1 // SPDX-License-Identifier: GPL-2.0
2 // Copyright (c) 2019 Mellanox Technologies.
3
4 #include "health.h"
5 #include "lib/eq.h"
6 #include "lib/mlx5.h"
7
mlx5e_health_fmsg_named_obj_nest_start(struct devlink_fmsg * fmsg,char * name)8 void mlx5e_health_fmsg_named_obj_nest_start(struct devlink_fmsg *fmsg, char *name)
9 {
10 devlink_fmsg_pair_nest_start(fmsg, name);
11 devlink_fmsg_obj_nest_start(fmsg);
12 }
13
mlx5e_health_fmsg_named_obj_nest_end(struct devlink_fmsg * fmsg)14 void mlx5e_health_fmsg_named_obj_nest_end(struct devlink_fmsg *fmsg)
15 {
16 devlink_fmsg_obj_nest_end(fmsg);
17 devlink_fmsg_pair_nest_end(fmsg);
18 }
19
mlx5e_health_cq_diag_fmsg(struct mlx5e_cq * cq,struct devlink_fmsg * fmsg)20 void mlx5e_health_cq_diag_fmsg(struct mlx5e_cq *cq, struct devlink_fmsg *fmsg)
21 {
22 u32 out[MLX5_ST_SZ_DW(query_cq_out)] = {};
23 u8 hw_status;
24 void *cqc;
25
26 mlx5_core_query_cq(cq->mdev, &cq->mcq, out);
27 cqc = MLX5_ADDR_OF(query_cq_out, out, cq_context);
28 hw_status = MLX5_GET(cqc, cqc, status);
29
30 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "CQ");
31 devlink_fmsg_u32_pair_put(fmsg, "cqn", cq->mcq.cqn);
32 devlink_fmsg_u8_pair_put(fmsg, "HW status", hw_status);
33 devlink_fmsg_u32_pair_put(fmsg, "ci", mlx5_cqwq_get_ci(&cq->wq));
34 devlink_fmsg_u32_pair_put(fmsg, "size", mlx5_cqwq_get_size(&cq->wq));
35 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
36 }
37
mlx5e_health_cq_common_diag_fmsg(struct mlx5e_cq * cq,struct devlink_fmsg * fmsg)38 void mlx5e_health_cq_common_diag_fmsg(struct mlx5e_cq *cq, struct devlink_fmsg *fmsg)
39 {
40 u8 cq_log_stride;
41 u32 cq_sz;
42
43 cq_sz = mlx5_cqwq_get_size(&cq->wq);
44 cq_log_stride = mlx5_cqwq_get_log_stride_size(&cq->wq);
45
46 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "CQ");
47 devlink_fmsg_u64_pair_put(fmsg, "stride size", BIT(cq_log_stride));
48 devlink_fmsg_u32_pair_put(fmsg, "size", cq_sz);
49 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
50 }
51
mlx5e_health_eq_diag_fmsg(struct mlx5_eq_comp * eq,struct devlink_fmsg * fmsg)52 void mlx5e_health_eq_diag_fmsg(struct mlx5_eq_comp *eq, struct devlink_fmsg *fmsg)
53 {
54 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "EQ");
55 devlink_fmsg_u8_pair_put(fmsg, "eqn", eq->core.eqn);
56 devlink_fmsg_u32_pair_put(fmsg, "irqn", eq->core.irqn);
57 devlink_fmsg_u32_pair_put(fmsg, "vecidx", eq->core.vecidx);
58 devlink_fmsg_u32_pair_put(fmsg, "ci", eq->core.cons_index);
59 devlink_fmsg_u32_pair_put(fmsg, "size", eq_get_size(&eq->core));
60 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
61 }
62
mlx5e_health_create_reporters(struct mlx5e_priv * priv)63 void mlx5e_health_create_reporters(struct mlx5e_priv *priv)
64 {
65 mlx5e_reporter_tx_create(priv);
66 mlx5e_reporter_rx_create(priv);
67 }
68
mlx5e_health_destroy_reporters(struct mlx5e_priv * priv)69 void mlx5e_health_destroy_reporters(struct mlx5e_priv *priv)
70 {
71 mlx5e_reporter_rx_destroy(priv);
72 mlx5e_reporter_tx_destroy(priv);
73 }
74
mlx5e_health_channels_update(struct mlx5e_priv * priv)75 void mlx5e_health_channels_update(struct mlx5e_priv *priv)
76 {
77 if (priv->tx_reporter)
78 devlink_health_reporter_state_update(priv->tx_reporter,
79 DEVLINK_HEALTH_REPORTER_STATE_HEALTHY);
80 if (priv->rx_reporter)
81 devlink_health_reporter_state_update(priv->rx_reporter,
82 DEVLINK_HEALTH_REPORTER_STATE_HEALTHY);
83 }
84
mlx5e_health_sq_to_ready(struct mlx5_core_dev * mdev,struct net_device * dev,u32 sqn)85 int mlx5e_health_sq_to_ready(struct mlx5_core_dev *mdev, struct net_device *dev, u32 sqn)
86 {
87 struct mlx5e_modify_sq_param msp = {};
88 int err;
89
90 msp.curr_state = MLX5_SQC_STATE_ERR;
91 msp.next_state = MLX5_SQC_STATE_RST;
92
93 err = mlx5e_modify_sq(mdev, sqn, &msp);
94 if (err) {
95 netdev_err(dev, "Failed to move sq 0x%x to reset\n", sqn);
96 return err;
97 }
98
99 memset(&msp, 0, sizeof(msp));
100 msp.curr_state = MLX5_SQC_STATE_RST;
101 msp.next_state = MLX5_SQC_STATE_RDY;
102
103 err = mlx5e_modify_sq(mdev, sqn, &msp);
104 if (err) {
