1 // SPDX-License-Identifier: GPL-2.0
2 // Copyright (c) 2019 Mellanox Technologies.
3
4 #include "health.h"
5 #include "params.h"
6
mlx5e_query_rq_state(struct mlx5_core_dev * dev,u32 rqn,u8 * state)7 static int mlx5e_query_rq_state(struct mlx5_core_dev *dev, u32 rqn, u8 *state)
8 {
9 int outlen = MLX5_ST_SZ_BYTES(query_rq_out);
10 void *out;
11 void *rqc;
12 int err;
13
14 out = kvzalloc(outlen, GFP_KERNEL);
15 if (!out)
16 return -ENOMEM;
17
18 err = mlx5_core_query_rq(dev, rqn, out);
19 if (err)
20 goto out;
21
22 rqc = MLX5_ADDR_OF(query_rq_out, out, rq_context);
23 *state = MLX5_GET(rqc, rqc, state);
24
25 out:
26 kvfree(out);
27 return err;
28 }
29
mlx5e_wait_for_icosq_flush(struct mlx5e_icosq * icosq)30 static int mlx5e_wait_for_icosq_flush(struct mlx5e_icosq *icosq)
31 {
32 unsigned long exp_time = jiffies + msecs_to_jiffies(2000);
33
34 while (time_before(jiffies, exp_time)) {
35 if (icosq->cc == icosq->pc)
36 return 0;
37
38 msleep(20);
39 }
40
41 netdev_err(icosq->channel->netdev,
42 "Wait for ICOSQ 0x%x flush timeout (cc = 0x%x, pc = 0x%x)\n",
43 icosq->sqn, icosq->cc, icosq->pc);
44
45 return -ETIMEDOUT;
46 }
47
mlx5e_reset_icosq_cc_pc(struct mlx5e_icosq * icosq)48 static void mlx5e_reset_icosq_cc_pc(struct mlx5e_icosq *icosq)
49 {
50 WARN_ONCE(icosq->cc != icosq->pc, "ICOSQ 0x%x: cc (0x%x) != pc (0x%x)\n",
51 icosq->sqn, icosq->cc, icosq->pc);
52 icosq->cc = 0;
53 icosq->pc = 0;
54 }
55
mlx5e_rx_reporter_err_icosq_cqe_recover(void * ctx)56 static int mlx5e_rx_reporter_err_icosq_cqe_recover(void *ctx)
57 {
58 struct mlx5_core_dev *mdev;
59 struct mlx5e_icosq *icosq;
60 struct net_device *dev;
61 struct mlx5e_rq *rq;
62 u8 state;
63 int err;
64
65 icosq = ctx;
66 rq = &icosq->channel->rq;
67 mdev = icosq->channel->mdev;
68 dev = icosq->channel->netdev;
69 err = mlx5_core_query_sq_state(mdev, icosq->sqn, &state);
70 if (err) {
71 netdev_err(dev, "Failed to query ICOSQ 0x%x state. err = %d\n",
72 icosq->sqn, err);
73 goto out;
74 }
75
76 if (state != MLX5_SQC_STATE_ERR)
77 goto out;
78
79 mlx5e_deactivate_rq(rq);
80 err = mlx5e_wait_for_icosq_flush(icosq);
81 if (err)
82 goto out;
83
84 mlx5e_deactivate_icosq(icosq);
85
86 /* At this point, both the rq and the icosq are disabled */
87
88 err = mlx5e_health_sq_to_ready(icosq->channel, icosq->sqn);
89 if (err)
90 goto out;
91
92 mlx5e_reset_icosq_cc_pc(icosq);
93 mlx5e_free_rx_descs(rq);
94 clear_bit(MLX5E_SQ_STATE_RECOVERING, &icosq->state);
95 mlx5e_activate_icosq(icosq);
96 mlx5e_activate_rq(rq);
97
98 rq->stats->recover++;
99 return 0;
100 out:
101 clear_bit(MLX5E_SQ_STATE_RECOVERING, &icosq->state);
102 return err;
103 }
104
mlx5e_reporter_icosq_cqe_err(struct mlx5e_icosq * icosq)105 void mlx5e_reporter_icosq_cqe_err(struct mlx5e_icosq *icosq)
106 {
107 struct mlx5e_priv *priv = icosq->channel->priv;
108 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
