xref: /linux/drivers/net/ethernet/marvell/octeontx2/nic/otx2_tc.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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 #define OTX2_UNSUPP_LSE_DEPTH		GENMASK(6, 4)
31 
32 #define MCAST_INVALID_GRP		(-1U)
33 
34 struct otx2_tc_flow_stats {
35 	u64 bytes;
36 	u64 pkts;
37 	u64 used;
38 };
39 
40 struct otx2_tc_flow {
41 	struct list_head		list;
42 	unsigned long			cookie;
43 	struct rcu_head			rcu;
44 	struct otx2_tc_flow_stats	stats;
45 	spinlock_t			lock; /* lock for stats */
46 	u16				rq;
47 	u16				entry;
48 	u16				leaf_profile;
49 	bool				is_act_police;
50 	u32				prio;
51 	struct npc_install_flow_req	req;
52 	u32				mcast_grp_idx;
53 	u64				rate;
54 	u32				burst;
55 	bool				is_pps;
56 };
57 
58 static void otx2_get_egress_burst_cfg(struct otx2_nic *nic, u32 burst,
59 				      u32 *burst_exp, u32 *burst_mantissa)
60 {
61 	int max_burst, max_mantissa;
62 	unsigned int tmp;
63 
64 	if (is_dev_otx2(nic->pdev)) {
65 		max_burst = MAX_BURST_SIZE;
66 		max_mantissa = MAX_BURST_MANTISSA;
67 	} else {
68 		max_burst = CN10K_MAX_BURST_SIZE;
69 		max_mantissa = CN10K_MAX_BURST_MANTISSA;
70 	}
71 
72 	/* Burst is calculated as
73 	 * ((256 + BURST_MANTISSA) << (1 + BURST_EXPONENT)) / 256
74 	 * Max supported burst size is 130,816 bytes.
75 	 */
76 	burst = min_t(u32, burst, max_burst);
77 	if (burst) {
78 		*burst_exp = ilog2(burst) ? ilog2(burst) - 1 : 0;
79 		tmp = burst - rounddown_pow_of_two(burst);
80 		if (burst < max_mantissa)
81 			*burst_mantissa = tmp * 2;
82 		else
83 			*burst_mantissa = tmp / (1ULL << (*burst_exp - 7));
84 	} else {
85 		*burst_exp = MAX_BURST_EXPONENT;
86 		*burst_mantissa = max_mantissa;
87 	}
88 }
89 
90 static void otx2_get_egress_rate_cfg(u64 maxrate, u32 *exp,
91 				     u32 *mantissa, u32 *div_exp)
92 {
93 	u64 tmp;
94 
95 	/* Rate calculation by hardware
96 	 *
97 	 * PIR_ADD = ((256 + mantissa) << exp) / 256
98 	 * rate = (2 * PIR_ADD) / ( 1 << div_exp)
99 	 * The resultant rate is in Mbps.
100 	 */
101 
102 	/* 2Mbps to 100Gbps can be expressed with div_exp = 0.
103 	 * Setting this to '0' will ease the calculation of
104 	 * exponent and mantissa.
105 	 */
106 	*div_exp = 0;
107 
108 	if (maxrate) {
109 		*exp = ilog2(maxrate) ? ilog2(maxrate) - 1 : 0;
110 		tmp = maxrate - rounddown_pow_of_two(maxrate);
111 		if (maxrate < MAX_RATE_MANTISSA)
112 			*mantissa = tmp * 2;
113 		else
114 			*mantissa = tmp / (1ULL << (*exp - 7));
115 	} else {
116 		/* Instead of disabling rate limiting, set all values to max */
117 		*exp = MAX_RATE_EXPONENT;
118 		*mantissa = MAX_RATE_MANTISSA;
119 	}
120 }
121 
122 u64 otx2_get_txschq_rate_regval(struct otx2_nic *nic,
123 				u64 maxrate, u32 burst)
124 {
125 	u32 burst_exp, burst_mantissa;
126 	u32 exp, mantissa, div_exp;
127 	u64 regval = 0;
128 
129 	/* Get exponent and mantissa values from the desired rate */
130 	otx2_get_egress_burst_cfg(nic, burst, &burst_exp, &burst_mantissa);
131 	otx2_get_egress_rate_cfg(maxrate, &exp, &mantissa, &div_exp);
132 
133 	if (is_dev_otx2(nic->pdev)) {
134 		regval = FIELD_PREP(TLX_BURST_EXPONENT, (u64)burst_exp) |
135 				FIELD_PREP(TLX_BURST_MANTISSA, (u64)burst_mantissa) |
136 				FIELD_PREP(TLX_RATE_DIVIDER_EXPONENT, div_exp) |
137 				FIELD_PREP(TLX_RATE_EXPONENT, exp) |
138 				FIELD_PREP(TLX_RATE_MANTISSA, mantissa) | BIT_ULL(0);
139 	} else {
140 		regval = FIELD_PREP(CN10K_TLX_BURST_EXPONENT, (u64)burst_exp) |
141 				FIELD_PREP(CN10K_TLX_BURST_MANTISSA, (u64)burst_mantissa) |
142 				FIELD_PREP(TLX_RATE_DIVIDER_EXPONENT, div_exp) |
143 				FIELD_PREP(TLX_RATE_EXPONENT, exp) |
144 				FIELD_PREP(TLX_RATE_MANTISSA, mantissa) | BIT_ULL(0);
145 	}
146 
147 	return regval;
148 }
149 
150 static int otx2_set_matchall_egress_rate(struct otx2_nic *nic,
151 					 u32 burst, u64 maxrate)
152 {
153 	struct otx2_hw *hw = &nic->hw;
154 	struct nix_txschq_config *req;
155 	int txschq, err;
156 
157 	/* All SQs share the same TL4, so pick the first scheduler */
158 	txschq = hw->txschq_list[NIX_TXSCH_LVL_TL4][0];
159 
160 	mutex_lock(&nic->mbox.lock);
161 	req = otx2_mbox_alloc_msg_nix_txschq_cfg(&nic->mbox);
162 	if (!req) {
163 		mutex_unlock(&nic->mbox.lock);
164 		return -ENOMEM;
165 	}
166 
167 	req->lvl = NIX_TXSCH_LVL_TL4;
168 	req->num_regs = 1;
169 	req->reg[0] = NIX_AF_TL4X_PIR(txschq);
170 	req->regval[0] = otx2_get_txschq_rate_regval(nic, maxrate, burst);
171 
172 	err = otx2_sync_mbox_msg(&nic->mbox);
173 	mutex_unlock(&nic->mbox.lock);
174 	return err;
175 }
176 
177 static int otx2_tc_validate_flow(struct otx2_nic *nic,
178 				 struct flow_action *actions,
179 				 struct netlink_ext_ack *extack)
180 {
181 	if (nic->flags & OTX2_FLAG_INTF_DOWN) {
182 		NL_SET_ERR_MSG_MOD(extack, "Interface not initialized");
183 		return -EINVAL;
184 	}
185 
186 	if (!flow_action_has_entries(actions)) {
187 		NL_SET_ERR_MSG_MOD(extack, "MATCHALL offload called with no action");
188 		return -EINVAL;
189 	}
190 
191 	if (!flow_offload_has_one_action(actions)) {
192 		NL_SET_ERR_MSG_MOD(extack,
193 				   "Egress MATCHALL offload supports only 1 policing action");
194 		return -EINVAL;
195 	}
196 	return 0;
197 }
198 
199 static int otx2_policer_validate(const struct flow_action *action,
200 				 const struct flow_action_entry *act,
201 				 struct netlink_ext_ack *extack)
202 {
203 	if (act->police.exceed.act_id != FLOW_ACTION_DROP) {
204 		NL_SET_ERR_MSG_MOD(extack,
205 				   "Offload not supported when exceed action is not drop");
206 		return -EOPNOTSUPP;
207 	}
208 
209 	if (act->police.notexceed.act_id != FLOW_ACTION_PIPE &&
210 	    act->police.notexceed.act_id != FLOW_ACTION_ACCEPT) {
211 		NL_SET_ERR_MSG_MOD(extack,
212 				   "Offload not supported when conform action is not pipe or ok");
213 		return -EOPNOTSUPP;
214 	}
215 
216 	if (act->police.notexceed.act_id == FLOW_ACTION_ACCEPT &&
217 	    !flow_action_is_last_entry(action, act)) {
218 		NL_SET_ERR_MSG_MOD(extack,
219 				   "Offload not supported when conform action is ok, but action is not last");
220 		return -EOPNOTSUPP;
221 	}
222 
223 	if (act->police.peakrate_bytes_ps ||
224 	    act->police.avrate || act->police.overhead) {
225 		NL_SET_ERR_MSG_MOD(extack,
226 				   "Offload not supported when peakrate/avrate/overhead is configured");
227 		return -EOPNOTSUPP;
228 	}
229 
230 	return 0;
231 }
232 
233 static int otx2_tc_egress_matchall_install(struct otx2_nic *nic,
234 					   struct tc_cls_matchall_offload *cls)
235 {
236 	struct netlink_ext_ack *extack = cls->common.extack;
237 	struct flow_action *actions = &cls->rule->action;
238 	struct flow_action_entry *entry;
239 	int err;
240 
241 	err = otx2_tc_validate_flow(nic, actions, extack);
242 	if (err)
243 		return err;
244 
245 	if (nic->flags & OTX2_FLAG_TC_MATCHALL_EGRESS_ENABLED) {
246 		NL_SET_ERR_MSG_MOD(extack,
247 				   "Only one Egress MATCHALL ratelimiter can be offloaded");
248 		return -ENOMEM;
249 	}
250 
