xref: /openbmc/linux/drivers/net/ethernet/marvell/octeontx2/nic/otx2_tc.c (revision f4356947f0297b0962fdd197672db7edf9f58be6)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Marvell RVU Ethernet driver
3  *
4  * Copyright (C) 2021 Marvell.
5  *
6  */
7 
8 #include <linux/netdevice.h>
9 #include <linux/etherdevice.h>
10 #include <linux/inetdevice.h>
11 #include <linux/rhashtable.h>
12 #include <linux/bitfield.h>
13 #include <net/flow_dissector.h>
14 #include <net/pkt_cls.h>
15 #include <net/tc_act/tc_gact.h>
16 #include <net/tc_act/tc_mirred.h>
17 #include <net/tc_act/tc_vlan.h>
18 #include <net/ipv6.h>
19 
20 #include "cn10k.h"
21 #include "otx2_common.h"
22 #include "qos.h"
23 
24 #define CN10K_MAX_BURST_MANTISSA	0x7FFFULL
25 #define CN10K_MAX_BURST_SIZE		8453888ULL
26 
27 #define CN10K_TLX_BURST_MANTISSA	GENMASK_ULL(43, 29)
28 #define CN10K_TLX_BURST_EXPONENT	GENMASK_ULL(47, 44)
29 
30 struct otx2_tc_flow_stats {
31 	u64 bytes;
32 	u64 pkts;
33 	u64 used;
34 };
35 
36 struct otx2_tc_flow {
37 	struct rhash_head		node;
38 	unsigned long			cookie;
39 	unsigned int			bitpos;
40 	struct rcu_head			rcu;
41 	struct otx2_tc_flow_stats	stats;
42 	spinlock_t			lock; /* lock for stats */
43 	u16				rq;
44 	u16				entry;
45 	u16				leaf_profile;
46 	bool				is_act_police;
47 };
48 
49 int otx2_tc_alloc_ent_bitmap(struct otx2_nic *nic)
50 {
51 	struct otx2_tc_info *tc = &nic->tc_info;
52 
53 	if (!nic->flow_cfg->max_flows)
54 		return 0;
55 
56 	/* Max flows changed, free the existing bitmap */
57 	kfree(tc->tc_entries_bitmap);
58 
59 	tc->tc_entries_bitmap =
60 			kcalloc(BITS_TO_LONGS(nic->flow_cfg->max_flows),
61 				sizeof(long), GFP_KERNEL);
62 	if (!tc->tc_entries_bitmap) {
63 		netdev_err(nic->netdev,
64 			   "Unable to alloc TC flow entries bitmap\n");
65 		return -ENOMEM;
66 	}
67 
68 	return 0;
69 }
70 EXPORT_SYMBOL(otx2_tc_alloc_ent_bitmap);
71 
72 static void otx2_get_egress_burst_cfg(struct otx2_nic *nic, u32 burst,
73 				      u32 *burst_exp, u32 *burst_mantissa)
74 {
75 	int max_burst, max_mantissa;
76 	unsigned int tmp;
77 
78 	if (is_dev_otx2(nic->pdev)) {
79 		max_burst = MAX_BURST_SIZE;
80 		max_mantissa = MAX_BURST_MANTISSA;
81 	} else {
82 		max_burst = CN10K_MAX_BURST_SIZE;
83 		max_mantissa = CN10K_MAX_BURST_MANTISSA;
84 	}
85 
86 	/* Burst is calculated as
87 	 * ((256 + BURST_MANTISSA) << (1 + BURST_EXPONENT)) / 256
88 	 * Max supported burst size is 130,816 bytes.
89 	 */
90 	burst = min_t(u32, burst, max_burst);
91 	if (burst) {
92 		*burst_exp = ilog2(burst) ? ilog2(burst) - 1 : 0;
93 		tmp = burst - rounddown_pow_of_two(burst);
94 		if (burst < max_mantissa)
95 			*burst_mantissa = tmp * 2;
96 		else
97 			*burst_mantissa = tmp / (1ULL << (*burst_exp - 7));
98 	} else {
99 		*burst_exp = MAX_BURST_EXPONENT;
100 		*burst_mantissa = max_mantissa;
101 	}
102 }
103 
104 static void otx2_get_egress_rate_cfg(u64 maxrate, u32 *exp,
105 				     u32 *mantissa, u32 *div_exp)
106 {
107 	u64 tmp;
108 
109 	/* Rate calculation by hardware
110 	 *
111 	 * PIR_ADD = ((256 + mantissa) << exp) / 256
112 	 * rate = (2 * PIR_ADD) / ( 1 << div_exp)
113 	 * The resultant rate is in Mbps.
114 	 */
115 
116 	/* 2Mbps to 100Gbps can be expressed with div_exp = 0.
117 	 * Setting this to '0' will ease the calculation of
118 	 * exponent and mantissa.
