2 * Copyright (c) 2016, Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <generated/utsrelease.h>
34 #include <linux/mlx5/fs.h>
35 #include <net/switchdev.h>
36 #include <net/pkt_cls.h>
37 #include <net/act_api.h>
38 #include <net/netevent.h>
47 #define MLX5E_REP_PARAMS_LOG_SQ_SIZE \
48 max(0x6, MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE)
49 #define MLX5E_REP_PARAMS_LOG_RQ_SIZE \
50 max(0x6, MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE)
52 static const char mlx5e_rep_driver_name[] = "mlx5e_rep";
54 static void mlx5e_rep_get_drvinfo(struct net_device *dev,
55 struct ethtool_drvinfo *drvinfo)
57 strlcpy(drvinfo->driver, mlx5e_rep_driver_name,
58 sizeof(drvinfo->driver));
59 strlcpy(drvinfo->version, UTS_RELEASE, sizeof(drvinfo->version));
62 static const struct counter_desc sw_rep_stats_desc[] = {
63 { MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, rx_packets) },
64 { MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, rx_bytes) },
65 { MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_packets) },
66 { MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_bytes) },
76 static const struct counter_desc vport_rep_stats_desc[] = {
77 { MLX5E_DECLARE_STAT(struct vport_stats, vport_rx_packets) },
78 { MLX5E_DECLARE_STAT(struct vport_stats, vport_rx_bytes) },
79 { MLX5E_DECLARE_STAT(struct vport_stats, vport_tx_packets) },
80 { MLX5E_DECLARE_STAT(struct vport_stats, vport_tx_bytes) },
83 #define NUM_VPORT_REP_SW_COUNTERS ARRAY_SIZE(sw_rep_stats_desc)
84 #define NUM_VPORT_REP_HW_COUNTERS ARRAY_SIZE(vport_rep_stats_desc)
86 static void mlx5e_rep_get_strings(struct net_device *dev,
87 u32 stringset, uint8_t *data)
93 for (i = 0; i < NUM_VPORT_REP_SW_COUNTERS; i++)
94 strcpy(data + (i * ETH_GSTRING_LEN),
95 sw_rep_stats_desc[i].format);
96 for (j = 0; j < NUM_VPORT_REP_HW_COUNTERS; j++, i++)
97 strcpy(data + (i * ETH_GSTRING_LEN),
98 vport_rep_stats_desc[j].format);
103 static void mlx5e_rep_update_hw_counters(struct mlx5e_priv *priv)
105 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
106 struct mlx5e_rep_priv *rpriv = priv->ppriv;
107 struct mlx5_eswitch_rep *rep = rpriv->rep;
108 struct rtnl_link_stats64 *vport_stats;
109 struct ifla_vf_stats vf_stats;
112 err = mlx5_eswitch_get_vport_stats(esw, rep->vport, &vf_stats);
114 pr_warn("vport %d error %d reading stats\n", rep->vport, err);
118 vport_stats = &priv->stats.vf_vport;
119 /* flip tx/rx as we are reporting the counters for the switch vport */
120 vport_stats->rx_packets = vf_stats.tx_packets;
121 vport_stats->rx_bytes = vf_stats.tx_bytes;
122 vport_stats->tx_packets = vf_stats.rx_packets;
123 vport_stats->tx_bytes = vf_stats.rx_bytes;
126 static void mlx5e_rep_update_sw_counters(struct mlx5e_priv *priv)
128 struct mlx5e_sw_stats *s = &priv->stats.sw;
129 struct mlx5e_rq_stats *rq_stats;
130 struct mlx5e_sq_stats *sq_stats;
133 memset(s, 0, sizeof(*s));
134 for (i = 0; i < priv->channels.num; i++) {
135 struct mlx5e_channel *c = priv->channels.c[i];
137 rq_stats = c->rq.stats;
139 s->rx_packets += rq_stats->packets;
140 s->rx_bytes += rq_stats->bytes;
142 for (j = 0; j < priv->channels.params.num_tc; j++) {
143 sq_stats = c->sq[j].stats;
145 s->tx_packets += sq_stats->packets;
146 s->tx_bytes += sq_stats->bytes;
151 static void mlx5e_rep_get_ethtool_stats(struct net_device *dev,
152 struct ethtool_stats *stats, u64 *data)
154 struct mlx5e_priv *priv = netdev_priv(dev);
160 mutex_lock(&priv->state_lock);
161 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
162 mlx5e_rep_update_sw_counters(priv);
163 mlx5e_rep_update_hw_counters(priv);
164 mutex_unlock(&priv->state_lock);
166 for (i = 0; i < NUM_VPORT_REP_SW_COUNTERS; i++)
167 data[i] = MLX5E_READ_CTR64_CPU(&priv->stats.sw,
168 sw_rep_stats_desc, i);
170 for (j = 0; j < NUM_VPORT_REP_HW_COUNTERS; j++, i++)
171 data[i] = MLX5E_READ_CTR64_CPU(&priv->stats.vf_vport,
172 vport_rep_stats_desc, j);
175 static int mlx5e_rep_get_sset_count(struct net_device *dev, int sset)