251 	entry = &cls->rule->action.entries[0];
252 	switch (entry->id) {
253 	case FLOW_ACTION_POLICE:
254 		err = otx2_policer_validate(&cls->rule->action, entry, extack);
255 		if (err)
256 			return err;
257 
258 		if (entry->police.rate_pkt_ps) {
259 			NL_SET_ERR_MSG_MOD(extack, "QoS offload not support packets per second");
260 			return -EOPNOTSUPP;
261 		}
262 		err = otx2_set_matchall_egress_rate(nic, entry->police.burst,
263 						    otx2_convert_rate(entry->police.rate_bytes_ps));
264 		if (err)
265 			return err;
266 		nic->flags |= OTX2_FLAG_TC_MATCHALL_EGRESS_ENABLED;
267 		break;
268 	default:
269 		NL_SET_ERR_MSG_MOD(extack,
270 				   "Only police action is supported with Egress MATCHALL offload");
271 		return -EOPNOTSUPP;
272 	}
273 
274 	return 0;
275 }
276 
277 static int otx2_tc_egress_matchall_delete(struct otx2_nic *nic,
278 					  struct tc_cls_matchall_offload *cls)
279 {
280 	struct netlink_ext_ack *extack = cls->common.extack;
281 	int err;
282 
283 	if (nic->flags & OTX2_FLAG_INTF_DOWN) {
284 		NL_SET_ERR_MSG_MOD(extack, "Interface not initialized");
285 		return -EINVAL;
286 	}
287 
288 	err = otx2_set_matchall_egress_rate(nic, 0, 0);
289 	nic->flags &= ~OTX2_FLAG_TC_MATCHALL_EGRESS_ENABLED;
290 	return err;
291 }
292 
293 static int otx2_tc_act_set_hw_police(struct otx2_nic *nic,
294 				     struct otx2_tc_flow *node)
295 {
296 	int rc;
297 
298 	mutex_lock(&nic->mbox.lock);
299 
300 	rc = cn10k_alloc_leaf_profile(nic, &node->leaf_profile);
301 	if (rc) {
302 		mutex_unlock(&nic->mbox.lock);
303 		return rc;
304 	}
305 
306 	rc = cn10k_set_ipolicer_rate(nic, node->leaf_profile,
307 				     node->burst, node->rate, node->is_pps);
308 	if (rc)
309 		goto free_leaf;
310 
311 	rc = cn10k_map_unmap_rq_policer(nic, node->rq, node->leaf_profile, true);
312 	if (rc)
313 		goto free_leaf;
314 
315 	mutex_unlock(&nic->mbox.lock);
316 
317 	return 0;
318 
319 free_leaf:
320 	if (cn10k_free_leaf_profile(nic, node->leaf_profile))
321 		netdev_err(nic->netdev,
322 			   "Unable to free leaf bandwidth profile(%d)\n",
323 			   node->leaf_profile);
324 	mutex_unlock(&nic->mbox.lock);
325 	return rc;
326 }
327 
328 static int otx2_tc_act_set_police(struct otx2_nic *nic,
329 				  struct otx2_tc_flow *node,
330 				  struct flow_cls_offload *f,
331 				  u64 rate, u32 burst, u32 mark,
332 				  struct npc_install_flow_req *req, bool pps)
333 {
334 	struct netlink_ext_ack *extack = f->common.extack;
335 	struct otx2_hw *hw = &nic->hw;
336 	int rq_idx, rc;
337 
338 	rq_idx = find_first_zero_bit(&nic->rq_bmap, hw->rx_queues);
339 	if (rq_idx >= hw->rx_queues) {
340 		NL_SET_ERR_MSG_MOD(extack, "Police action rules exceeded");
341 		return -EINVAL;
342 	}
343 
344 	req->match_id = mark & 0xFFFFULL;
345 	req->index = rq_idx;
346 	req->op = NIX_RX_ACTIONOP_UCAST;
347 
348 	node->is_act_police = true;
349 	node->rq = rq_idx;
350 	node->burst = burst;
351 	node->rate = rate;
352 	node->is_pps = pps;
353 
354 	rc = otx2_tc_act_set_hw_police(nic, node);
355 	if (!rc)
356 		set_bit(rq_idx, &nic->rq_bmap);
357 
358 	return rc;
359 }
360 
361 static int otx2_tc_update_mcast(struct otx2_nic *nic,
362 				struct npc_install_flow_req *req,
363 				struct netlink_ext_ack *extack,
364 				struct otx2_tc_flow *node,
365 				struct nix_mcast_grp_update_req *ureq,
366 				u8 num_intf)
367 {
368 	struct nix_mcast_grp_update_req *grp_update_req;
369 	struct nix_mcast_grp_create_req *creq;
370 	struct nix_mcast_grp_create_rsp *crsp;
371 	u32 grp_index;
372 	int rc;
373 
374 	mutex_lock(&nic->mbox.lock);
375 	creq = otx2_mbox_alloc_msg_nix_mcast_grp_create(&nic->mbox);
376 	if (!creq) {
377 		rc = -ENOMEM;
378 		goto error;
379 	}
380 
381 	creq->dir = NIX_MCAST_INGRESS;
382 	/* Send message to AF */
383 	rc = otx2_sync_mbox_msg(&nic->mbox);
384 	if (rc) {
385 		NL_SET_ERR_MSG_MOD(extack, "Failed to create multicast group");
386 		goto error;
387 	}
388 
389 	crsp = (struct nix_mcast_grp_create_rsp *)otx2_mbox_get_rsp(&nic->mbox.mbox,
390 			0,
391 			&creq->hdr);
392 	if (IS_ERR(crsp)) {
393 		rc = PTR_ERR(crsp);
394 		goto error;
395 	}
396 
397 	grp_index = crsp->mcast_grp_idx;
398 	grp_update_req = otx2_mbox_alloc_msg_nix_mcast_grp_update(&nic->mbox);
399 	if (!grp_update_req) {
400 		NL_SET_ERR_MSG_MOD(extack, "Failed to update multicast group");
401 		rc = -ENOMEM;
402 		goto error;
403 	}
404 
405 	ureq->op = NIX_MCAST_OP_ADD_ENTRY;
406 	ureq->mcast_grp_idx = grp_index;
407 	ureq->num_mce_entry = num_intf;
408 	ureq->pcifunc[0] = nic->pcifunc;
409 	ureq->channel[0] = nic->hw.tx_chan_base;
410 
411 	ureq->dest_type[0] = NIX_RX_RSS;
412 	ureq->rq_rss_index[0] = 0;
413 	memcpy(&ureq->hdr, &grp_update_req->hdr, sizeof(struct mbox_msghdr));
414 	memcpy(grp_update_req, ureq, sizeof(struct nix_mcast_grp_update_req));
415 
416 	/* Send message to AF */
417 	rc = otx2_sync_mbox_msg(&nic->mbox);
418 	if (rc) {
419 		NL_SET_ERR_MSG_MOD(extack, "Failed to update multicast group");
420 		goto error;
421 	}
422 
423 	mutex_unlock(&nic->mbox.lock);
424 	req->op = NIX_RX_ACTIONOP_MCAST;
425 	req->index = grp_index;
426 	node->mcast_grp_idx = grp_index;
427 	return 0;
428 
429 error:
430 	mutex_unlock(&nic->mbox.lock);
431 	return rc;
432 }
433 
434 static int otx2_tc_parse_actions(struct otx2_nic *nic,
435 				 struct flow_action *flow_action,
436 				 struct npc_install_flow_req *req,
437 				 struct flow_cls_offload *f,
438 				 struct otx2_tc_flow *node)
439 {
440 	struct nix_mcast_grp_update_req dummy_grp_update_req = { 0 };
441 	struct netlink_ext_ack *extack = f->common.extack;
442 	bool pps = false, mcast = false;
443 	struct flow_action_entry *act;
444 	struct net_device *target;
445 	struct otx2_nic *priv;
446 	u32 burst, mark = 0;
447 	u8 nr_police = 0;
448 	u8 num_intf = 1;
449 	int err, i;
450 	u64 rate;
451 
452 	if (!flow_action_has_entries(flow_action)) {
453 		NL_SET_ERR_MSG_MOD(extack, "no tc actions specified");
454 		return -EINVAL;
455 	}
456 
457 	flow_action_for_each(i, act, flow_action) {
458 		switch (act->id) {
459 		case FLOW_ACTION_DROP:
460 			req->op = NIX_RX_ACTIONOP_DROP;
461 			return 0;
462 		case FLOW_ACTION_ACCEPT:
463 			req->op = NIX_RX_ACTION_DEFAULT;
464 			return 0;
465 		case FLOW_ACTION_REDIRECT_INGRESS:
466 			target = act->dev;
467 			priv = netdev_priv(target);
468 			/* npc_install_flow_req doesn't support passing a target pcifunc */
469 			if (rvu_get_pf(nic->pcifunc) != rvu_get_pf(priv->pcifunc)) {
470 				NL_SET_ERR_MSG_MOD(extack,
471 						   "can't redirect to other pf/vf");
472 				return -EOPNOTSUPP;
473 			}
474 			req->vf = priv->pcifunc & RVU_PFVF_FUNC_MASK;
475 
476 			/* if op is already set; avoid overwriting the same */
477 			if (!req->op)
478 				req->op = NIX_RX_ACTION_DEFAULT;
479 			break;
480 
481 		case FLOW_ACTION_VLAN_POP:
482 			req->vtag0_valid = true;
483 			/* use RX_VTAG_TYPE7 which is initialized to strip vlan tag */
484 			req->vtag0_type = NIX_AF_LFX_RX_VTAG_TYPE7;
