1 // SPDX-License-Identifier: GPL-2.0
2 // Copyright (c) 2019 Mellanox Technologies.
3
4 #include "health.h"
5 #include "params.h"
6 #include "txrx.h"
7 #include "devlink.h"
8 #include "ptp.h"
9 #include "lib/tout.h"
10
11 /* Keep this string array consistent with the MLX5E_RQ_STATE_* enums in en.h */
12 static const char * const rq_sw_state_type_name[] = {
13 [MLX5E_RQ_STATE_ENABLED] = "enabled",
14 [MLX5E_RQ_STATE_RECOVERING] = "recovering",
15 [MLX5E_RQ_STATE_DIM] = "dim",
16 [MLX5E_RQ_STATE_NO_CSUM_COMPLETE] = "no_csum_complete",
17 [MLX5E_RQ_STATE_CSUM_FULL] = "csum_full",
18 [MLX5E_RQ_STATE_MINI_CQE_HW_STRIDX] = "mini_cqe_hw_stridx",
19 [MLX5E_RQ_STATE_SHAMPO] = "shampo",
20 [MLX5E_RQ_STATE_MINI_CQE_ENHANCED] = "mini_cqe_enhanced",
21 [MLX5E_RQ_STATE_XSK] = "xsk",
22 };
23
mlx5e_query_rq_state(struct mlx5_core_dev * dev,u32 rqn,u8 * state)24 static int mlx5e_query_rq_state(struct mlx5_core_dev *dev, u32 rqn, u8 *state)
25 {
26 int outlen = MLX5_ST_SZ_BYTES(query_rq_out);
27 void *out;
28 void *rqc;
29 int err;
30
31 out = kvzalloc(outlen, GFP_KERNEL);
32 if (!out)
33 return -ENOMEM;
34
35 err = mlx5_core_query_rq(dev, rqn, out);
36 if (err)
37 goto out;
38
39 rqc = MLX5_ADDR_OF(query_rq_out, out, rq_context);
40 *state = MLX5_GET(rqc, rqc, state);
41
42 out:
43 kvfree(out);
44 return err;
45 }
46
mlx5e_wait_for_icosq_flush(struct mlx5e_icosq * icosq)47 static int mlx5e_wait_for_icosq_flush(struct mlx5e_icosq *icosq)
48 {
49 struct mlx5_core_dev *dev = icosq->channel->mdev;
50 unsigned long exp_time;
51
52 exp_time = jiffies + msecs_to_jiffies(mlx5_tout_ms(dev, FLUSH_ON_ERROR));
53
54 while (time_before(jiffies, exp_time)) {
55 if (icosq->cc == icosq->pc)
56 return 0;
57
58 msleep(20);
59 }
60
61 netdev_err(icosq->channel->netdev,
62 "Wait for ICOSQ 0x%x flush timeout (cc = 0x%x, pc = 0x%x)\n",
63 icosq->sqn, icosq->cc, icosq->pc);
64
65 return -ETIMEDOUT;
66 }
67
mlx5e_reset_icosq_cc_pc(struct mlx5e_icosq * icosq)68 static void mlx5e_reset_icosq_cc_pc(struct mlx5e_icosq *icosq)
69 {
70 WARN_ONCE(icosq->cc != icosq->pc, "ICOSQ 0x%x: cc (0x%x) != pc (0x%x)\n",
71 icosq->sqn, icosq->cc, icosq->pc);
72 icosq->cc = 0;
73 icosq->pc = 0;
74 }
75
mlx5e_rx_reporter_err_icosq_cqe_recover(void * ctx)76 static int mlx5e_rx_reporter_err_icosq_cqe_recover(void *ctx)
77 {
78 struct mlx5e_rq *xskrq = NULL;
79 struct mlx5_core_dev *mdev;
80 struct mlx5e_icosq *icosq;
81 struct net_device *dev;
82 struct mlx5e_rq *rq;
83 u8 state;
84 int err;
85
86 icosq = ctx;
87
88 mutex_lock(&icosq->channel->icosq_recovery_lock);
89
90 /* mlx5e_close_rq cancels this work before RQ and ICOSQ are killed. */
91 rq = &icosq->channel->rq;
92 if (test_bit(MLX5E_RQ_STATE_ENABLED, &icosq->channel->xskrq.state))
93 xskrq = &icosq->channel->xskrq;
94 mdev = icosq->channel->mdev;
95 dev = icosq->channel->netdev;
96 err = mlx5_core_query_sq_state(mdev, icosq->sqn, &state);
97 if (err) {
98 netdev_err(dev, "Failed to query ICOSQ 0x%x state. err = %d\n",
99 icosq->sqn, err);
100 goto out;
101 }
102
103 if (state != MLX5_SQC_STATE_ERR)