105 netdev_err(dev, "Failed to move sq 0x%x to ready\n", sqn);
106 return err;
107 }
108
109 return 0;
110 }
111
mlx5e_health_recover_channels(struct mlx5e_priv * priv)112 int mlx5e_health_recover_channels(struct mlx5e_priv *priv)
113 {
114 int err = 0;
115
116 rtnl_lock();
117 netdev_lock(priv->netdev);
118 mutex_lock(&priv->state_lock);
119
120 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
121 goto out;
122
123 err = mlx5e_safe_reopen_channels(priv);
124
125 out:
126 mutex_unlock(&priv->state_lock);
127 netdev_unlock(priv->netdev);
128 rtnl_unlock();
129
130 return err;
131 }
132
mlx5e_health_channel_eq_recover(struct net_device * dev,struct mlx5_eq_comp * eq,struct mlx5e_ch_stats * stats)133 int mlx5e_health_channel_eq_recover(struct net_device *dev, struct mlx5_eq_comp *eq,
134 struct mlx5e_ch_stats *stats)
135 {
136 u32 eqe_count;
137
138 netdev_err(dev, "EQ 0x%x: Cons = 0x%x, irqn = 0x%x\n",
139 eq->core.eqn, eq->core.cons_index, eq->core.irqn);
140
141 eqe_count = mlx5_eq_poll_irq_disabled(eq);
142 if (!eqe_count)
143 return -EIO;
144
145 netdev_err(dev, "Recovered %d eqes on EQ 0x%x\n",
146 eqe_count, eq->core.eqn);
147
148 stats->eq_rearm++;
149 return 0;
150 }
151
mlx5e_health_report(struct mlx5e_priv * priv,struct devlink_health_reporter * reporter,char * err_str,struct mlx5e_err_ctx * err_ctx)152 int mlx5e_health_report(struct mlx5e_priv *priv,
153 struct devlink_health_reporter *reporter, char *err_str,
154 struct mlx5e_err_ctx *err_ctx)
155 {
156 netdev_err(priv->netdev, "%s\n", err_str);
157
158 if (!reporter)
159 return err_ctx->recover(err_ctx->ctx);
160
161 return devlink_health_report(reporter, err_str, err_ctx);
162 }
163
164 #define MLX5_HEALTH_DEVLINK_MAX_SIZE 1024
mlx5e_health_rsc_fmsg_binary(struct devlink_fmsg * fmsg,const void * value,u32 value_len)165 static void mlx5e_health_rsc_fmsg_binary(struct devlink_fmsg *fmsg,
166 const void *value, u32 value_len)
167
168 {
169 u32 data_size;
170 u32 offset;
171
172 for (offset = 0; offset < value_len; offset += data_size) {
173 data_size = value_len - offset;
174 if (data_size > MLX5_HEALTH_DEVLINK_MAX_SIZE)
175 data_size = MLX5_HEALTH_DEVLINK_MAX_SIZE;
176 devlink_fmsg_binary_put(fmsg, value + offset, data_size);
177 }
178 }
179
mlx5e_health_rsc_fmsg_dump(struct mlx5e_priv * priv,struct mlx5_rsc_key * key,struct devlink_fmsg * fmsg)180 int mlx5e_health_rsc_fmsg_dump(struct mlx5e_priv *priv, struct mlx5_rsc_key *key,
181 struct devlink_fmsg *fmsg)
182 {
183 struct mlx5_core_dev *mdev = priv->mdev;
184 struct mlx5_rsc_dump_cmd *cmd;
185 int cmd_err, err = 0;
186 struct page *page;
187 int size;
188
189 if (IS_ERR_OR_NULL(mdev->rsc_dump))
190 return -EOPNOTSUPP;
191
192 page = alloc_page(GFP_KERNEL);
193 if (!page)
194 return -ENOMEM;
195
196 devlink_fmsg_binary_pair_nest_start(fmsg, "data");
197
198 cmd = mlx5_rsc_dump_cmd_create(mdev, key);
199 if (IS_ERR(cmd)) {
200 err = PTR_ERR(cmd);
201 goto free_page;
202 }
203
204 do {
205 cmd_err = mlx5_rsc_dump_next(mdev, cmd, page, &size);
206 if (cmd_err < 0) {
207 err = cmd_err;
208 goto destroy_cmd;
209 }
210
211 mlx5e_health_rsc_fmsg_binary(fmsg, page_address(page), size);
212 } while (cmd_err > 0);
213
214 destroy_cmd:
215 mlx5_rsc_dump_cmd_destroy(cmd);
216 devlink_fmsg_binary_pair_nest_end(fmsg);
217 free_page:
218 __free_page(page);
219 return err;
220 }
221
mlx5e_health_queue_dump(struct mlx5e_priv * priv,struct devlink_fmsg * fmsg,int queue_idx,char * lbl)222 void mlx5e_health_queue_dump(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg,
223 int queue_idx, char *lbl)
224 {
225 struct mlx5_rsc_key key = {};
226
227 key.rsc = MLX5_SGMT_TYPE_FULL_QPC;
228 key.index1 = queue_idx;
229 key.size = PAGE_SIZE;
230 key.num_of_obj1 = 1;
231
232 devlink_fmsg_obj_nest_start(fmsg);
233 mlx5e_health_fmsg_named_obj_nest_start(fmsg, lbl);
234 devlink_fmsg_u32_pair_put(fmsg, "index", queue_idx);
235 mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
236 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
237 devlink_fmsg_obj_nest_end(fmsg);
238 }
239