1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /*
3 * Copyright (c) 2018 Mellanox Technologies. All rights reserved.
4 */
5
6 #include <linux/mlx5/vport.h>
7 #include "ib_rep.h"
8 #include "srq.h"
9
10 static int
mlx5_ib_set_vport_rep(struct mlx5_core_dev * dev,struct mlx5_eswitch_rep * rep,int vport_index)11 mlx5_ib_set_vport_rep(struct mlx5_core_dev *dev,
12 struct mlx5_eswitch_rep *rep,
13 int vport_index)
14 {
15 struct mlx5_ib_dev *ibdev;
16 struct net_device *ndev;
17
18 ibdev = mlx5_eswitch_uplink_get_proto_dev(dev->priv.eswitch, REP_IB);
19 if (!ibdev)
20 return -EINVAL;
21
22 ibdev->port[vport_index].rep = rep;
23 rep->rep_data[REP_IB].priv = ibdev;
24 ndev = mlx5_ib_get_rep_netdev(rep->esw, rep->vport);
25
26 return ib_device_set_netdev(&ibdev->ib_dev, ndev, vport_index + 1);
27 }
28
29 static void mlx5_ib_register_peer_vport_reps(struct mlx5_core_dev *mdev);
30
mlx5_ib_num_ports_update(struct mlx5_core_dev * dev,u32 * num_ports)31 static void mlx5_ib_num_ports_update(struct mlx5_core_dev *dev, u32 *num_ports)
32 {
33 struct mlx5_core_dev *peer_dev;
34 int i;
35
36 mlx5_lag_for_each_peer_mdev(dev, peer_dev, i) {
37 u32 peer_num_ports = mlx5_eswitch_get_total_vports(peer_dev);
38
39 if (mlx5_lag_is_mpesw(peer_dev))
40 *num_ports += peer_num_ports;
41 else
42 /* Only 1 ib port is the representor for all uplinks */
43 *num_ports += peer_num_ports - 1;
44 }
45 }
46
47 static int
mlx5_ib_vport_rep_load(struct mlx5_core_dev * dev,struct mlx5_eswitch_rep * rep)48 mlx5_ib_vport_rep_load(struct mlx5_core_dev *dev, struct mlx5_eswitch_rep *rep)
49 {
50 u32 num_ports = mlx5_eswitch_get_total_vports(dev);
51 struct mlx5_core_dev *lag_master = dev;
52 const struct mlx5_ib_profile *profile;
53 struct mlx5_core_dev *peer_dev;
54 struct mlx5_ib_dev *ibdev;
55 int new_uplink = false;
56 int vport_index;
57 int ret;
58 int i;
59
60 vport_index = rep->vport_index;
61
62 if (mlx5_lag_is_shared_fdb(dev)) {
63 if (mlx5_lag_is_master(dev)) {
64 mlx5_ib_num_ports_update(dev, &num_ports);
65 } else {
66 if (rep->vport == MLX5_VPORT_UPLINK) {
67 if (!mlx5_lag_is_mpesw(dev))
68 return 0;
69 new_uplink = true;
70 }
71 mlx5_lag_for_each_peer_mdev(dev, peer_dev, i) {
72 u32 peer_n_ports = mlx5_eswitch_get_total_vports(peer_dev);
73
74 if (mlx5_lag_is_master(peer_dev))
75 lag_master = peer_dev;
76 else if (!mlx5_lag_is_mpesw(dev))
77 /* Only 1 ib port is the representor for all uplinks */
78 peer_n_ports--;
79
80 if (mlx5_get_dev_index(peer_dev) < mlx5_get_dev_index(dev))
81 vport_index += peer_n_ports;
82 }
83 }
84 }
85
86 if (rep->vport == MLX5_VPORT_UPLINK && !new_uplink)
87 profile = &raw_eth_profile;
88 else
89 return mlx5_ib_set_vport_rep(lag_master, rep, vport_index);