104 goto out;
105
106 mlx5e_deactivate_rq(rq);
107 if (xskrq)
108 mlx5e_deactivate_rq(xskrq);
109
110 err = mlx5e_wait_for_icosq_flush(icosq);
111 if (err)
112 goto out;
113
114 mlx5e_deactivate_icosq(icosq);
115
116 /* At this point, both the rq and the icosq are disabled */
117
118 err = mlx5e_health_sq_to_ready(mdev, dev, icosq->sqn);
119 if (err)
120 goto out;
121
122 mlx5e_reset_icosq_cc_pc(icosq);
123
124 mlx5e_free_rx_missing_descs(rq);
125 if (xskrq)
126 mlx5e_free_rx_missing_descs(xskrq);
127
128 clear_bit(MLX5E_SQ_STATE_RECOVERING, &icosq->state);
129 mlx5e_activate_icosq(icosq);
130
131 mlx5e_activate_rq(rq);
132 rq->stats->recover++;
133
134 if (xskrq) {
135 mlx5e_activate_rq(xskrq);
136 xskrq->stats->recover++;
137 }
138
139 mlx5e_trigger_napi_icosq(icosq->channel);
140
141 mutex_unlock(&icosq->channel->icosq_recovery_lock);
142
143 return 0;
144 out:
145 clear_bit(MLX5E_SQ_STATE_RECOVERING, &icosq->state);
146 mutex_unlock(&icosq->channel->icosq_recovery_lock);
147 return err;
148 }
149
mlx5e_rx_reporter_err_rq_cqe_recover(void * ctx)150 static int mlx5e_rx_reporter_err_rq_cqe_recover(void *ctx)
151 {
152 struct mlx5e_rq *rq = ctx;
153 int err;
154
155 mlx5e_deactivate_rq(rq);
156 err = mlx5e_flush_rq(rq, MLX5_RQC_STATE_ERR);
157 clear_bit(MLX5E_RQ_STATE_RECOVERING, &rq->state);
158 if (err)
159 return err;
160
161 mlx5e_activate_rq(rq);
162 rq->stats->recover++;
163 if (rq->channel)
164 mlx5e_trigger_napi_icosq(rq->channel);
165 else
166 mlx5e_trigger_napi_sched(rq->cq.napi);
167 return 0;
168 }
169
mlx5e_rx_reporter_timeout_recover(void * ctx)170 static int mlx5e_rx_reporter_timeout_recover(void *ctx)
171 {
172 struct mlx5_eq_comp *eq;
173 struct mlx5e_rq *rq;
174 int err;
175
176 rq = ctx;
177 eq = rq->cq.mcq.eq;
178
179 err = mlx5e_health_channel_eq_recover(rq->netdev, eq, rq->cq.ch_stats);
180 if (err && rq->icosq)
181 clear_bit(MLX5E_SQ_STATE_ENABLED, &rq->icosq->state);
182
183 return err;
184 }
185
mlx5e_rx_reporter_recover_from_ctx(struct mlx5e_err_ctx * err_ctx)186 static int mlx5e_rx_reporter_recover_from_ctx(struct mlx5e_err_ctx *err_ctx)
187 {
188 return err_ctx->recover(err_ctx->ctx);
189 }
190
mlx5e_rx_reporter_recover(struct devlink_health_reporter * reporter,void * context,struct netlink_ext_ack * extack)191 static int mlx5e_rx_reporter_recover(struct devlink_health_reporter *reporter,
192 void *context,
193 struct netlink_ext_ack *extack)
194 {
195 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
196 struct mlx5e_err_ctx *err_ctx = context;
197
198 return err_ctx ? mlx5e_rx_reporter_recover_from_ctx(err_ctx) :
199 mlx5e_health_recover_channels(priv);
200 }
201
mlx5e_reporter_icosq_diagnose(struct mlx5e_icosq * icosq,u8 hw_state,struct devlink_fmsg * fmsg)202 static void mlx5e_reporter_icosq_diagnose(struct mlx5e_icosq *icosq, u8 hw_state,
203 struct devlink_fmsg *fmsg)
204 {
205 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "ICOSQ");
206 devlink_fmsg_u32_pair_put(fmsg, "sqn", icosq->sqn);
207 devlink_fmsg_u8_pair_put(fmsg, "HW state", hw_state);