119 	 */
120 	*div_exp = 0;
121 
122 	if (maxrate) {
123 		*exp = ilog2(maxrate) ? ilog2(maxrate) - 1 : 0;
124 		tmp = maxrate - rounddown_pow_of_two(maxrate);
125 		if (maxrate < MAX_RATE_MANTISSA)
126 			*mantissa = tmp * 2;
127 		else
128 			*mantissa = tmp / (1ULL << (*exp - 7));
129 	} else {
130 		/* Instead of disabling rate limiting, set all values to max */
131 		*exp = MAX_RATE_EXPONENT;
132 		*mantissa = MAX_RATE_MANTISSA;
133 	}
134 }
135 
136 u64 otx2_get_txschq_rate_regval(struct otx2_nic *nic,
137 				u64 maxrate, u32 burst)
138 {
139 	u32 burst_exp, burst_mantissa;
140 	u32 exp, mantissa, div_exp;
141 	u64 regval = 0;
142 
143 	/* Get exponent and mantissa values from the desired rate */
144 	otx2_get_egress_burst_cfg(nic, burst, &burst_exp, &burst_mantissa);
145 	otx2_get_egress_rate_cfg(maxrate, &exp, &mantissa, &div_exp);
146 
147 	if (is_dev_otx2(nic->pdev)) {
148 		regval = FIELD_PREP(TLX_BURST_EXPONENT, (u64)burst_exp) |
149 				FIELD_PREP(TLX_BURST_MANTISSA, (u64)burst_mantissa) |
150 				FIELD_PREP(TLX_RATE_DIVIDER_EXPONENT, div_exp) |
151 				FIELD_PREP(TLX_RATE_EXPONENT, exp) |
152 				FIELD_PREP(TLX_RATE_MANTISSA, mantissa) | BIT_ULL(0);
153 	} else {
154 		regval = FIELD_PREP(CN10K_TLX_BURST_EXPONENT, (u64)burst_exp) |
155 				FIELD_PREP(CN10K_TLX_BURST_MANTISSA, (u64)burst_mantissa) |
156 				FIELD_PREP(TLX_RATE_DIVIDER_EXPONENT, div_exp) |
157 				FIELD_PREP(TLX_RATE_EXPONENT, exp) |
158 				FIELD_PREP(TLX_RATE_MANTISSA, mantissa) | BIT_ULL(0);
159 	}
160 
161 	return regval;
162 }
163 
164 static int otx2_set_matchall_egress_rate(struct otx2_nic *nic,
165 					 u32 burst, u64 maxrate)
166 {
167 	struct otx2_hw *hw = &nic->hw;
168 	struct nix_txschq_config *req;
169 	int txschq, err;
170 
171 	/* All SQs share the same TL4, so pick the first scheduler */
172 	txschq = hw->txschq_list[NIX_TXSCH_LVL_TL4][0];
173 
174 	mutex_lock(&nic->mbox.lock);
175 	req = otx2_mbox_alloc_msg_nix_txschq_cfg(&nic->mbox);
176 	if (!req) {
177 		mutex_unlock(&nic->mbox.lock);
178 		return -ENOMEM;
179 	}
180 
181 	req->lvl = NIX_TXSCH_LVL_TL4;
182 	req->num_regs = 1;
183 	req->reg[0] = NIX_AF_TL4X_PIR(txschq);
184 	req->regval[0] = otx2_get_txschq_rate_regval(nic, maxrate, burst);
185 
186 	err = otx2_sync_mbox_msg(&nic->mbox);
187 	mutex_unlock(&nic->mbox.lock);
188 	return err;
189 }
190 
191 static int otx2_tc_validate_flow(struct otx2_nic *nic,
192 				 struct flow_action *actions,
193 				 struct netlink_ext_ack *extack)
194 {
195 	if (nic->flags & OTX2_FLAG_INTF_DOWN) {
196 		NL_SET_ERR_MSG_MOD(extack, "Interface not initialized");
197 		return -EINVAL;
198 	}
199 
200 	if (!flow_action_has_entries(actions)) {
201 		NL_SET_ERR_MSG_MOD(extack, "MATCHALL offload called with no action");
202 		return -EINVAL;
203 	}
204 
205 	if (!flow_offload_has_one_action(actions)) {
206 		NL_SET_ERR_MSG_MOD(extack,
207 				   "Egress MATCHALL offload supports only 1 policing action");
208 		return -EINVAL;
209 	}
210 	return 0;
211 }
212 
213 static int otx2_policer_validate(const struct flow_action *action,
214 				 const struct flow_action_entry *act,
215 				 struct netlink_ext_ack *extack)
216 {
217 	if (act->police.exceed.act_id != FLOW_ACTION_DROP) {
218 		NL_SET_ERR_MSG_MOD(extack,
219 				   "Offload not supported when exceed action is not drop");
220 		return -EOPNOTSUPP;
221 	}
222 
223 	if (act->police.notexceed.act_id != FLOW_ACTION_PIPE &&
224 	    act->police.notexceed.act_id != FLOW_ACTION_ACCEPT) {
225 		NL_SET_ERR_MSG_MOD(extack,
226 				   "Offload not supported when conform action is not pipe or ok");
227 		return -EOPNOTSUPP;
228 	}
229 
230 	if (act->police.notexceed.act_id == FLOW_ACTION_ACCEPT &&
231 	    !flow_action_is_last_entry(action, act)) {
232 		NL_SET_ERR_MSG_MOD(extack,
233 				   "Offload not supported when conform action is ok, but action is not last");
234 		return -EOPNOTSUPP;
235 	}
236 
237 	if (act->police.peakrate_bytes_ps ||
238 	    act->police.avrate || act->police.overhead) {
239 		NL_SET_ERR_MSG_MOD(extack,
240 				   "Offload not supported when peakrate/avrate/overhead is configured");
241 		return -EOPNOTSUPP;
242 	}
243 
244 	return 0;
245 }
246 
247 static int otx2_tc_egress_matchall_install(struct otx2_nic *nic,
248 					   struct tc_cls_matchall_offload *cls)
249 {
250 	struct netlink_ext_ack *extack = cls->common.extack;
251 	struct flow_action *actions = &cls->rule->action;
252 	struct flow_action_entry *entry;
253 	int err;
254 
255 	err = otx2_tc_validate_flow(nic, actions, extack);
256 	if (err)
257 		return err;
258 
259 	if (nic->flags & OTX2_FLAG_TC_MATCHALL_EGRESS_ENABLED) {
260 		NL_SET_ERR_MSG_MOD(extack,
261 				   "Only one Egress MATCHALL ratelimiter can be offloaded");
262 		return -ENOMEM;
263 	}
264 
265 	entry = &cls->rule->action.entries[0];
266 	switch (entry->id) {
267 	case FLOW_ACTION_POLICE:
268 		err = otx2_policer_validate(&cls->rule->action, entry, extack);