485 			break;
486 		case FLOW_ACTION_POLICE:
487 			/* Ingress ratelimiting is not supported on OcteonTx2 */
488 			if (is_dev_otx2(nic->pdev)) {
489 				NL_SET_ERR_MSG_MOD(extack,
490 					"Ingress policing not supported on this platform");
491 				return -EOPNOTSUPP;
492 			}
493 
494 			err = otx2_policer_validate(flow_action, act, extack);
495 			if (err)
496 				return err;
497 
498 			if (act->police.rate_bytes_ps > 0) {
499 				rate = act->police.rate_bytes_ps * 8;
500 				burst = act->police.burst;
501 			} else if (act->police.rate_pkt_ps > 0) {
502 				/* The algorithm used to calculate rate
503 				 * mantissa, exponent values for a given token
504 				 * rate (token can be byte or packet) requires
505 				 * token rate to be mutiplied by 8.
506 				 */
507 				rate = act->police.rate_pkt_ps * 8;
508 				burst = act->police.burst_pkt;
509 				pps = true;
510 			}
511 			nr_police++;
512 			break;
513 		case FLOW_ACTION_MARK:
514 			if (act->mark & ~OTX2_RX_MATCH_ID_MASK) {
515 				NL_SET_ERR_MSG_MOD(extack, "Bad flow mark, only 16 bit supported");
516 				return -EOPNOTSUPP;
517 			}
518 			mark = act->mark;
519 			req->match_id = mark & OTX2_RX_MATCH_ID_MASK;
520 			req->op = NIX_RX_ACTION_DEFAULT;
521 			nic->flags |= OTX2_FLAG_TC_MARK_ENABLED;
522 			refcount_inc(&nic->flow_cfg->mark_flows);
523 			break;
524 
525 		case FLOW_ACTION_RX_QUEUE_MAPPING:
526 			req->op = NIX_RX_ACTIONOP_UCAST;
527 			req->index = act->rx_queue;
528 			break;
529 
530 		case FLOW_ACTION_MIRRED_INGRESS:
531 			target = act->dev;
532 			priv = netdev_priv(target);
533 			dummy_grp_update_req.pcifunc[num_intf] = priv->pcifunc;
534 			dummy_grp_update_req.channel[num_intf] = priv->hw.tx_chan_base;
535 			dummy_grp_update_req.dest_type[num_intf] = NIX_RX_RSS;
536 			dummy_grp_update_req.rq_rss_index[num_intf] = 0;
537 			mcast = true;
538 			num_intf++;
539 			break;
540 
541 		default:
542 			return -EOPNOTSUPP;
543 		}
544 	}
545 
546 	if (mcast) {
547 		err = otx2_tc_update_mcast(nic, req, extack, node,
548 					   &dummy_grp_update_req,
549 					   num_intf);
550 		if (err)
551 			return err;
552 	}
553 
554 	if (nr_police > 1) {
555 		NL_SET_ERR_MSG_MOD(extack,
556 				   "rate limit police offload requires a single action");
557 		return -EOPNOTSUPP;
558 	}
559 
560 	if (nr_police)
561 		return otx2_tc_act_set_police(nic, node, f, rate, burst,
562 					      mark, req, pps);
563 
564 	return 0;
565 }
566 
567 static int otx2_tc_process_vlan(struct otx2_nic *nic, struct flow_msg *flow_spec,
568 				struct flow_msg *flow_mask, struct flow_rule *rule,
569 				struct npc_install_flow_req *req, bool is_inner)
570 {
571 	struct flow_match_vlan match;
572 	u16 vlan_tci, vlan_tci_mask;
573 
574 	if (is_inner)
575 		flow_rule_match_cvlan(rule, &match);
576 	else
577 		flow_rule_match_vlan(rule, &match);
578 
579 	if (!eth_type_vlan(match.key->vlan_tpid)) {
580 		netdev_err(nic->netdev, "vlan tpid 0x%x not supported\n",
581 			   ntohs(match.key->vlan_tpid));
582 		return -EOPNOTSUPP;
583 	}
584 
585 	if (!match.mask->vlan_id) {
586 		struct flow_action_entry *act;
587 		int i;
588 
589 		flow_action_for_each(i, act, &rule->action) {
590 			if (act->id == FLOW_ACTION_DROP) {
591 				netdev_err(nic->netdev,
592 					   "vlan tpid 0x%x with vlan_id %d is not supported for DROP rule.\n",
593 					   ntohs(match.key->vlan_tpid), match.key->vlan_id);
594 				return -EOPNOTSUPP;
595 			}
596 		}
597 	}
598 
599 	if (match.mask->vlan_id ||
600 	    match.mask->vlan_dei ||
601 	    match.mask->vlan_priority) {
602 		vlan_tci = match.key->vlan_id |
603 			   match.key->vlan_dei << 12 |
604 			   match.key->vlan_priority << 13;
605 
606 		vlan_tci_mask = match.mask->vlan_id |
607 				match.mask->vlan_dei << 12 |
608 				match.mask->vlan_priority << 13;
609 		if (is_inner) {
610 			flow_spec->vlan_itci = htons(vlan_tci);
611 			flow_mask->vlan_itci = htons(vlan_tci_mask);
612 			req->features |= BIT_ULL(NPC_INNER_VID);
613 		} else {
614 			flow_spec->vlan_tci = htons(vlan_tci);
615 			flow_mask->vlan_tci = htons(vlan_tci_mask);
616 			req->features |= BIT_ULL(NPC_OUTER_VID);
617 		}
618 	}
619 
620 	return 0;
621 }
622 
623 static int otx2_tc_prepare_flow(struct otx2_nic *nic, struct otx2_tc_flow *node,
624 				struct flow_cls_offload *f,
625 				struct npc_install_flow_req *req)
626 {
627 	struct netlink_ext_ack *extack = f->common.extack;
628 	struct flow_msg *flow_spec = &req->packet;
629 	struct flow_msg *flow_mask = &req->mask;
630 	struct flow_dissector *dissector;
631 	struct flow_rule *rule;
632 	u8 ip_proto = 0;
633 
634 	rule = flow_cls_offload_flow_rule(f);
635 	dissector = rule->match.dissector;
636 
637 	if ((dissector->used_keys &
638 	    ~(BIT_ULL(FLOW_DISSECTOR_KEY_CONTROL) |
639 	      BIT_ULL(FLOW_DISSECTOR_KEY_BASIC) |
640 	      BIT_ULL(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
641 	      BIT_ULL(FLOW_DISSECTOR_KEY_VLAN) |
642 	      BIT(FLOW_DISSECTOR_KEY_CVLAN) |
643 	      BIT_ULL(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
644 	      BIT_ULL(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
645 	      BIT_ULL(FLOW_DISSECTOR_KEY_PORTS) |
646 	      BIT(FLOW_DISSECTOR_KEY_IPSEC) |
647 	      BIT_ULL(FLOW_DISSECTOR_KEY_MPLS) |
648 	      BIT_ULL(FLOW_DISSECTOR_KEY_ICMP) |
649 	      BIT_ULL(FLOW_DISSECTOR_KEY_TCP) |
650 	      BIT_ULL(FLOW_DISSECTOR_KEY_IP))))  {
651 		netdev_info(nic->netdev, "unsupported flow used key 0x%llx",
652 			    dissector->used_keys);
653 		return -EOPNOTSUPP;
654 	}
655 
656 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
657 		struct flow_match_basic match;
658 
659 		flow_rule_match_basic(rule, &match);
660 
661 		/* All EtherTypes can be matched, no hw limitation */
662 		flow_spec->etype = match.key->n_proto;
663 		flow_mask->etype = match.mask->n_proto;
664 		req->features |= BIT_ULL(NPC_ETYPE);
665 
666 		if (match.mask->ip_proto &&
667 		    (match.key->ip_proto != IPPROTO_TCP &&
668 		     match.key->ip_proto != IPPROTO_UDP &&
669 		     match.key->ip_proto != IPPROTO_SCTP &&
670 		     match.key->ip_proto != IPPROTO_ICMP &&
671 		     match.key->ip_proto != IPPROTO_ESP &&
672 		     match.key->ip_proto != IPPROTO_AH &&
673 		     match.key->ip_proto != IPPROTO_ICMPV6)) {
674 			netdev_info(nic->netdev,
675 				    "ip_proto=0x%x not supported\n",
676 				    match.key->ip_proto);
677 			return -EOPNOTSUPP;
678 		}
679 		if (match.mask->ip_proto)
680 			ip_proto = match.key->ip_proto;
681 
682 		if (ip_proto == IPPROTO_UDP)
683 			req->features |= BIT_ULL(NPC_IPPROTO_UDP);
684 		else if (ip_proto == IPPROTO_TCP)
685 			req->features |= BIT_ULL(NPC_IPPROTO_TCP);
686 		else if (ip_proto == IPPROTO_SCTP)
687 			req->features |= BIT_ULL(NPC_IPPROTO_SCTP);
688 		else if (ip_proto == IPPROTO_ICMP)
689 			req->features |= BIT_ULL(NPC_IPPROTO_ICMP);
690 		else if (ip_proto == IPPROTO_ICMPV6)
691 			req->features |= BIT_ULL(NPC_IPPROTO_ICMP6);
