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(mlx5_ib_dev, ib_dev);
92 if (!ibdev)
93 return -ENOMEM;
94
95 ibdev->port = kcalloc(num_ports, sizeof(*ibdev->port),
96 GFP_KERNEL);
97 if (!ibdev->port) {
98 ret = -ENOMEM;
99 goto fail_port;
100 }
101
102 ibdev->is_rep = true;
103 vport_index = rep->vport_index;
104 ibdev->port[vport_index].rep = rep;
105 ibdev->mdev = lag_master;
106 ibdev->num_ports = num_ports;
107 ibdev->ib_dev.phys_port_cnt = num_ports;
108 ret = ib_device_set_netdev(&ibdev->ib_dev,
109 mlx5_ib_get_rep_netdev(lag_master->priv.eswitch,
110 rep->vport),
111 vport_index + 1);
112 if (ret)
113 goto fail_add;
114
115 ret = __mlx5_ib_add(ibdev, profile);
116 if (ret)
117 goto fail_add;
118
119 rep->rep_data[REP_IB].priv = ibdev;
120 if (mlx5_lag_is_shared_fdb(lag_master))
121 mlx5_ib_register_peer_vport_reps(lag_master);
122
123 return 0;
124
125 fail_add:
126 kfree(ibdev->port);
127 fail_port:
128 ib_dealloc_device(&ibdev->ib_dev);
129 return ret;
130 }
131
mlx5_ib_rep_to_dev(struct mlx5_eswitch_rep * rep)132 static void *mlx5_ib_rep_to_dev(struct mlx5_eswitch_rep *rep)
133 {
134 return rep->rep_data[REP_IB].priv;
135 }
136
137 static void
mlx5_ib_vport_rep_unload(struct mlx5_eswitch_rep * rep)138 mlx5_ib_vport_rep_unload(struct mlx5_eswitch_rep *rep)
139 {
140 struct mlx5_core_dev *mdev = mlx5_eswitch_get_core_dev(rep->esw);
141 struct mlx5_ib_dev *dev = mlx5_ib_rep_to_dev(rep);
142 int vport_index = rep->vport_index;
143 struct mlx5_ib_port *port;
144 int i;
145
146 if (WARN_ON(!mdev))
147 return;
148
149 if (!dev)
150 return;
151
152 if (mlx5_lag_is_shared_fdb(mdev) &&
153 !mlx5_lag_is_master(mdev)) {
154 if (rep->vport == MLX5_VPORT_UPLINK && !mlx5_lag_is_mpesw(mdev))
155 return;
156 for (i = 0; i < dev->num_ports; i++) {
157 if (dev->port[i].rep == rep)
158 break;
159 }
160 if (WARN_ON(i == dev->num_ports))
161 return;
162 vport_index = i;
163 }
164
165 port = &dev->port[vport_index];
166
167 ib_device_set_netdev(&dev->ib_dev, NULL, vport_index + 1);
168 rep->rep_data[REP_IB].priv = NULL;
169 port->rep = NULL;
170
171 if (rep->vport == MLX5_VPORT_UPLINK) {
172
173 if (mlx5_lag_is_shared_fdb(mdev) && !mlx5_lag_is_master(mdev))
174 return;
175
176 if (mlx5_lag_is_shared_fdb(mdev)) {
177 struct mlx5_core_dev *peer_mdev;
178 struct mlx5_eswitch *esw;
179
180 mlx5_lag_for_each_peer_mdev(mdev, peer_mdev, i) {
181 esw = peer_mdev->priv.eswitch;
182 mlx5_eswitch_unregister_vport_reps(esw, REP_IB);
183 }
184 }
185 __mlx5_ib_remove(dev, dev->profile, MLX5_IB_STAGE_MAX);
186 }
187 }
188
189 static const struct mlx5_eswitch_rep_ops rep_ops = {
190 .load = mlx5_ib_vport_rep_load,
191 .unload = mlx5_ib_vport_rep_unload,