269 		if (err)
270 			return err;
271 
272 		if (entry->police.rate_pkt_ps) {
273 			NL_SET_ERR_MSG_MOD(extack, "QoS offload not support packets per second");
274 			return -EOPNOTSUPP;
275 		}
276 		err = otx2_set_matchall_egress_rate(nic, entry->police.burst,
277 						    otx2_convert_rate(entry->police.rate_bytes_ps));
278 		if (err)
279 			return err;
280 		nic->flags |= OTX2_FLAG_TC_MATCHALL_EGRESS_ENABLED;
281 		break;
282 	default:
283 		NL_SET_ERR_MSG_MOD(extack,
284 				   "Only police action is supported with Egress MATCHALL offload");
285 		return -EOPNOTSUPP;
286 	}
287 
288 	return 0;
289 }
290 
291 static int otx2_tc_egress_matchall_delete(struct otx2_nic *nic,
292 					  struct tc_cls_matchall_offload *cls)
293 {
294 	struct netlink_ext_ack *extack = cls->common.extack;
295 	int err;
296 
297 	if (nic->flags & OTX2_FLAG_INTF_DOWN) {
298 		NL_SET_ERR_MSG_MOD(extack, "Interface not initialized");
299 		return -EINVAL;
300 	}
301 
302 	err = otx2_set_matchall_egress_rate(nic, 0, 0);
303 	nic->flags &= ~OTX2_FLAG_TC_MATCHALL_EGRESS_ENABLED;
304 	return err;
305 }
306 
307 static int otx2_tc_act_set_police(struct otx2_nic *nic,
308 				  struct otx2_tc_flow *node,
309 				  struct flow_cls_offload *f,
310 				  u64 rate, u32 burst, u32 mark,
311 				  struct npc_install_flow_req *req, bool pps)
312 {
313 	struct netlink_ext_ack *extack = f->common.extack;
314 	struct otx2_hw *hw = &nic->hw;
315 	int rq_idx, rc;
316 
317 	rq_idx = find_first_zero_bit(&nic->rq_bmap, hw->rx_queues);
318 	if (rq_idx >= hw->rx_queues) {
319 		NL_SET_ERR_MSG_MOD(extack, "Police action rules exceeded");
320 		return -EINVAL;
321 	}
322 
323 	mutex_lock(&nic->mbox.lock);
324 
325 	rc = cn10k_alloc_leaf_profile(nic, &node->leaf_profile);
326 	if (rc) {
327 		mutex_unlock(&nic->mbox.lock);
328 		return rc;
329 	}
330 
331 	rc = cn10k_set_ipolicer_rate(nic, node->leaf_profile, burst, rate, pps);
332 	if (rc)
333 		goto free_leaf;
334 
335 	rc = cn10k_map_unmap_rq_policer(nic, rq_idx, node->leaf_profile, true);
336 	if (rc)
337 		goto free_leaf;
338 
339 	mutex_unlock(&nic->mbox.lock);
340 
341 	req->match_id = mark & 0xFFFFULL;
342 	req->index = rq_idx;
343 	req->op = NIX_RX_ACTIONOP_UCAST;
344 	set_bit(rq_idx, &nic->rq_bmap);
345 	node->is_act_police = true;
346 	node->rq = rq_idx;
347 
348 	return 0;
349 
350 free_leaf:
351 	if (cn10k_free_leaf_profile(nic, node->leaf_profile))
352 		netdev_err(nic->netdev,
353 			   "Unable to free leaf bandwidth profile(%d)\n",
354 			   node->leaf_profile);
355 	mutex_unlock(&nic->mbox.lock);
356 	return rc;
357 }
358 
359 static int otx2_tc_parse_actions(struct otx2_nic *nic,
360 				 struct flow_action *flow_action,
361 				 struct npc_install_flow_req *req,
362 				 struct flow_cls_offload *f,
363 				 struct otx2_tc_flow *node)
364 {
365 	struct netlink_ext_ack *extack = f->common.extack;
366 	struct flow_action_entry *act;
367 	struct net_device *target;
368 	struct otx2_nic *priv;
369 	u32 burst, mark = 0;
370 	u8 nr_police = 0;
371 	bool pps = false;
372 	u64 rate;
373 	int err;
374 	int i;
375 
376 	if (!flow_action_has_entries(flow_action)) {
377 		NL_SET_ERR_MSG_MOD(extack, "no tc actions specified");
378 		return -EINVAL;
379 	}
380 
381 	flow_action_for_each(i, act, flow_action) {
382 		switch (act->id) {
383 		case FLOW_ACTION_DROP:
384 			req->op = NIX_RX_ACTIONOP_DROP;
385 			return 0;
386 		case FLOW_ACTION_ACCEPT:
387 			req->op = NIX_RX_ACTION_DEFAULT;
388 			return 0;
389 		case FLOW_ACTION_REDIRECT_INGRESS:
390 			target = act->dev;
391 			priv = netdev_priv(target);
392 			/* npc_install_flow_req doesn't support passing a target pcifunc */
393 			if (rvu_get_pf(nic->pcifunc) != rvu_get_pf(priv->pcifunc)) {
394 				NL_SET_ERR_MSG_MOD(extack,
395 						   "can't redirect to other pf/vf");
396 				return -EOPNOTSUPP;
397 			}
398 			req->vf = priv->pcifunc & RVU_PFVF_FUNC_MASK;
399 			req->op = NIX_RX_ACTION_DEFAULT;
400 			return 0;
401 		case FLOW_ACTION_VLAN_POP:
402 			req->vtag0_valid = true;
403 			/* use RX_VTAG_TYPE7 which is initialized to strip vlan tag */
404 			req->vtag0_type = NIX_AF_LFX_RX_VTAG_TYPE7;
405 			break;
406 		case FLOW_ACTION_POLICE:
407 			/* Ingress ratelimiting is not supported on OcteonTx2 */
408 			if (is_dev_otx2(nic->pdev)) {
409 				NL_SET_ERR_MSG_MOD(extack,
410 					"Ingress policing not supported on this platform");
411 				return -EOPNOTSUPP;
412 			}
413 
414 			err = otx2_policer_validate(flow_action, act, extack);
415 			if (err)
416 				return err;
417 
418 			if (act->police.rate_bytes_ps > 0) {
419 				rate = act->police.rate_bytes_ps * 8;
420 				burst = act->police.burst;
421 			} else if (act->police.rate_pkt_ps > 0) {
422 				/* The algorithm used to calculate rate
423 				 * mantissa, exponent values for a given token
424 				 * rate (token can be byte or packet) requires
425 				 * token rate to be mutiplied by 8.