208 devlink_fmsg_u32_pair_put(fmsg, "cc", icosq->cc);
209 devlink_fmsg_u32_pair_put(fmsg, "pc", icosq->pc);
210 devlink_fmsg_u32_pair_put(fmsg, "WQE size", mlx5_wq_cyc_get_size(&icosq->wq));
211
212 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "CQ");
213 devlink_fmsg_u32_pair_put(fmsg, "cqn", icosq->cq.mcq.cqn);
214 devlink_fmsg_u32_pair_put(fmsg, "cc", icosq->cq.wq.cc);
215 devlink_fmsg_u32_pair_put(fmsg, "size", mlx5_cqwq_get_size(&icosq->cq.wq));
216 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
217
218 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
219 }
220
mlx5e_health_rq_put_sw_state(struct devlink_fmsg * fmsg,struct mlx5e_rq * rq)221 static void mlx5e_health_rq_put_sw_state(struct devlink_fmsg *fmsg, struct mlx5e_rq *rq)
222 {
223 int i;
224
225 BUILD_BUG_ON_MSG(ARRAY_SIZE(rq_sw_state_type_name) != MLX5E_NUM_RQ_STATES,
226 "rq_sw_state_type_name string array must be consistent with MLX5E_RQ_STATE_* enum in en.h");
227 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "SW State");
228
229 for (i = 0; i < ARRAY_SIZE(rq_sw_state_type_name); ++i)
230 devlink_fmsg_u32_pair_put(fmsg, rq_sw_state_type_name[i],
231 test_bit(i, &rq->state));
232
233 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
234 }
235
236 static int
mlx5e_rx_reporter_build_diagnose_output_rq_common(struct mlx5e_rq * rq,struct devlink_fmsg * fmsg)237 mlx5e_rx_reporter_build_diagnose_output_rq_common(struct mlx5e_rq *rq,
238 struct devlink_fmsg *fmsg)
239 {
240 u16 wqe_counter;
241 int wqes_sz;
242 u8 hw_state;
243 u16 wq_head;
244 int err;
245
246 err = mlx5e_query_rq_state(rq->mdev, rq->rqn, &hw_state);
247 if (err)
248 return err;
249
250 wqes_sz = mlx5e_rqwq_get_cur_sz(rq);
251 wq_head = mlx5e_rqwq_get_head(rq);
252 wqe_counter = mlx5e_rqwq_get_wqe_counter(rq);
253
254 devlink_fmsg_u32_pair_put(fmsg, "rqn", rq->rqn);
255 devlink_fmsg_u8_pair_put(fmsg, "HW state", hw_state);
256 devlink_fmsg_u32_pair_put(fmsg, "WQE counter", wqe_counter);
257 devlink_fmsg_u32_pair_put(fmsg, "posted WQEs", wqes_sz);
258 devlink_fmsg_u32_pair_put(fmsg, "cc", wq_head);
259 mlx5e_health_rq_put_sw_state(fmsg, rq);
260 mlx5e_health_cq_diag_fmsg(&rq->cq, fmsg);
261 mlx5e_health_eq_diag_fmsg(rq->cq.mcq.eq, fmsg);
262
263 if (rq->icosq) {
264 struct mlx5e_icosq *icosq = rq->icosq;
265 u8 icosq_hw_state;
266 int err;
267
268 err = mlx5_core_query_sq_state(rq->mdev, icosq->sqn, &icosq_hw_state);
269 if (err)
270 return err;
271
272 mlx5e_reporter_icosq_diagnose(icosq, icosq_hw_state, fmsg);
273 }
274
275 return 0;
276 }
277
mlx5e_rx_reporter_build_diagnose_output(struct mlx5e_rq * rq,struct devlink_fmsg * fmsg)278 static void mlx5e_rx_reporter_build_diagnose_output(struct mlx5e_rq *rq,
279 struct devlink_fmsg *fmsg)
280 {
281 devlink_fmsg_obj_nest_start(fmsg);
282 devlink_fmsg_u32_pair_put(fmsg, "channel ix", rq->ix);
283 mlx5e_rx_reporter_build_diagnose_output_rq_common(rq, fmsg);
284 devlink_fmsg_obj_nest_end(fmsg);
285 }
286
