1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /* Copyright (c) 2022, NVIDIA CORPORATION & AFFILIATES. All rights reserved. */
3
4 #include <linux/netdevice.h>
5 #include <net/nexthop.h>
6 #include "lag/lag.h"
7 #include "eswitch.h"
8 #include "esw/acl/ofld.h"
9 #include "lib/events.h"
10
mlx5_mpesw_metadata_cleanup(struct mlx5_lag * ldev)11 static void mlx5_mpesw_metadata_cleanup(struct mlx5_lag *ldev)
12 {
13 struct mlx5_core_dev *dev;
14 struct mlx5_eswitch *esw;
15 u32 pf_metadata;
16 int i;
17
18 mlx5_ldev_for_each(i, 0, ldev) {
19 dev = ldev->pf[i].dev;
20 esw = dev->priv.eswitch;
21 pf_metadata = ldev->lag_mpesw.pf_metadata[i];
22 if (!pf_metadata)
23 continue;
24 mlx5_esw_acl_ingress_vport_metadata_update(esw, MLX5_VPORT_UPLINK, 0);
25 mlx5_notifier_call_chain(dev->priv.events, MLX5_DEV_EVENT_MULTIPORT_ESW,
26 (void *)0);
27 mlx5_esw_match_metadata_free(esw, pf_metadata);
28 ldev->lag_mpesw.pf_metadata[i] = 0;
29 }
30 }
31
mlx5_mpesw_metadata_set(struct mlx5_lag * ldev)32 static int mlx5_mpesw_metadata_set(struct mlx5_lag *ldev)
33 {
34 struct mlx5_core_dev *dev;
35 struct mlx5_eswitch *esw;
36 u32 pf_metadata;
37 int i, err;
38
39 mlx5_ldev_for_each(i, 0, ldev) {
40 dev = ldev->pf[i].dev;
41 esw = dev->priv.eswitch;
42 pf_metadata = mlx5_esw_match_metadata_alloc(esw);
43 if (!pf_metadata) {
44 err = -ENOSPC;
45 goto err_metadata;
46 }
47
48 ldev->lag_mpesw.pf_metadata[i] = pf_metadata;
49 err = mlx5_esw_acl_ingress_vport_metadata_update(esw, MLX5_VPORT_UPLINK,
50 pf_metadata);
51 if (err)
52 goto err_metadata;
53 }
54
55 mlx5_ldev_for_each(i, 0, ldev) {
56 dev = ldev->pf[i].dev;
57 mlx5_notifier_call_chain(dev->priv.events, MLX5_DEV_EVENT_MULTIPORT_ESW,
58 (void *)0);
59 }
60
61 return 0;
62
63 err_metadata:
64 mlx5_mpesw_metadata_cleanup(ldev);
65 return err;
66 }
67
enable_mpesw(struct mlx5_lag * ldev)68 static int enable_mpesw(struct mlx5_lag *ldev)
69 {
70 int idx = mlx5_lag_get_dev_index_by_seq(ldev, MLX5_LAG_P1);
71 struct mlx5_core_dev *dev0;
72 int err;
73 int i;
74
75 if (idx < 0 || ldev->mode != MLX5_LAG_MODE_NONE)
76 return -EINVAL;
77
78 dev0 = ldev->pf[idx].dev;
79 if (mlx5_eswitch_mode(dev0) != MLX5_ESWITCH_OFFLOADS ||
80 !MLX5_CAP_PORT_SELECTION(dev0, port_select_flow_table) ||
81 !MLX5_CAP_GEN(dev0, create_lag_when_not_master_up) ||
82 !mlx5_lag_check_prereq(ldev) ||
83 !mlx5_lag_shared_fdb_supported(ldev))
84 return -EOPNOTSUPP;
85
86 err = mlx5_mpesw_metadata_set(ldev);
87 if (err)
88 return err;
89
90 mlx5_lag_remove_devices(ldev);
91
92 err = mlx5_activate_lag(ldev, NULL, MLX5_LAG_MODE_MPESW, true);
93 if (err) {
94 mlx5_core_warn(dev0, "Failed to create LAG in MPESW mode (%d)\n", err);
95 goto err_add_devices;
96 }
97
98 dev0->priv.flags &= ~MLX5_PRIV_FLAGS_DISABLE_IB_ADEV;
99 mlx5_rescan_drivers_locked(dev0);
100 mlx5_ldev_for_each(i, 0, ldev) {
101 err = mlx5_eswitch_reload_ib_reps(ldev->pf[i].dev->priv.eswitch);
102 if (err)
103 goto err_rescan_drivers;
104 }
105
106 return 0;
107
108 err_rescan_drivers:
109 dev0->priv.flags |= MLX5_PRIV_FLAGS_DISABLE_IB_ADEV;
110 mlx5_rescan_drivers_locked(dev0);
111 mlx5_deactivate_lag(ldev);
112 err_add_devices:
113 mlx5_lag_add_devices(ldev);
114 mlx5_ldev_for_each(i, 0, ldev)
115 mlx5_eswitch_reload_ib_reps(ldev->pf[i].dev->priv.eswitch);
116 mlx5_mpesw_metadata_cleanup(ldev);
117 return err;
118 }
119
disable_mpesw(struct mlx5_lag * ldev)120 static void disable_mpesw(struct mlx5_lag *ldev)
121 {
122 if (ldev->mode == MLX5_LAG_MODE_MPESW) {
123 mlx5_mpesw_metadata_cleanup(ldev);
124 mlx5_disable_lag(ldev);
125 }
126 }
127
mlx5_mpesw_work(struct work_struct * work)128 static void mlx5_mpesw_work(struct work_struct *work)
129 {
130 struct mlx5_mpesw_work_st *mpesww = container_of(work, struct mlx5_mpesw_work_st, work);
131 struct mlx5_devcom_comp_dev *devcom;
132 struct mlx5_lag *ldev = mpesww->lag;
133
134 devcom = mlx5_lag_get_devcom_comp(ldev);
135 if (!devcom)
136 return;
137
138 mlx5_devcom_comp_lock(devcom);
139 mutex_lock(&ldev->lock);
140 if (ldev->mode_changes_in_progress) {
141 mpesww->result = -EAGAIN;
142 goto unlock;
143 }
144
145 if (mpesww->op == MLX5_MPESW_OP_ENABLE)
146 mpesww->result = enable_mpesw(ldev);
147 else if (mpesww->op == MLX5_MPESW_OP_DISABLE)
148 disable_mpesw(ldev);
149 unlock:
150 mutex_unlock(&ldev->lock);
151 mlx5_devcom_comp_unlock(devcom);
152 complete(&mpesww->comp);
153 }
154
mlx5_lag_mpesw_queue_work(struct mlx5_core_dev * dev,enum mpesw_op op)155 static int mlx5_lag_mpesw_queue_work(struct mlx5_core_dev *dev,
156 enum mpesw_op op)
157 {
158 struct mlx5_lag *ldev = mlx5_lag_dev(dev);
159 struct mlx5_mpesw_work_st *work;
160 int err = 0;
161
162 if (!ldev)
163 return 0;
164
165 work = kzalloc(sizeof(*work), GFP_KERNEL);
166 if (!work)
167 return -ENOMEM;
168
169 INIT_WORK(&work->work, mlx5_mpesw_work);
170 init_completion(&work->comp);
171 work->op = op;
172 work->lag = ldev;
173
174 if (!queue_work(ldev->wq, &work->work)) {
175 mlx5_core_warn(dev, "failed to queue mpesw work\n");
176 err = -EINVAL;
177 goto out;
178 }
179 wait_for_completion(&work->comp);
180 err = work->result;
181 out:
182 kfree(work);
183 return err;
184 }
185
mlx5_lag_mpesw_disable(struct mlx5_core_dev * dev)186 void mlx5_lag_mpesw_disable(struct mlx5_core_dev *dev)
187 {
188 mlx5_lag_mpesw_queue_work(dev, MLX5_MPESW_OP_DISABLE);
189 }
190
mlx5_lag_mpesw_enable(struct mlx5_core_dev * dev)191 int mlx5_lag_mpesw_enable(struct mlx5_core_dev *dev)
192 {
193 return mlx5_lag_mpesw_queue_work(dev, MLX5_MPESW_OP_ENABLE);
194 }
195
mlx5_lag_mpesw_do_mirred(struct mlx5_core_dev * mdev,struct net_device * out_dev,struct netlink_ext_ack * extack)196 int mlx5_lag_mpesw_do_mirred(struct mlx5_core_dev *mdev,
197 struct net_device *out_dev,
198 struct netlink_ext_ack *extack)
199 {
200 struct mlx5_lag *ldev = mlx5_lag_dev(mdev);
201
202 if (!netif_is_bond_master(out_dev) || !ldev)
203 return 0;
204
205 if (ldev->mode != MLX5_LAG_MODE_MPESW)
206 return 0;
207
208 NL_SET_ERR_MSG_MOD(extack, "can't forward to bond in mpesw mode");
209 return -EOPNOTSUPP;
210 }
211
mlx5_lag_is_mpesw(struct mlx5_core_dev * dev)212 bool mlx5_lag_is_mpesw(struct mlx5_core_dev *dev)
213 {
214 struct mlx5_lag *ldev = mlx5_lag_dev(dev);
215
216 return ldev && ldev->mode == MLX5_LAG_MODE_MPESW;
217 }
218 EXPORT_SYMBOL(mlx5_lag_is_mpesw);
219