426 				 */
427 				rate = act->police.rate_pkt_ps * 8;
428 				burst = act->police.burst_pkt;
429 				pps = true;
430 			}
431 			nr_police++;
432 			break;
433 		case FLOW_ACTION_MARK:
434 			mark = act->mark;
435 			break;
436 		default:
437 			return -EOPNOTSUPP;
438 		}
439 	}
440 
441 	if (nr_police > 1) {
442 		NL_SET_ERR_MSG_MOD(extack,
443 				   "rate limit police offload requires a single action");
444 		return -EOPNOTSUPP;
445 	}
446 
447 	if (nr_police)
448 		return otx2_tc_act_set_police(nic, node, f, rate, burst,
449 					      mark, req, pps);
450 
451 	return 0;
452 }
453 
454 static int otx2_tc_prepare_flow(struct otx2_nic *nic, struct otx2_tc_flow *node,
455 				struct flow_cls_offload *f,
456 				struct npc_install_flow_req *req)
457 {
458 	struct netlink_ext_ack *extack = f->common.extack;
459 	struct flow_msg *flow_spec = &req->packet;
460 	struct flow_msg *flow_mask = &req->mask;
461 	struct flow_dissector *dissector;
462 	struct flow_rule *rule;
463 	u8 ip_proto = 0;
464 
465 	rule = flow_cls_offload_flow_rule(f);
466 	dissector = rule->match.dissector;
467 
468 	if ((dissector->used_keys &
469 	    ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
470 	      BIT(FLOW_DISSECTOR_KEY_BASIC) |
471 	      BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
472 	      BIT(FLOW_DISSECTOR_KEY_VLAN) |
473 	      BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
474 	      BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
475 	      BIT(FLOW_DISSECTOR_KEY_PORTS) |
476 	      BIT(FLOW_DISSECTOR_KEY_IP))))  {
477 		netdev_info(nic->netdev, "unsupported flow used key 0x%x",
478 			    dissector->used_keys);
479 		return -EOPNOTSUPP;
480 	}
481 
482 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
483 		struct flow_match_basic match;
484 
485 		flow_rule_match_basic(rule, &match);
486 
487 		/* All EtherTypes can be matched, no hw limitation */
488 		flow_spec->etype = match.key->n_proto;
489 		flow_mask->etype = match.mask->n_proto;
490 		req->features |= BIT_ULL(NPC_ETYPE);
491 
492 		if (match.mask->ip_proto &&
493 		    (match.key->ip_proto != IPPROTO_TCP &&
494 		     match.key->ip_proto != IPPROTO_UDP &&
495 		     match.key->ip_proto != IPPROTO_SCTP &&
496 		     match.key->ip_proto != IPPROTO_ICMP &&
497 		     match.key->ip_proto != IPPROTO_ICMPV6)) {
498 			netdev_info(nic->netdev,
499 				    "ip_proto=0x%x not supported\n",
500 				    match.key->ip_proto);
501 			return -EOPNOTSUPP;
502 		}
503 		if (match.mask->ip_proto)
504 			ip_proto = match.key->ip_proto;
505 
506 		if (ip_proto == IPPROTO_UDP)
507 			req->features |= BIT_ULL(NPC_IPPROTO_UDP);
508 		else if (ip_proto == IPPROTO_TCP)
509 			req->features |= BIT_ULL(NPC_IPPROTO_TCP);
510 		else if (ip_proto == IPPROTO_SCTP)
511 			req->features |= BIT_ULL(NPC_IPPROTO_SCTP);
512 		else if (ip_proto == IPPROTO_ICMP)
513 			req->features |= BIT_ULL(NPC_IPPROTO_ICMP);
514 		else if (ip_proto == IPPROTO_ICMPV6)
515 			req->features |= BIT_ULL(NPC_IPPROTO_ICMP6);
516 	}
517 
518 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
519 		struct flow_match_control match;
520 
521 		flow_rule_match_control(rule, &match);
522 		if (match.mask->flags & FLOW_DIS_FIRST_FRAG) {
523 			NL_SET_ERR_MSG_MOD(extack, "HW doesn't support frag first/later");
524 			return -EOPNOTSUPP;
525 		}
526 
527 		if (match.mask->flags & FLOW_DIS_IS_FRAGMENT) {
528 			if (ntohs(flow_spec->etype) == ETH_P_IP) {
529 				flow_spec->ip_flag = IPV4_FLAG_MORE;
530 				flow_mask->ip_flag = IPV4_FLAG_MORE;
531 				req->features |= BIT_ULL(NPC_IPFRAG_IPV4);
532 			} else if (ntohs(flow_spec->etype) == ETH_P_IPV6) {
533 				flow_spec->next_header = IPPROTO_FRAGMENT;
534 				flow_mask->next_header = 0xff;
535 				req->features |= BIT_ULL(NPC_IPFRAG_IPV6);
536 			} else {
537 				NL_SET_ERR_MSG_MOD(extack, "flow-type should be either IPv4 and IPv6");
538 				return -EOPNOTSUPP;
539 			}
540 		}
541 	}
542 
543 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
544 		struct flow_match_eth_addrs match;
545 
546 		flow_rule_match_eth_addrs(rule, &match);
547 		if (!is_zero_ether_addr(match.mask->src)) {
548 			NL_SET_ERR_MSG_MOD(extack, "src mac match not supported");
549 			return -EOPNOTSUPP;
550 		}
551 
552 		if (!is_zero_ether_addr(match.mask->dst)) {
553 			ether_addr_copy(flow_spec->dmac, (u8 *)&match.key->dst);
554 			ether_addr_copy(flow_mask->dmac,
555 					(u8 *)&match.mask->dst);
556 			req->features |= BIT_ULL(NPC_DMAC);
557 		}
558 	}
559 
560 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IP)) {