179 return NUM_VPORT_REP_SW_COUNTERS + NUM_VPORT_REP_HW_COUNTERS;
185 static const struct ethtool_ops mlx5e_rep_ethtool_ops = {
186 .get_drvinfo = mlx5e_rep_get_drvinfo,
187 .get_link = ethtool_op_get_link,
188 .get_strings = mlx5e_rep_get_strings,
189 .get_sset_count = mlx5e_rep_get_sset_count,
190 .get_ethtool_stats = mlx5e_rep_get_ethtool_stats,
193 int mlx5e_attr_get(struct net_device *dev, struct switchdev_attr *attr)
195 struct mlx5e_priv *priv = netdev_priv(dev);
196 struct mlx5e_rep_priv *rpriv = priv->ppriv;
197 struct mlx5_eswitch_rep *rep = rpriv->rep;
198 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
200 if (esw->mode == SRIOV_NONE)
204 case SWITCHDEV_ATTR_ID_PORT_PARENT_ID:
205 attr->u.ppid.id_len = ETH_ALEN;
206 ether_addr_copy(attr->u.ppid.id, rep->hw_id);
215 static void mlx5e_sqs2vport_stop(struct mlx5_eswitch *esw,
216 struct mlx5_eswitch_rep *rep)
218 struct mlx5e_rep_sq *rep_sq, *tmp;
219 struct mlx5e_rep_priv *rpriv;
221 if (esw->mode != SRIOV_OFFLOADS)
224 rpriv = mlx5e_rep_to_rep_priv(rep);
225 list_for_each_entry_safe(rep_sq, tmp, &rpriv->vport_sqs_list, list) {
226 mlx5_eswitch_del_send_to_vport_rule(rep_sq->send_to_vport_rule);
227 list_del(&rep_sq->list);
232 static int mlx5e_sqs2vport_start(struct mlx5_eswitch *esw,
233 struct mlx5_eswitch_rep *rep,
234 u32 *sqns_array, int sqns_num)
236 struct mlx5_flow_handle *flow_rule;
237 struct mlx5e_rep_priv *rpriv;
238 struct mlx5e_rep_sq *rep_sq;
242 if (esw->mode != SRIOV_OFFLOADS)
245 rpriv = mlx5e_rep_to_rep_priv(rep);
246 for (i = 0; i < sqns_num; i++) {
247 rep_sq = kzalloc(sizeof(*rep_sq), GFP_KERNEL);
253 /* Add re-inject rule to the PF/representor sqs */
254 flow_rule = mlx5_eswitch_add_send_to_vport_rule(esw,
257 if (IS_ERR(flow_rule)) {
258 err = PTR_ERR(flow_rule);
262 rep_sq->send_to_vport_rule = flow_rule;
263 list_add(&rep_sq->list, &rpriv->vport_sqs_list);
268 mlx5e_sqs2vport_stop(esw, rep);
272 int mlx5e_add_sqs_fwd_rules(struct mlx5e_priv *priv)
274 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
275 struct mlx5e_rep_priv *rpriv = priv->ppriv;
276 struct mlx5_eswitch_rep *rep = rpriv->rep;
277 struct mlx5e_channel *c;
278 int n, tc, num_sqs = 0;
282 sqs = kcalloc(priv->channels.num * priv->channels.params.num_tc, sizeof(*sqs), GFP_KERNEL);
286 for (n = 0; n < priv->channels.num; n++) {
287 c = priv->channels.c[n];
288 for (tc = 0; tc < c->num_tc; tc++)
289 sqs[num_sqs++] = c->sq[tc].sqn;
292 err = mlx5e_sqs2vport_start(esw, rep, sqs, num_sqs);
297 netdev_warn(priv->netdev, "Failed to add SQs FWD rules %d\n", err);
301 void mlx5e_remove_sqs_fwd_rules(struct mlx5e_priv *priv)
303 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
304 struct mlx5e_rep_priv *rpriv = priv->ppriv;
305 struct mlx5_eswitch_rep *rep = rpriv->rep;
307 mlx5e_sqs2vport_stop(esw, rep);
310 static void mlx5e_rep_neigh_update_init_interval(struct mlx5e_rep_priv *rpriv)
312 #if IS_ENABLED(CONFIG_IPV6)
313 unsigned long ipv6_interval = NEIGH_VAR(&nd_tbl.parms,
316 unsigned long ipv6_interval = ~0UL;
318 unsigned long ipv4_interval = NEIGH_VAR(&arp_tbl.parms,
320 struct net_device *netdev = rpriv->netdev;
321 struct mlx5e_priv *priv = netdev_priv(netdev);
323 rpriv->neigh_update.min_interval = min_t(unsigned long, ipv6_interval, ipv4_interval);
324 mlx5_fc_update_sampling_interval(priv->mdev, rpriv->neigh_update.min_interval);
327 void mlx5e_rep_queue_neigh_stats_work(struct mlx5e_priv *priv)
329 struct mlx5e_rep_priv *rpriv = priv->ppriv;
330 struct mlx5e_neigh_update_table *neigh_update = &rpriv->neigh_update;
332 mlx5_fc_queue_stats_work(priv->mdev,
333 &neigh_update->neigh_stats_work,
334 neigh_update->min_interval);
337 static void mlx5e_rep_neigh_stats_work(struct work_struct *work)
339 struct mlx5e_rep_priv *rpriv = container_of(work, struct mlx5e_rep_priv,
340 neigh_update.neigh_stats_work.work);
341 struct net_device *netdev = rpriv->netdev;
342 struct mlx5e_priv *priv = netdev_priv(netdev);