mlx5e_rx_reporter_diagnose_generic_rq(struct mlx5e_rq * rq,struct devlink_fmsg * fmsg)287 static void mlx5e_rx_reporter_diagnose_generic_rq(struct mlx5e_rq *rq,
288 struct devlink_fmsg *fmsg)
289 {
290 struct mlx5e_priv *priv = rq->priv;
291 struct mlx5e_params *params;
292 u32 rq_stride, rq_sz;
293 bool real_time;
294
295 params = &priv->channels.params;
296 rq_sz = mlx5e_rqwq_get_size(rq);
297 real_time = mlx5_is_real_time_rq(rq->mdev);
298 rq_stride = BIT(mlx5e_mpwqe_get_log_stride_size(rq->mdev, params, NULL));
299
300 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RQ");
301 devlink_fmsg_u8_pair_put(fmsg, "type", params->rq_wq_type);
302 devlink_fmsg_u64_pair_put(fmsg, "stride size", rq_stride);
303 devlink_fmsg_u32_pair_put(fmsg, "size", rq_sz);
304 devlink_fmsg_string_pair_put(fmsg, "ts_format", real_time ? "RT" : "FRC");
305 mlx5e_health_cq_common_diag_fmsg(&rq->cq, fmsg);
306 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
307 }
308
309 static void
mlx5e_rx_reporter_diagnose_common_ptp_config(struct mlx5e_priv * priv,struct mlx5e_ptp * ptp_ch,struct devlink_fmsg * fmsg)310 mlx5e_rx_reporter_diagnose_common_ptp_config(struct mlx5e_priv *priv, struct mlx5e_ptp *ptp_ch,
311 struct devlink_fmsg *fmsg)
312 {
313 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "PTP");
314 devlink_fmsg_u32_pair_put(fmsg, "filter_type", priv->tstamp.rx_filter);
315 mlx5e_rx_reporter_diagnose_generic_rq(&ptp_ch->rq, fmsg);
316 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
317 }
318
319 static void
mlx5e_rx_reporter_diagnose_common_config(struct mlx5e_priv * priv,struct devlink_fmsg * fmsg)320 mlx5e_rx_reporter_diagnose_common_config(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg)
321 {
322 struct mlx5e_rq *generic_rq = &priv->channels.c[0]->rq;
323 struct mlx5e_ptp *ptp_ch = priv->channels.ptp;
324
325 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "Common config");
326 mlx5e_rx_reporter_diagnose_generic_rq(generic_rq, fmsg);
327 if (ptp_ch && test_bit(MLX5E_PTP_STATE_RX, ptp_ch->state))
328 mlx5e_rx_reporter_diagnose_common_ptp_config(priv, ptp_ch, fmsg);
329 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
330 }
331
mlx5e_rx_reporter_build_diagnose_output_ptp_rq(struct mlx5e_rq * rq,struct devlink_fmsg * fmsg)332 static void mlx5e_rx_reporter_build_diagnose_output_ptp_rq(struct mlx5e_rq *rq,
333 struct devlink_fmsg *fmsg)
334 {
335 devlink_fmsg_obj_nest_start(fmsg);
336 devlink_fmsg_string_pair_put(fmsg, "channel", "ptp");
337 mlx5e_rx_reporter_build_diagnose_output_rq_common(rq, fmsg);
338 devlink_fmsg_obj_nest_end(fmsg);
339 }
340
mlx5e_rx_reporter_diagnose_rx_res_dir_tirns(struct mlx5e_rx_res * rx_res,struct devlink_fmsg * fmsg)341 static void mlx5e_rx_reporter_diagnose_rx_res_dir_tirns(struct mlx5e_rx_res *rx_res,
342 struct devlink_fmsg *fmsg)
343 {
344 unsigned int max_nch = mlx5e_rx_res_get_max_nch(rx_res);
345 int i;
346
347 devlink_fmsg_arr_pair_nest_start(fmsg, "Direct TIRs");
348
349 for (i = 0; i < max_nch; i++) {
350 devlink_fmsg_obj_nest_start(fmsg);
351
352 devlink_fmsg_u32_pair_put(fmsg, "ix", i);
