1 /* Broadcom NetXtreme-C/E network driver.
3 * Copyright (c) 2017 Broadcom Limited
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation.
10 #include <linux/netdevice.h>
11 #include <linux/inetdevice.h>
12 #include <linux/if_vlan.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_skbedit.h>
17 #include <net/tc_act/tc_mirred.h>
18 #include <net/tc_act/tc_vlan.h>
22 #include "bnxt_sriov.h"
26 #ifdef CONFIG_BNXT_FLOWER_OFFLOAD
28 #define BNXT_FID_INVALID 0xffff
29 #define VLAN_TCI(vid, prio) ((vid) | ((prio) << VLAN_PRIO_SHIFT))
31 /* Return the dst fid of the func for flow forwarding
32 * For PFs: src_fid is the fid of the PF
33 * For VF-reps: src_fid the fid of the VF
35 static u16 bnxt_flow_get_dst_fid(struct bnxt *pf_bp, struct net_device *dev)
39 /* check if dev belongs to the same switch */
40 if (!switchdev_port_same_parent_id(pf_bp->dev, dev)) {
41 netdev_info(pf_bp->dev, "dev(ifindex=%d) not on same switch",
43 return BNXT_FID_INVALID;
46 /* Is dev a VF-rep? */
47 if (dev != pf_bp->dev)
48 return bnxt_vf_rep_get_fid(dev);
50 bp = netdev_priv(dev);
54 static int bnxt_tc_parse_redir(struct bnxt *bp,
55 struct bnxt_tc_actions *actions,
56 const struct tc_action *tc_act)
58 int ifindex = tcf_mirred_ifindex(tc_act);
59 struct net_device *dev;
62 dev = __dev_get_by_index(dev_net(bp->dev), ifindex);
64 netdev_info(bp->dev, "no dev for ifindex=%d", ifindex);
68 /* find the FID from dev */
69 dst_fid = bnxt_flow_get_dst_fid(bp, dev);
70 if (dst_fid == BNXT_FID_INVALID) {
71 netdev_info(bp->dev, "can't get fid for ifindex=%d", ifindex);
75 actions->flags |= BNXT_TC_ACTION_FLAG_FWD;
76 actions->dst_fid = dst_fid;
77 actions->dst_dev = dev;
81 static int bnxt_tc_parse_vlan(struct bnxt *bp,
82 struct bnxt_tc_actions *actions,
83 const struct tc_action *tc_act)
85 switch (tcf_vlan_action(tc_act)) {
86 case TCA_VLAN_ACT_POP:
87 actions->flags |= BNXT_TC_ACTION_FLAG_POP_VLAN;
89 case TCA_VLAN_ACT_PUSH:
90 actions->flags |= BNXT_TC_ACTION_FLAG_PUSH_VLAN;
91 actions->push_vlan_tci = htons(tcf_vlan_push_vid(tc_act));
92 actions->push_vlan_tpid = tcf_vlan_push_proto(tc_act);
100 static int bnxt_tc_parse_actions(struct bnxt *bp,
101 struct bnxt_tc_actions *actions,
102 struct tcf_exts *tc_exts)
104 const struct tc_action *tc_act;
105 LIST_HEAD(tc_actions);
108 if (!tcf_exts_has_actions(tc_exts)) {
109 netdev_info(bp->dev, "no actions");
113 tcf_exts_to_list(tc_exts, &tc_actions);
114 list_for_each_entry(tc_act, &tc_actions, list) {
116 if (is_tcf_gact_shot(tc_act)) {
117 actions->flags |= BNXT_TC_ACTION_FLAG_DROP;
118 return 0; /* don't bother with other actions */
121 /* Redirect action */
122 if (is_tcf_mirred_egress_redirect(tc_act)) {
123 rc = bnxt_tc_parse_redir(bp, actions, tc_act);
130 if (is_tcf_vlan(tc_act)) {
131 rc = bnxt_tc_parse_vlan(bp, actions, tc_act);
141 #define GET_KEY(flow_cmd, key_type) \
142 skb_flow_dissector_target((flow_cmd)->dissector, key_type,\
144 #define GET_MASK(flow_cmd, key_type) \
145 skb_flow_dissector_target((flow_cmd)->dissector, key_type,\
148 static int bnxt_tc_parse_flow(struct bnxt *bp,
149 struct tc_cls_flower_offload *tc_flow_cmd,
150 struct bnxt_tc_flow *flow)
152 struct flow_dissector *dissector = tc_flow_cmd->dissector;