109 struct mlx5e_err_ctx err_ctx = {};
110
111 err_ctx.ctx = icosq;
112 err_ctx.recover = mlx5e_rx_reporter_err_icosq_cqe_recover;
113 sprintf(err_str, "ERR CQE on ICOSQ: 0x%x", icosq->sqn);
114
115 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx);
116 }
117
mlx5e_rq_to_ready(struct mlx5e_rq * rq,int curr_state)118 static int mlx5e_rq_to_ready(struct mlx5e_rq *rq, int curr_state)
119 {
120 struct net_device *dev = rq->netdev;
121 int err;
122
123 err = mlx5e_modify_rq_state(rq, curr_state, MLX5_RQC_STATE_RST);
124 if (err) {
125 netdev_err(dev, "Failed to move rq 0x%x to reset\n", rq->rqn);
126 return err;
127 }
128 err = mlx5e_modify_rq_state(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
129 if (err) {
130 netdev_err(dev, "Failed to move rq 0x%x to ready\n", rq->rqn);
131 return err;
132 }
133
134 return 0;
135 }
136
mlx5e_rx_reporter_err_rq_cqe_recover(void * ctx)137 static int mlx5e_rx_reporter_err_rq_cqe_recover(void *ctx)
138 {
139 struct mlx5_core_dev *mdev;
140 struct net_device *dev;
141 struct mlx5e_rq *rq;
142 u8 state;
143 int err;
144
145 rq = ctx;
146 mdev = rq->mdev;
147 dev = rq->netdev;
148 err = mlx5e_query_rq_state(mdev, rq->rqn, &state);
149 if (err) {
150 netdev_err(dev, "Failed to query RQ 0x%x state. err = %d\n",
151 rq->rqn, err);
152 goto out;
153 }
154
155 if (state != MLX5_RQC_STATE_ERR)
156 goto out;
157
158 mlx5e_deactivate_rq(rq);
159 mlx5e_free_rx_descs(rq);
160
161 err = mlx5e_rq_to_ready(rq, MLX5_RQC_STATE_ERR);
162 if (err)
163 goto out;
164
165 clear_bit(MLX5E_RQ_STATE_RECOVERING, &rq->state);
166 mlx5e_activate_rq(rq);
167 rq->stats->recover++;
168 return 0;
169 out:
170 clear_bit(MLX5E_RQ_STATE_RECOVERING, &rq->state);
171 return err;
172 }
173
mlx5e_reporter_rq_cqe_err(struct mlx5e_rq * rq)174 void mlx5e_reporter_rq_cqe_err(struct mlx5e_rq *rq)
175 {
176 struct mlx5e_priv *priv = rq->channel->priv;
177 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
178 struct mlx5e_err_ctx err_ctx = {};
179
180 err_ctx.ctx = rq;
181 err_ctx.recover = mlx5e_rx_reporter_err_rq_cqe_recover;
182 sprintf(err_str, "ERR CQE on RQ: 0x%x", rq->rqn);
183
184 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx);
185 }
186
mlx5e_rx_reporter_timeout_recover(void * ctx)187 static int mlx5e_rx_reporter_timeout_recover(void *ctx)
188 {
189 struct mlx5e_icosq *icosq;
190 struct mlx5_eq_comp *eq;
191 struct mlx5e_rq *rq;
192 int err;
193
194 rq = ctx;
195 icosq = &rq->channel->icosq;
196 eq = rq->cq.mcq.eq;
197 err = mlx5e_health_channel_eq_recover(eq, rq->channel);
198 if (err)
199 clear_bit(MLX5E_SQ_STATE_ENABLED, &icosq->state);
200
201 return err;
202 }
203
mlx5e_reporter_rx_timeout(struct mlx5e_rq * rq)204 void mlx5e_reporter_rx_timeout(struct mlx5e_rq *rq)
205 {
206 struct mlx5e_icosq *icosq = &rq->channel->icosq;
207 struct mlx5e_priv *priv = rq->channel->priv;
208 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
