1114 					rc = -EOPNOTSUPP;
1115 					NL_SET_ERR_MSG_MOD(extack, "Encap action violates action order");
1116 					goto release;
1117 				}
1118 				encap = efx_tc_flower_create_encap_md(
1119 						efx, encap_info, fa->dev, extack);
1120 				if (IS_ERR_OR_NULL(encap)) {
1121 					rc = PTR_ERR(encap);
1122 					if (!rc)
1123 						rc = -EIO; /* arbitrary */
1124 					goto release;
1125 				}
1126 				act->encap_md = encap;
1127 				list_add_tail(&act->encap_user, &encap->users);
1128 				act->dest_mport = encap->dest_mport;
1129 				act->deliver = 1;
1130 				if (act->count && !WARN_ON(!act->count->cnt)) {
1131 					/* This counter is used by an encap
1132 					 * action, which needs a reference back
1133 					 * so it can prod neighbouring whenever
1134 					 * traffic is seen.
1135 					 */
1136 					spin_lock_bh(&act->count->cnt->lock);
1137 					list_add_tail(&act->count_user,
1138 						      &act->count->cnt->users);
1139 					spin_unlock_bh(&act->count->cnt->lock);
1140 				}
1141 				rc = efx_mae_alloc_action_set(efx, act);
1142 				if (rc) {
1143 					NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (encap)");
1144 					goto release;
1145 				}
1146 				list_add_tail(&act->list, &rule->acts.list);
1147 				act->user = &rule->acts;
1148 				act = NULL;
1149 				if (fa->id == FLOW_ACTION_REDIRECT)
1150 					break; /* end of the line */
1151 				/* Mirror, so continue on with saved act */
1152 				save.count = NULL;
1153 				act = kzalloc(sizeof(*act), GFP_USER);
1154 				if (!act) {
1155 					rc = -ENOMEM;
1156 					goto release;
1157 				}
1158 				*act = save;
1159 				break;
1160 			}
1161 
1162 			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_DELIVER)) {
1163 				/* can't happen */
1164 				rc = -EOPNOTSUPP;
1165 				NL_SET_ERR_MSG_MOD(extack, "Deliver action violates action order (can't happen)");
1166 				goto release;
1167 			}
1168 
1169 			to_efv = efx_tc_flower_lookup_efv(efx, fa->dev);
1170 			if (IS_ERR(to_efv)) {
1171 				NL_SET_ERR_MSG_MOD(extack, "Mirred egress device not on switch");
1172 				rc = PTR_ERR(to_efv);
1173 				goto release;
1174 			}
1175 			rc = efx_tc_flower_external_mport(efx, to_efv);
1176 			if (rc < 0) {
1177 				NL_SET_ERR_MSG_MOD(extack, "Failed to identify egress m-port");
1178 				goto release;
1179 			}
1180 			act->dest_mport = rc;
1181 			act->deliver = 1;
1182 			rc = efx_mae_alloc_action_set(efx, act);
1183 			if (rc) {
1184 				NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (mirred)");
1185 				goto release;
1186 			}
1187 			list_add_tail(&act->list, &rule->acts.list);
1188 			act = NULL;
1189 			if (fa->id == FLOW_ACTION_REDIRECT)
1190 				break; /* end of the line */
1191 			/* Mirror, so continue on with saved act */
1192 			save.count = NULL;
1193 			act = kzalloc(sizeof(*act), GFP_USER);
1194 			if (!act) {
1195 				rc = -ENOMEM;
1196 				goto release;
1197 			}
1198 			*act = save;
1199 			break;
1200 		case FLOW_ACTION_VLAN_POP:
1201 			if (act->vlan_push) {
1202 				act->vlan_push--;
1203 			} else if (efx_tc_flower_action_order_ok(act, EFX_TC_AO_VLAN_POP)) {
1204 				act->vlan_pop++;
1205 			} else {
1206 				NL_SET_ERR_MSG_MOD(extack,
1207 						   "More than two VLAN pops, or action order violated");
1208 				rc = -EINVAL;
1209 				goto release;
1210 			}
1211 			break;
1212 		case FLOW_ACTION_VLAN_PUSH:
1213 			if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_VLAN_PUSH)) {
1214 				rc = -EINVAL;
1215 				NL_SET_ERR_MSG_MOD(extack,
1216 						   "More than two VLAN pushes, or action order violated");
1217 				goto release;
1218 			}
1219 			tci = fa->vlan.vid & VLAN_VID_MASK;
1220 			tci |= fa->vlan.prio << VLAN_PRIO_SHIFT;
1221 			act->vlan_tci[act->vlan_push] = cpu_to_be16(tci);
1222 			act->vlan_proto[act->vlan_push] = fa->vlan.proto;
1223 			act->vlan_push++;
1224 			break;
1225 		case FLOW_ACTION_TUNNEL_ENCAP:
1226 			if (encap_info) {
1227 				/* Can't specify encap multiple times.
