1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /* Copyright (c) 2020 Mellanox Technologies Inc. All rights reserved. */
4 #include <linux/netdevice.h>
5 #include <linux/list.h>
10 #include "esw/acl/ofld.h"
13 struct mlx5e_rep_bond {
14 struct notifier_block nb;
15 struct netdev_net_notifier nn;
16 struct list_head metadata_list;
19 struct mlx5e_rep_bond_slave_entry {
20 struct list_head list;
21 struct net_device *netdev;
24 struct mlx5e_rep_bond_metadata {
25 struct list_head list; /* link to global list of rep_bond_metadata */
26 struct mlx5_eswitch *esw;
27 /* private of uplink holding rep bond metadata list */
28 struct net_device *lag_dev;
31 struct list_head slaves_list; /* slaves list */
35 static struct mlx5e_rep_bond_metadata *
36 mlx5e_lookup_rep_bond_metadata(struct mlx5_rep_uplink_priv *uplink_priv,
37 const struct net_device *lag_dev)
39 struct mlx5e_rep_bond_metadata *found = NULL;
40 struct mlx5e_rep_bond_metadata *cur;
42 list_for_each_entry(cur, &uplink_priv->bond->metadata_list, list) {
43 if (cur->lag_dev == lag_dev) {
52 static struct mlx5e_rep_bond_slave_entry *
53 mlx5e_lookup_rep_bond_slave_entry(struct mlx5e_rep_bond_metadata *mdata,
54 const struct net_device *netdev)
56 struct mlx5e_rep_bond_slave_entry *found = NULL;
57 struct mlx5e_rep_bond_slave_entry *cur;
59 list_for_each_entry(cur, &mdata->slaves_list, list) {
60 if (cur->netdev == netdev) {
69 static void mlx5e_rep_bond_metadata_release(struct mlx5e_rep_bond_metadata *mdata)
71 netdev_dbg(mdata->lag_dev, "destroy rep_bond_metadata(%d)\n",
72 mdata->metadata_reg_c_0);
73 list_del(&mdata->list);
74 mlx5_esw_match_metadata_free(mdata->esw, mdata->metadata_reg_c_0);
75 WARN_ON(!list_empty(&mdata->slaves_list));
79 /* This must be called under rtnl_lock */
80 int mlx5e_rep_bond_enslave(struct mlx5_eswitch *esw, struct net_device *netdev,
81 struct net_device *lag_dev)
83 struct mlx5e_rep_bond_slave_entry *s_entry;
84 struct mlx5e_rep_bond_metadata *mdata;
85 struct mlx5e_rep_priv *rpriv;
86 struct mlx5e_priv *priv;
91 rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
92 mdata = mlx5e_lookup_rep_bond_metadata(&rpriv->uplink_priv, lag_dev);
94 /* First netdev becomes slave, no metadata presents the lag_dev. Create one */
95 mdata = kzalloc(sizeof(*mdata), GFP_KERNEL);
99 mdata->lag_dev = lag_dev;
101 INIT_LIST_HEAD(&mdata->slaves_list);
102 mdata->metadata_reg_c_0 = mlx5_esw_match_metadata_alloc(esw);
103 if (!mdata->metadata_reg_c_0) {
107 list_add(&mdata->list, &rpriv->uplink_priv.bond->metadata_list);
109 netdev_dbg(lag_dev, "create rep_bond_metadata(%d)\n",
110 mdata->metadata_reg_c_0);
113 s_entry = kzalloc(sizeof(*s_entry), GFP_KERNEL);
116 goto entry_alloc_err;
119 s_entry->netdev = netdev;
120 priv = netdev_priv(netdev);
123 err = mlx5_esw_acl_ingress_vport_bond_update(esw, rpriv->rep->vport,
124 mdata->metadata_reg_c_0);
129 list_add_tail(&s_entry->list, &mdata->slaves_list);
130 netdev_dbg(netdev, "enslave rep vport(%d) lag_dev(%s) metadata(0x%x)\n",
131 rpriv->rep->vport, lag_dev->name, mdata->metadata_reg_c_0);
139 mlx5e_rep_bond_metadata_release(mdata);
143 /* This must be called under rtnl_lock */
144 void mlx5e_rep_bond_unslave(struct mlx5_eswitch *esw,
145 const struct net_device *netdev,
146 const struct net_device *lag_dev)
148 struct mlx5e_rep_bond_slave_entry *s_entry;
149 struct mlx5e_rep_bond_metadata *mdata;
150 struct mlx5e_rep_priv *rpriv;