209 struct mlx5e_err_ctx err_ctx = {};
210
211 err_ctx.ctx = rq;
212 err_ctx.recover = mlx5e_rx_reporter_timeout_recover;
213 sprintf(err_str, "RX timeout on channel: %d, ICOSQ: 0x%x RQ: 0x%x, CQ: 0x%x\n",
214 icosq->channel->ix, icosq->sqn, rq->rqn, rq->cq.mcq.cqn);
215
216 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx);
217 }
218
mlx5e_rx_reporter_recover_from_ctx(struct mlx5e_err_ctx * err_ctx)219 static int mlx5e_rx_reporter_recover_from_ctx(struct mlx5e_err_ctx *err_ctx)
220 {
221 return err_ctx->recover(err_ctx->ctx);
222 }
223
mlx5e_rx_reporter_recover(struct devlink_health_reporter * reporter,void * context)224 static int mlx5e_rx_reporter_recover(struct devlink_health_reporter *reporter,
225 void *context)
226 {
227 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
228 struct mlx5e_err_ctx *err_ctx = context;
229
230 return err_ctx ? mlx5e_rx_reporter_recover_from_ctx(err_ctx) :
231 mlx5e_health_recover_channels(priv);
232 }
233
mlx5e_rx_reporter_build_diagnose_output(struct mlx5e_rq * rq,struct devlink_fmsg * fmsg)234 static int mlx5e_rx_reporter_build_diagnose_output(struct mlx5e_rq *rq,
235 struct devlink_fmsg *fmsg)
236 {
237 struct mlx5e_priv *priv = rq->channel->priv;
238 struct mlx5e_params *params;
239 struct mlx5e_icosq *icosq;
240 u8 icosq_hw_state;
241 int wqes_sz;
242 u8 hw_state;
243 u16 wq_head;
244 int err;
245
246 params = &priv->channels.params;
247 icosq = &rq->channel->icosq;
248 err = mlx5e_query_rq_state(priv->mdev, rq->rqn, &hw_state);
249 if (err)
250 return err;
251
252 err = mlx5_core_query_sq_state(priv->mdev, icosq->sqn, &icosq_hw_state);
253 if (err)
254 return err;
255
256 wqes_sz = mlx5e_rqwq_get_cur_sz(rq);
257 wq_head = params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ ?
258 rq->mpwqe.wq.head : mlx5_wq_cyc_get_head(&rq->wqe.wq);
259
260 err = devlink_fmsg_obj_nest_start(fmsg);
261 if (err)
262 return err;
263
264 err = devlink_fmsg_u32_pair_put(fmsg, "channel ix", rq->channel->ix);
265 if (err)
266 return err;
267
268 err = devlink_fmsg_u32_pair_put(fmsg, "rqn", rq->rqn);
269 if (err)
270 return err;
271
272 err = devlink_fmsg_u8_pair_put(fmsg, "HW state", hw_state);
273 if (err)
274 return err;
275
276 err = devlink_fmsg_u8_pair_put(fmsg, "SW state", rq->state);
277 if (err)
278 return err;
279
280 err = devlink_fmsg_u32_pair_put(fmsg, "posted WQEs", wqes_sz);
281 if (err)
282 return err;
283
284 err = devlink_fmsg_u32_pair_put(fmsg, "cc", wq_head);
285 if (err)
286 return err;
287
288 err = devlink_fmsg_u8_pair_put(fmsg, "ICOSQ HW state", icosq_hw_state);
289 if (err)
290 return err;
291
292 err = mlx5e_reporter_cq_diagnose(&rq->cq, fmsg);
293 if (err)
294 return err;
295
296 err = devlink_fmsg_obj_nest_end(fmsg);
297 if (err)
298 return err;
299
300 return 0;
301 }
302