1228 				 * If you want to overwrite an existing
1229 				 * encap_info, use an intervening
1230 				 * FLOW_ACTION_TUNNEL_DECAP to clear it.
1231 				 */
1232 				NL_SET_ERR_MSG_MOD(extack, "Tunnel key set when already set");
1233 				rc = -EINVAL;
1234 				goto release;
1235 			}
1236 			if (!fa->tunnel) {
1237 				NL_SET_ERR_MSG_MOD(extack, "Tunnel key set is missing key");
1238 				rc = -EOPNOTSUPP;
1239 				goto release;
1240 			}
1241 			encap_info = fa->tunnel;
1242 			break;
1243 		case FLOW_ACTION_TUNNEL_DECAP:
1244 			if (encap_info) {
1245 				encap_info = NULL;
1246 				break;
1247 			}
1248 			/* Since we don't support enc_key matches on ingress
1249 			 * (and if we did there'd be no tunnel-device to give
1250 			 * us a type), we can't offload a decap that's not
1251 			 * just undoing a previous encap action.
1252 			 */
1253 			NL_SET_ERR_MSG_MOD(extack, "Cannot offload tunnel decap action without tunnel device");
1254 			rc = -EOPNOTSUPP;
1255 			goto release;
1256 		default:
1257 			NL_SET_ERR_MSG_FMT_MOD(extack, "Unhandled action %u",
1258 					       fa->id);
1259 			rc = -EOPNOTSUPP;
1260 			goto release;
1261 		}
1262 	}
1263 
1264 	if (act) {
1265 		/* Not shot/redirected, so deliver to default dest */
1266 		if (from_efv == EFX_EFV_PF)
1267 			/* Rule applies to traffic from the wire,
1268 			 * and default dest is thus the PF
1269 			 */
1270 			efx_mae_mport_uplink(efx, &act->dest_mport);
1271 		else
1272 			/* Representor, so rule applies to traffic from
1273 			 * representee, and default dest is thus the rep.