192 .get_proto_dev = mlx5_ib_rep_to_dev,
193 };
194
mlx5_ib_register_peer_vport_reps(struct mlx5_core_dev * mdev)195 static void mlx5_ib_register_peer_vport_reps(struct mlx5_core_dev *mdev)
196 {
197 struct mlx5_core_dev *peer_mdev;
198 struct mlx5_eswitch *esw;
199 int i;
200
201 mlx5_lag_for_each_peer_mdev(mdev, peer_mdev, i) {
202 esw = peer_mdev->priv.eswitch;
203 mlx5_eswitch_register_vport_reps(esw, &rep_ops, REP_IB);
204 }
205 }
206
mlx5_ib_get_rep_netdev(struct mlx5_eswitch * esw,u16 vport_num)207 struct net_device *mlx5_ib_get_rep_netdev(struct mlx5_eswitch *esw,
208 u16 vport_num)
209 {
210 return mlx5_eswitch_get_proto_dev(esw, vport_num, REP_ETH);
211 }
212
create_flow_rule_vport_sq(struct mlx5_ib_dev * dev,struct mlx5_ib_sq * sq,u32 port)213 struct mlx5_flow_handle *create_flow_rule_vport_sq(struct mlx5_ib_dev *dev,
214 struct mlx5_ib_sq *sq,
215 u32 port)
216 {
217 struct mlx5_eswitch *esw = dev->mdev->priv.eswitch;
218 struct mlx5_eswitch_rep *rep;
219
220 if (!dev->is_rep || !port)
221 return NULL;
222
223 if (!dev->port[port - 1].rep)
224 return ERR_PTR(-EINVAL);
225
226 rep = dev->port[port - 1].rep;
227
228 return mlx5_eswitch_add_send_to_vport_rule(esw, esw, rep, sq->base.mqp.qpn);
229 }
230
mlx5r_rep_probe(struct auxiliary_device * adev,const struct auxiliary_device_id * id)231 static int mlx5r_rep_probe(struct auxiliary_device *adev,
232 const struct auxiliary_device_id *id)
233 {
234 struct mlx5_adev *idev = container_of(adev, struct mlx5_adev, adev);
235 struct mlx5_core_dev *mdev = idev->mdev;
236 struct mlx5_eswitch *esw;
237
238 esw = mdev->priv.eswitch;
239 mlx5_eswitch_register_vport_reps(esw, &rep_ops, REP_IB);
240 return 0;
241 }
242
mlx5r_rep_remove(struct auxiliary_device * adev)243 static void mlx5r_rep_remove(struct auxiliary_device *adev)
244 {
245 struct mlx5_adev *idev = container_of(adev, struct mlx5_adev, adev);
246 struct mlx5_core_dev *mdev = idev->mdev;
247 struct mlx5_eswitch *esw;
248
249 esw = mdev->priv.eswitch;
250 mlx5_eswitch_unregister_vport_reps(esw, REP_IB);
251 }
252
253 static const struct auxiliary_device_id mlx5r_rep_id_table[] = {
254 { .name = MLX5_ADEV_NAME ".rdma-rep", },
255 {},
256 };
257
258 MODULE_DEVICE_TABLE(auxiliary, mlx5r_rep_id_table);
259
260 static struct auxiliary_driver mlx5r_rep_driver = {
261 .name = "rep",
262 .probe = mlx5r_rep_probe,
263 .remove = mlx5r_rep_remove,
264 .id_table = mlx5r_rep_id_table,
265 };
266
mlx5r_rep_init(void)267 int mlx5r_rep_init(void)
268 {
269 return auxiliary_driver_register(&mlx5r_rep_driver);
270 }
271
mlx5r_rep_cleanup(void)272 void mlx5r_rep_cleanup(void)
273 {
274 auxiliary_driver_unregister(&mlx5r_rep_driver);
275 }
276