155 /* KEY_CONTROL and KEY_BASIC are needed for forming a meaningful key */
156 if ((dissector->used_keys & BIT(FLOW_DISSECTOR_KEY_CONTROL)) == 0 ||
157 (dissector->used_keys & BIT(FLOW_DISSECTOR_KEY_BASIC)) == 0) {
158 netdev_info(bp->dev, "cannot form TC key: used_keys = 0x%x",
159 dissector->used_keys);
163 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_CONTROL)) {
164 struct flow_dissector_key_control *key =
165 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_CONTROL);
167 addr_type = key->addr_type;
170 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_BASIC)) {
171 struct flow_dissector_key_basic *key =
172 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_BASIC);
173 struct flow_dissector_key_basic *mask =
174 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_BASIC);
176 flow->l2_key.ether_type = key->n_proto;
177 flow->l2_mask.ether_type = mask->n_proto;
179 if (key->n_proto == htons(ETH_P_IP) ||
180 key->n_proto == htons(ETH_P_IPV6)) {
181 flow->l4_key.ip_proto = key->ip_proto;
182 flow->l4_mask.ip_proto = mask->ip_proto;
186 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
187 struct flow_dissector_key_eth_addrs *key =
188 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_ETH_ADDRS);
189 struct flow_dissector_key_eth_addrs *mask =
190 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_ETH_ADDRS);
192 flow->flags |= BNXT_TC_FLOW_FLAGS_ETH_ADDRS;
193 ether_addr_copy(flow->l2_key.dmac, key->dst);
194 ether_addr_copy(flow->l2_mask.dmac, mask->dst);
195 ether_addr_copy(flow->l2_key.smac, key->src);
196 ether_addr_copy(flow->l2_mask.smac, mask->src);
199 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_VLAN)) {
200 struct flow_dissector_key_vlan *key =
201 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_VLAN);
202 struct flow_dissector_key_vlan *mask =
203 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_VLAN);
205 flow->l2_key.inner_vlan_tci =
206 cpu_to_be16(VLAN_TCI(key->vlan_id, key->vlan_priority));
207 flow->l2_mask.inner_vlan_tci =
208 cpu_to_be16((VLAN_TCI(mask->vlan_id, mask->vlan_priority)));
209 flow->l2_key.inner_vlan_tpid = htons(ETH_P_8021Q);
210 flow->l2_mask.inner_vlan_tpid = htons(0xffff);
211 flow->l2_key.num_vlans = 1;
214 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_IPV4_ADDRS)) {
215 struct flow_dissector_key_ipv4_addrs *key =
216 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_IPV4_ADDRS);
217 struct flow_dissector_key_ipv4_addrs *mask =
218 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_IPV4_ADDRS);
220 flow->flags |= BNXT_TC_FLOW_FLAGS_IPV4_ADDRS;
221 flow->l3_key.ipv4.daddr.s_addr = key->dst;
222 flow->l3_mask.ipv4.daddr.s_addr = mask->dst;
223 flow->l3_key.ipv4.saddr.s_addr = key->src;
224 flow->l3_mask.ipv4.saddr.s_addr = mask->src;
225 } else if (dissector_uses_key(dissector,
226 FLOW_DISSECTOR_KEY_IPV6_ADDRS)) {
227 struct flow_dissector_key_ipv6_addrs *key =
228 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_IPV6_ADDRS);
229 struct flow_dissector_key_ipv6_addrs *mask =
230 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_IPV6_ADDRS);
232 flow->flags |= BNXT_TC_FLOW_FLAGS_IPV6_ADDRS;
233 flow->l3_key.ipv6.daddr = key->dst;
234 flow->l3_mask.ipv6.daddr = mask->dst;