1274 			 * All reps use the same mport for delivery
1275 			 */
1276 			efx_mae_mport_mport(efx, efx->tc->reps_mport_id,
1277 					    &act->dest_mport);
1278 		act->deliver = 1;
1279 		rc = efx_mae_alloc_action_set(efx, act);
1280 		if (rc) {
1281 			NL_SET_ERR_MSG_MOD(extack, "Failed to write action set to hw (deliver)");
1282 			goto release;
1283 		}
1284 		list_add_tail(&act->list, &rule->acts.list);
1285 		act = NULL; /* Prevent double-free in error path */
1286 	}
1287 
1288 	netif_dbg(efx, drv, efx->net_dev,
1289 		  "Successfully parsed filter (cookie %lx)\n",
1290 		  tc->cookie);
1291 
1292 	rule->match = match;
1293 
1294 	rc = efx_mae_alloc_action_set_list(efx, &rule->acts);
1295 	if (rc) {
1296 		NL_SET_ERR_MSG_MOD(extack, "Failed to write action set list to hw");
1297 		goto release;
1298 	}
1299 	if (from_efv == EFX_EFV_PF)
1300 		/* PF netdev, so rule applies to traffic from wire */
1301 		rule->fallback = &efx->tc->facts.pf;
1302 	else
1303 		/* repdev, so rule applies to traffic from representee */
1304 		rule->fallback = &efx->tc->facts.reps;
1305 	if (!efx_tc_check_ready(efx, rule)) {
1306 		netif_dbg(efx, drv, efx->net_dev, "action not ready for hw\n");
1307 		acts_id = rule->fallback->fw_id;
1308 	} else {
1309 		netif_dbg(efx, drv, efx->net_dev, "ready for hw\n");
1310 		acts_id = rule->acts.fw_id;
1311 	}
1312 	rc = efx_mae_insert_rule(efx, &rule->match, EFX_TC_PRIO_TC,
1313 				 acts_id, &rule->fw_id);
1314 	if (rc) {
1315 		NL_SET_ERR_MSG_MOD(extack, "Failed to insert rule in hw");
1316 		goto release_acts;
1317 	}
1318 	return 0;
1319 
1320 release_acts:
1321 	efx_mae_free_action_set_list(efx, &rule->acts);
1322 release:
1323 	/* We failed to insert the rule, so free up any entries we created in
1324 	 * subsidiary tables.
1325 	 */
1326 	if (act)
1327 		efx_tc_free_action_set(efx, act, false);
1328 	if (rule) {
1329 		rhashtable_remove_fast(&efx->tc->match_action_ht,
1330 				       &rule->linkage,
1331 				       efx_tc_match_action_ht_params);
1332 		efx_tc_free_action_set_list(efx, &rule->acts, false);
1333 	}
1334 	kfree(rule);
1335 	return rc;
1336 }
1337 
1338 static int efx_tc_flower_destroy(struct efx_nic *efx,
1339 				 struct net_device *net_dev,
1340 				 struct flow_cls_offload *tc)
1341 {
1342 	struct netlink_ext_ack *extack = tc->common.extack;
1343 	struct efx_tc_flow_rule *rule;
1344 
1345 	rule = rhashtable_lookup_fast(&efx->tc->match_action_ht, &tc->cookie,
1346 				      efx_tc_match_action_ht_params);
1347 	if (!rule) {
1348 		/* Only log a message if we're the ingress device.  Otherwise
1349 		 * it's a foreign filter and we might just not have been
1350 		 * interested (e.g. we might not have been the egress device
1351 		 * either).
1352 		 */
1353 		if (!IS_ERR(efx_tc_flower_lookup_efv(efx, net_dev)))
1354 			netif_warn(efx, drv, efx->net_dev,
1355 				   "Filter %lx not found to remove\n", tc->cookie);
1356 		NL_SET_ERR_MSG_MOD(extack, "Flow cookie not found in offloaded rules");