353 devlink_fmsg_u32_pair_put(fmsg, "tirn", mlx5e_rx_res_get_tirn_direct(rx_res, i));
354 devlink_fmsg_u32_pair_put(fmsg, "rqtn", mlx5e_rx_res_get_rqtn_direct(rx_res, i));
355
356 devlink_fmsg_obj_nest_end(fmsg);
357 }
358
359 devlink_fmsg_arr_pair_nest_end(fmsg);
360 }
361
mlx5e_rx_reporter_diagnose_rx_res_rss_tirn(struct mlx5e_rss * rss,bool inner,struct devlink_fmsg * fmsg)362 static void mlx5e_rx_reporter_diagnose_rx_res_rss_tirn(struct mlx5e_rss *rss, bool inner,
363 struct devlink_fmsg *fmsg)
364 {
365 bool found_valid_tir = false;
366 int tt;
367
368 for (tt = 0; tt < MLX5E_NUM_INDIR_TIRS; tt++) {
369 if (!mlx5e_rss_valid_tir(rss, tt, inner))
370 continue;
371
372 if (!found_valid_tir) {
373 char *tir_msg = inner ? "Inner TIRs Numbers" : "TIRs Numbers";
374
375 found_valid_tir = true;
376 devlink_fmsg_arr_pair_nest_start(fmsg, tir_msg);
377 }
378
379 devlink_fmsg_obj_nest_start(fmsg);
380 devlink_fmsg_string_pair_put(fmsg, "tt", mlx5_ttc_get_name(tt));
381 devlink_fmsg_u32_pair_put(fmsg, "tirn", mlx5e_rss_get_tirn(rss, tt, inner));
382 devlink_fmsg_obj_nest_end(fmsg);
383 }
384
385 if (found_valid_tir)
386 devlink_fmsg_arr_pair_nest_end(fmsg);
387 }
388
mlx5e_rx_reporter_diagnose_rx_res_rss_ix(struct mlx5e_rx_res * rx_res,u32 rss_idx,struct devlink_fmsg * fmsg)389 static void mlx5e_rx_reporter_diagnose_rx_res_rss_ix(struct mlx5e_rx_res *rx_res, u32 rss_idx,
390 struct devlink_fmsg *fmsg)
391 {
392 struct mlx5e_rss *rss = mlx5e_rx_res_rss_get(rx_res, rss_idx);
393
394 if (!rss)
395 return;
396
397 devlink_fmsg_obj_nest_start(fmsg);
398
399 devlink_fmsg_u32_pair_put(fmsg, "Index", rss_idx);
400 devlink_fmsg_u32_pair_put(fmsg, "rqtn", mlx5e_rss_get_rqtn(rss));
401 mlx5e_rx_reporter_diagnose_rx_res_rss_tirn(rss, false, fmsg);
402 if (mlx5e_rss_get_inner_ft_support(rss))
403 mlx5e_rx_reporter_diagnose_rx_res_rss_tirn(rss, true, fmsg);
404
405 devlink_fmsg_obj_nest_end(fmsg);
406 }
407
mlx5e_rx_reporter_diagnose_rx_res_rss(struct mlx5e_rx_res * rx_res,struct devlink_fmsg * fmsg)408 static void mlx5e_rx_reporter_diagnose_rx_res_rss(struct mlx5e_rx_res *rx_res,
409 struct devlink_fmsg *fmsg)
410 {
411 int rss_ix;
412
413 devlink_fmsg_arr_pair_nest_start(fmsg, "RSS");
414 for (rss_ix = 0; rss_ix < MLX5E_MAX_NUM_RSS; rss_ix++)
415 mlx5e_rx_reporter_diagnose_rx_res_rss_ix(rx_res, rss_ix, fmsg);
416 devlink_fmsg_arr_pair_nest_end(fmsg);
417 }
418
mlx5e_rx_reporter_diagnose_rx_res(struct mlx5e_priv * priv,struct devlink_fmsg * fmsg)419 static void mlx5e_rx_reporter_diagnose_rx_res(struct mlx5e_priv *priv,
420 struct devlink_fmsg *fmsg)
421 {
422 struct mlx5e_rx_res *rx_res = priv->rx_res;
423
424 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RX resources");
425 mlx5e_rx_reporter_diagnose_rx_res_dir_tirns(rx_res, fmsg);
426 mlx5e_rx_reporter_diagnose_rx_res_rss(rx_res, fmsg);
427 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
428 }
429
mlx5e_rx_reporter_diagnose_rqs(struct mlx5e_priv * priv,struct devlink_fmsg * fmsg)430 static void mlx5e_rx_reporter_diagnose_rqs(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg)