692 		else if (ip_proto == IPPROTO_ESP)
693 			req->features |= BIT_ULL(NPC_IPPROTO_ESP);
694 		else if (ip_proto == IPPROTO_AH)
695 			req->features |= BIT_ULL(NPC_IPPROTO_AH);
696 	}
697 
698 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL)) {
699 		struct flow_match_control match;
700 		u32 val;
701 
702 		flow_rule_match_control(rule, &match);
703 
704 		if (match.mask->flags & FLOW_DIS_IS_FRAGMENT) {
705 			val = match.key->flags & FLOW_DIS_IS_FRAGMENT;
706 			if (ntohs(flow_spec->etype) == ETH_P_IP) {
707 				flow_spec->ip_flag = val ? IPV4_FLAG_MORE : 0;
708 				flow_mask->ip_flag = IPV4_FLAG_MORE;
709 				req->features |= BIT_ULL(NPC_IPFRAG_IPV4);
710 			} else if (ntohs(flow_spec->etype) == ETH_P_IPV6) {
711 				flow_spec->next_header = val ?
712 							 IPPROTO_FRAGMENT : 0;
713 				flow_mask->next_header = 0xff;
714 				req->features |= BIT_ULL(NPC_IPFRAG_IPV6);
715 			} else {
716 				NL_SET_ERR_MSG_MOD(extack, "flow-type should be either IPv4 and IPv6");
717 				return -EOPNOTSUPP;
718 			}
719 		}
720 
721 		if (!flow_rule_is_supp_control_flags(FLOW_DIS_IS_FRAGMENT,
722 						     match.mask->flags, extack))
723 			return -EOPNOTSUPP;
724 	}
725 
726 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
727 		struct flow_match_eth_addrs match;
728 
729 		flow_rule_match_eth_addrs(rule, &match);
730 		if (!is_zero_ether_addr(match.mask->src)) {
731 			NL_SET_ERR_MSG_MOD(extack, "src mac match not supported");
732 			return -EOPNOTSUPP;
733 		}
734 
735 		if (!is_zero_ether_addr(match.mask->dst)) {
736 			ether_addr_copy(flow_spec->dmac, (u8 *)&match.key->dst);
737 			ether_addr_copy(flow_mask->dmac,
738 					(u8 *)&match.mask->dst);
739 			req->features |= BIT_ULL(NPC_DMAC);
740 		}
741 	}
742 
743 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IPSEC)) {
744 		struct flow_match_ipsec match;
745 
746 		flow_rule_match_ipsec(rule, &match);
747 		if (!match.mask->spi) {
748 			NL_SET_ERR_MSG_MOD(extack, "spi index not specified");
749 			return -EOPNOTSUPP;
750 		}
751 		if (ip_proto != IPPROTO_ESP &&
752 		    ip_proto != IPPROTO_AH) {
753 			NL_SET_ERR_MSG_MOD(extack,
754 					   "SPI index is valid only for ESP/AH proto");
755 			return -EOPNOTSUPP;
756 		}
757 
758 		flow_spec->spi = match.key->spi;
759 		flow_mask->spi = match.mask->spi;
760 		req->features |= BIT_ULL(NPC_IPSEC_SPI);
761 	}
762 
763 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IP)) {
764 		struct flow_match_ip match;
765 
766 		flow_rule_match_ip(rule, &match);
767 		if ((ntohs(flow_spec->etype) != ETH_P_IP) &&
768 		    match.mask->tos) {
769 			NL_SET_ERR_MSG_MOD(extack, "tos not supported");
770 			return -EOPNOTSUPP;
771 		}
772 		if (match.mask->ttl) {
773 			NL_SET_ERR_MSG_MOD(extack, "ttl not supported");
774 			return -EOPNOTSUPP;
775 		}
776 		flow_spec->tos = match.key->tos;
777 		flow_mask->tos = match.mask->tos;
778 		req->features |= BIT_ULL(NPC_TOS);
779 	}
780 
781 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
782 		int ret;
783 
784 		ret = otx2_tc_process_vlan(nic, flow_spec, flow_mask, rule, req, false);
785 		if (ret)
786 			return ret;
787 	}
788 
789 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CVLAN)) {
790 		int ret;
791 
792 		ret = otx2_tc_process_vlan(nic, flow_spec, flow_mask, rule, req, true);
793 		if (ret)
794 			return ret;
795 	}
796 
797 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IPV4_ADDRS)) {
798 		struct flow_match_ipv4_addrs match;
799 
800 		flow_rule_match_ipv4_addrs(rule, &match);
801 
802 		flow_spec->ip4dst = match.key->dst;
803 		flow_mask->ip4dst = match.mask->dst;
804 		req->features |= BIT_ULL(NPC_DIP_IPV4);
805 
806 		flow_spec->ip4src = match.key->src;
807 		flow_mask->ip4src = match.mask->src;
808 		req->features |= BIT_ULL(NPC_SIP_IPV4);
809 	} else if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_IPV6_ADDRS)) {
810 		struct flow_match_ipv6_addrs match;
811 
812 		flow_rule_match_ipv6_addrs(rule, &match);
813 
814 		if (ipv6_addr_loopback(&match.key->dst) ||
815 		    ipv6_addr_loopback(&match.key->src)) {
816 			NL_SET_ERR_MSG_MOD(extack,
817 					   "Flow matching IPv6 loopback addr not supported");
818 			return -EOPNOTSUPP;
819 		}
820 
821 		if (!ipv6_addr_any(&match.mask->dst)) {
822 			memcpy(&flow_spec->ip6dst,
823 			       (struct in6_addr *)&match.key->dst,
824 			       sizeof(flow_spec->ip6dst));
825 			memcpy(&flow_mask->ip6dst,
826 			       (struct in6_addr *)&match.mask->dst,
827 			       sizeof(flow_spec->ip6dst));
828 			req->features |= BIT_ULL(NPC_DIP_IPV6);
829 		}
830 
831 		if (!ipv6_addr_any(&match.mask->src)) {
832 			memcpy(&flow_spec->ip6src,
833 			       (struct in6_addr *)&match.key->src,
834 			       sizeof(flow_spec->ip6src));
835 			memcpy(&flow_mask->ip6src,
836 			       (struct in6_addr *)&match.mask->src,
837 			       sizeof(flow_spec->ip6src));
838 			req->features |= BIT_ULL(NPC_SIP_IPV6);
839 		}
840 	}
841 
842 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_PORTS)) {
843 		struct flow_match_ports match;
844 
845 		flow_rule_match_ports(rule, &match);
846 
847 		flow_spec->dport = match.key->dst;
848 		flow_mask->dport = match.mask->dst;
849 
850 		if (flow_mask->dport) {
851 			if (ip_proto == IPPROTO_UDP)
852 				req->features |= BIT_ULL(NPC_DPORT_UDP);
853 			else if (ip_proto == IPPROTO_TCP)
854 				req->features |= BIT_ULL(NPC_DPORT_TCP);
855 			else if (ip_proto == IPPROTO_SCTP)
856 				req->features |= BIT_ULL(NPC_DPORT_SCTP);
857 		}
858 
859 		flow_spec->sport = match.key->src;
860 		flow_mask->sport = match.mask->src;
861 
862 		if (flow_mask->sport) {
863 			if (ip_proto == IPPROTO_UDP)
864 				req->features |= BIT_ULL(NPC_SPORT_UDP);
865 			else if (ip_proto == IPPROTO_TCP)
866 				req->features |= BIT_ULL(NPC_SPORT_TCP);
867 			else if (ip_proto == IPPROTO_SCTP)
868 				req->features |= BIT_ULL(NPC_SPORT_SCTP);
869 		}
870 	}
871 
872 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_TCP)) {
873 		struct flow_match_tcp match;
874 
875 		flow_rule_match_tcp(rule, &match);
876 
877 		flow_spec->tcp_flags = match.key->flags;
878 		flow_mask->tcp_flags = match.mask->flags;
879 		req->features |= BIT_ULL(NPC_TCP_FLAGS);
880 	}
881 
882 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_MPLS)) {
883 		struct flow_match_mpls match;
884 		u8 bit;
885 
886 		flow_rule_match_mpls(rule, &match);
887 
888 		if (match.mask->used_lses & OTX2_UNSUPP_LSE_DEPTH) {
889 			NL_SET_ERR_MSG_MOD(extack,
890 					   "unsupported LSE depth for MPLS match offload");
891 			return -EOPNOTSUPP;
892 		}
893 
894 		for_each_set_bit(bit, (unsigned long *)&match.mask->used_lses,
895 				 FLOW_DIS_MPLS_MAX)  {
896 			/* check if any of the fields LABEL,TC,BOS are set */
897 			if (*((u32 *)&match.mask->ls[bit]) &
898 			    OTX2_FLOWER_MASK_MPLS_NON_TTL) {
899 				/* Hardware will capture 4 byte MPLS header into
900 				 * two fields NPC_MPLSX_LBTCBOS and NPC_MPLSX_TTL.