1357 		return -ENOENT;
1358 	}
1359 
1360 	/* Remove it from HW */
1361 	efx_tc_delete_rule(efx, rule);
1362 	/* Delete it from SW */
1363 	rhashtable_remove_fast(&efx->tc->match_action_ht, &rule->linkage,
1364 			       efx_tc_match_action_ht_params);
1365 	netif_dbg(efx, drv, efx->net_dev, "Removed filter %lx\n", rule->cookie);
1366 	kfree(rule);
1367 	return 0;
1368 }
1369 
1370 static int efx_tc_flower_stats(struct efx_nic *efx, struct net_device *net_dev,
1371 			       struct flow_cls_offload *tc)
1372 {
1373 	struct netlink_ext_ack *extack = tc->common.extack;
1374 	struct efx_tc_counter_index *ctr;
1375 	struct efx_tc_counter *cnt;
1376 	u64 packets, bytes;
1377 
1378 	ctr = efx_tc_flower_find_counter_index(efx, tc->cookie);
1379 	if (!ctr) {
1380 		/* See comment in efx_tc_flower_destroy() */
1381 		if (!IS_ERR(efx_tc_flower_lookup_efv(efx, net_dev)))
1382 			if (net_ratelimit())
1383 				netif_warn(efx, drv, efx->net_dev,
1384 					   "Filter %lx not found for stats\n",
1385 					   tc->cookie);
1386 		NL_SET_ERR_MSG_MOD(extack, "Flow cookie not found in offloaded rules");
1387 		return -ENOENT;
1388 	}
1389 	if (WARN_ON(!ctr->cnt)) /* can't happen */
1390 		return -EIO;
1391 	cnt = ctr->cnt;
1392 
1393 	spin_lock_bh(&cnt->lock);
1394 	/* Report only new pkts/bytes since last time TC asked */
1395 	packets = cnt->packets;
1396 	bytes = cnt->bytes;
1397 	flow_stats_update(&tc->stats, bytes - cnt->old_bytes,
1398 			  packets - cnt->old_packets, 0, cnt->touched,
1399 			  FLOW_ACTION_HW_STATS_DELAYED);
1400 	cnt->old_packets = packets;
1401 	cnt->old_bytes = bytes;
1402 	spin_unlock_bh(&cnt->lock);
1403 	return 0;
1404 }
1405 
1406 int efx_tc_flower(struct efx_nic *efx, struct net_device *net_dev,
1407 		  struct flow_cls_offload *tc, struct efx_rep *efv)
1408 {
1409 	int rc;
1410 
1411 	if (!efx->tc)
1412 		return -EOPNOTSUPP;
1413 
1414 	mutex_lock(&efx->tc->mutex);
1415 	switch (tc->command) {
1416 	case FLOW_CLS_REPLACE:
1417 		rc = efx_tc_flower_replace(efx, net_dev, tc, efv);
1418 		break;
1419 	case FLOW_CLS_DESTROY:
1420 		rc = efx_tc_flower_destroy(efx, net_dev, tc);
1421 		break;
1422 	case FLOW_CLS_STATS:
1423 		rc = efx_tc_flower_stats(efx, net_dev, tc);
1424 		break;
1425 	default:
1426 		rc = -EOPNOTSUPP;
1427 		break;
1428 	}
1429 	mutex_unlock(&efx->tc->mutex);
1430 	return rc;
1431 }
1432 
1433 static int efx_tc_configure_default_rule(struct efx_nic *efx, u32 ing_port,
1434 					 u32 eg_port, struct efx_tc_flow_rule *rule)
1435 {
1436 	struct efx_tc_action_set_list *acts = &rule->acts;
1437 	struct efx_tc_match *match = &rule->match;
1438 	struct efx_tc_action_set *act;
1439 	int rc;
1440 
1441 	match->value.ingress_port = ing_port;
1442 	match->mask.ingress_port = ~0;
1443 	act = kzalloc(sizeof(*act), GFP_KERNEL);
1444 	if (!act)
1445 		return -ENOMEM;
1446 	act->deliver = 1;
1447 	act->dest_mport = eg_port;
1448 	rc = efx_mae_alloc_action_set(efx, act);
1449 	if (rc)
1450 		goto fail1;