561 		struct flow_match_ip match;
562 
563 		flow_rule_match_ip(rule, &match);
564 		if ((ntohs(flow_spec->etype) != ETH_P_IP) &&
565 		    match.mask->tos) {
566 			NL_SET_ERR_MSG_MOD(extack, "tos not supported");
567 			return -EOPNOTSUPP;
568 		}
569 		if (match.mask->ttl) {
570 			NL_SET_ERR_MSG_MOD(extack, "ttl not supported");
571 			return -EOPNOTSUPP;
572 		}
573 		flow_spec->tos = match.key->tos;
574 		flow_mask->tos = match.mask->tos;
575 		req->features |= BIT_ULL(NPC_TOS);
576 	}
577 
578 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
579 		struct flow_match_vlan match;
580 		u16 vlan_tci, vlan_tci_mask;
581 
582 		flow_rule_match_vlan(rule, &match);
583 
584 		if (ntohs(match.key->vlan_tpid) != ETH_P_8021Q) {
585 			netdev_err(nic->netdev, "vlan tpid 0x%x not supported\n",
586 				   ntohs(match.key->vlan_tpid));
587 			return -EOPNOTSUPP;
588 		}
589 
590 		if (match.mask->vlan_id ||
591 		    match.mask->vlan_dei ||
592 		    match.mask->vlan_priority) {
593 			vlan_tci = match.key->vlan_id |
594 				   match.key->vlan_dei << 12 |
595 				   match.key->vlan_priority << 13;
596 
597 			vlan_tci_mask = match.mask->vlan_id |
598 					match.mask->vlan_dei << 12 |
599 					match.mask->vlan_priority << 13;
600 
601 			flow_spec->vlan_tci = htons(vlan_tci);
602 			flow_mask->vlan_tci = htons(vlan_tci_mask);
603 			req->features |= BIT_ULL(NPC_OUTER_VID);
604 		}
605 	}
606 
607 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IPV4_ADDRS)) {
608 		struct flow_match_ipv4_addrs match;
609 
610 		flow_rule_match_ipv4_addrs(rule, &match);
611 
612 		flow_spec->ip4dst = match.key->dst;
613 		flow_mask->ip4dst = match.mask->dst;
614 		req->features |= BIT_ULL(NPC_DIP_IPV4);
615 
616 		flow_spec->ip4src = match.key->src;
617 		flow_mask->ip4src = match.mask->src;
618 		req->features |= BIT_ULL(NPC_SIP_IPV4);
619 	} else if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IPV6_ADDRS)) {
620 		struct flow_match_ipv6_addrs match;
621 
622 		flow_rule_match_ipv6_addrs(rule, &match);
623 
624 		if (ipv6_addr_loopback(&match.key->dst) ||
625 		    ipv6_addr_loopback(&match.key->src)) {
626 			NL_SET_ERR_MSG_MOD(extack,
627 					   "Flow matching IPv6 loopback addr not supported");
628 			return -EOPNOTSUPP;
629 		}
630 
631 		if (!ipv6_addr_any(&match.mask->dst)) {
632 			memcpy(&flow_spec->ip6dst,
633 			       (struct in6_addr *)&match.key->dst,
634 			       sizeof(flow_spec->ip6dst));
635 			memcpy(&flow_mask->ip6dst,
636 			       (struct in6_addr *)&match.mask->dst,
637 			       sizeof(flow_spec->ip6dst));
638 			req->features |= BIT_ULL(NPC_DIP_IPV6);
639 		}
640 
641 		if (!ipv6_addr_any(&match.mask->src)) {
642 			memcpy(&flow_spec->ip6src,
643 			       (struct in6_addr *)&match.key->src,
644 			       sizeof(flow_spec->ip6src));
645 			memcpy(&flow_mask->ip6src,
646 			       (struct in6_addr *)&match.mask->src,
647 			       sizeof(flow_spec->ip6src));
648 			req->features |= BIT_ULL(NPC_SIP_IPV6);
649 		}
650 	}
651 
652 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_PORTS)) {
653 		struct flow_match_ports match;
654 
655 		flow_rule_match_ports(rule, &match);
656 
657 		flow_spec->dport = match.key->dst;
658 		flow_mask->dport = match.mask->dst;
659 
660 		if (flow_mask->dport) {
661 			if (ip_proto == IPPROTO_UDP)
662 				req->features |= BIT_ULL(NPC_DPORT_UDP);
663 			else if (ip_proto == IPPROTO_TCP)
664 				req->features |= BIT_ULL(NPC_DPORT_TCP);
665 			else if (ip_proto == IPPROTO_SCTP)
666 				req->features |= BIT_ULL(NPC_DPORT_SCTP);
667 		}
668 
669 		flow_spec->sport = match.key->src;
670 		flow_mask->sport = match.mask->src;
671 
672 		if (flow_mask->sport) {
673 			if (ip_proto == IPPROTO_UDP)
674 				req->features |= BIT_ULL(NPC_SPORT_UDP);
675 			else if (ip_proto == IPPROTO_TCP)
676 				req->features |= BIT_ULL(NPC_SPORT_TCP);
677 			else if (ip_proto == IPPROTO_SCTP)
678 				req->features |= BIT_ULL(NPC_SPORT_SCTP);
679 		}
680 	}
681 
682 	return otx2_tc_parse_actions(nic, &rule->action, req, f, node);
683 }
684 
685 static int otx2_del_mcam_flow_entry(struct otx2_nic *nic, u16 entry)
686 {
687 	struct npc_delete_flow_req *req;
688 	int err;
689 
690 	mutex_lock(&nic->mbox.lock);
691 	req = otx2_mbox_alloc_msg_npc_delete_flow(&nic->mbox);
692 	if (!req) {
693 		mutex_unlock(&nic->mbox.lock);
694 		return -ENOMEM;
695 	}
696 
697 	req->entry = entry;
698 
699 	/* Send message to AF */
700 	err = otx2_sync_mbox_msg(&nic->mbox);
701 	if (err) {
702 		netdev_err(nic->netdev, "Failed to delete MCAM flow entry %d\n",
703 			   entry);