343 struct mlx5e_neigh_hash_entry *nhe;
346 if (!list_empty(&rpriv->neigh_update.neigh_list))
347 mlx5e_rep_queue_neigh_stats_work(priv);
349 list_for_each_entry(nhe, &rpriv->neigh_update.neigh_list, neigh_list)
350 mlx5e_tc_update_neigh_used_value(nhe);
355 static void mlx5e_rep_neigh_entry_hold(struct mlx5e_neigh_hash_entry *nhe)
357 refcount_inc(&nhe->refcnt);
360 static void mlx5e_rep_neigh_entry_release(struct mlx5e_neigh_hash_entry *nhe)
362 if (refcount_dec_and_test(&nhe->refcnt))
366 static void mlx5e_rep_update_flows(struct mlx5e_priv *priv,
367 struct mlx5e_encap_entry *e,
368 bool neigh_connected,
369 unsigned char ha[ETH_ALEN])
371 struct ethhdr *eth = (struct ethhdr *)e->encap_header;
375 if ((!neigh_connected && (e->flags & MLX5_ENCAP_ENTRY_VALID)) ||
376 !ether_addr_equal(e->h_dest, ha))
377 mlx5e_tc_encap_flows_del(priv, e);
379 if (neigh_connected && !(e->flags & MLX5_ENCAP_ENTRY_VALID)) {
380 ether_addr_copy(e->h_dest, ha);
381 ether_addr_copy(eth->h_dest, ha);
383 mlx5e_tc_encap_flows_add(priv, e);
387 static void mlx5e_rep_neigh_update(struct work_struct *work)
389 struct mlx5e_neigh_hash_entry *nhe =
390 container_of(work, struct mlx5e_neigh_hash_entry, neigh_update_work);
391 struct neighbour *n = nhe->n;
392 struct mlx5e_encap_entry *e;
393 unsigned char ha[ETH_ALEN];
394 struct mlx5e_priv *priv;
395 bool neigh_connected;
396 bool encap_connected;
401 /* If these parameters are changed after we release the lock,
402 * we'll receive another event letting us know about it.
403 * We use this lock to avoid inconsistency between the neigh validity
404 * and it's hw address.
406 read_lock_bh(&n->lock);
407 memcpy(ha, n->ha, ETH_ALEN);
408 nud_state = n->nud_state;
410 read_unlock_bh(&n->lock);
412 neigh_connected = (nud_state & NUD_VALID) && !dead;
414 list_for_each_entry(e, &nhe->encap_list, encap_list) {
415 encap_connected = !!(e->flags & MLX5_ENCAP_ENTRY_VALID);
416 priv = netdev_priv(e->out_dev);
418 if (encap_connected != neigh_connected ||
419 !ether_addr_equal(e->h_dest, ha))
420 mlx5e_rep_update_flows(priv, e, neigh_connected, ha);
422 mlx5e_rep_neigh_entry_release(nhe);
427 static struct mlx5e_neigh_hash_entry *
428 mlx5e_rep_neigh_entry_lookup(struct mlx5e_priv *priv,
429 struct mlx5e_neigh *m_neigh);
431 static int mlx5e_rep_netevent_event(struct notifier_block *nb,
432 unsigned long event, void *ptr)
434 struct mlx5e_rep_priv *rpriv = container_of(nb, struct mlx5e_rep_priv,
435 neigh_update.netevent_nb);
436 struct mlx5e_neigh_update_table *neigh_update = &rpriv->neigh_update;
437 struct net_device *netdev = rpriv->netdev;
438 struct mlx5e_priv *priv = netdev_priv(netdev);
439 struct mlx5e_neigh_hash_entry *nhe = NULL;
440 struct mlx5e_neigh m_neigh = {};
441 struct neigh_parms *p;
446 case NETEVENT_NEIGH_UPDATE:
448 #if IS_ENABLED(CONFIG_IPV6)
449 if (n->tbl != &nd_tbl && n->tbl != &arp_tbl)
451 if (n->tbl != &arp_tbl)
455 m_neigh.dev = n->dev;
456 m_neigh.family = n->ops->family;
457 memcpy(&m_neigh.dst_ip, n->primary_key, n->tbl->key_len);
459 /* We are in atomic context and can't take RTNL mutex, so use
460 * spin_lock_bh to lookup the neigh table. bh is used since
461 * netevent can be called from a softirq context.
463 spin_lock_bh(&neigh_update->encap_lock);
464 nhe = mlx5e_rep_neigh_entry_lookup(priv, &m_neigh);
466 spin_unlock_bh(&neigh_update->encap_lock);
470 /* This assignment is valid as long as the the neigh reference
475 /* Take a reference to ensure the neighbour and mlx5 encap
476 * entry won't be destructed until we drop the reference in
480 mlx5e_rep_neigh_entry_hold(nhe);
482 if (!queue_work(priv->wq, &nhe->neigh_update_work)) {
483 mlx5e_rep_neigh_entry_release(nhe);
486 spin_unlock_bh(&neigh_update->encap_lock);
489 case NETEVENT_DELAY_PROBE_TIME_UPDATE:
492 /* We check the device is present since we don't care about
493 * changes in the default table, we only care about changes
494 * done per device delay prob time parameter.