1451 	EFX_WARN_ON_PARANOID(!list_empty(&acts->list));
1452 	list_add_tail(&act->list, &acts->list);
1453 	rc = efx_mae_alloc_action_set_list(efx, acts);
1454 	if (rc)
1455 		goto fail2;
1456 	rc = efx_mae_insert_rule(efx, match, EFX_TC_PRIO_DFLT,
1457 				 acts->fw_id, &rule->fw_id);
1458 	if (rc)
1459 		goto fail3;
1460 	return 0;
1461 fail3:
1462 	efx_mae_free_action_set_list(efx, acts);
1463 fail2:
1464 	list_del(&act->list);
1465 	efx_mae_free_action_set(efx, act->fw_id);
1466 fail1:
1467 	kfree(act);
1468 	return rc;
1469 }
1470 
1471 static int efx_tc_configure_default_rule_pf(struct efx_nic *efx)
1472 {
1473 	struct efx_tc_flow_rule *rule = &efx->tc->dflt.pf;
1474 	u32 ing_port, eg_port;
1475 
1476 	efx_mae_mport_uplink(efx, &ing_port);
1477 	efx_mae_mport_wire(efx, &eg_port);
1478 	return efx_tc_configure_default_rule(efx, ing_port, eg_port, rule);
1479 }
1480 
1481 static int efx_tc_configure_default_rule_wire(struct efx_nic *efx)
1482 {
1483 	struct efx_tc_flow_rule *rule = &efx->tc->dflt.wire;
1484 	u32 ing_port, eg_port;
1485 
1486 	efx_mae_mport_wire(efx, &ing_port);
1487 	efx_mae_mport_uplink(efx, &eg_port);
1488 	return efx_tc_configure_default_rule(efx, ing_port, eg_port, rule);
1489 }
1490 
1491 int efx_tc_configure_default_rule_rep(struct efx_rep *efv)
1492 {
1493 	struct efx_tc_flow_rule *rule = &efv->dflt;
1494 	struct efx_nic *efx = efv->parent;
1495 	u32 ing_port, eg_port;
1496 
1497 	efx_mae_mport_mport(efx, efv->mport, &ing_port);
1498 	efx_mae_mport_mport(efx, efx->tc->reps_mport_id, &eg_port);
1499 	return efx_tc_configure_default_rule(efx, ing_port, eg_port, rule);
1500 }
1501 
1502 void efx_tc_deconfigure_default_rule(struct efx_nic *efx,
1503 				     struct efx_tc_flow_rule *rule)
1504 {
1505 	if (rule->fw_id != MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL)
1506 		efx_tc_delete_rule(efx, rule);
1507 	rule->fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
1508 }
1509 
1510 static int efx_tc_configure_fallback_acts(struct efx_nic *efx, u32 eg_port,
1511 					  struct efx_tc_action_set_list *acts)
1512 {
1513 	struct efx_tc_action_set *act;
1514 	int rc;
1515 
1516 	act = kzalloc(sizeof(*act), GFP_KERNEL);
1517 	if (!act)
1518 		return -ENOMEM;
1519 	act->deliver = 1;
1520 	act->dest_mport = eg_port;
1521 	rc = efx_mae_alloc_action_set(efx, act);
1522 	if (rc)
1523 		goto fail1;
1524 	EFX_WARN_ON_PARANOID(!list_empty(&acts->list));
1525 	list_add_tail(&act->list, &acts->list);
1526 	rc = efx_mae_alloc_action_set_list(efx, acts);
1527 	if (rc)
1528 		goto fail2;
1529 	return 0;
1530 fail2:
1531 	list_del(&act->list);
1532 	efx_mae_free_action_set(efx, act->fw_id);
1533 fail1:
1534 	kfree(act);
1535 	return rc;
1536 }
1537 
1538 static int efx_tc_configure_fallback_acts_pf(struct efx_nic *efx)
1539 {
1540 	struct efx_tc_action_set_list *acts = &efx->tc->facts.pf;
1541 	u32 eg_port;
1542 
1543 	efx_mae_mport_uplink(efx, &eg_port);
1544 	return efx_tc_configure_fallback_acts(efx, eg_port, acts);
1545 }
1546 