704 		mutex_unlock(&nic->mbox.lock);
705 		return -EFAULT;
706 	}
707 	mutex_unlock(&nic->mbox.lock);
708 
709 	return 0;
710 }
711 
712 static int otx2_tc_del_flow(struct otx2_nic *nic,
713 			    struct flow_cls_offload *tc_flow_cmd)
714 {
715 	struct otx2_flow_config *flow_cfg = nic->flow_cfg;
716 	struct otx2_tc_info *tc_info = &nic->tc_info;
717 	struct otx2_tc_flow *flow_node;
718 	int err;
719 
720 	flow_node = rhashtable_lookup_fast(&tc_info->flow_table,
721 					   &tc_flow_cmd->cookie,
722 					   tc_info->flow_ht_params);
723 	if (!flow_node) {
724 		netdev_err(nic->netdev, "tc flow not found for cookie 0x%lx\n",
725 			   tc_flow_cmd->cookie);
726 		return -EINVAL;
727 	}
728 
729 	if (flow_node->is_act_police) {
730 		mutex_lock(&nic->mbox.lock);
731 
732 		err = cn10k_map_unmap_rq_policer(nic, flow_node->rq,
733 						 flow_node->leaf_profile, false);
734 		if (err)
735 			netdev_err(nic->netdev,
736 				   "Unmapping RQ %d & profile %d failed\n",
737 				   flow_node->rq, flow_node->leaf_profile);
738 
739 		err = cn10k_free_leaf_profile(nic, flow_node->leaf_profile);
740 		if (err)
741 			netdev_err(nic->netdev,
742 				   "Unable to free leaf bandwidth profile(%d)\n",
743 				   flow_node->leaf_profile);
744 
745 		__clear_bit(flow_node->rq, &nic->rq_bmap);
746 
747 		mutex_unlock(&nic->mbox.lock);
748 	}
749 
750 	otx2_del_mcam_flow_entry(nic, flow_node->entry);
751 
752 	WARN_ON(rhashtable_remove_fast(&nic->tc_info.flow_table,
753 				       &flow_node->node,
754 				       nic->tc_info.flow_ht_params));
755 	kfree_rcu(flow_node, rcu);
756 
757 	clear_bit(flow_node->bitpos, tc_info->tc_entries_bitmap);
758 	flow_cfg->nr_flows--;
759 
760 	return 0;
761 }
762 
763 static int otx2_tc_add_flow(struct otx2_nic *nic,
764 			    struct flow_cls_offload *tc_flow_cmd)
765 {
766 	struct netlink_ext_ack *extack = tc_flow_cmd->common.extack;
767 	struct otx2_flow_config *flow_cfg = nic->flow_cfg;
768 	struct otx2_tc_info *tc_info = &nic->tc_info;
769 	struct otx2_tc_flow *new_node, *old_node;
770 	struct npc_install_flow_req *req, dummy;
771 	int rc, err;
772 
773 	if (!(nic->flags & OTX2_FLAG_TC_FLOWER_SUPPORT))
774 		return -ENOMEM;
775 
776 	if (bitmap_full(tc_info->tc_entries_bitmap, flow_cfg->max_flows)) {
777 		NL_SET_ERR_MSG_MOD(extack,
778 				   "Free MCAM entry not available to add the flow");
779 		return -ENOMEM;
780 	}
781 
782 	/* allocate memory for the new flow and it's node */
783 	new_node = kzalloc(sizeof(*new_node), GFP_KERNEL);
784 	if (!new_node)
785 		return -ENOMEM;
786 	spin_lock_init(&new_node->lock);
787 	new_node->cookie = tc_flow_cmd->cookie;
788 
789 	memset(&dummy, 0, sizeof(struct npc_install_flow_req));
790 
791 	rc = otx2_tc_prepare_flow(nic, new_node, tc_flow_cmd, &dummy);
792 	if (rc) {
793 		kfree_rcu(new_node, rcu);
794 		return rc;
795 	}
796 
797 	/* If a flow exists with the same cookie, delete it */
798 	old_node = rhashtable_lookup_fast(&tc_info->flow_table,
799 					  &tc_flow_cmd->cookie,
800 					  tc_info->flow_ht_params);
801 	if (old_node)
802 		otx2_tc_del_flow(nic, tc_flow_cmd);
803 
804 	mutex_lock(&nic->mbox.lock);
805 	req = otx2_mbox_alloc_msg_npc_install_flow(&nic->mbox);
806 	if (!req) {
807 		mutex_unlock(&nic->mbox.lock);
808 		rc = -ENOMEM;
809 		goto free_leaf;
810 	}
811 
812 	memcpy(&dummy.hdr, &req->hdr, sizeof(struct mbox_msghdr));
813 	memcpy(req, &dummy, sizeof(struct npc_install_flow_req));
814 
815 	new_node->bitpos = find_first_zero_bit(tc_info->tc_entries_bitmap,
816 					       flow_cfg->max_flows);
817 	req->channel = nic->hw.rx_chan_base;
818 	req->entry = flow_cfg->flow_ent[flow_cfg->max_flows - new_node->bitpos - 1];
819 	req->intf = NIX_INTF_RX;
820 	req->set_cntr = 1;
821 	new_node->entry = req->entry;
822 
823 	/* Send message to AF */
824 	rc = otx2_sync_mbox_msg(&nic->mbox);
825 	if (rc) {
826 		NL_SET_ERR_MSG_MOD(extack, "Failed to install MCAM flow entry");
827 		mutex_unlock(&nic->mbox.lock);
828 		kfree_rcu(new_node, rcu);
829 		goto free_leaf;
830 	}
831 	mutex_unlock(&nic->mbox.lock);
832 
833 	/* add new flow to flow-table */
834 	rc = rhashtable_insert_fast(&nic->tc_info.flow_table, &new_node->node,
835 				    nic->tc_info.flow_ht_params);
836 	if (rc) {
837 		otx2_del_mcam_flow_entry(nic, req->entry);
838 		kfree_rcu(new_node, rcu);
839 		goto free_leaf;
840 	}
841 
842 	set_bit(new_node->bitpos, tc_info->tc_entries_bitmap);
843 	flow_cfg->nr_flows++;
844 
845 	return 0;
846 
847 free_leaf:
848 	if (new_node->is_act_police) {
849 		mutex_lock(&nic->mbox.lock);
850 