901 				 * Derive the associated NPC key based on header
902 				 * index and offset.
903 				 */
904 
905 				req->features |= BIT_ULL(NPC_MPLS1_LBTCBOS +
906 							 2 * bit);
907 				flow_spec->mpls_lse[bit] =
908 					FIELD_PREP(OTX2_FLOWER_MASK_MPLS_LB,
909 						   match.key->ls[bit].mpls_label) |
910 					FIELD_PREP(OTX2_FLOWER_MASK_MPLS_TC,
911 						   match.key->ls[bit].mpls_tc) |
912 					FIELD_PREP(OTX2_FLOWER_MASK_MPLS_BOS,
913 						   match.key->ls[bit].mpls_bos);
914 
915 				flow_mask->mpls_lse[bit] =
916 					FIELD_PREP(OTX2_FLOWER_MASK_MPLS_LB,
917 						   match.mask->ls[bit].mpls_label) |
918 					FIELD_PREP(OTX2_FLOWER_MASK_MPLS_TC,
919 						   match.mask->ls[bit].mpls_tc) |
920 					FIELD_PREP(OTX2_FLOWER_MASK_MPLS_BOS,
921 						   match.mask->ls[bit].mpls_bos);
922 			}
923 
924 			if (match.mask->ls[bit].mpls_ttl) {
925 				req->features |= BIT_ULL(NPC_MPLS1_TTL +
926 							 2 * bit);
927 				flow_spec->mpls_lse[bit] |=
928 					FIELD_PREP(OTX2_FLOWER_MASK_MPLS_TTL,
929 						   match.key->ls[bit].mpls_ttl);
930 				flow_mask->mpls_lse[bit] |=
931 					FIELD_PREP(OTX2_FLOWER_MASK_MPLS_TTL,
932 						   match.mask->ls[bit].mpls_ttl);
933 			}
934 		}
935 	}
936 
937 	if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ICMP)) {
938 		struct flow_match_icmp match;
939 
940 		flow_rule_match_icmp(rule, &match);
941 
942 		flow_spec->icmp_type = match.key->type;
943 		flow_mask->icmp_type = match.mask->type;
944 		req->features |= BIT_ULL(NPC_TYPE_ICMP);
945 
946 		flow_spec->icmp_code = match.key->code;
947 		flow_mask->icmp_code = match.mask->code;
948 		req->features |= BIT_ULL(NPC_CODE_ICMP);
949 	}
950 	return otx2_tc_parse_actions(nic, &rule->action, req, f, node);
951 }
952 
953 static void otx2_destroy_tc_flow_list(struct otx2_nic *pfvf)
954 {
955 	struct otx2_flow_config *flow_cfg = pfvf->flow_cfg;
956 	struct otx2_tc_flow *iter, *tmp;
957 
958 	if (!(pfvf->flags & OTX2_FLAG_MCAM_ENTRIES_ALLOC))
959 		return;
960 
961 	list_for_each_entry_safe(iter, tmp, &flow_cfg->flow_list_tc, list) {
962 		list_del(&iter->list);
963 		kfree(iter);
964 		flow_cfg->nr_flows--;
965 	}
966 }
967 
968 static struct otx2_tc_flow *otx2_tc_get_entry_by_cookie(struct otx2_flow_config *flow_cfg,
969 							unsigned long cookie)
970 {
971 	struct otx2_tc_flow *tmp;
972 
973 	list_for_each_entry(tmp, &flow_cfg->flow_list_tc, list) {
974 		if (tmp->cookie == cookie)
975 			return tmp;
976 	}
977 
978 	return NULL;
979 }
980 
981 static struct otx2_tc_flow *otx2_tc_get_entry_by_index(struct otx2_flow_config *flow_cfg,
982 						       int index)
983 {
984 	struct otx2_tc_flow *tmp;
985 	int i = 0;
986 
987 	list_for_each_entry(tmp, &flow_cfg->flow_list_tc, list) {
988 		if (i == index)
989 			return tmp;
990 		i++;
991 	}
992 
993 	return NULL;
994 }
995 
996 static void otx2_tc_del_from_flow_list(struct otx2_flow_config *flow_cfg,
997 				       struct otx2_tc_flow *node)
998 {
999 	struct list_head *pos, *n;
1000 	struct otx2_tc_flow *tmp;
1001 
1002 	list_for_each_safe(pos, n, &flow_cfg->flow_list_tc) {
1003 		tmp = list_entry(pos, struct otx2_tc_flow, list);
1004 		if (node == tmp) {
1005 			list_del(&node->list);
1006 			return;
1007 		}
1008 	}
1009 }
1010 
1011 static int otx2_tc_add_to_flow_list(struct otx2_flow_config *flow_cfg,
1012 				    struct otx2_tc_flow *node)
1013 {
1014 	struct list_head *pos, *n;
1015 	struct otx2_tc_flow *tmp;
1016 	int index = 0;
1017 
1018 	/* If the flow list is empty then add the new node */
1019 	if (list_empty(&flow_cfg->flow_list_tc)) {
1020 		list_add(&node->list, &flow_cfg->flow_list_tc);
1021 		return index;
1022 	}
1023 
1024 	list_for_each_safe(pos, n, &flow_cfg->flow_list_tc) {
1025 		tmp = list_entry(pos, struct otx2_tc_flow, list);
1026 		if (node->prio < tmp->prio)
1027 			break;
1028 		index++;
1029 	}
1030 
1031 	list_add(&node->list, pos->prev);
1032 	return index;
1033 }
1034 
1035 static int otx2_add_mcam_flow_entry(struct otx2_nic *nic, struct npc_install_flow_req *req)
1036 {
1037 	struct npc_install_flow_req *tmp_req;
1038 	int err;
1039 
1040 	mutex_lock(&nic->mbox.lock);
1041 	tmp_req = otx2_mbox_alloc_msg_npc_install_flow(&nic->mbox);
1042 	if (!tmp_req) {
1043 		mutex_unlock(&nic->mbox.lock);
1044 		return -ENOMEM;
1045 	}
1046 
1047 	memcpy(tmp_req, req, sizeof(struct npc_install_flow_req));
1048 	/* Send message to AF */
1049 	err = otx2_sync_mbox_msg(&nic->mbox);
1050 	if (err) {
1051 		netdev_err(nic->netdev, "Failed to install MCAM flow entry %d\n",
1052 			   req->entry);
1053 		mutex_unlock(&nic->mbox.lock);
1054 		return -EFAULT;
1055 	}
1056 
1057 	mutex_unlock(&nic->mbox.lock);
1058 	return 0;
1059 }
1060 
1061 static int otx2_del_mcam_flow_entry(struct otx2_nic *nic, u16 entry, u16 *cntr_val)
1062 {
1063 	struct npc_delete_flow_rsp *rsp;
1064 	struct npc_delete_flow_req *req;
1065 	int err;
1066 
1067 	mutex_lock(&nic->mbox.lock);
1068 	req = otx2_mbox_alloc_msg_npc_delete_flow(&nic->mbox);
1069 	if (!req) {
1070 		mutex_unlock(&nic->mbox.lock);
1071 		return -ENOMEM;
1072 	}
1073 
1074 	req->entry = entry;
1075 
1076 	/* Send message to AF */
1077 	err = otx2_sync_mbox_msg(&nic->mbox);
1078 	if (err) {
1079 		netdev_err(nic->netdev, "Failed to delete MCAM flow entry %d\n",
1080 			   entry);
1081 		mutex_unlock(&nic->mbox.lock);
1082 		return -EFAULT;
1083 	}
1084 
1085 	if (cntr_val) {
1086 		rsp = (struct npc_delete_flow_rsp *)otx2_mbox_get_rsp(&nic->mbox.mbox,
1087 								      0, &req->hdr);
1088 		if (IS_ERR(rsp)) {
1089 			netdev_err(nic->netdev, "Failed to get MCAM delete response for entry %d\n",
1090 				   entry);
1091 			mutex_unlock(&nic->mbox.lock);
1092 			return -EFAULT;
1093 		}
1094 
1095 		*cntr_val = rsp->cntr_val;
1096 	}
1097 
1098 	mutex_unlock(&nic->mbox.lock);
1099 	return 0;
1100 }
1101 
1102 static int otx2_tc_update_mcam_table_del_req(struct otx2_nic *nic,
1103 					     struct otx2_flow_config *flow_cfg,
1104 					     struct otx2_tc_flow *node)
1105 {
1106 	struct list_head *pos, *n;
1107 	struct otx2_tc_flow *tmp;
1108 	int i = 0, index = 0;
1109 	u16 cntr_val = 0;
1110 
1111 	/* Find and delete the entry from the list and re-install
1112 	 * all the entries from beginning to the index of the
1113 	 * deleted entry to higher mcam indexes.
