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 int mlx5e_health_fmsg_named_obj_nest_start(struct devlink_fmsg *fmsg, char *name)
9 {
10 int err;
11
12 err = devlink_fmsg_pair_nest_start(fmsg, name);
13 if (err)
14 return err;
15
16 err = devlink_fmsg_obj_nest_start(fmsg);
17 if (err)
18 return err;
19
20 return 0;
21 }
22
mlx5e_health_fmsg_named_obj_nest_end(struct devlink_fmsg * fmsg)23 int mlx5e_health_fmsg_named_obj_nest_end(struct devlink_fmsg *fmsg)
24 {
25 int err;
26
27 err = devlink_fmsg_obj_nest_end(fmsg);
28 if (err)
29 return err;
30
31 err = devlink_fmsg_pair_nest_end(fmsg);
32 if (err)
33 return err;
34
35 return 0;
36 }
37
mlx5e_health_cq_diag_fmsg(struct mlx5e_cq * cq,struct devlink_fmsg * fmsg)38 int mlx5e_health_cq_diag_fmsg(struct mlx5e_cq *cq, struct devlink_fmsg *fmsg)
39 {
40 struct mlx5e_priv *priv = cq->channel->priv;
41 u32 out[MLX5_ST_SZ_DW(query_cq_out)] = {};
42 u8 hw_status;
43 void *cqc;
44 int err;
45
46 err = mlx5_core_query_cq(priv->mdev, &cq->mcq, out);
47 if (err)
48 return err;
49
50 cqc = MLX5_ADDR_OF(query_cq_out, out, cq_context);
51 hw_status = MLX5_GET(cqc, cqc, status);
52
53 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "CQ");
54 if (err)
55 return err;
56
57 err = devlink_fmsg_u32_pair_put(fmsg, "cqn", cq->mcq.cqn);
58 if (err)
59 return err;
60
61 err = devlink_fmsg_u8_pair_put(fmsg, "HW status", hw_status);
62 if (err)
63 return err;
64
65 err = devlink_fmsg_u32_pair_put(fmsg, "ci", mlx5_cqwq_get_ci(&cq->wq));
66 if (err)
67 return err;
68
69 err = devlink_fmsg_u32_pair_put(fmsg, "size", mlx5_cqwq_get_size(&cq->wq));
70 if (err)
71 return err;
72
73 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
74 if (err)
75 return err;
76
77 return 0;
78 }
79
mlx5e_health_cq_common_diag_fmsg(struct mlx5e_cq * cq,struct devlink_fmsg * fmsg)80 int mlx5e_health_cq_common_diag_fmsg(struct mlx5e_cq *cq, struct devlink_fmsg *fmsg)
81 {
82 u8 cq_log_stride;
83 u32 cq_sz;
84 int err;
85
86 cq_sz = mlx5_cqwq_get_size(&cq->wq);
87 cq_log_stride = mlx5_cqwq_get_log_stride_size(&cq->wq);
88
89 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "CQ");
90 if (err)
91 return err;
92
93 err = devlink_fmsg_u64_pair_put(fmsg, "stride size", BIT(cq_log_stride));
94 if (err)
95 return err;
96
97 err = devlink_fmsg_u32_pair_put(fmsg, "size", cq_sz);
98 if (err)
99 return err;
100
101 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
102 if (err)
103 return err;
104
105 return 0;
106 }
107
mlx5e_health_eq_diag_fmsg(struct mlx5_eq_comp * eq,struct devlink_fmsg * fmsg)108 int mlx5e_health_eq_diag_fmsg(struct mlx5_eq_comp *eq, struct devlink_fmsg *fmsg)
109 {
110 int err;
111
112 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "EQ");
113 if (err)
114 return err;
115
116 err = devlink_fmsg_u8_pair_put(fmsg, "eqn", eq->core.eqn);
117 if (err)
118 return err;
119
120 err = devlink_fmsg_u32_pair_put(fmsg, "irqn", eq->core.irqn);
121 if (err)
122 return err;
123
124 err = devlink_fmsg_u32_pair_put(fmsg, "vecidx", eq->core.vecidx);
125 if (err)
126 return err;
127