496 #if IS_ENABLED(CONFIG_IPV6)
497 if (!p->dev || (p->tbl != &nd_tbl && p->tbl != &arp_tbl))
499 if (!p->dev || p->tbl != &arp_tbl)
503 /* We are in atomic context and can't take RTNL mutex,
504 * so use spin_lock_bh to walk the neigh list and look for
505 * the relevant device. bh is used since netevent can be
506 * called from a softirq context.
508 spin_lock_bh(&neigh_update->encap_lock);
509 list_for_each_entry(nhe, &neigh_update->neigh_list, neigh_list) {
510 if (p->dev == nhe->m_neigh.dev) {
515 spin_unlock_bh(&neigh_update->encap_lock);
519 neigh_update->min_interval = min_t(unsigned long,
520 NEIGH_VAR(p, DELAY_PROBE_TIME),
521 neigh_update->min_interval);
522 mlx5_fc_update_sampling_interval(priv->mdev,
523 neigh_update->min_interval);
529 static const struct rhashtable_params mlx5e_neigh_ht_params = {
530 .head_offset = offsetof(struct mlx5e_neigh_hash_entry, rhash_node),
531 .key_offset = offsetof(struct mlx5e_neigh_hash_entry, m_neigh),
532 .key_len = sizeof(struct mlx5e_neigh),
533 .automatic_shrinking = true,
536 static int mlx5e_rep_neigh_init(struct mlx5e_rep_priv *rpriv)
538 struct mlx5e_neigh_update_table *neigh_update = &rpriv->neigh_update;
541 err = rhashtable_init(&neigh_update->neigh_ht, &mlx5e_neigh_ht_params);
545 INIT_LIST_HEAD(&neigh_update->neigh_list);
546 spin_lock_init(&neigh_update->encap_lock);
547 INIT_DELAYED_WORK(&neigh_update->neigh_stats_work,
548 mlx5e_rep_neigh_stats_work);
549 mlx5e_rep_neigh_update_init_interval(rpriv);
551 rpriv->neigh_update.netevent_nb.notifier_call = mlx5e_rep_netevent_event;
552 err = register_netevent_notifier(&rpriv->neigh_update.netevent_nb);
558 rhashtable_destroy(&neigh_update->neigh_ht);
562 static void mlx5e_rep_neigh_cleanup(struct mlx5e_rep_priv *rpriv)
564 struct mlx5e_neigh_update_table *neigh_update = &rpriv->neigh_update;
565 struct mlx5e_priv *priv = netdev_priv(rpriv->netdev);
567 unregister_netevent_notifier(&neigh_update->netevent_nb);
569 flush_workqueue(priv->wq); /* flush neigh update works */
571 cancel_delayed_work_sync(&rpriv->neigh_update.neigh_stats_work);
573 rhashtable_destroy(&neigh_update->neigh_ht);
576 static int mlx5e_rep_neigh_entry_insert(struct mlx5e_priv *priv,
577 struct mlx5e_neigh_hash_entry *nhe)
579 struct mlx5e_rep_priv *rpriv = priv->ppriv;
582 err = rhashtable_insert_fast(&rpriv->neigh_update.neigh_ht,
584 mlx5e_neigh_ht_params);
588 list_add(&nhe->neigh_list, &rpriv->neigh_update.neigh_list);
593 static void mlx5e_rep_neigh_entry_remove(struct mlx5e_priv *priv,
594 struct mlx5e_neigh_hash_entry *nhe)
596 struct mlx5e_rep_priv *rpriv = priv->ppriv;
598 spin_lock_bh(&rpriv->neigh_update.encap_lock);
600 list_del(&nhe->neigh_list);
602 rhashtable_remove_fast(&rpriv->neigh_update.neigh_ht,
604 mlx5e_neigh_ht_params);
605 spin_unlock_bh(&rpriv->neigh_update.encap_lock);
608 /* This function must only be called under RTNL lock or under the
609 * representor's encap_lock in case RTNL mutex can't be held.
611 static struct mlx5e_neigh_hash_entry *
612 mlx5e_rep_neigh_entry_lookup(struct mlx5e_priv *priv,
613 struct mlx5e_neigh *m_neigh)
615 struct mlx5e_rep_priv *rpriv = priv->ppriv;
616 struct mlx5e_neigh_update_table *neigh_update = &rpriv->neigh_update;
618 return rhashtable_lookup_fast(&neigh_update->neigh_ht, m_neigh,
619 mlx5e_neigh_ht_params);
622 static int mlx5e_rep_neigh_entry_create(struct mlx5e_priv *priv,
623 struct mlx5e_encap_entry *e,
624 struct mlx5e_neigh_hash_entry **nhe)
628 *nhe = kzalloc(sizeof(**nhe), GFP_KERNEL);
632 memcpy(&(*nhe)->m_neigh, &e->m_neigh, sizeof(e->m_neigh));
633 INIT_WORK(&(*nhe)->neigh_update_work, mlx5e_rep_neigh_update);
634 INIT_LIST_HEAD(&(*nhe)->encap_list);
635 refcount_set(&(*nhe)->refcnt, 1);
637 err = mlx5e_rep_neigh_entry_insert(priv, *nhe);
647 static void mlx5e_rep_neigh_entry_destroy(struct mlx5e_priv *priv,
648 struct mlx5e_neigh_hash_entry *nhe)
650 /* The neigh hash entry must be removed from the hash table regardless
651 * of the reference count value, so it won't be found by the next
652 * neigh notification call. The neigh hash entry reference count is
653 * incremented only during creation and neigh notification calls and
654 * protects from freeing the nhe struct.