1114 	 */
1115 	list_for_each_safe(pos, n, &flow_cfg->flow_list_tc) {
1116 		tmp = list_entry(pos, struct otx2_tc_flow, list);
1117 		if (node == tmp) {
1118 			list_del(&tmp->list);
1119 			break;
1120 		}
1121 
1122 		otx2_del_mcam_flow_entry(nic, tmp->entry, &cntr_val);
1123 		tmp->entry++;
1124 		tmp->req.entry = tmp->entry;
1125 		tmp->req.cntr_val = cntr_val;
1126 		index++;
1127 	}
1128 
1129 	list_for_each_safe(pos, n, &flow_cfg->flow_list_tc) {
1130 		if (i == index)
1131 			break;
1132 
1133 		tmp = list_entry(pos, struct otx2_tc_flow, list);
1134 		otx2_add_mcam_flow_entry(nic, &tmp->req);
1135 		i++;
1136 	}
1137 
1138 	return 0;
1139 }
1140 
1141 static int otx2_tc_update_mcam_table_add_req(struct otx2_nic *nic,
1142 					     struct otx2_flow_config *flow_cfg,
1143 					     struct otx2_tc_flow *node)
1144 {
1145 	int mcam_idx = flow_cfg->max_flows - flow_cfg->nr_flows - 1;
1146 	struct otx2_tc_flow *tmp;
1147 	int list_idx, i;
1148 	u16 cntr_val = 0;
1149 
1150 	/* Find the index of the entry(list_idx) whose priority
1151 	 * is greater than the new entry and re-install all
1152 	 * the entries from beginning to list_idx to higher
1153 	 * mcam indexes.
1154 	 */
1155 	list_idx = otx2_tc_add_to_flow_list(flow_cfg, node);
1156 	for (i = 0; i < list_idx; i++) {
1157 		tmp = otx2_tc_get_entry_by_index(flow_cfg, i);
1158 		if (!tmp)
1159 			return -ENOMEM;
1160 
1161 		otx2_del_mcam_flow_entry(nic, tmp->entry, &cntr_val);
1162 		tmp->entry = flow_cfg->flow_ent[mcam_idx];
1163 		tmp->req.entry = tmp->entry;
1164 		tmp->req.cntr_val = cntr_val;
1165 		otx2_add_mcam_flow_entry(nic, &tmp->req);
1166 		mcam_idx++;
1167 	}
1168 
1169 	return mcam_idx;
1170 }
1171 
1172 static int otx2_tc_update_mcam_table(struct otx2_nic *nic,
1173 				     struct otx2_flow_config *flow_cfg,
1174 				     struct otx2_tc_flow *node,
1175 				     bool add_req)
1176 {
1177 	if (add_req)
1178 		return otx2_tc_update_mcam_table_add_req(nic, flow_cfg, node);
1179 
1180 	return otx2_tc_update_mcam_table_del_req(nic, flow_cfg, node);
1181 }
1182 
1183 static int otx2_tc_del_flow(struct otx2_nic *nic,
1184 			    struct flow_cls_offload *tc_flow_cmd)
1185 {
1186 	struct otx2_flow_config *flow_cfg = nic->flow_cfg;
1187 	struct nix_mcast_grp_destroy_req *grp_destroy_req;
1188 	struct otx2_tc_flow *flow_node;
1189 	int err;
1190 
1191 	flow_node = otx2_tc_get_entry_by_cookie(flow_cfg, tc_flow_cmd->cookie);
1192 	if (!flow_node) {
1193 		netdev_err(nic->netdev, "tc flow not found for cookie 0x%lx\n",
1194 			   tc_flow_cmd->cookie);
1195 		return -EINVAL;
1196 	}
1197 
1198 	/* Disable TC MARK flag if they are no rules with skbedit mark action */
1199 	if (flow_node->req.match_id)
1200 		if (!refcount_dec_and_test(&flow_cfg->mark_flows))
1201 			nic->flags &= ~OTX2_FLAG_TC_MARK_ENABLED;
1202 
1203 	if (flow_node->is_act_police) {
1204 		__clear_bit(flow_node->rq, &nic->rq_bmap);
1205 
1206 		if (nic->flags & OTX2_FLAG_INTF_DOWN)
1207 			goto free_mcam_flow;
1208 
1209 		mutex_lock(&nic->mbox.lock);
1210 
1211 		err = cn10k_map_unmap_rq_policer(nic, flow_node->rq,
1212 						 flow_node->leaf_profile, false);
1213 		if (err)
1214 			netdev_err(nic->netdev,
1215 				   "Unmapping RQ %d & profile %d failed\n",
1216 				   flow_node->rq, flow_node->leaf_profile);
1217 
1218 		err = cn10k_free_leaf_profile(nic, flow_node->leaf_profile);
1219 		if (err)
1220 			netdev_err(nic->netdev,
1221 				   "Unable to free leaf bandwidth profile(%d)\n",
1222 				   flow_node->leaf_profile);
1223 
1224 		mutex_unlock(&nic->mbox.lock);
1225 	}
1226 	/* Remove the multicast/mirror related nodes */
1227 	if (flow_node->mcast_grp_idx != MCAST_INVALID_GRP) {
1228 		mutex_lock(&nic->mbox.lock);
1229 		grp_destroy_req = otx2_mbox_alloc_msg_nix_mcast_grp_destroy(&nic->mbox);
1230 		grp_destroy_req->mcast_grp_idx = flow_node->mcast_grp_idx;
1231 		otx2_sync_mbox_msg(&nic->mbox);
1232 		mutex_unlock(&nic->mbox.lock);
1233 	}
1234 
1235 
1236 free_mcam_flow:
1237 	otx2_del_mcam_flow_entry(nic, flow_node->entry, NULL);
1238 	otx2_tc_update_mcam_table(nic, flow_cfg, flow_node, false);
1239 	kfree_rcu(flow_node, rcu);
1240 	flow_cfg->nr_flows--;
1241 	return 0;
1242 }
1243 
1244 static int otx2_tc_add_flow(struct otx2_nic *nic,
1245 			    struct flow_cls_offload *tc_flow_cmd)
1246 {
1247 	struct netlink_ext_ack *extack = tc_flow_cmd->common.extack;
1248 	struct otx2_flow_config *flow_cfg = nic->flow_cfg;
1249 	struct otx2_tc_flow *new_node, *old_node;
1250 	struct npc_install_flow_req *req, dummy;
1251 	int rc, err, mcam_idx;
1252 
1253 	if (!(nic->flags & OTX2_FLAG_TC_FLOWER_SUPPORT))
1254 		return -ENOMEM;
1255 
1256 	if (nic->flags & OTX2_FLAG_INTF_DOWN) {
1257 		NL_SET_ERR_MSG_MOD(extack, "Interface not initialized");
1258 		return -EINVAL;
1259 	}
1260 
1261 	if (flow_cfg->nr_flows == flow_cfg->max_flows) {
1262 		NL_SET_ERR_MSG_MOD(extack,
1263 				   "Free MCAM entry not available to add the flow");
1264 		return -ENOMEM;
1265 	}
1266 
1267 	/* allocate memory for the new flow and it's node */
1268 	new_node = kzalloc(sizeof(*new_node), GFP_KERNEL);
1269 	if (!new_node)
1270 		return -ENOMEM;
1271 	spin_lock_init(&new_node->lock);
1272 	new_node->cookie = tc_flow_cmd->cookie;
1273 	new_node->prio = tc_flow_cmd->common.prio;
1274 	new_node->mcast_grp_idx = MCAST_INVALID_GRP;
1275 
1276 	memset(&dummy, 0, sizeof(struct npc_install_flow_req));
1277 
1278 	rc = otx2_tc_prepare_flow(nic, new_node, tc_flow_cmd, &dummy);
1279 	if (rc) {
1280 		kfree_rcu(new_node, rcu);
1281 		return rc;
1282 	}
1283 
1284 	/* If a flow exists with the same cookie, delete it */
1285 	old_node = otx2_tc_get_entry_by_cookie(flow_cfg, tc_flow_cmd->cookie);