128 err = devlink_fmsg_u32_pair_put(fmsg, "ci", eq->core.cons_index);
129 if (err)
130 return err;
131
132 err = devlink_fmsg_u32_pair_put(fmsg, "size", eq->core.nent);
133 if (err)
134 return err;
135
136 return mlx5e_health_fmsg_named_obj_nest_end(fmsg);
137 }
138
mlx5e_health_create_reporters(struct mlx5e_priv * priv)139 void mlx5e_health_create_reporters(struct mlx5e_priv *priv)
140 {
141 mlx5e_reporter_tx_create(priv);
142 mlx5e_reporter_rx_create(priv);
143 }
144
mlx5e_health_destroy_reporters(struct mlx5e_priv * priv)145 void mlx5e_health_destroy_reporters(struct mlx5e_priv *priv)
146 {
147 mlx5e_reporter_rx_destroy(priv);
148 mlx5e_reporter_tx_destroy(priv);
149 }
150
mlx5e_health_channels_update(struct mlx5e_priv * priv)151 void mlx5e_health_channels_update(struct mlx5e_priv *priv)
152 {
153 if (priv->tx_reporter)
154 devlink_health_reporter_state_update(priv->tx_reporter,
155 DEVLINK_HEALTH_REPORTER_STATE_HEALTHY);
156 if (priv->rx_reporter)
157 devlink_health_reporter_state_update(priv->rx_reporter,
158 DEVLINK_HEALTH_REPORTER_STATE_HEALTHY);
159 }
160
mlx5e_health_sq_to_ready(struct mlx5e_channel * channel,u32 sqn)161 int mlx5e_health_sq_to_ready(struct mlx5e_channel *channel, u32 sqn)
162 {
163 struct mlx5_core_dev *mdev = channel->mdev;
164 struct net_device *dev = channel->netdev;
165 struct mlx5e_modify_sq_param msp = {};
166 int err;
167
168 msp.curr_state = MLX5_SQC_STATE_ERR;
169 msp.next_state = MLX5_SQC_STATE_RST;
170
171 err = mlx5e_modify_sq(mdev, sqn, &msp);
172 if (err) {
173 netdev_err(dev, "Failed to move sq 0x%x to reset\n", sqn);
174 return err;
175 }
176
177 memset(&msp, 0, sizeof(msp));
178 msp.curr_state = MLX5_SQC_STATE_RST;
179 msp.next_state = MLX5_SQC_STATE_RDY;
180
181 err = mlx5e_modify_sq(mdev, sqn, &msp);
182 if (err) {
183 netdev_err(dev, "Failed to move sq 0x%x to ready\n", sqn);
184 return err;
185 }
186
187 return 0;
188 }
189
mlx5e_health_recover_channels(struct mlx5e_priv * priv)190 int mlx5e_health_recover_channels(struct mlx5e_priv *priv)
191 {
192 int err = 0;
193
194 rtnl_lock();
195 mutex_lock(&priv->state_lock);
196
197 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
198 goto out;
199
200 err = mlx5e_safe_reopen_channels(priv);
201
202 out:
203 mutex_unlock(&priv->state_lock);
204 rtnl_unlock();
205
206 return err;
207 }
208
mlx5e_health_channel_eq_recover(struct mlx5_eq_comp * eq,struct mlx5e_channel * channel)209 int mlx5e_health_channel_eq_recover(struct mlx5_eq_comp *eq, struct mlx5e_channel *channel)
210 {
211 u32 eqe_count;
212
213 netdev_err(channel->netdev, "EQ 0x%x: Cons = 0x%x, irqn = 0x%x\n",
214 eq->core.eqn, eq->core.cons_index, eq->core.irqn);
215
216 eqe_count = mlx5_eq_poll_irq_disabled(eq);
217 if (!eqe_count)
218 return -EIO;
219
220 netdev_err(channel->netdev, "Recovered %d eqes on EQ 0x%x\n",
221 eqe_count, eq->core.eqn);
222
223 channel->stats->eq_rearm++;
224 return 0;
225 }
226
mlx5e_health_report(struct mlx5e_priv * priv,struct devlink_health_reporter * reporter,char * err_str,struct mlx5e_err_ctx * err_ctx)227 int mlx5e_health_report(struct mlx5e_priv *priv,