656 mlx5e_rep_neigh_entry_remove(priv, nhe);
657 mlx5e_rep_neigh_entry_release(nhe);
660 int mlx5e_rep_encap_entry_attach(struct mlx5e_priv *priv,
661 struct mlx5e_encap_entry *e)
663 struct mlx5e_neigh_hash_entry *nhe;
666 nhe = mlx5e_rep_neigh_entry_lookup(priv, &e->m_neigh);
668 err = mlx5e_rep_neigh_entry_create(priv, e, &nhe);
672 list_add(&e->encap_list, &nhe->encap_list);
676 void mlx5e_rep_encap_entry_detach(struct mlx5e_priv *priv,
677 struct mlx5e_encap_entry *e)
679 struct mlx5e_neigh_hash_entry *nhe;
681 list_del(&e->encap_list);
682 nhe = mlx5e_rep_neigh_entry_lookup(priv, &e->m_neigh);
684 if (list_empty(&nhe->encap_list))
685 mlx5e_rep_neigh_entry_destroy(priv, nhe);
688 static int mlx5e_rep_open(struct net_device *dev)
690 struct mlx5e_priv *priv = netdev_priv(dev);
691 struct mlx5e_rep_priv *rpriv = priv->ppriv;
692 struct mlx5_eswitch_rep *rep = rpriv->rep;
695 mutex_lock(&priv->state_lock);
696 err = mlx5e_open_locked(dev);
700 if (!mlx5_modify_vport_admin_state(priv->mdev,
701 MLX5_QUERY_VPORT_STATE_IN_OP_MOD_ESW_VPORT,
702 rep->vport, MLX5_ESW_VPORT_ADMIN_STATE_UP))
703 netif_carrier_on(dev);
706 mutex_unlock(&priv->state_lock);
710 static int mlx5e_rep_close(struct net_device *dev)
712 struct mlx5e_priv *priv = netdev_priv(dev);
713 struct mlx5e_rep_priv *rpriv = priv->ppriv;
714 struct mlx5_eswitch_rep *rep = rpriv->rep;
717 mutex_lock(&priv->state_lock);
718 mlx5_modify_vport_admin_state(priv->mdev,
719 MLX5_QUERY_VPORT_STATE_IN_OP_MOD_ESW_VPORT,
720 rep->vport, MLX5_ESW_VPORT_ADMIN_STATE_DOWN);
721 ret = mlx5e_close_locked(dev);
722 mutex_unlock(&priv->state_lock);
726 static int mlx5e_rep_get_phys_port_name(struct net_device *dev,
727 char *buf, size_t len)
729 struct mlx5e_priv *priv = netdev_priv(dev);
730 struct mlx5e_rep_priv *rpriv = priv->ppriv;
731 struct mlx5_eswitch_rep *rep = rpriv->rep;
734 ret = snprintf(buf, len, "%d", rep->vport - 1);
742 mlx5e_rep_setup_tc_cls_flower(struct mlx5e_priv *priv,
743 struct tc_cls_flower_offload *cls_flower, int flags)
745 switch (cls_flower->command) {
746 case TC_CLSFLOWER_REPLACE:
747 return mlx5e_configure_flower(priv, cls_flower, flags);
748 case TC_CLSFLOWER_DESTROY:
749 return mlx5e_delete_flower(priv, cls_flower, flags);
750 case TC_CLSFLOWER_STATS:
751 return mlx5e_stats_flower(priv, cls_flower, flags);
757 static int mlx5e_rep_setup_tc_cb_egdev(enum tc_setup_type type, void *type_data,
760 struct mlx5e_priv *priv = cb_priv;
762 if (!tc_cls_can_offload_and_chain0(priv->netdev, type_data))
766 case TC_SETUP_CLSFLOWER:
767 return mlx5e_rep_setup_tc_cls_flower(priv, type_data, MLX5E_TC_EGRESS);
773 static int mlx5e_rep_setup_tc_cb(enum tc_setup_type type, void *type_data,
776 struct mlx5e_priv *priv = cb_priv;
778 if (!tc_cls_can_offload_and_chain0(priv->netdev, type_data))
782 case TC_SETUP_CLSFLOWER:
783 return mlx5e_rep_setup_tc_cls_flower(priv, type_data, MLX5E_TC_INGRESS);
789 static int mlx5e_rep_setup_tc_block(struct net_device *dev,
790 struct tc_block_offload *f)
792 struct mlx5e_priv *priv = netdev_priv(dev);
794 if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
797 switch (f->command) {
799 return tcf_block_cb_register(f->block, mlx5e_rep_setup_tc_cb,
800 priv, priv, f->extack);
801 case TC_BLOCK_UNBIND:
802 tcf_block_cb_unregister(f->block, mlx5e_rep_setup_tc_cb, priv);
809 static int mlx5e_rep_setup_tc(struct net_device *dev, enum tc_setup_type type,
814 return mlx5e_rep_setup_tc_block(dev, type_data);
820 bool mlx5e_is_uplink_rep(struct mlx5e_priv *priv)
822 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
823 struct mlx5e_rep_priv *rpriv = priv->ppriv;
824 struct mlx5_eswitch_rep *rep;
826 if (!MLX5_CAP_GEN(priv->mdev, vport_group_manager))