431 {
432 struct mlx5e_ptp *ptp_ch = priv->channels.ptp;
433 int i;
434
435 devlink_fmsg_arr_pair_nest_start(fmsg, "RQs");
436
437 for (i = 0; i < priv->channels.num; i++) {
438 struct mlx5e_channel *c = priv->channels.c[i];
439 struct mlx5e_rq *rq;
440
441 rq = test_bit(MLX5E_CHANNEL_STATE_XSK, c->state) ?
442 &c->xskrq : &c->rq;
443
444 mlx5e_rx_reporter_build_diagnose_output(rq, fmsg);
445 }
446 if (ptp_ch && test_bit(MLX5E_PTP_STATE_RX, ptp_ch->state))
447 mlx5e_rx_reporter_build_diagnose_output_ptp_rq(&ptp_ch->rq, fmsg);
448
449 devlink_fmsg_arr_pair_nest_end(fmsg);
450 }
451
mlx5e_rx_reporter_diagnose(struct devlink_health_reporter * reporter,struct devlink_fmsg * fmsg,struct netlink_ext_ack * extack)452 static int mlx5e_rx_reporter_diagnose(struct devlink_health_reporter *reporter,
453 struct devlink_fmsg *fmsg,
454 struct netlink_ext_ack *extack)
455 {
456 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
457
458 mutex_lock(&priv->state_lock);
459
460 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
461 goto unlock;
462
463 mlx5e_rx_reporter_diagnose_common_config(priv, fmsg);
464 mlx5e_rx_reporter_diagnose_rqs(priv, fmsg);
465 mlx5e_rx_reporter_diagnose_rx_res(priv, fmsg);
466 unlock:
467 mutex_unlock(&priv->state_lock);
468 return 0;
469 }
470
mlx5e_rx_reporter_dump_icosq(struct mlx5e_priv * priv,struct devlink_fmsg * fmsg,void * ctx)471 static int mlx5e_rx_reporter_dump_icosq(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg,
472 void *ctx)
473 {
474 struct mlx5e_txqsq *icosq = ctx;
475 struct mlx5_rsc_key key = {};
476
477 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
478 return 0;
479
480 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "SX Slice");
481 key.size = PAGE_SIZE;
482 key.rsc = MLX5_SGMT_TYPE_SX_SLICE_ALL;
483 mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
484 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
485
486 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "ICOSQ");
487
488 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "QPC");
489 key.rsc = MLX5_SGMT_TYPE_FULL_QPC;
490 key.index1 = icosq->sqn;
491 key.num_of_obj1 = 1;
492 mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
493 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
494
495 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "send_buff");
496 key.rsc = MLX5_SGMT_TYPE_SND_BUFF;
497 key.num_of_obj2 = MLX5_RSC_DUMP_ALL;
498 mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
499 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
500
501 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
502
503 return 0;
504 }
505
mlx5e_rx_reporter_dump_rq(struct mlx5e_priv * priv,struct devlink_fmsg * fmsg,void * ctx)506 static int mlx5e_rx_reporter_dump_rq(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg,
507 void *ctx)
508 {
509 struct mlx5_rsc_key key = {};
510 struct mlx5e_rq *rq = ctx;
511
512 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
513 return 0;
514
515 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RX Slice");