1547 static int efx_tc_configure_fallback_acts_reps(struct efx_nic *efx)
1548 {
1549 	struct efx_tc_action_set_list *acts = &efx->tc->facts.reps;
1550 	u32 eg_port;
1551 
1552 	efx_mae_mport_mport(efx, efx->tc->reps_mport_id, &eg_port);
1553 	return efx_tc_configure_fallback_acts(efx, eg_port, acts);
1554 }
1555 
1556 static void efx_tc_deconfigure_fallback_acts(struct efx_nic *efx,
1557 					     struct efx_tc_action_set_list *acts)
1558 {
1559 	efx_tc_free_action_set_list(efx, acts, true);
1560 }
1561 
1562 static int efx_tc_configure_rep_mport(struct efx_nic *efx)
1563 {
1564 	u32 rep_mport_label;
1565 	int rc;
1566 
1567 	rc = efx_mae_allocate_mport(efx, &efx->tc->reps_mport_id, &rep_mport_label);
1568 	if (rc)
1569 		return rc;
1570 	pci_dbg(efx->pci_dev, "created rep mport 0x%08x (0x%04x)\n",
1571 		efx->tc->reps_mport_id, rep_mport_label);
1572 	/* Use mport *selector* as vport ID */
1573 	efx_mae_mport_mport(efx, efx->tc->reps_mport_id,
1574 			    &efx->tc->reps_mport_vport_id);
1575 	return 0;
1576 }
1577 
1578 static void efx_tc_deconfigure_rep_mport(struct efx_nic *efx)
1579 {
1580 	efx_mae_free_mport(efx, efx->tc->reps_mport_id);
1581 	efx->tc->reps_mport_id = MAE_MPORT_SELECTOR_NULL;
1582 }
1583 
1584 int efx_tc_insert_rep_filters(struct efx_nic *efx)
1585 {
1586 	struct efx_filter_spec promisc, allmulti;
1587 	int rc;
1588 
1589 	if (efx->type->is_vf)
1590 		return 0;
1591 	if (!efx->tc)
1592 		return 0;
1593 	efx_filter_init_rx(&promisc, EFX_FILTER_PRI_REQUIRED, 0, 0);
1594 	efx_filter_set_uc_def(&promisc);
1595 	efx_filter_set_vport_id(&promisc, efx->tc->reps_mport_vport_id);
1596 	rc = efx_filter_insert_filter(efx, &promisc, false);
1597 	if (rc < 0)
1598 		return rc;
1599 	efx->tc->reps_filter_uc = rc;
1600 	efx_filter_init_rx(&allmulti, EFX_FILTER_PRI_REQUIRED, 0, 0);
1601 	efx_filter_set_mc_def(&allmulti);
1602 	efx_filter_set_vport_id(&allmulti, efx->tc->reps_mport_vport_id);
1603 	rc = efx_filter_insert_filter(efx, &allmulti, false);
1604 	if (rc < 0)
1605 		return rc;
1606 	efx->tc->reps_filter_mc = rc;
1607 	return 0;
1608 }
1609 
1610 void efx_tc_remove_rep_filters(struct efx_nic *efx)
1611 {
1612 	if (efx->type->is_vf)
1613 		return;
1614 	if (!efx->tc)
1615 		return;
1616 	if (efx->tc->reps_filter_mc >= 0)
1617 		efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED, efx->tc->reps_filter_mc);
1618 	efx->tc->reps_filter_mc = -1;
1619 	if (efx->tc->reps_filter_uc >= 0)
1620 		efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED, efx->tc->reps_filter_uc);
1621 	efx->tc->reps_filter_uc = -1;
1622 }
1623 
1624 int efx_init_tc(struct efx_nic *efx)
1625 {
1626 	int rc;
1627 
1628 	rc = efx_mae_get_caps(efx, efx->tc->caps);
1629 	if (rc)
1630 		return rc;
1631 	if (efx->tc->caps->match_field_count > MAE_NUM_FIELDS)
1632 		/* Firmware supports some match fields the driver doesn't know
1633 		 * about.  Not fatal, unless any of those fields are required
1634 		 * (MAE_FIELD_SUPPORTED_MATCH_ALWAYS) but if so we don't know.