235 flow->l3_key.ipv6.saddr = key->src;
236 flow->l3_mask.ipv6.saddr = mask->src;
239 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_PORTS)) {
240 struct flow_dissector_key_ports *key =
241 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_PORTS);
242 struct flow_dissector_key_ports *mask =
243 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_PORTS);
245 flow->flags |= BNXT_TC_FLOW_FLAGS_PORTS;
246 flow->l4_key.ports.dport = key->dst;
247 flow->l4_mask.ports.dport = mask->dst;
248 flow->l4_key.ports.sport = key->src;
249 flow->l4_mask.ports.sport = mask->src;
252 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_ICMP)) {
253 struct flow_dissector_key_icmp *key =
254 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_ICMP);
255 struct flow_dissector_key_icmp *mask =
256 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_ICMP);
258 flow->flags |= BNXT_TC_FLOW_FLAGS_ICMP;
259 flow->l4_key.icmp.type = key->type;
260 flow->l4_key.icmp.code = key->code;
261 flow->l4_mask.icmp.type = mask->type;
262 flow->l4_mask.icmp.code = mask->code;
265 return bnxt_tc_parse_actions(bp, &flow->actions, tc_flow_cmd->exts);
268 static int bnxt_hwrm_cfa_flow_free(struct bnxt *bp, __le16 flow_handle)
270 struct hwrm_cfa_flow_free_input req = { 0 };
273 bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_FLOW_FREE, -1, -1);
274 req.flow_handle = flow_handle;
276 rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
278 netdev_info(bp->dev, "Error: %s: flow_handle=0x%x rc=%d",
279 __func__, flow_handle, rc);
283 static int ipv6_mask_len(struct in6_addr *mask)
287 for (i = 0; i < 4; i++)
288 mask_len += inet_mask_len(mask->s6_addr32[i]);
293 static bool is_wildcard(void *mask, int len)
298 for (i = 0; i < len; i++) {
305 static int bnxt_hwrm_cfa_flow_alloc(struct bnxt *bp, struct bnxt_tc_flow *flow,
306 __le16 ref_flow_handle, __le16 *flow_handle)
308 struct hwrm_cfa_flow_alloc_output *resp = bp->hwrm_cmd_resp_addr;
309 struct bnxt_tc_actions *actions = &flow->actions;
310 struct bnxt_tc_l3_key *l3_mask = &flow->l3_mask;
311 struct bnxt_tc_l3_key *l3_key = &flow->l3_key;
312 struct hwrm_cfa_flow_alloc_input req = { 0 };
313 u16 flow_flags = 0, action_flags = 0;
316 bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_FLOW_ALLOC, -1, -1);
318 req.src_fid = cpu_to_le16(flow->src_fid);
319 req.ref_flow_handle = ref_flow_handle;
320 req.ethertype = flow->l2_key.ether_type;
321 req.ip_proto = flow->l4_key.ip_proto;
323 if (flow->flags & BNXT_TC_FLOW_FLAGS_ETH_ADDRS) {
324 memcpy(req.dmac, flow->l2_key.dmac, ETH_ALEN);
325 memcpy(req.smac, flow->l2_key.smac, ETH_ALEN);
328 if (flow->l2_key.num_vlans > 0) {
329 flow_flags |= CFA_FLOW_ALLOC_REQ_FLAGS_NUM_VLAN_ONE;
330 /* FW expects the inner_vlan_tci value to be set
331 * in outer_vlan_tci when num_vlans is 1 (which is
332 * always the case in TC.)
334 req.outer_vlan_tci = flow->l2_key.inner_vlan_tci;
337 /* If all IP and L4 fields are wildcarded then this is an L2 flow */
338 if (is_wildcard(l3_mask, sizeof(*l3_mask)) &&
339 is_wildcard(&flow->l4_mask, sizeof(flow->l4_mask))) {
340 flow_flags |= CFA_FLOW_ALLOC_REQ_FLAGS_FLOWTYPE_L2;
342 flow_flags |= flow->l2_key.ether_type == htons(ETH_P_IP) ?