516 key.size = PAGE_SIZE;
517 key.rsc = MLX5_SGMT_TYPE_RX_SLICE_ALL;
518 mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
519 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
520
521 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RQ");
522
523 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "QPC");
524 key.rsc = MLX5_SGMT_TYPE_FULL_QPC;
525 key.index1 = rq->rqn;
526 key.num_of_obj1 = 1;
527 mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
528 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
529
530 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "receive_buff");
531 key.rsc = MLX5_SGMT_TYPE_RCV_BUFF;
532 key.num_of_obj2 = MLX5_RSC_DUMP_ALL;
533 mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
534 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
535
536 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
537
538 return 0;
539 }
540
mlx5e_rx_reporter_dump_all_rqs(struct mlx5e_priv * priv,struct devlink_fmsg * fmsg)541 static int mlx5e_rx_reporter_dump_all_rqs(struct mlx5e_priv *priv,
542 struct devlink_fmsg *fmsg)
543 {
544 struct mlx5e_ptp *ptp_ch = priv->channels.ptp;
545 struct mlx5_rsc_key key = {};
546
547 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
548 return 0;
549
550 mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RX Slice");
551 key.size = PAGE_SIZE;
552 key.rsc = MLX5_SGMT_TYPE_RX_SLICE_ALL;
553 mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
554 mlx5e_health_fmsg_named_obj_nest_end(fmsg);
555 devlink_fmsg_arr_pair_nest_start(fmsg, "RQs");
556
557 for (int i = 0; i < priv->channels.num; i++) {
558 struct mlx5e_rq *rq = &priv->channels.c[i]->rq;
559
560 mlx5e_health_queue_dump(priv, fmsg, rq->rqn, "RQ");
561 }
562
563 if (ptp_ch && test_bit(MLX5E_PTP_STATE_RX, ptp_ch->state))
564 mlx5e_health_queue_dump(priv, fmsg, ptp_ch->rq.rqn, "PTP RQ");
565
566 devlink_fmsg_arr_pair_nest_end(fmsg);
567 return 0;
568 }
569
mlx5e_rx_reporter_dump_from_ctx(struct mlx5e_priv * priv,struct mlx5e_err_ctx * err_ctx,struct devlink_fmsg * fmsg)570 static int mlx5e_rx_reporter_dump_from_ctx(struct mlx5e_priv *priv,
571 struct mlx5e_err_ctx *err_ctx,
572 struct devlink_fmsg *fmsg)
573 {
574 return err_ctx->dump(priv, fmsg, err_ctx->ctx);
575 }
576
mlx5e_rx_reporter_dump(struct devlink_health_reporter * reporter,struct devlink_fmsg * fmsg,void * context,struct netlink_ext_ack * extack)577 static int mlx5e_rx_reporter_dump(struct devlink_health_reporter *reporter,
578 struct devlink_fmsg *fmsg, void *context,
579 struct netlink_ext_ack *extack)
580 {
581 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
582 struct mlx5e_err_ctx *err_ctx = context;
583
584 return err_ctx ? mlx5e_rx_reporter_dump_from_ctx(priv, err_ctx, fmsg) :
585 mlx5e_rx_reporter_dump_all_rqs(priv, fmsg);
586 }
587
mlx5e_reporter_rx_timeout(struct mlx5e_rq * rq)588 void mlx5e_reporter_rx_timeout(struct mlx5e_rq *rq)
589 {
590 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
591 struct mlx5e_icosq *icosq = rq->icosq;
592 struct mlx5e_priv *priv = rq->priv;