1635 		 */
1636 		netif_warn(efx, probe, efx->net_dev,
1637 			   "FW reports additional match fields %u\n",
1638 			   efx->tc->caps->match_field_count);
1639 	if (efx->tc->caps->action_prios < EFX_TC_PRIO__NUM) {
1640 		netif_err(efx, probe, efx->net_dev,
1641 			  "Too few action prios supported (have %u, need %u)\n",
1642 			  efx->tc->caps->action_prios, EFX_TC_PRIO__NUM);
1643 		return -EIO;
1644 	}
1645 	rc = efx_tc_configure_default_rule_pf(efx);
1646 	if (rc)
1647 		return rc;
1648 	rc = efx_tc_configure_default_rule_wire(efx);
1649 	if (rc)
1650 		return rc;
1651 	rc = efx_tc_configure_rep_mport(efx);
1652 	if (rc)
1653 		return rc;
1654 	rc = efx_tc_configure_fallback_acts_pf(efx);
1655 	if (rc)
1656 		return rc;
1657 	rc = efx_tc_configure_fallback_acts_reps(efx);
1658 	if (rc)
1659 		return rc;
1660 	efx->tc->up = true;
1661 	rc = flow_indr_dev_register(efx_tc_indr_setup_cb, efx);
1662 	if (rc)
1663 		return rc;
1664 	return 0;
1665 }
1666 
1667 void efx_fini_tc(struct efx_nic *efx)
1668 {
1669 	/* We can get called even if efx_init_struct_tc() failed */
1670 	if (!efx->tc)
1671 		return;
1672 	if (efx->tc->up)
1673 		flow_indr_dev_unregister(efx_tc_indr_setup_cb, efx, efx_tc_block_unbind);
1674 	efx_tc_deconfigure_rep_mport(efx);
1675 	efx_tc_deconfigure_default_rule(efx, &efx->tc->dflt.pf);
1676 	efx_tc_deconfigure_default_rule(efx, &efx->tc->dflt.wire);
1677 	efx_tc_deconfigure_fallback_acts(efx, &efx->tc->facts.pf);
1678 	efx_tc_deconfigure_fallback_acts(efx, &efx->tc->facts.reps);
1679 	efx->tc->up = false;
1680 }
1681 
1682 /* At teardown time, all TC filter rules (and thus all resources they created)
1683  * should already have been removed.  If we find any in our hashtables, make a
1684  * cursory attempt to clean up the software side.
1685  */
1686 static void efx_tc_encap_match_free(void *ptr, void *__unused)
1687 {
1688 	struct efx_tc_encap_match *encap = ptr;
1689 
1690 	WARN_ON(refcount_read(&encap->ref));
1691 	kfree(encap);
1692 }
1693 
1694 static void efx_tc_flow_free(void *ptr, void *arg)
1695 {
1696 	struct efx_tc_flow_rule *rule = ptr;
1697 	struct efx_nic *efx = arg;
1698 
1699 	netif_err(efx, drv, efx->net_dev,
1700 		  "tc rule %lx still present at teardown, removing\n",
1701 		  rule->cookie);
1702 
1703 	/* Also releases entries in subsidiary tables */
1704 	efx_tc_delete_rule(efx, rule);
1705 
1706 	kfree(rule);
1707 }
1708 
1709 int efx_init_struct_tc(struct efx_nic *efx)
1710 {
1711 	int rc;
1712 
1713 	if (efx->type->is_vf)
1714 		return 0;
1715 
1716 	efx->tc = kzalloc(sizeof(*efx->tc), GFP_KERNEL);
1717 	if (!efx->tc)
1718 		return -ENOMEM;
1719 	efx->tc->caps = kzalloc(sizeof(struct mae_caps), GFP_KERNEL);
1720 	if (!efx->tc->caps) {
1721 		rc = -ENOMEM;
1722 		goto fail_alloc_caps;