mlx5e_rx_reporter_diagnose(struct devlink_health_reporter * reporter,struct devlink_fmsg * fmsg)303 static int mlx5e_rx_reporter_diagnose(struct devlink_health_reporter *reporter,
304 struct devlink_fmsg *fmsg)
305 {
306 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
307 struct mlx5e_params *params = &priv->channels.params;
308 struct mlx5e_rq *generic_rq;
309 u32 rq_stride, rq_sz;
310 int i, err = 0;
311
312 mutex_lock(&priv->state_lock);
313
314 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
315 goto unlock;
316
317 generic_rq = &priv->channels.c[0]->rq;
318 rq_sz = mlx5e_rqwq_get_size(generic_rq);
319 rq_stride = BIT(mlx5e_mpwqe_get_log_stride_size(priv->mdev, params, NULL));
320
321 err = mlx5e_reporter_named_obj_nest_start(fmsg, "Common config");
322 if (err)
323 goto unlock;
324
325 err = mlx5e_reporter_named_obj_nest_start(fmsg, "RQ");
326 if (err)
327 goto unlock;
328
329 err = devlink_fmsg_u8_pair_put(fmsg, "type", params->rq_wq_type);
330 if (err)
331 goto unlock;
332
333 err = devlink_fmsg_u64_pair_put(fmsg, "stride size", rq_stride);
334 if (err)
335 goto unlock;
336
337 err = devlink_fmsg_u32_pair_put(fmsg, "size", rq_sz);
338 if (err)
339 goto unlock;
340
341 err = mlx5e_reporter_named_obj_nest_end(fmsg);
342 if (err)
343 goto unlock;
344
345 err = mlx5e_reporter_cq_common_diagnose(&generic_rq->cq, fmsg);
346 if (err)
347 goto unlock;
348
349 err = mlx5e_reporter_named_obj_nest_end(fmsg);
350 if (err)
351 goto unlock;
352
353 err = devlink_fmsg_arr_pair_nest_start(fmsg, "RQs");
354 if (err)
355 goto unlock;
356
357 for (i = 0; i < priv->channels.num; i++) {
358 struct mlx5e_rq *rq = &priv->channels.c[i]->rq;
359
360 err = mlx5e_rx_reporter_build_diagnose_output(rq, fmsg);
361 if (err)
362 goto unlock;
363 }
364 err = devlink_fmsg_arr_pair_nest_end(fmsg);
365 if (err)
366 goto unlock;
367 unlock:
368 mutex_unlock(&priv->state_lock);
369 return err;
370 }
371
372 static const struct devlink_health_reporter_ops mlx5_rx_reporter_ops = {
373 .name = "rx",
374 .recover = mlx5e_rx_reporter_recover,
375 .diagnose = mlx5e_rx_reporter_diagnose,
376 };
377
378 #define MLX5E_REPORTER_RX_GRACEFUL_PERIOD 500
379
mlx5e_reporter_rx_create(struct mlx5e_priv * priv)380 int mlx5e_reporter_rx_create(struct mlx5e_priv *priv)
381 {
382 struct devlink *devlink = priv_to_devlink(priv->mdev);
383 struct devlink_health_reporter *reporter;
384
385 reporter = devlink_health_reporter_create(devlink,
386 &mlx5_rx_reporter_ops,
387 MLX5E_REPORTER_RX_GRACEFUL_PERIOD,
388 true, priv);
389 if (IS_ERR(reporter)) {
390 netdev_warn(priv->netdev, "Failed to create rx reporter, err = %ld\n",
391 PTR_ERR(reporter));
392 return PTR_ERR(reporter);
393 }
394 priv->rx_reporter = reporter;
395 return 0;
396 }
397
mlx5e_reporter_rx_destroy(struct mlx5e_priv * priv)398 void mlx5e_reporter_rx_destroy(struct mlx5e_priv *priv)
399 {
400 if (!priv->rx_reporter)
401 return;
402
403 devlink_health_reporter_destroy(priv->rx_reporter);
404 }
405