830 if (esw->mode == SRIOV_OFFLOADS &&
831 rep && rep->vport == FDB_UPLINK_VPORT)
837 static bool mlx5e_is_vf_vport_rep(struct mlx5e_priv *priv)
839 struct mlx5e_rep_priv *rpriv = priv->ppriv;
840 struct mlx5_eswitch_rep *rep = rpriv->rep;
842 if (rep && rep->vport != FDB_UPLINK_VPORT)
848 bool mlx5e_has_offload_stats(const struct net_device *dev, int attr_id)
850 struct mlx5e_priv *priv = netdev_priv(dev);
853 case IFLA_OFFLOAD_XSTATS_CPU_HIT:
854 if (mlx5e_is_vf_vport_rep(priv) || mlx5e_is_uplink_rep(priv))
862 mlx5e_get_sw_stats64(const struct net_device *dev,
863 struct rtnl_link_stats64 *stats)
865 struct mlx5e_priv *priv = netdev_priv(dev);
866 struct mlx5e_sw_stats *sstats = &priv->stats.sw;
868 mlx5e_rep_update_sw_counters(priv);
870 stats->rx_packets = sstats->rx_packets;
871 stats->rx_bytes = sstats->rx_bytes;
872 stats->tx_packets = sstats->tx_packets;
873 stats->tx_bytes = sstats->tx_bytes;
875 stats->tx_dropped = sstats->tx_queue_dropped;
880 int mlx5e_get_offload_stats(int attr_id, const struct net_device *dev,
884 case IFLA_OFFLOAD_XSTATS_CPU_HIT:
885 return mlx5e_get_sw_stats64(dev, sp);
892 mlx5e_rep_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats)
894 struct mlx5e_priv *priv = netdev_priv(dev);
896 memcpy(stats, &priv->stats.vf_vport, sizeof(*stats));
899 static const struct switchdev_ops mlx5e_rep_switchdev_ops = {
900 .switchdev_port_attr_get = mlx5e_attr_get,
903 static int mlx5e_change_rep_mtu(struct net_device *netdev, int new_mtu)
905 return mlx5e_change_mtu(netdev, new_mtu, NULL);
908 static const struct net_device_ops mlx5e_netdev_ops_rep = {
909 .ndo_open = mlx5e_rep_open,
910 .ndo_stop = mlx5e_rep_close,
911 .ndo_start_xmit = mlx5e_xmit,
912 .ndo_get_phys_port_name = mlx5e_rep_get_phys_port_name,
913 .ndo_setup_tc = mlx5e_rep_setup_tc,
914 .ndo_get_stats64 = mlx5e_rep_get_stats,
915 .ndo_has_offload_stats = mlx5e_has_offload_stats,
916 .ndo_get_offload_stats = mlx5e_get_offload_stats,
917 .ndo_change_mtu = mlx5e_change_rep_mtu,
920 static void mlx5e_build_rep_params(struct mlx5_core_dev *mdev,
921 struct mlx5e_params *params, u16 mtu)
923 u8 cq_period_mode = MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
924 MLX5_CQ_PERIOD_MODE_START_FROM_CQE :
925 MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
927 params->hard_mtu = MLX5E_ETH_HARD_MTU;
928 params->sw_mtu = mtu;
929 params->log_sq_size = MLX5E_REP_PARAMS_LOG_SQ_SIZE;
930 params->rq_wq_type = MLX5_WQ_TYPE_CYCLIC;
931 params->log_rq_mtu_frames = MLX5E_REP_PARAMS_LOG_RQ_SIZE;
933 params->rx_dim_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
934 mlx5e_set_rx_cq_mode_params(params, cq_period_mode);
937 params->lro_wqe_sz = MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ;
939 mlx5_query_min_inline(mdev, ¶ms->tx_min_inline_mode);
942 static void mlx5e_build_rep_netdev(struct net_device *netdev)
944 struct mlx5e_priv *priv = netdev_priv(netdev);
945 struct mlx5_core_dev *mdev = priv->mdev;
948 netdev->netdev_ops = &mlx5e_netdev_ops_rep;
950 netdev->watchdog_timeo = 15 * HZ;
952 netdev->ethtool_ops = &mlx5e_rep_ethtool_ops;
954 netdev->switchdev_ops = &mlx5e_rep_switchdev_ops;
956 netdev->features |= NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_TC | NETIF_F_NETNS_LOCAL;
957 netdev->hw_features |= NETIF_F_HW_TC;
959 eth_hw_addr_random(netdev);
961 netdev->min_mtu = ETH_MIN_MTU;
962 mlx5_query_port_max_mtu(mdev, &max_mtu, 1);
963 netdev->max_mtu = MLX5E_HW2SW_MTU(&priv->channels.params, max_mtu);
966 static void mlx5e_init_rep(struct mlx5_core_dev *mdev,
967 struct net_device *netdev,
968 const struct mlx5e_profile *profile,
971 struct mlx5e_priv *priv = netdev_priv(netdev);
974 priv->netdev = netdev;
975 priv->profile = profile;