851 		err = cn10k_map_unmap_rq_policer(nic, new_node->rq,
852 						 new_node->leaf_profile, false);
853 		if (err)
854 			netdev_err(nic->netdev,
855 				   "Unmapping RQ %d & profile %d failed\n",
856 				   new_node->rq, new_node->leaf_profile);
857 		err = cn10k_free_leaf_profile(nic, new_node->leaf_profile);
858 		if (err)
859 			netdev_err(nic->netdev,
860 				   "Unable to free leaf bandwidth profile(%d)\n",
861 				   new_node->leaf_profile);
862 
863 		__clear_bit(new_node->rq, &nic->rq_bmap);
864 
865 		mutex_unlock(&nic->mbox.lock);
866 	}
867 
868 	return rc;
869 }
870 
871 static int otx2_tc_get_flow_stats(struct otx2_nic *nic,
872 				  struct flow_cls_offload *tc_flow_cmd)
873 {
874 	struct otx2_tc_info *tc_info = &nic->tc_info;
875 	struct npc_mcam_get_stats_req *req;
876 	struct npc_mcam_get_stats_rsp *rsp;
877 	struct otx2_tc_flow_stats *stats;
878 	struct otx2_tc_flow *flow_node;
879 	int err;
880 
881 	flow_node = rhashtable_lookup_fast(&tc_info->flow_table,
882 					   &tc_flow_cmd->cookie,
883 					   tc_info->flow_ht_params);
884 	if (!flow_node) {
885 		netdev_info(nic->netdev, "tc flow not found for cookie %lx",
886 			    tc_flow_cmd->cookie);
887 		return -EINVAL;
888 	}
889 
890 	mutex_lock(&nic->mbox.lock);
891 
892 	req = otx2_mbox_alloc_msg_npc_mcam_entry_stats(&nic->mbox);
893 	if (!req) {
894 		mutex_unlock(&nic->mbox.lock);
895 		return -ENOMEM;
896 	}
897 
898 	req->entry = flow_node->entry;
899 
900 	err = otx2_sync_mbox_msg(&nic->mbox);
901 	if (err) {
902 		netdev_err(nic->netdev, "Failed to get stats for MCAM flow entry %d\n",
903 			   req->entry);
904 		mutex_unlock(&nic->mbox.lock);
905 		return -EFAULT;
906 	}
907 
908 	rsp = (struct npc_mcam_get_stats_rsp *)otx2_mbox_get_rsp
909 		(&nic->mbox.mbox, 0, &req->hdr);
910 	if (IS_ERR(rsp)) {
911 		mutex_unlock(&nic->mbox.lock);
912 		return PTR_ERR(rsp);
913 	}
914 
915 	mutex_unlock(&nic->mbox.lock);
916 
917 	if (!rsp->stat_ena)
918 		return -EINVAL;
919 
920 	stats = &flow_node->stats;
921 
922 	spin_lock(&flow_node->lock);
923 	flow_stats_update(&tc_flow_cmd->stats, 0x0, rsp->stat - stats->pkts, 0x0, 0x0,
924 			  FLOW_ACTION_HW_STATS_IMMEDIATE);
925 	stats->pkts = rsp->stat;
926 	spin_unlock(&flow_node->lock);
927 
928 	return 0;
929 }
930 
931 static int otx2_setup_tc_cls_flower(struct otx2_nic *nic,
932 				    struct flow_cls_offload *cls_flower)
933 {
934 	switch (cls_flower->command) {
935 	case FLOW_CLS_REPLACE:
936 		return otx2_tc_add_flow(nic, cls_flower);
937 	case FLOW_CLS_DESTROY:
938 		return otx2_tc_del_flow(nic, cls_flower);
939 	case FLOW_CLS_STATS:
940 		return otx2_tc_get_flow_stats(nic, cls_flower);
941 	default:
942 		return -EOPNOTSUPP;
943 	}
944 }
945 
946 static int otx2_tc_ingress_matchall_install(struct otx2_nic *nic,
947 					    struct tc_cls_matchall_offload *cls)
948 {
949 	struct netlink_ext_ack *extack = cls->common.extack;
950 	struct flow_action *actions = &cls->rule->action;
951 	struct flow_action_entry *entry;
952 	u64 rate;
953 	int err;
954 
955 	err = otx2_tc_validate_flow(nic, actions, extack);
956 	if (err)
957 		return err;
958 
959 	if (nic->flags & OTX2_FLAG_TC_MATCHALL_INGRESS_ENABLED) {
960 		NL_SET_ERR_MSG_MOD(extack,
961 				   "Only one ingress MATCHALL ratelimitter can be offloaded");
962 		return -ENOMEM;
963 	}
964 
965 	entry = &cls->rule->action.entries[0];
966 	switch (entry->id) {
967 	case FLOW_ACTION_POLICE:
968 		/* Ingress ratelimiting is not supported on OcteonTx2 */
969 		if (is_dev_otx2(nic->pdev)) {
970 			NL_SET_ERR_MSG_MOD(extack,
971 					   "Ingress policing not supported on this platform");
972 			return -EOPNOTSUPP;
973 		}
974 
975 		err = cn10k_alloc_matchall_ipolicer(nic);
976 		if (err)
977 			return err;
978 
979 		/* Convert to bits per second */
980 		rate = entry->police.rate_bytes_ps * 8;
981 		err = cn10k_set_matchall_ipolicer_rate(nic, entry->police.burst, rate);
982 		if (err)
983 			return err;
984 		nic->flags |= OTX2_FLAG_TC_MATCHALL_INGRESS_ENABLED;
985 		break;
986 	default:
987 		NL_SET_ERR_MSG_MOD(extack,
988 				   "Only police action supported with Ingress MATCHALL offload");
989 		return -EOPNOTSUPP;
990 	}
991 
992 	return 0;
993 }
994 
995 static int otx2_tc_ingress_matchall_delete(struct otx2_nic *nic,
996 					   struct tc_cls_matchall_offload *cls)
997 {
998 	struct netlink_ext_ack *extack = cls->common.extack;
999 	int err;
1000 
1001 	if (nic->flags & OTX2_FLAG_INTF_DOWN) {
1002 		NL_SET_ERR_MSG_MOD(extack, "Interface not initialized");
1003 		return -EINVAL;
1004 	}
1005 