1723 	}
1724 	INIT_LIST_HEAD(&efx->tc->block_list);
1725 
1726 	mutex_init(&efx->tc->mutex);
1727 	init_waitqueue_head(&efx->tc->flush_wq);
1728 	rc = efx_tc_init_encap_actions(efx);
1729 	if (rc < 0)
1730 		goto fail_encap_actions;
1731 	rc = efx_tc_init_counters(efx);
1732 	if (rc < 0)
1733 		goto fail_counters;
1734 	rc = rhashtable_init(&efx->tc->encap_match_ht, &efx_tc_encap_match_ht_params);
1735 	if (rc < 0)
1736 		goto fail_encap_match_ht;
1737 	rc = rhashtable_init(&efx->tc->match_action_ht, &efx_tc_match_action_ht_params);
1738 	if (rc < 0)
1739 		goto fail_match_action_ht;
1740 	efx->tc->reps_filter_uc = -1;
1741 	efx->tc->reps_filter_mc = -1;
1742 	INIT_LIST_HEAD(&efx->tc->dflt.pf.acts.list);
1743 	efx->tc->dflt.pf.fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
1744 	INIT_LIST_HEAD(&efx->tc->dflt.wire.acts.list);
1745 	efx->tc->dflt.wire.fw_id = MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL;
1746 	INIT_LIST_HEAD(&efx->tc->facts.pf.list);
1747 	efx->tc->facts.pf.fw_id = MC_CMD_MAE_ACTION_SET_ALLOC_OUT_ACTION_SET_ID_NULL;
1748 	INIT_LIST_HEAD(&efx->tc->facts.reps.list);
1749 	efx->tc->facts.reps.fw_id = MC_CMD_MAE_ACTION_SET_ALLOC_OUT_ACTION_SET_ID_NULL;
1750 	efx->extra_channel_type[EFX_EXTRA_CHANNEL_TC] = &efx_tc_channel_type;
1751 	return 0;
1752 fail_match_action_ht:
1753 	rhashtable_destroy(&efx->tc->encap_match_ht);
1754 fail_encap_match_ht:
1755 	efx_tc_destroy_counters(efx);
1756 fail_counters:
1757 	efx_tc_destroy_encap_actions(efx);
1758 fail_encap_actions:
1759 	mutex_destroy(&efx->tc->mutex);
1760 	kfree(efx->tc->caps);
1761 fail_alloc_caps:
1762 	kfree(efx->tc);
1763 	efx->tc = NULL;
1764 	return rc;
1765 }
1766 
1767 void efx_fini_struct_tc(struct efx_nic *efx)
1768 {
1769 	if (!efx->tc)
1770 		return;
1771 
1772 	mutex_lock(&efx->tc->mutex);
1773 	EFX_WARN_ON_PARANOID(efx->tc->dflt.pf.fw_id !=
1774 			     MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL);
1775 	EFX_WARN_ON_PARANOID(efx->tc->dflt.wire.fw_id !=
1776 			     MC_CMD_MAE_ACTION_RULE_INSERT_OUT_ACTION_RULE_ID_NULL);
1777 	EFX_WARN_ON_PARANOID(efx->tc->facts.pf.fw_id !=
1778 			     MC_CMD_MAE_ACTION_SET_LIST_ALLOC_OUT_ACTION_SET_LIST_ID_NULL);
1779 	EFX_WARN_ON_PARANOID(efx->tc->facts.reps.fw_id !=
1780 			     MC_CMD_MAE_ACTION_SET_LIST_ALLOC_OUT_ACTION_SET_LIST_ID_NULL);
1781 	rhashtable_free_and_destroy(&efx->tc->match_action_ht, efx_tc_flow_free,
1782 				    efx);
1783 	rhashtable_free_and_destroy(&efx->tc->encap_match_ht,
1784 				    efx_tc_encap_match_free, NULL);
1785 	efx_tc_fini_counters(efx);
1786 	efx_tc_fini_encap_actions(efx);
1787 	mutex_unlock(&efx->tc->mutex);
1788 	mutex_destroy(&efx->tc->mutex);
1789 	kfree(efx->tc->caps);
1790 	kfree(efx->tc);
1791 	efx->tc = NULL;
1792 }
1793