228 struct devlink_health_reporter *reporter, char *err_str,
229 struct mlx5e_err_ctx *err_ctx)
230 {
231 netdev_err(priv->netdev, "%s\n", err_str);
232
233 if (!reporter)
234 return err_ctx->recover(err_ctx->ctx);
235
236 return devlink_health_report(reporter, err_str, err_ctx);
237 }
238
239 #define MLX5_HEALTH_DEVLINK_MAX_SIZE 1024
mlx5e_health_rsc_fmsg_binary(struct devlink_fmsg * fmsg,const void * value,u32 value_len)240 static int mlx5e_health_rsc_fmsg_binary(struct devlink_fmsg *fmsg,
241 const void *value, u32 value_len)
242
243 {
244 u32 data_size;
245 int err = 0;
246 u32 offset;
247
248 for (offset = 0; offset < value_len; offset += data_size) {
249 data_size = value_len - offset;
250 if (data_size > MLX5_HEALTH_DEVLINK_MAX_SIZE)
251 data_size = MLX5_HEALTH_DEVLINK_MAX_SIZE;
252 err = devlink_fmsg_binary_put(fmsg, value + offset, data_size);
253 if (err)
254 break;
255 }
256 return err;
257 }
258
mlx5e_health_rsc_fmsg_dump(struct mlx5e_priv * priv,struct mlx5_rsc_key * key,struct devlink_fmsg * fmsg)259 int mlx5e_health_rsc_fmsg_dump(struct mlx5e_priv *priv, struct mlx5_rsc_key *key,
260 struct devlink_fmsg *fmsg)
261 {
262 struct mlx5_core_dev *mdev = priv->mdev;
263 struct mlx5_rsc_dump_cmd *cmd;
264 struct page *page;
265 int cmd_err, err;
266 int end_err;
267 int size;
268
269 if (IS_ERR_OR_NULL(mdev->rsc_dump))
270 return -EOPNOTSUPP;
271
272 page = alloc_page(GFP_KERNEL);
273 if (!page)
274 return -ENOMEM;
275
276 err = devlink_fmsg_binary_pair_nest_start(fmsg, "data");
277 if (err)
278 goto free_page;
279
280 cmd = mlx5_rsc_dump_cmd_create(mdev, key);
281 if (IS_ERR(cmd)) {
282 err = PTR_ERR(cmd);
283 goto free_page;
284 }
285
286 do {
287 cmd_err = mlx5_rsc_dump_next(mdev, cmd, page, &size);
288 if (cmd_err < 0) {
289 err = cmd_err;
290 goto destroy_cmd;
291 }
292
293 err = mlx5e_health_rsc_fmsg_binary(fmsg, page_address(page), size);
294 if (err)
295 goto destroy_cmd;
296
297 } while (cmd_err > 0);
298
299 destroy_cmd:
300 mlx5_rsc_dump_cmd_destroy(cmd);
301 end_err = devlink_fmsg_binary_pair_nest_end(fmsg);
302 if (end_err)
303 err = end_err;
304 free_page:
305 __free_page(page);
306 return err;
307 }
308
mlx5e_health_queue_dump(struct mlx5e_priv * priv,struct devlink_fmsg * fmsg,int queue_idx,char * lbl)309 int mlx5e_health_queue_dump(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg,
310 int queue_idx, char *lbl)
311 {
312 struct mlx5_rsc_key key = {};
313 int err;
314
315 key.rsc = MLX5_SGMT_TYPE_FULL_QPC;
316 key.index1 = queue_idx;
317 key.size = PAGE_SIZE;
318 key.num_of_obj1 = 1;
319
320 err = devlink_fmsg_obj_nest_start(fmsg);
321 if (err)
322 return err;
323
324 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, lbl);
325 if (err)
326 return err;
327
328 err = devlink_fmsg_u32_pair_put(fmsg, "index", queue_idx);
329 if (err)
330 return err;
331
332 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
333 if (err)
334 return err;
335
336 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
337 if (err)
338 return err;
339
340 return devlink_fmsg_obj_nest_end(fmsg);
341 }
342