343 CFA_FLOW_ALLOC_REQ_FLAGS_FLOWTYPE_IPV4 :
344 CFA_FLOW_ALLOC_REQ_FLAGS_FLOWTYPE_IPV6;
346 if (flow->flags & BNXT_TC_FLOW_FLAGS_IPV4_ADDRS) {
347 req.ip_dst[0] = l3_key->ipv4.daddr.s_addr;
348 req.ip_dst_mask_len =
349 inet_mask_len(l3_mask->ipv4.daddr.s_addr);
350 req.ip_src[0] = l3_key->ipv4.saddr.s_addr;
351 req.ip_src_mask_len =
352 inet_mask_len(l3_mask->ipv4.saddr.s_addr);
353 } else if (flow->flags & BNXT_TC_FLOW_FLAGS_IPV6_ADDRS) {
354 memcpy(req.ip_dst, l3_key->ipv6.daddr.s6_addr32,
356 req.ip_dst_mask_len =
357 ipv6_mask_len(&l3_mask->ipv6.daddr);
358 memcpy(req.ip_src, l3_key->ipv6.saddr.s6_addr32,
360 req.ip_src_mask_len =
361 ipv6_mask_len(&l3_mask->ipv6.saddr);
365 if (flow->flags & BNXT_TC_FLOW_FLAGS_PORTS) {
366 req.l4_src_port = flow->l4_key.ports.sport;
367 req.l4_src_port_mask = flow->l4_mask.ports.sport;
368 req.l4_dst_port = flow->l4_key.ports.dport;
369 req.l4_dst_port_mask = flow->l4_mask.ports.dport;
370 } else if (flow->flags & BNXT_TC_FLOW_FLAGS_ICMP) {
371 /* l4 ports serve as type/code when ip_proto is ICMP */
372 req.l4_src_port = htons(flow->l4_key.icmp.type);
373 req.l4_src_port_mask = htons(flow->l4_mask.icmp.type);
374 req.l4_dst_port = htons(flow->l4_key.icmp.code);
375 req.l4_dst_port_mask = htons(flow->l4_mask.icmp.code);
377 req.flags = cpu_to_le16(flow_flags);
379 if (actions->flags & BNXT_TC_ACTION_FLAG_DROP) {
380 action_flags |= CFA_FLOW_ALLOC_REQ_ACTION_FLAGS_DROP;
382 if (actions->flags & BNXT_TC_ACTION_FLAG_FWD) {
383 action_flags |= CFA_FLOW_ALLOC_REQ_ACTION_FLAGS_FWD;
384 req.dst_fid = cpu_to_le16(actions->dst_fid);
386 if (actions->flags & BNXT_TC_ACTION_FLAG_PUSH_VLAN) {
388 CFA_FLOW_ALLOC_REQ_ACTION_FLAGS_L2_HEADER_REWRITE;
389 req.l2_rewrite_vlan_tpid = actions->push_vlan_tpid;
390 req.l2_rewrite_vlan_tci = actions->push_vlan_tci;
391 memcpy(&req.l2_rewrite_dmac, &req.dmac, ETH_ALEN);
392 memcpy(&req.l2_rewrite_smac, &req.smac, ETH_ALEN);
394 if (actions->flags & BNXT_TC_ACTION_FLAG_POP_VLAN) {
396 CFA_FLOW_ALLOC_REQ_ACTION_FLAGS_L2_HEADER_REWRITE;
397 /* Rewrite config with tpid = 0 implies vlan pop */
398 req.l2_rewrite_vlan_tpid = 0;
399 memcpy(&req.l2_rewrite_dmac, &req.dmac, ETH_ALEN);
400 memcpy(&req.l2_rewrite_smac, &req.smac, ETH_ALEN);
403 req.action_flags = cpu_to_le16(action_flags);
405 mutex_lock(&bp->hwrm_cmd_lock);
407 rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
409 *flow_handle = resp->flow_handle;
411 mutex_unlock(&bp->hwrm_cmd_lock);
416 /* Add val to accum while handling a possible wraparound
417 * of val. Eventhough val is of type u64, its actual width
418 * is denoted by mask and will wrap-around beyond that width.
420 static void accumulate_val(u64 *accum, u64 val, u64 mask)
422 #define low_bits(x, mask) ((x) & (mask))
423 #define high_bits(x, mask) ((x) & ~(mask))
424 bool wrapped = val < low_bits(*accum, mask);
426 *accum = high_bits(*accum, mask) + val;
428 *accum += (mask + 1);
431 /* The HW counters' width is much less than 64bits.
432 * Handle possible wrap-around while updating the stat counters
434 static void bnxt_flow_stats_fix_wraparound(struct bnxt_tc_info *tc_info,
435 struct bnxt_tc_flow_stats *stats,
436 struct bnxt_tc_flow_stats *hw_stats)
438 accumulate_val(&stats->bytes, hw_stats->bytes, tc_info->bytes_mask);
439 accumulate_val(&stats->packets, hw_stats->packets,
440 tc_info->packets_mask);
443 /* Fix possible wraparound of the stats queried from HW, calculate
444 * the delta from prev_stats, and also update the prev_stats.