978 mutex_init(&priv->state_lock);
980 INIT_DELAYED_WORK(&priv->update_stats_work, mlx5e_update_stats_work);
982 priv->channels.params.num_channels = profile->max_nch(mdev);
984 mlx5e_build_rep_params(mdev, &priv->channels.params, netdev->mtu);
985 mlx5e_build_rep_netdev(netdev);
987 mlx5e_timestamp_init(priv);
990 static int mlx5e_init_rep_rx(struct mlx5e_priv *priv)
992 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
993 struct mlx5e_rep_priv *rpriv = priv->ppriv;
994 struct mlx5_eswitch_rep *rep = rpriv->rep;
995 struct mlx5_flow_handle *flow_rule;
998 mlx5e_init_l2_addr(priv);
1000 err = mlx5e_create_direct_rqts(priv);
1004 err = mlx5e_create_direct_tirs(priv);
1006 goto err_destroy_direct_rqts;
1008 flow_rule = mlx5_eswitch_create_vport_rx_rule(esw,
1010 priv->direct_tir[0].tirn);
1011 if (IS_ERR(flow_rule)) {
1012 err = PTR_ERR(flow_rule);
1013 goto err_destroy_direct_tirs;
1015 rpriv->vport_rx_rule = flow_rule;
1019 err_destroy_direct_tirs:
1020 mlx5e_destroy_direct_tirs(priv);
1021 err_destroy_direct_rqts:
1022 mlx5e_destroy_direct_rqts(priv);
1026 static void mlx5e_cleanup_rep_rx(struct mlx5e_priv *priv)
1028 struct mlx5e_rep_priv *rpriv = priv->ppriv;
1030 mlx5_del_flow_rules(rpriv->vport_rx_rule);
1031 mlx5e_destroy_direct_tirs(priv);
1032 mlx5e_destroy_direct_rqts(priv);
1035 static int mlx5e_init_rep_tx(struct mlx5e_priv *priv)
1039 err = mlx5e_create_tises(priv);
1041 mlx5_core_warn(priv->mdev, "create tises failed, %d\n", err);
1047 static int mlx5e_get_rep_max_num_channels(struct mlx5_core_dev *mdev)
1049 #define MLX5E_PORT_REPRESENTOR_NCH 1
1050 return MLX5E_PORT_REPRESENTOR_NCH;
1053 static const struct mlx5e_profile mlx5e_rep_profile = {
1054 .init = mlx5e_init_rep,
1055 .init_rx = mlx5e_init_rep_rx,
1056 .cleanup_rx = mlx5e_cleanup_rep_rx,
1057 .init_tx = mlx5e_init_rep_tx,
1058 .cleanup_tx = mlx5e_cleanup_nic_tx,
1059 .update_stats = mlx5e_rep_update_hw_counters,
1060 .max_nch = mlx5e_get_rep_max_num_channels,
1061 .update_carrier = NULL,
1062 .rx_handlers.handle_rx_cqe = mlx5e_handle_rx_cqe_rep,
1063 .rx_handlers.handle_rx_cqe_mpwqe = NULL /* Not supported */,
1067 /* e-Switch vport representors */
1070 mlx5e_nic_rep_load(struct mlx5_core_dev *dev, struct mlx5_eswitch_rep *rep)
1072 struct mlx5e_rep_priv *rpriv = mlx5e_rep_to_rep_priv(rep);
1073 struct mlx5e_priv *priv = netdev_priv(rpriv->netdev);
1077 if (test_bit(MLX5E_STATE_OPENED, &priv->state)) {
1078 err = mlx5e_add_sqs_fwd_rules(priv);
1083 err = mlx5e_rep_neigh_init(rpriv);
1085 goto err_remove_sqs;
1087 /* init shared tc flow table */
1088 err = mlx5e_tc_esw_init(&rpriv->tc_ht);
1090 goto err_neigh_cleanup;
1095 mlx5e_rep_neigh_cleanup(rpriv);
1097 mlx5e_remove_sqs_fwd_rules(priv);
1102 mlx5e_nic_rep_unload(struct mlx5_eswitch_rep *rep)
1104 struct mlx5e_rep_priv *rpriv = mlx5e_rep_to_rep_priv(rep);
1105 struct mlx5e_priv *priv = netdev_priv(rpriv->netdev);
1107 if (test_bit(MLX5E_STATE_OPENED, &priv->state))
1108 mlx5e_remove_sqs_fwd_rules(priv);
1110 /* clean uplink offloaded TC rules, delete shared tc flow table */
1111 mlx5e_tc_esw_cleanup(&rpriv->tc_ht);
1113 mlx5e_rep_neigh_cleanup(rpriv);
1117 mlx5e_vport_rep_load(struct mlx5_core_dev *dev, struct mlx5_eswitch_rep *rep)
1119 struct mlx5e_rep_priv *uplink_rpriv;
1120 struct mlx5e_rep_priv *rpriv;
1121 struct net_device *netdev;
1122 struct mlx5e_priv *upriv;
1125 rpriv = kzalloc(sizeof(*rpriv), GFP_KERNEL);
1129 netdev = mlx5e_create_netdev(dev, &mlx5e_rep_profile, rpriv);
1131 pr_warn("Failed to create representor netdev for vport %d\n",
1137 rpriv->netdev = netdev;
1139 rep->rep_if[REP_ETH].priv = rpriv;
1140 INIT_LIST_HEAD(&rpriv->vport_sqs_list);
1142 err = mlx5e_attach_netdev(netdev_priv(netdev));