151 struct mlx5e_priv *priv;
155 rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
156 mdata = mlx5e_lookup_rep_bond_metadata(&rpriv->uplink_priv, lag_dev);
160 s_entry = mlx5e_lookup_rep_bond_slave_entry(mdata, netdev);
164 priv = netdev_priv(netdev);
167 /* Reset bond_metadata to zero first then reset all ingress/egress
168 * acls and rx rules of unslave representor's vport
170 mlx5_esw_acl_ingress_vport_bond_update(esw, rpriv->rep->vport, 0);
171 mlx5_esw_acl_egress_vport_unbond(esw, rpriv->rep->vport);
172 mlx5e_rep_bond_update(priv, false);
174 list_del(&s_entry->list);
176 netdev_dbg(netdev, "unslave rep vport(%d) lag_dev(%s) metadata(0x%x)\n",
177 rpriv->rep->vport, lag_dev->name, mdata->metadata_reg_c_0);
179 if (--mdata->slaves == 0)
180 mlx5e_rep_bond_metadata_release(mdata);
184 static bool mlx5e_rep_is_lag_netdev(struct net_device *netdev)
186 return netif_is_lag_port(netdev) && mlx5e_eswitch_vf_rep(netdev);
189 static void mlx5e_rep_changelowerstate_event(struct net_device *netdev, void *ptr)
191 struct netdev_notifier_changelowerstate_info *info;
192 struct netdev_lag_lower_state_info *lag_info;
193 struct mlx5e_rep_priv *rpriv;
194 struct net_device *lag_dev;
195 struct mlx5e_priv *priv;
196 struct list_head *iter;
197 struct net_device *dev;
203 lag_info = info->lower_state_info;
204 /* This is not an event of a representor becoming active slave */
205 if (!lag_info->tx_enabled)
208 priv = netdev_priv(netdev);
210 fwd_vport_num = rpriv->rep->vport;
211 lag_dev = netdev_master_upper_dev_get(netdev);
215 netdev_dbg(netdev, "lag_dev(%s)'s slave vport(%d) is txable(%d)\n",
216 lag_dev->name, fwd_vport_num, net_lag_port_dev_txable(netdev));
218 /* Point everyone's egress acl to the vport of the active representor */
219 netdev_for_each_lower_dev(lag_dev, dev, iter) {
220 priv = netdev_priv(dev);
222 acl_vport_num = rpriv->rep->vport;
223 if (acl_vport_num != fwd_vport_num) {
224 /* Only single rx_rule for unique bond_metadata should be
225 * present, delete it if it's saved as passive vport's
226 * rx_rule with destination as passive vport's root_ft
228 mlx5e_rep_bond_update(priv, true);
229 err = mlx5_esw_acl_egress_vport_bond(priv->mdev->priv.eswitch,
234 "configure slave vport(%d) egress fwd, err(%d)",
239 /* Insert new rx_rule for unique bond_metadata, save it as active vport's
240 * rx_rule with new destination as active vport's root_ft
242 err = mlx5e_rep_bond_update(netdev_priv(netdev), false);
244 netdev_warn(netdev, "configure active slave vport(%d) rx_rule, err(%d)",
248 static void mlx5e_rep_changeupper_event(struct net_device *netdev, void *ptr)
250 struct netdev_notifier_changeupper_info *info = ptr;
251 struct mlx5e_rep_priv *rpriv;
252 struct net_device *lag_dev;
253 struct mlx5e_priv *priv;
255 priv = netdev_priv(netdev);
257 lag_dev = info->upper_dev;
259 netdev_dbg(netdev, "%sslave vport(%d) lag(%s)\n",
260 info->linking ? "en" : "un", rpriv->rep->vport, lag_dev->name);
263 mlx5e_rep_bond_enslave(priv->mdev->priv.eswitch, netdev, lag_dev);
265 mlx5e_rep_bond_unslave(priv->mdev->priv.eswitch, netdev, lag_dev);
268 /* Bond device of representors and netdev events are used here in specific way
269 * to support eswitch vports bonding and to perform failover of eswitch vport
270 * by modifying the vport's egress acl of lower dev representors. Thus this
271 * also change the traditional behavior of lower dev under bond device.