1286 	if (old_node)
1287 		otx2_tc_del_flow(nic, tc_flow_cmd);
1288 
1289 	mcam_idx = otx2_tc_update_mcam_table(nic, flow_cfg, new_node, true);
1290 	mutex_lock(&nic->mbox.lock);
1291 	req = otx2_mbox_alloc_msg_npc_install_flow(&nic->mbox);
1292 	if (!req) {
1293 		mutex_unlock(&nic->mbox.lock);
1294 		rc = -ENOMEM;
1295 		goto free_leaf;
1296 	}
1297 
1298 	memcpy(&dummy.hdr, &req->hdr, sizeof(struct mbox_msghdr));
1299 	memcpy(req, &dummy, sizeof(struct npc_install_flow_req));
1300 	req->channel = nic->hw.rx_chan_base;
1301 	req->entry = flow_cfg->flow_ent[mcam_idx];
1302 	req->intf = NIX_INTF_RX;
1303 	req->set_cntr = 1;
1304 	new_node->entry = req->entry;
1305 
1306 	/* Send message to AF */
1307 	rc = otx2_sync_mbox_msg(&nic->mbox);
1308 	if (rc) {
1309 		NL_SET_ERR_MSG_MOD(extack, "Failed to install MCAM flow entry");
1310 		mutex_unlock(&nic->mbox.lock);
1311 		goto free_leaf;
1312 	}
1313 
1314 	mutex_unlock(&nic->mbox.lock);
1315 	memcpy(&new_node->req, req, sizeof(struct npc_install_flow_req));
1316 
1317 	flow_cfg->nr_flows++;
1318 	return 0;
1319 
1320 free_leaf:
1321 	otx2_tc_del_from_flow_list(flow_cfg, new_node);
1322 	kfree_rcu(new_node, rcu);
1323 	if (new_node->is_act_police) {
1324 		mutex_lock(&nic->mbox.lock);
1325 
1326 		err = cn10k_map_unmap_rq_policer(nic, new_node->rq,
1327 						 new_node->leaf_profile, false);
1328 		if (err)
1329 			netdev_err(nic->netdev,
1330 				   "Unmapping RQ %d & profile %d failed\n",
1331 				   new_node->rq, new_node->leaf_profile);
1332 		err = cn10k_free_leaf_profile(nic, new_node->leaf_profile);
1333 		if (err)
1334 			netdev_err(nic->netdev,
1335 				   "Unable to free leaf bandwidth profile(%d)\n",
1336 				   new_node->leaf_profile);
1337 
1338 		__clear_bit(new_node->rq, &nic->rq_bmap);
1339 
1340 		mutex_unlock(&nic->mbox.lock);
1341 	}
1342 
1343 	return rc;
1344 }
1345 
1346 static int otx2_tc_get_flow_stats(struct otx2_nic *nic,
1347 				  struct flow_cls_offload *tc_flow_cmd)
1348 {
1349 	struct npc_mcam_get_stats_req *req;
1350 	struct npc_mcam_get_stats_rsp *rsp;
1351 	struct otx2_tc_flow_stats *stats;
1352 	struct otx2_tc_flow *flow_node;
1353 	int err;
1354 
1355 	flow_node = otx2_tc_get_entry_by_cookie(nic->flow_cfg, tc_flow_cmd->cookie);
1356 	if (!flow_node) {
1357 		netdev_info(nic->netdev, "tc flow not found for cookie %lx",
1358 			    tc_flow_cmd->cookie);
1359 		return -EINVAL;
1360 	}
1361 
1362 	mutex_lock(&nic->mbox.lock);
1363 
1364 	req = otx2_mbox_alloc_msg_npc_mcam_entry_stats(&nic->mbox);
1365 	if (!req) {
1366 		mutex_unlock(&nic->mbox.lock);
1367 		return -ENOMEM;
1368 	}
1369 
1370 	req->entry = flow_node->entry;
1371 
1372 	err = otx2_sync_mbox_msg(&nic->mbox);
1373 	if (err) {
1374 		netdev_err(nic->netdev, "Failed to get stats for MCAM flow entry %d\n",
1375 			   req->entry);
1376 		mutex_unlock(&nic->mbox.lock);
1377 		return -EFAULT;
1378 	}
1379 
1380 	rsp = (struct npc_mcam_get_stats_rsp *)otx2_mbox_get_rsp
1381 		(&nic->mbox.mbox, 0, &req->hdr);
1382 	if (IS_ERR(rsp)) {
1383 		mutex_unlock(&nic->mbox.lock);
1384 		return PTR_ERR(rsp);
1385 	}
1386 
1387 	mutex_unlock(&nic->mbox.lock);
1388 
1389 	if (!rsp->stat_ena)
1390 		return -EINVAL;
1391 
1392 	stats = &flow_node->stats;
1393 
1394 	spin_lock(&flow_node->lock);
1395 	flow_stats_update(&tc_flow_cmd->stats, 0x0, rsp->stat - stats->pkts, 0x0, 0x0,
1396 			  FLOW_ACTION_HW_STATS_IMMEDIATE);
1397 	stats->pkts = rsp->stat;
1398 	spin_unlock(&flow_node->lock);
1399 
1400 	return 0;
1401 }
1402 
1403 static int otx2_setup_tc_cls_flower(struct otx2_nic *nic,
1404 				    struct flow_cls_offload *cls_flower)
1405 {
1406 	switch (cls_flower->command) {
1407 	case FLOW_CLS_REPLACE:
1408 		return otx2_tc_add_flow(nic, cls_flower);
1409 	case FLOW_CLS_DESTROY:
1410 		return otx2_tc_del_flow(nic, cls_flower);
1411 	case FLOW_CLS_STATS:
1412 		return otx2_tc_get_flow_stats(nic, cls_flower);
1413 	default:
1414 		return -EOPNOTSUPP;
1415 	}
1416 }
1417 
1418 static int otx2_tc_ingress_matchall_install(struct otx2_nic *nic,
1419 					    struct tc_cls_matchall_offload *cls)
1420 {
1421 	struct netlink_ext_ack *extack = cls->common.extack;
1422 	struct flow_action *actions = &cls->rule->action;
1423 	struct flow_action_entry *entry;
1424 	u64 rate;
1425 	int err;
1426 
1427 	err = otx2_tc_validate_flow(nic, actions, extack);
1428 	if (err)
1429 		return err;
1430 
1431 	if (nic->flags & OTX2_FLAG_TC_MATCHALL_INGRESS_ENABLED) {
1432 		NL_SET_ERR_MSG_MOD(extack,
1433 				   "Only one ingress MATCHALL ratelimitter can be offloaded");
1434 		return -ENOMEM;
1435 	}
1436 
1437 	entry = &cls->rule->action.entries[0];
1438 	switch (entry->id) {
1439 	case FLOW_ACTION_POLICE:
1440 		/* Ingress ratelimiting is not supported on OcteonTx2 */
1441 		if (is_dev_otx2(nic->pdev)) {
1442 			NL_SET_ERR_MSG_MOD(extack,
1443 					   "Ingress policing not supported on this platform");
1444 			return -EOPNOTSUPP;
1445 		}
1446 
1447 		err = cn10k_alloc_matchall_ipolicer(nic);
1448 		if (err)
1449 			return err;
1450 
1451 		/* Convert to bits per second */
1452 		rate = entry->police.rate_bytes_ps * 8;
1453 		err = cn10k_set_matchall_ipolicer_rate(nic, entry->police.burst, rate);
1454 		if (err)
1455 			return err;
1456 		nic->flags |= OTX2_FLAG_TC_MATCHALL_INGRESS_ENABLED;
1457 		break;
1458 	default:
1459 		NL_SET_ERR_MSG_MOD(extack,
1460 				   "Only police action supported with Ingress MATCHALL offload");
1461 		return -EOPNOTSUPP;
1462 	}
1463 
1464 	return 0;
1465 }
1466 
1467 static int otx2_tc_ingress_matchall_delete(struct otx2_nic *nic,
1468 					   struct tc_cls_matchall_offload *cls)
1469 {
1470 	struct netlink_ext_ack *extack = cls->common.extack;