1006 	err = cn10k_free_matchall_ipolicer(nic);
1007 	nic->flags &= ~OTX2_FLAG_TC_MATCHALL_INGRESS_ENABLED;
1008 	return err;
1009 }
1010 
1011 static int otx2_setup_tc_ingress_matchall(struct otx2_nic *nic,
1012 					  struct tc_cls_matchall_offload *cls_matchall)
1013 {
1014 	switch (cls_matchall->command) {
1015 	case TC_CLSMATCHALL_REPLACE:
1016 		return otx2_tc_ingress_matchall_install(nic, cls_matchall);
1017 	case TC_CLSMATCHALL_DESTROY:
1018 		return otx2_tc_ingress_matchall_delete(nic, cls_matchall);
1019 	case TC_CLSMATCHALL_STATS:
1020 	default:
1021 		break;
1022 	}
1023 
1024 	return -EOPNOTSUPP;
1025 }
1026 
1027 static int otx2_setup_tc_block_ingress_cb(enum tc_setup_type type,
1028 					  void *type_data, void *cb_priv)
1029 {
1030 	struct otx2_nic *nic = cb_priv;
1031 
1032 	if (!tc_cls_can_offload_and_chain0(nic->netdev, type_data))
1033 		return -EOPNOTSUPP;
1034 
1035 	switch (type) {
1036 	case TC_SETUP_CLSFLOWER:
1037 		return otx2_setup_tc_cls_flower(nic, type_data);
1038 	case TC_SETUP_CLSMATCHALL:
1039 		return otx2_setup_tc_ingress_matchall(nic, type_data);
1040 	default:
1041 		break;
1042 	}
1043 
1044 	return -EOPNOTSUPP;
1045 }
1046 
1047 static int otx2_setup_tc_egress_matchall(struct otx2_nic *nic,
1048 					 struct tc_cls_matchall_offload *cls_matchall)
1049 {
1050 	switch (cls_matchall->command) {
1051 	case TC_CLSMATCHALL_REPLACE:
1052 		return otx2_tc_egress_matchall_install(nic, cls_matchall);
1053 	case TC_CLSMATCHALL_DESTROY:
1054 		return otx2_tc_egress_matchall_delete(nic, cls_matchall);
1055 	case TC_CLSMATCHALL_STATS:
1056 	default:
1057 		break;
1058 	}
1059 
1060 	return -EOPNOTSUPP;
1061 }
1062 
1063 static int otx2_setup_tc_block_egress_cb(enum tc_setup_type type,
1064 					 void *type_data, void *cb_priv)
1065 {
1066 	struct otx2_nic *nic = cb_priv;
1067 
1068 	if (!tc_cls_can_offload_and_chain0(nic->netdev, type_data))
1069 		return -EOPNOTSUPP;
1070 
1071 	switch (type) {
1072 	case TC_SETUP_CLSMATCHALL:
1073 		return otx2_setup_tc_egress_matchall(nic, type_data);
1074 	default:
1075 		break;
1076 	}
1077 
1078 	return -EOPNOTSUPP;
1079 }
1080 
1081 static LIST_HEAD(otx2_block_cb_list);
1082 
1083 static int otx2_setup_tc_block(struct net_device *netdev,
1084 			       struct flow_block_offload *f)
1085 {
1086 	struct otx2_nic *nic = netdev_priv(netdev);
1087 	flow_setup_cb_t *cb;
1088 	bool ingress;
1089 
1090 	if (f->block_shared)
1091 		return -EOPNOTSUPP;
1092 
1093 	if (f->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS) {
1094 		cb = otx2_setup_tc_block_ingress_cb;
1095 		ingress = true;
1096 	} else if (f->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS) {
1097 		cb = otx2_setup_tc_block_egress_cb;
1098 		ingress = false;
1099 	} else {
1100 		return -EOPNOTSUPP;
1101 	}
1102 
1103 	return flow_block_cb_setup_simple(f, &otx2_block_cb_list, cb,
1104 					  nic, nic, ingress);
1105 }
1106 
1107 int otx2_setup_tc(struct net_device *netdev, enum tc_setup_type type,
1108 		  void *type_data)
1109 {
1110 	switch (type) {
1111 	case TC_SETUP_BLOCK:
1112 		return otx2_setup_tc_block(netdev, type_data);
1113 	case TC_SETUP_QDISC_HTB:
1114 		return otx2_setup_tc_htb(netdev, type_data);
1115 	default:
1116 		return -EOPNOTSUPP;
1117 	}
1118 }
1119 EXPORT_SYMBOL(otx2_setup_tc);
1120 
1121 static const struct rhashtable_params tc_flow_ht_params = {
1122 	.head_offset = offsetof(struct otx2_tc_flow, node),
1123 	.key_offset = offsetof(struct otx2_tc_flow, cookie),
1124 	.key_len = sizeof(((struct otx2_tc_flow *)0)->cookie),
1125 	.automatic_shrinking = true,
1126 };
1127 
1128 int otx2_init_tc(struct otx2_nic *nic)
1129 {
1130 	struct otx2_tc_info *tc = &nic->tc_info;
1131 	int err;
1132 
1133 	/* Exclude receive queue 0 being used for police action */
1134 	set_bit(0, &nic->rq_bmap);
1135 
1136 	if (!nic->flow_cfg) {
1137 		netdev_err(nic->netdev,
1138 			   "Can't init TC, nic->flow_cfg is not setup\n");
1139 		return -EINVAL;
1140 	}
1141 
1142 	err = otx2_tc_alloc_ent_bitmap(nic);
1143 	if (err)
1144 		return err;
1145 
1146 	tc->flow_ht_params = tc_flow_ht_params;
1147 	err = rhashtable_init(&tc->flow_table, &tc->flow_ht_params);
1148 	if (err) {
1149 		kfree(tc->tc_entries_bitmap);
1150 		tc->tc_entries_bitmap = NULL;
1151 	}
1152 	return err;
1153 }
1154 EXPORT_SYMBOL(otx2_init_tc);
1155 
1156 void otx2_shutdown_tc(struct otx2_nic *nic)
1157 {
1158 	struct otx2_tc_info *tc = &nic->tc_info;
1159 
1160 	kfree(tc->tc_entries_bitmap);
1161 	rhashtable_destroy(&tc->flow_table);
1162 }
1163 EXPORT_SYMBOL(otx2_shutdown_tc);
1164