272 * All non-representor netdevs or representors of other vendors as lower dev
273 * of bond device are not supported.
275 static int mlx5e_rep_esw_bond_netevent(struct notifier_block *nb,
276 unsigned long event, void *ptr)
278 struct net_device *netdev = netdev_notifier_info_to_dev(ptr);
279 struct mlx5e_rep_priv *rpriv;
280 struct mlx5e_rep_bond *bond;
281 struct mlx5e_priv *priv;
283 if (!mlx5e_rep_is_lag_netdev(netdev))
286 bond = container_of(nb, struct mlx5e_rep_bond, nb);
287 priv = netdev_priv(netdev);
288 rpriv = mlx5_eswitch_get_uplink_priv(priv->mdev->priv.eswitch, REP_ETH);
289 /* Verify VF representor is on the same device of the bond handling the netevent. */
290 if (rpriv->uplink_priv.bond != bond)
294 case NETDEV_CHANGELOWERSTATE:
295 mlx5e_rep_changelowerstate_event(netdev, ptr);
297 case NETDEV_CHANGEUPPER:
298 mlx5e_rep_changeupper_event(netdev, ptr);
304 /* If HW support eswitch vports bonding, register a specific notifier to
305 * handle it when two or more representors are bonded
307 int mlx5e_rep_bond_init(struct mlx5e_rep_priv *rpriv)
309 struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
310 struct net_device *netdev = rpriv->netdev;
311 struct mlx5e_priv *priv;
314 priv = netdev_priv(netdev);
315 if (!mlx5_esw_acl_egress_fwd2vport_supported(priv->mdev->priv.eswitch))
318 uplink_priv->bond = kvzalloc(sizeof(*uplink_priv->bond), GFP_KERNEL);
319 if (!uplink_priv->bond) {
324 INIT_LIST_HEAD(&uplink_priv->bond->metadata_list);
325 uplink_priv->bond->nb.notifier_call = mlx5e_rep_esw_bond_netevent;
326 ret = register_netdevice_notifier_dev_net(netdev,
327 &uplink_priv->bond->nb,
328 &uplink_priv->bond->nn);
330 netdev_err(netdev, "register bonding netevent notifier, err(%d)\n", ret);
331 kvfree(uplink_priv->bond);
332 uplink_priv->bond = NULL;
339 void mlx5e_rep_bond_cleanup(struct mlx5e_rep_priv *rpriv)
341 struct mlx5e_priv *priv = netdev_priv(rpriv->netdev);
343 if (!mlx5_esw_acl_egress_fwd2vport_supported(priv->mdev->priv.eswitch) ||
344 !rpriv->uplink_priv.bond)
347 unregister_netdevice_notifier_dev_net(rpriv->netdev,
348 &rpriv->uplink_priv.bond->nb,
349 &rpriv->uplink_priv.bond->nn);
350 kvfree(rpriv->uplink_priv.bond);