90
91 ibdev = ib_alloc_device_with_net(mlx5_ib_dev, ib_dev,
92 mlx5_core_net(lag_master));
93 if (!ibdev)
94 return -ENOMEM;
95
96 ibdev->port = kcalloc(num_ports, sizeof(*ibdev->port),
97 GFP_KERNEL);
98 if (!ibdev->port) {
99 ret = -ENOMEM;
100 goto fail_port;
101 }
102
103 ibdev->is_rep = true;
104 vport_index = rep->vport_index;
105 ibdev->port[vport_index].rep = rep;
106 ibdev->mdev = lag_master;
107 ibdev->num_ports = num_ports;
108 ibdev->ib_dev.phys_port_cnt = num_ports;
109 ret = ib_device_set_netdev(&ibdev->ib_dev,
110 mlx5_ib_get_rep_netdev(lag_master->priv.eswitch,
111 rep->vport),
112 vport_index + 1);
113 if (ret)
114 goto fail_add;
115
116 ret = __mlx5_ib_add(ibdev, profile);
117 if (ret)
118 goto fail_add;
119
120 rep->rep_data[REP_IB].priv = ibdev;
121 if (mlx5_lag_is_shared_fdb(lag_master))
122 mlx5_ib_register_peer_vport_reps(lag_master);
123
124 return 0;
125
126 fail_add:
127 kfree(ibdev->port);
128 fail_port:
129 ib_dealloc_device(&ibdev->ib_dev);
130 return ret;
131 }
132
mlx5_ib_rep_to_dev(struct mlx5_eswitch_rep * rep)133 static void *mlx5_ib_rep_to_dev(struct mlx5_eswitch_rep *rep)
134 {
135 return rep->rep_data[REP_IB].priv;
136 }
137
138 static void
mlx5_ib_vport_rep_unload(struct mlx5_eswitch_rep * rep)139 mlx5_ib_vport_rep_unload(struct mlx5_eswitch_rep *rep)
140 {
141 struct mlx5_core_dev *mdev = mlx5_eswitch_get_core_dev(rep->esw);
142 struct mlx5_ib_dev *dev = mlx5_ib_rep_to_dev(rep);
143 int vport_index = rep->vport_index;
144 struct mlx5_ib_port *port;
145 int i;
146
147 if (WARN_ON(!mdev))
148 return;
149
150 if (!dev)
151 return;
152
153 if (mlx5_lag_is_shared_fdb(mdev) &&
154 !mlx5_lag_is_master(mdev)) {
155 if (rep->vport == MLX5_VPORT_UPLINK && !mlx5_lag_is_mpesw(mdev))
156 return;
157 for (i = 0; i < dev->num_ports; i++) {
158 if (dev->port[i].rep == rep)
159 break;
160 }
161 if (WARN_ON(i == dev->num_ports))
162 return;
163 vport_index = i;
164 }
165
166 port = &dev->port[vport_index];
167
168 ib_device_set_netdev(&dev->ib_dev, NULL, vport_index + 1);
169 rep->rep_data[REP_IB].priv = NULL;
170 port->rep = NULL;
171
172 if (rep->vport == MLX5_VPORT_UPLINK) {
173
174 if (mlx5_lag_is_shared_fdb(mdev) && !mlx5_lag_is_master(mdev))
175 return;
176
177 if (mlx5_lag_is_shared_fdb(mdev)) {
178 struct mlx5_core_dev *peer_mdev;
179 struct mlx5_eswitch *esw;
180
181 mlx5_lag_for_each_peer_mdev(mdev, peer_mdev, i) {
182 esw = peer_mdev->priv.eswitch;
183 mlx5_eswitch_unregister_vport_reps(esw, REP_IB);
184 }
185 }
186 __mlx5_ib_remove(dev, dev->profile, MLX5_IB_STAGE_MAX);
187 }
188 }
189
190 static const struct mlx5_eswitch_rep_ops rep_ops = {
191 .load = mlx5_ib_vport_rep_load,
192 .unload = mlx5_ib_vport_rep_unload,