445 * The HW flow stats are fetched under the hwrm_cmd_lock mutex.
446 * This routine is best called while under the mutex so that the
447 * stats processing happens atomically.
449 static void bnxt_flow_stats_calc(struct bnxt_tc_info *tc_info,
450 struct bnxt_tc_flow *flow,
451 struct bnxt_tc_flow_stats *stats)
453 struct bnxt_tc_flow_stats *acc_stats, *prev_stats;
455 acc_stats = &flow->stats;
456 bnxt_flow_stats_fix_wraparound(tc_info, acc_stats, stats);
458 prev_stats = &flow->prev_stats;
459 stats->bytes = acc_stats->bytes - prev_stats->bytes;
460 stats->packets = acc_stats->packets - prev_stats->packets;
461 *prev_stats = *acc_stats;
464 static int bnxt_hwrm_cfa_flow_stats_get(struct bnxt *bp,
466 struct bnxt_tc_flow *flow,
467 struct bnxt_tc_flow_stats *stats)
469 struct hwrm_cfa_flow_stats_output *resp = bp->hwrm_cmd_resp_addr;
470 struct hwrm_cfa_flow_stats_input req = { 0 };
473 bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_FLOW_STATS, -1, -1);
474 req.num_flows = cpu_to_le16(1);
475 req.flow_handle_0 = flow_handle;
477 mutex_lock(&bp->hwrm_cmd_lock);
478 rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
480 stats->packets = le64_to_cpu(resp->packet_0);
481 stats->bytes = le64_to_cpu(resp->byte_0);
482 bnxt_flow_stats_calc(&bp->tc_info, flow, stats);
484 netdev_info(bp->dev, "error rc=%d", rc);
487 mutex_unlock(&bp->hwrm_cmd_lock);
491 static int bnxt_tc_put_l2_node(struct bnxt *bp,
492 struct bnxt_tc_flow_node *flow_node)
494 struct bnxt_tc_l2_node *l2_node = flow_node->l2_node;
495 struct bnxt_tc_info *tc_info = &bp->tc_info;
498 /* remove flow_node from the L2 shared flow list */
499 list_del(&flow_node->l2_list_node);
500 if (--l2_node->refcount == 0) {
501 rc = rhashtable_remove_fast(&tc_info->l2_table, &l2_node->node,
502 tc_info->l2_ht_params);
505 "Error: %s: rhashtable_remove_fast: %d",
507 kfree_rcu(l2_node, rcu);
512 static struct bnxt_tc_l2_node *
513 bnxt_tc_get_l2_node(struct bnxt *bp, struct rhashtable *l2_table,
514 struct rhashtable_params ht_params,
515 struct bnxt_tc_l2_key *l2_key)
517 struct bnxt_tc_l2_node *l2_node;
520 l2_node = rhashtable_lookup_fast(l2_table, l2_key, ht_params);
522 l2_node = kzalloc(sizeof(*l2_node), GFP_KERNEL);
528 l2_node->key = *l2_key;
529 rc = rhashtable_insert_fast(l2_table, &l2_node->node,
534 "Error: %s: rhashtable_insert_fast: %d",
538 INIT_LIST_HEAD(&l2_node->common_l2_flows);
543 /* Get the ref_flow_handle for a flow by checking if there are any other
544 * flows that share the same L2 key as this flow.