1471 	int err;
1472 
1473 	if (nic->flags & OTX2_FLAG_INTF_DOWN) {
1474 		NL_SET_ERR_MSG_MOD(extack, "Interface not initialized");
1475 		return -EINVAL;
1476 	}
1477 
1478 	err = cn10k_free_matchall_ipolicer(nic);
1479 	nic->flags &= ~OTX2_FLAG_TC_MATCHALL_INGRESS_ENABLED;
1480 	return err;
1481 }
1482 
1483 static int otx2_setup_tc_ingress_matchall(struct otx2_nic *nic,
1484 					  struct tc_cls_matchall_offload *cls_matchall)
1485 {
1486 	switch (cls_matchall->command) {
1487 	case TC_CLSMATCHALL_REPLACE:
1488 		return otx2_tc_ingress_matchall_install(nic, cls_matchall);
1489 	case TC_CLSMATCHALL_DESTROY:
1490 		return otx2_tc_ingress_matchall_delete(nic, cls_matchall);
1491 	case TC_CLSMATCHALL_STATS:
1492 	default:
1493 		break;
1494 	}
1495 
1496 	return -EOPNOTSUPP;
1497 }
1498 
1499 static int otx2_setup_tc_block_ingress_cb(enum tc_setup_type type,
1500 					  void *type_data, void *cb_priv)
1501 {
1502 	struct otx2_nic *nic = cb_priv;
1503 	bool ntuple;
1504 
1505 	if (!tc_cls_can_offload_and_chain0(nic->netdev, type_data))
1506 		return -EOPNOTSUPP;
1507 
1508 	ntuple = nic->netdev->features & NETIF_F_NTUPLE;
1509 	switch (type) {
1510 	case TC_SETUP_CLSFLOWER:
1511 		if (ntuple) {
1512 			netdev_warn(nic->netdev,
1513 				    "Can't install TC flower offload rule when NTUPLE is active");
1514 			return -EOPNOTSUPP;
1515 		}
1516 
1517 		return otx2_setup_tc_cls_flower(nic, type_data);
1518 	case TC_SETUP_CLSMATCHALL:
1519 		return otx2_setup_tc_ingress_matchall(nic, type_data);
1520 	default:
1521 		break;
1522 	}
1523 
1524 	return -EOPNOTSUPP;
1525 }
1526 
1527 static int otx2_setup_tc_egress_matchall(struct otx2_nic *nic,
1528 					 struct tc_cls_matchall_offload *cls_matchall)
1529 {
1530 	switch (cls_matchall->command) {
1531 	case TC_CLSMATCHALL_REPLACE:
1532 		return otx2_tc_egress_matchall_install(nic, cls_matchall);
1533 	case TC_CLSMATCHALL_DESTROY:
1534 		return otx2_tc_egress_matchall_delete(nic, cls_matchall);
1535 	case TC_CLSMATCHALL_STATS:
1536 	default:
1537 		break;
1538 	}
1539 
1540 	return -EOPNOTSUPP;
1541 }
1542 
1543 static int otx2_setup_tc_block_egress_cb(enum tc_setup_type type,
1544 					 void *type_data, void *cb_priv)
1545 {
1546 	struct otx2_nic *nic = cb_priv;
1547 
1548 	if (!tc_cls_can_offload_and_chain0(nic->netdev, type_data))
1549 		return -EOPNOTSUPP;
1550 
1551 	switch (type) {
1552 	case TC_SETUP_CLSMATCHALL:
1553 		return otx2_setup_tc_egress_matchall(nic, type_data);
1554 	default:
1555 		break;
1556 	}
1557 
1558 	return -EOPNOTSUPP;
1559 }
1560 
1561 static LIST_HEAD(otx2_block_cb_list);
1562 
1563 static int otx2_setup_tc_block(struct net_device *netdev,
1564 			       struct flow_block_offload *f)
1565 {
1566 	struct otx2_nic *nic = netdev_priv(netdev);
1567 	flow_setup_cb_t *cb;
1568 	bool ingress;
1569 
1570 	if (f->block_shared)
1571 		return -EOPNOTSUPP;
1572 
1573 	if (f->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS) {
1574 		cb = otx2_setup_tc_block_ingress_cb;
1575 		ingress = true;
1576 	} else if (f->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS) {
1577 		cb = otx2_setup_tc_block_egress_cb;
1578 		ingress = false;
1579 	} else {
1580 		return -EOPNOTSUPP;
1581 	}
1582 
1583 	return flow_block_cb_setup_simple(f, &otx2_block_cb_list, cb,
1584 					  nic, nic, ingress);
1585 }
1586 
1587 int otx2_setup_tc(struct net_device *netdev, enum tc_setup_type type,
1588 		  void *type_data)
1589 {
1590 	switch (type) {
1591 	case TC_SETUP_BLOCK:
1592 		return otx2_setup_tc_block(netdev, type_data);
1593 	case TC_SETUP_QDISC_HTB:
1594 		return otx2_setup_tc_htb(netdev, type_data);
1595 	default:
1596 		return -EOPNOTSUPP;
1597 	}
1598 }
1599 EXPORT_SYMBOL(otx2_setup_tc);
1600 
1601 int otx2_init_tc(struct otx2_nic *nic)
1602 {
1603 	/* Exclude receive queue 0 being used for police action */
1604 	set_bit(0, &nic->rq_bmap);
1605 
1606 	if (!nic->flow_cfg) {
1607 		netdev_err(nic->netdev,
1608 			   "Can't init TC, nic->flow_cfg is not setup\n");
1609 		return -EINVAL;
1610 	}
1611 
1612 	return 0;
1613 }
1614 EXPORT_SYMBOL(otx2_init_tc);
1615 
1616 void otx2_shutdown_tc(struct otx2_nic *nic)
1617 {
1618 	otx2_destroy_tc_flow_list(nic);
1619 }
1620 EXPORT_SYMBOL(otx2_shutdown_tc);
1621 
1622 static void otx2_tc_config_ingress_rule(struct otx2_nic *nic,
1623 					struct otx2_tc_flow *node)
1624 {
1625 	struct npc_install_flow_req *req;
1626 
1627 	if (otx2_tc_act_set_hw_police(nic, node))
1628 		return;
1629 
1630 	mutex_lock(&nic->mbox.lock);
1631 
1632 	req = otx2_mbox_alloc_msg_npc_install_flow(&nic->mbox);
1633 	if (!req)
1634 		goto err;
1635 
1636 	memcpy(req, &node->req, sizeof(struct npc_install_flow_req));
1637 
1638 	if (otx2_sync_mbox_msg(&nic->mbox))
1639 		netdev_err(nic->netdev,
1640 			   "Failed to install MCAM flow entry for ingress rule");
1641 err:
1642 	mutex_unlock(&nic->mbox.lock);
1643 }
1644 
1645 void otx2_tc_apply_ingress_police_rules(struct otx2_nic *nic)
1646 {
1647 	struct otx2_flow_config *flow_cfg = nic->flow_cfg;
1648 	struct otx2_tc_flow *node;
1649 
1650 	/* If any ingress policer rules exist for the interface then
1651 	 * apply those rules. Ingress policer rules depend on bandwidth
1652 	 * profiles linked to the receive queues. Since no receive queues
1653 	 * exist when interface is down, ingress policer rules are stored
1654 	 * and configured in hardware after all receive queues are allocated
1655 	 * in otx2_open.
1656 	 */
1657 	list_for_each_entry(node, &flow_cfg->flow_list_tc, list) {
1658 		if (node->is_act_police)
1659 			otx2_tc_config_ingress_rule(nic, node);
1660 	}
1661 }
1662 EXPORT_SYMBOL(otx2_tc_apply_ingress_police_rules);
1663