193 .get_proto_dev = mlx5_ib_rep_to_dev,
194 };
195
mlx5_ib_register_peer_vport_reps(struct mlx5_core_dev * mdev)196 static void mlx5_ib_register_peer_vport_reps(struct mlx5_core_dev *mdev)
197 {
198 struct mlx5_core_dev *peer_mdev;
199 struct mlx5_eswitch *esw;
200 int i;
201
202 mlx5_lag_for_each_peer_mdev(mdev, peer_mdev, i) {
203 esw = peer_mdev->priv.eswitch;
204 mlx5_eswitch_register_vport_reps(esw, &rep_ops, REP_IB);
205 }
206 }
207
mlx5_ib_get_rep_netdev(struct mlx5_eswitch * esw,u16 vport_num)208 struct net_device *mlx5_ib_get_rep_netdev(struct mlx5_eswitch *esw,
209 u16 vport_num)
210 {
211 return mlx5_eswitch_get_proto_dev(esw, vport_num, REP_ETH);
212 }
213
create_flow_rule_vport_sq(struct mlx5_ib_dev * dev,struct mlx5_ib_sq * sq,u32 port)214 struct mlx5_flow_handle *create_flow_rule_vport_sq(struct mlx5_ib_dev *dev,
215 struct mlx5_ib_sq *sq,
216 u32 port)
217 {
218 struct mlx5_eswitch *esw = dev->mdev->priv.eswitch;
219 struct mlx5_eswitch_rep *rep;
220
221 if (!dev->is_rep || !port)
222 return NULL;
223
224 if (!dev->port[port - 1].rep)
225 return ERR_PTR(-EINVAL);
226
227 rep = dev->port[port - 1].rep;
228
229 return mlx5_eswitch_add_send_to_vport_rule(esw, esw, rep, sq->base.mqp.qpn);
230 }
231
mlx5r_rep_probe(struct auxiliary_device * adev,const struct auxiliary_device_id * id)232 static int mlx5r_rep_probe(struct auxiliary_device *adev,
233 const struct auxiliary_device_id *id)
234 {
235 struct mlx5_adev *idev = container_of(adev, struct mlx5_adev, adev);
236 struct mlx5_core_dev *mdev = idev->mdev;
237 struct mlx5_eswitch *esw;
238
239 esw = mdev->priv.eswitch;
240 mlx5_eswitch_register_vport_reps(esw, &rep_ops, REP_IB);
241 return 0;
242 }
243
mlx5r_rep_remove(struct auxiliary_device * adev)244 static void mlx5r_rep_remove(struct auxiliary_device *adev)
245 {
246 struct mlx5_adev *idev = container_of(adev, struct mlx5_adev, adev);
247 struct mlx5_core_dev *mdev = idev->mdev;
248 struct mlx5_eswitch *esw;
249
250 esw = mdev->priv.eswitch;
251 mlx5_eswitch_unregister_vport_reps(esw, REP_IB);
252 }
253
254 static const struct auxiliary_device_id mlx5r_rep_id_table[] = {
255 { .name = MLX5_ADEV_NAME ".rdma-rep", },
256 {},
257 };
258
259 MODULE_DEVICE_TABLE(auxiliary, mlx5r_rep_id_table);
260
261 static struct auxiliary_driver mlx5r_rep_driver = {
262 .name = "rep",
263 .probe = mlx5r_rep_probe,
264 .remove = mlx5r_rep_remove,
265 .id_table = mlx5r_rep_id_table,
266 };
267
mlx5r_rep_init(void)268 int mlx5r_rep_init(void)
269 {
270 return auxiliary_driver_register(&mlx5r_rep_driver);
271 }
272
mlx5r_rep_cleanup(void)273 void mlx5r_rep_cleanup(void)
274 {
275 auxiliary_driver_unregister(&mlx5r_rep_driver);
276 }
277