547 bnxt_tc_get_ref_flow_handle(struct bnxt *bp, struct bnxt_tc_flow *flow,
548 struct bnxt_tc_flow_node *flow_node,
549 __le16 *ref_flow_handle)
551 struct bnxt_tc_info *tc_info = &bp->tc_info;
552 struct bnxt_tc_flow_node *ref_flow_node;
553 struct bnxt_tc_l2_node *l2_node;
555 l2_node = bnxt_tc_get_l2_node(bp, &tc_info->l2_table,
556 tc_info->l2_ht_params,
561 /* If any other flow is using this l2_node, use it's flow_handle
562 * as the ref_flow_handle
564 if (l2_node->refcount > 0) {
565 ref_flow_node = list_first_entry(&l2_node->common_l2_flows,
566 struct bnxt_tc_flow_node,
568 *ref_flow_handle = ref_flow_node->flow_handle;
570 *ref_flow_handle = cpu_to_le16(0xffff);
573 /* Insert the l2_node into the flow_node so that subsequent flows
574 * with a matching l2 key can use the flow_handle of this flow
575 * as their ref_flow_handle
577 flow_node->l2_node = l2_node;
578 list_add(&flow_node->l2_list_node, &l2_node->common_l2_flows);
583 /* After the flow parsing is done, this routine is used for checking
584 * if there are any aspects of the flow that prevent it from being
587 static bool bnxt_tc_can_offload(struct bnxt *bp, struct bnxt_tc_flow *flow)
589 /* If L4 ports are specified then ip_proto must be TCP or UDP */
590 if ((flow->flags & BNXT_TC_FLOW_FLAGS_PORTS) &&
591 (flow->l4_key.ip_proto != IPPROTO_TCP &&
592 flow->l4_key.ip_proto != IPPROTO_UDP)) {
593 netdev_info(bp->dev, "Cannot offload non-TCP/UDP (%d) ports",
594 flow->l4_key.ip_proto);
601 static int __bnxt_tc_del_flow(struct bnxt *bp,
602 struct bnxt_tc_flow_node *flow_node)
604 struct bnxt_tc_info *tc_info = &bp->tc_info;
607 /* send HWRM cmd to free the flow-id */
608 bnxt_hwrm_cfa_flow_free(bp, flow_node->flow_handle);
610 mutex_lock(&tc_info->lock);
612 /* release reference to l2 node */
613 bnxt_tc_put_l2_node(bp, flow_node);
615 mutex_unlock(&tc_info->lock);
617 rc = rhashtable_remove_fast(&tc_info->flow_table, &flow_node->node,
618 tc_info->flow_ht_params);
620 netdev_err(bp->dev, "Error: %s: rhashtable_remove_fast rc=%d",
623 kfree_rcu(flow_node, rcu);
627 /* Add a new flow or replace an existing flow.
629 * There are essentially two critical sections here.
630 * 1. while adding a new flow
632 * b) issue HWRM cmd and get flow_handle
633 * c) link l2-key with flow
634 * 2. while deleting a flow
635 * a) unlinking l2-key from flow
636 * A lock is needed to protect these two critical sections.
638 * The hash-tables are already protected by the rhashtable API.
640 static int bnxt_tc_add_flow(struct bnxt *bp, u16 src_fid,
641 struct tc_cls_flower_offload *tc_flow_cmd)
643 struct bnxt_tc_flow_node *new_node, *old_node;
644 struct bnxt_tc_info *tc_info = &bp->tc_info;
645 struct bnxt_tc_flow *flow;
646 __le16 ref_flow_handle;
649 /* allocate memory for the new flow and it's node */
650 new_node = kzalloc(sizeof(*new_node), GFP_KERNEL);
655 new_node->cookie = tc_flow_cmd->cookie;
656 flow = &new_node->flow;
658 rc = bnxt_tc_parse_flow(bp, tc_flow_cmd, flow);
661 flow->src_fid = src_fid;
663 if (!bnxt_tc_can_offload(bp, flow)) {
668 /* If a flow exists with the same cookie, delete it */
669 old_node = rhashtable_lookup_fast(&tc_info->flow_table,
670 &tc_flow_cmd->cookie,
671 tc_info->flow_ht_params);
673 __bnxt_tc_del_flow(bp, old_node);
675 /* Check if the L2 part of the flow has been offloaded already.
676 * If so, bump up it's refcnt and get it's reference handle.