1144 pr_warn("Failed to attach representor netdev for vport %d\n",
1146 goto err_destroy_netdev;
1149 err = mlx5e_rep_neigh_init(rpriv);
1151 pr_warn("Failed to initialized neighbours handling for vport %d\n",
1153 goto err_detach_netdev;
1156 uplink_rpriv = mlx5_eswitch_get_uplink_priv(dev->priv.eswitch, REP_ETH);
1157 upriv = netdev_priv(uplink_rpriv->netdev);
1158 err = tc_setup_cb_egdev_register(netdev, mlx5e_rep_setup_tc_cb_egdev,
1161 goto err_neigh_cleanup;
1163 err = register_netdev(netdev);
1165 pr_warn("Failed to register representor netdev for vport %d\n",
1167 goto err_egdev_cleanup;
1173 tc_setup_cb_egdev_unregister(netdev, mlx5e_rep_setup_tc_cb_egdev,
1177 mlx5e_rep_neigh_cleanup(rpriv);
1180 mlx5e_detach_netdev(netdev_priv(netdev));
1183 mlx5e_destroy_netdev(netdev_priv(netdev));
1189 mlx5e_vport_rep_unload(struct mlx5_eswitch_rep *rep)
1191 struct mlx5e_rep_priv *rpriv = mlx5e_rep_to_rep_priv(rep);
1192 struct net_device *netdev = rpriv->netdev;
1193 struct mlx5e_priv *priv = netdev_priv(netdev);
1194 struct mlx5e_rep_priv *uplink_rpriv;
1195 void *ppriv = priv->ppriv;
1196 struct mlx5e_priv *upriv;
1198 unregister_netdev(netdev);
1199 uplink_rpriv = mlx5_eswitch_get_uplink_priv(priv->mdev->priv.eswitch,
1201 upriv = netdev_priv(uplink_rpriv->netdev);
1202 tc_setup_cb_egdev_unregister(netdev, mlx5e_rep_setup_tc_cb_egdev,
1204 mlx5e_rep_neigh_cleanup(rpriv);
1205 mlx5e_detach_netdev(priv);
1206 mlx5e_destroy_netdev(priv);
1207 kfree(ppriv); /* mlx5e_rep_priv */
1210 static void *mlx5e_vport_rep_get_proto_dev(struct mlx5_eswitch_rep *rep)
1212 struct mlx5e_rep_priv *rpriv;
1214 rpriv = mlx5e_rep_to_rep_priv(rep);
1216 return rpriv->netdev;
1219 static void mlx5e_rep_register_vf_vports(struct mlx5e_priv *priv)
1221 struct mlx5_core_dev *mdev = priv->mdev;
1222 struct mlx5_eswitch *esw = mdev->priv.eswitch;
1223 int total_vfs = MLX5_TOTAL_VPORTS(mdev);
1226 for (vport = 1; vport < total_vfs; vport++) {
1227 struct mlx5_eswitch_rep_if rep_if = {};
1229 rep_if.load = mlx5e_vport_rep_load;
1230 rep_if.unload = mlx5e_vport_rep_unload;
1231 rep_if.get_proto_dev = mlx5e_vport_rep_get_proto_dev;
1232 mlx5_eswitch_register_vport_rep(esw, vport, &rep_if, REP_ETH);
1236 static void mlx5e_rep_unregister_vf_vports(struct mlx5e_priv *priv)
1238 struct mlx5_core_dev *mdev = priv->mdev;
1239 struct mlx5_eswitch *esw = mdev->priv.eswitch;
1240 int total_vfs = MLX5_TOTAL_VPORTS(mdev);
1243 for (vport = 1; vport < total_vfs; vport++)
1244 mlx5_eswitch_unregister_vport_rep(esw, vport, REP_ETH);
1247 void mlx5e_register_vport_reps(struct mlx5e_priv *priv)
1249 struct mlx5_core_dev *mdev = priv->mdev;
1250 struct mlx5_eswitch *esw = mdev->priv.eswitch;
1251 struct mlx5_eswitch_rep_if rep_if;
1252 struct mlx5e_rep_priv *rpriv;
1254 rpriv = priv->ppriv;
1255 rpriv->netdev = priv->netdev;
1257 rep_if.load = mlx5e_nic_rep_load;
1258 rep_if.unload = mlx5e_nic_rep_unload;
1259 rep_if.get_proto_dev = mlx5e_vport_rep_get_proto_dev;
1260 rep_if.priv = rpriv;
1261 INIT_LIST_HEAD(&rpriv->vport_sqs_list);
1262 mlx5_eswitch_register_vport_rep(esw, 0, &rep_if, REP_ETH); /* UPLINK PF vport*/
1264 mlx5e_rep_register_vf_vports(priv); /* VFs vports */
1267 void mlx5e_unregister_vport_reps(struct mlx5e_priv *priv)
1269 struct mlx5_core_dev *mdev = priv->mdev;
1270 struct mlx5_eswitch *esw = mdev->priv.eswitch;
1272 mlx5e_rep_unregister_vf_vports(priv); /* VFs vports */
1273 mlx5_eswitch_unregister_vport_rep(esw, 0, REP_ETH); /* UPLINK PF*/
1276 void *mlx5e_alloc_nic_rep_priv(struct mlx5_core_dev *mdev)
1278 struct mlx5_eswitch *esw = mdev->priv.eswitch;
1279 struct mlx5e_rep_priv *rpriv;
1281 rpriv = kzalloc(sizeof(*rpriv), GFP_KERNEL);
1285 rpriv->rep = &esw->offloads.vport_reps[0];