593 struct mlx5e_err_ctx err_ctx = {};
594 char icosq_str[32] = {};
595
596 err_ctx.ctx = rq;
597 err_ctx.recover = mlx5e_rx_reporter_timeout_recover;
598 err_ctx.dump = mlx5e_rx_reporter_dump_rq;
599
600 if (icosq)
601 snprintf(icosq_str, sizeof(icosq_str), "ICOSQ: 0x%x, ", icosq->sqn);
602 snprintf(err_str, sizeof(err_str),
603 "RX timeout on channel: %d, %s RQ: 0x%x, CQ: 0x%x",
604 rq->ix, icosq_str, rq->rqn, rq->cq.mcq.cqn);
605
606 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx);
607 }
608
mlx5e_reporter_rq_cqe_err(struct mlx5e_rq * rq)609 void mlx5e_reporter_rq_cqe_err(struct mlx5e_rq *rq)
610 {
611 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
612 struct mlx5e_priv *priv = rq->priv;
613 struct mlx5e_err_ctx err_ctx = {};
614
615 err_ctx.ctx = rq;
616 err_ctx.recover = mlx5e_rx_reporter_err_rq_cqe_recover;
617 err_ctx.dump = mlx5e_rx_reporter_dump_rq;
618 snprintf(err_str, sizeof(err_str), "ERR CQE on RQ: 0x%x", rq->rqn);
619
620 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx);
621 }
622
mlx5e_reporter_icosq_cqe_err(struct mlx5e_icosq * icosq)623 void mlx5e_reporter_icosq_cqe_err(struct mlx5e_icosq *icosq)
624 {
625 struct mlx5e_priv *priv = icosq->channel->priv;
626 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
627 struct mlx5e_err_ctx err_ctx = {};
628
629 err_ctx.ctx = icosq;
630 err_ctx.recover = mlx5e_rx_reporter_err_icosq_cqe_recover;
631 err_ctx.dump = mlx5e_rx_reporter_dump_icosq;
632 snprintf(err_str, sizeof(err_str), "ERR CQE on ICOSQ: 0x%x", icosq->sqn);
633
634 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx);
635 }
636
mlx5e_reporter_icosq_suspend_recovery(struct mlx5e_channel * c)637 void mlx5e_reporter_icosq_suspend_recovery(struct mlx5e_channel *c)
638 {
639 mutex_lock(&c->icosq_recovery_lock);
640 }
641
mlx5e_reporter_icosq_resume_recovery(struct mlx5e_channel * c)642 void mlx5e_reporter_icosq_resume_recovery(struct mlx5e_channel *c)
643 {
644 mutex_unlock(&c->icosq_recovery_lock);
645 }
646
647 static const struct devlink_health_reporter_ops mlx5_rx_reporter_ops = {
648 .name = "rx",
649 .recover = mlx5e_rx_reporter_recover,
650 .diagnose = mlx5e_rx_reporter_diagnose,
651 .dump = mlx5e_rx_reporter_dump,
652 };
653
654 #define MLX5E_REPORTER_RX_GRACEFUL_PERIOD 500
655
mlx5e_reporter_rx_create(struct mlx5e_priv * priv)656 void mlx5e_reporter_rx_create(struct mlx5e_priv *priv)
657 {
658 struct devlink_health_reporter *reporter;
659
660 reporter = devlink_port_health_reporter_create(priv->netdev->devlink_port,
661 &mlx5_rx_reporter_ops,
662 MLX5E_REPORTER_RX_GRACEFUL_PERIOD, priv);
663 if (IS_ERR(reporter)) {
664 netdev_warn(priv->netdev, "Failed to create rx reporter, err = %ld\n",
665 PTR_ERR(reporter));
666 return;
667 }
668 priv->rx_reporter = reporter;
669 }
670
mlx5e_reporter_rx_destroy(struct mlx5e_priv * priv)671 void mlx5e_reporter_rx_destroy(struct mlx5e_priv *priv)
672 {
673 if (!priv->rx_reporter)
674 return;
675
676 devlink_health_reporter_destroy(priv->rx_reporter);
677 priv->rx_reporter = NULL;
678 }
679