678 mutex_lock(&tc_info->lock);
679 rc = bnxt_tc_get_ref_flow_handle(bp, flow, new_node, &ref_flow_handle);
683 /* send HWRM cmd to alloc the flow */
684 rc = bnxt_hwrm_cfa_flow_alloc(bp, flow, ref_flow_handle,
685 &new_node->flow_handle);
689 /* add new flow to flow-table */
690 rc = rhashtable_insert_fast(&tc_info->flow_table, &new_node->node,
691 tc_info->flow_ht_params);
695 mutex_unlock(&tc_info->lock);
699 bnxt_hwrm_cfa_flow_free(bp, new_node->flow_handle);
701 bnxt_tc_put_l2_node(bp, new_node);
703 mutex_unlock(&tc_info->lock);
707 netdev_err(bp->dev, "Error: %s: cookie=0x%lx error=%d",
708 __func__, tc_flow_cmd->cookie, rc);
712 static int bnxt_tc_del_flow(struct bnxt *bp,
713 struct tc_cls_flower_offload *tc_flow_cmd)
715 struct bnxt_tc_info *tc_info = &bp->tc_info;
716 struct bnxt_tc_flow_node *flow_node;
718 flow_node = rhashtable_lookup_fast(&tc_info->flow_table,
719 &tc_flow_cmd->cookie,
720 tc_info->flow_ht_params);
722 netdev_info(bp->dev, "ERROR: no flow_node for cookie %lx",
723 tc_flow_cmd->cookie);
727 return __bnxt_tc_del_flow(bp, flow_node);
730 static int bnxt_tc_get_flow_stats(struct bnxt *bp,
731 struct tc_cls_flower_offload *tc_flow_cmd)
733 struct bnxt_tc_info *tc_info = &bp->tc_info;
734 struct bnxt_tc_flow_node *flow_node;
735 struct bnxt_tc_flow_stats stats;
738 flow_node = rhashtable_lookup_fast(&tc_info->flow_table,
739 &tc_flow_cmd->cookie,
740 tc_info->flow_ht_params);
742 netdev_info(bp->dev, "Error: no flow_node for cookie %lx",
743 tc_flow_cmd->cookie);
747 rc = bnxt_hwrm_cfa_flow_stats_get(bp, flow_node->flow_handle,
748 &flow_node->flow, &stats);
752 tcf_exts_stats_update(tc_flow_cmd->exts, stats.bytes, stats.packets, 0);
756 int bnxt_tc_setup_flower(struct bnxt *bp, u16 src_fid,
757 struct tc_cls_flower_offload *cls_flower)
761 if (!is_classid_clsact_ingress(cls_flower->common.classid) ||
762 cls_flower->common.chain_index)
765 switch (cls_flower->command) {
766 case TC_CLSFLOWER_REPLACE:
767 rc = bnxt_tc_add_flow(bp, src_fid, cls_flower);
770 case TC_CLSFLOWER_DESTROY:
771 rc = bnxt_tc_del_flow(bp, cls_flower);
774 case TC_CLSFLOWER_STATS:
775 rc = bnxt_tc_get_flow_stats(bp, cls_flower);
781 static const struct rhashtable_params bnxt_tc_flow_ht_params = {
782 .head_offset = offsetof(struct bnxt_tc_flow_node, node),
783 .key_offset = offsetof(struct bnxt_tc_flow_node, cookie),
784 .key_len = sizeof(((struct bnxt_tc_flow_node *)0)->cookie),
785 .automatic_shrinking = true
788 static const struct rhashtable_params bnxt_tc_l2_ht_params = {
789 .head_offset = offsetof(struct bnxt_tc_l2_node, node),
790 .key_offset = offsetof(struct bnxt_tc_l2_node, key),
791 .key_len = BNXT_TC_L2_KEY_LEN,
792 .automatic_shrinking = true
795 /* convert counter width in bits to a mask */
796 #define mask(width) ((u64)~0 >> (64 - (width)))
798 int bnxt_init_tc(struct bnxt *bp)
800 struct bnxt_tc_info *tc_info = &bp->tc_info;
803 if (bp->hwrm_spec_code < 0x10800) {
805 "Firmware does not support TC flower offload.\n");
808 mutex_init(&tc_info->lock);
810 /* Counter widths are programmed by FW */
811 tc_info->bytes_mask = mask(36);
812 tc_info->packets_mask = mask(28);
814 tc_info->flow_ht_params = bnxt_tc_flow_ht_params;
815 rc = rhashtable_init(&tc_info->flow_table, &tc_info->flow_ht_params);
819 tc_info->l2_ht_params = bnxt_tc_l2_ht_params;
820 rc = rhashtable_init(&tc_info->l2_table, &tc_info->l2_ht_params);
822 goto destroy_flow_table;
824 tc_info->enabled = true;
825 bp->dev->hw_features |= NETIF_F_HW_TC;
826 bp->dev->features |= NETIF_F_HW_TC;
830 rhashtable_destroy(&tc_info->flow_table);
834 void bnxt_shutdown_tc(struct bnxt *bp)
836 struct bnxt_tc_info *tc_info = &bp->tc_info;
838 if (!tc_info->enabled)
841 rhashtable_destroy(&tc_info->flow_table);
842 rhashtable_destroy(&tc_info->l2_table);