1 // SPDX-License-Identifier: GPL-2.0
2 // Copyright (c) 2019 Mellanox Technologies.
8 static int mlx5e_query_rq_state(struct mlx5_core_dev *dev, u32 rqn, u8 *state)
10 int outlen = MLX5_ST_SZ_BYTES(query_rq_out);
15 out = kvzalloc(outlen, GFP_KERNEL);
19 err = mlx5_core_query_rq(dev, rqn, out);
23 rqc = MLX5_ADDR_OF(query_rq_out, out, rq_context);
24 *state = MLX5_GET(rqc, rqc, state);
31 static int mlx5e_wait_for_icosq_flush(struct mlx5e_icosq *icosq)
33 unsigned long exp_time = jiffies +
34 msecs_to_jiffies(MLX5E_REPORTER_FLUSH_TIMEOUT_MSEC);
36 while (time_before(jiffies, exp_time)) {
37 if (icosq->cc == icosq->pc)
43 netdev_err(icosq->channel->netdev,
44 "Wait for ICOSQ 0x%x flush timeout (cc = 0x%x, pc = 0x%x)\n",
45 icosq->sqn, icosq->cc, icosq->pc);
50 static void mlx5e_reset_icosq_cc_pc(struct mlx5e_icosq *icosq)
52 WARN_ONCE(icosq->cc != icosq->pc, "ICOSQ 0x%x: cc (0x%x) != pc (0x%x)\n",
53 icosq->sqn, icosq->cc, icosq->pc);
58 static int mlx5e_rx_reporter_err_icosq_cqe_recover(void *ctx)
60 struct mlx5_core_dev *mdev;
61 struct mlx5e_icosq *icosq;
62 struct net_device *dev;
68 rq = &icosq->channel->rq;
69 mdev = icosq->channel->mdev;
70 dev = icosq->channel->netdev;
71 err = mlx5_core_query_sq_state(mdev, icosq->sqn, &state);
73 netdev_err(dev, "Failed to query ICOSQ 0x%x state. err = %d\n",
78 if (state != MLX5_SQC_STATE_ERR)
81 mlx5e_deactivate_rq(rq);
82 err = mlx5e_wait_for_icosq_flush(icosq);
86 mlx5e_deactivate_icosq(icosq);
88 /* At this point, both the rq and the icosq are disabled */
90 err = mlx5e_health_sq_to_ready(mdev, dev, icosq->sqn);
94 mlx5e_reset_icosq_cc_pc(icosq);
95 mlx5e_free_rx_in_progress_descs(rq);
96 clear_bit(MLX5E_SQ_STATE_RECOVERING, &icosq->state);
97 mlx5e_activate_icosq(icosq);
98 mlx5e_activate_rq(rq);
100 rq->stats->recover++;
103 clear_bit(MLX5E_SQ_STATE_RECOVERING, &icosq->state);
107 static int mlx5e_rq_to_ready(struct mlx5e_rq *rq, int curr_state)
109 struct net_device *dev = rq->netdev;
112 err = mlx5e_modify_rq_state(rq, curr_state, MLX5_RQC_STATE_RST);
114 netdev_err(dev, "Failed to move rq 0x%x to reset\n", rq->rqn);
117 err = mlx5e_modify_rq_state(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
119 netdev_err(dev, "Failed to move rq 0x%x to ready\n", rq->rqn);
126 static int mlx5e_rx_reporter_err_rq_cqe_recover(void *ctx)
128 struct mlx5e_rq *rq = ctx;
131 mlx5e_deactivate_rq(rq);
132 mlx5e_free_rx_descs(rq);
134 err = mlx5e_rq_to_ready(rq, MLX5_RQC_STATE_ERR);
138 clear_bit(MLX5E_RQ_STATE_RECOVERING, &rq->state);
139 mlx5e_activate_rq(rq);
140 rq->stats->recover++;
143 clear_bit(MLX5E_RQ_STATE_RECOVERING, &rq->state);
147 static int mlx5e_rx_reporter_timeout_recover(void *ctx)
149 struct mlx5_eq_comp *eq;
156 err = mlx5e_health_channel_eq_recover(rq->netdev, eq, rq->cq.ch_stats);
157 if (err && rq->icosq)
158 clear_bit(MLX5E_SQ_STATE_ENABLED, &rq->icosq->state);
163 static int mlx5e_rx_reporter_recover_from_ctx(struct mlx5e_err_ctx *err_ctx)
165 return err_ctx->recover(err_ctx->ctx);
168 static int mlx5e_rx_reporter_recover(struct devlink_health_reporter *reporter,
170 struct netlink_ext_ack *extack)
172 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
173 struct mlx5e_err_ctx *err_ctx = context;
175 return err_ctx ? mlx5e_rx_reporter_recover_from_ctx(err_ctx) :
176 mlx5e_health_recover_channels(priv);
179 static int mlx5e_reporter_icosq_diagnose(struct mlx5e_icosq *icosq, u8 hw_state,
180 struct devlink_fmsg *fmsg)
184 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "ICOSQ");
188 err = devlink_fmsg_u32_pair_put(fmsg, "sqn", icosq->sqn);
192 err = devlink_fmsg_u8_pair_put(fmsg, "HW state", hw_state);
196 err = devlink_fmsg_u32_pair_put(fmsg, "cc", icosq->cc);
200 err = devlink_fmsg_u32_pair_put(fmsg, "pc", icosq->pc);
204 err = devlink_fmsg_u32_pair_put(fmsg, "WQE size",
205 mlx5_wq_cyc_get_size(&icosq->wq));
209 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "CQ");
213 err = devlink_fmsg_u32_pair_put(fmsg, "cqn", icosq->cq.mcq.cqn);
217 err = devlink_fmsg_u32_pair_put(fmsg, "cc", icosq->cq.wq.cc);
221 err = devlink_fmsg_u32_pair_put(fmsg, "size", mlx5_cqwq_get_size(&icosq->cq.wq));
225 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
229 return mlx5e_health_fmsg_named_obj_nest_end(fmsg);
232 static int mlx5e_rx_reporter_build_diagnose_output(struct mlx5e_rq *rq,
233 struct devlink_fmsg *fmsg)
241 err = mlx5e_query_rq_state(rq->mdev, rq->rqn, &hw_state);
245 wqes_sz = mlx5e_rqwq_get_cur_sz(rq);
246 wq_head = mlx5e_rqwq_get_head(rq);
247 wqe_counter = mlx5e_rqwq_get_wqe_counter(rq);
249 err = devlink_fmsg_obj_nest_start(fmsg);
253 err = devlink_fmsg_u32_pair_put(fmsg, "channel ix", rq->ix);
257 err = devlink_fmsg_u32_pair_put(fmsg, "rqn", rq->rqn);
261 err = devlink_fmsg_u8_pair_put(fmsg, "HW state", hw_state);
265 err = devlink_fmsg_u8_pair_put(fmsg, "SW state", rq->state);
269 err = devlink_fmsg_u32_pair_put(fmsg, "WQE counter", wqe_counter);
273 err = devlink_fmsg_u32_pair_put(fmsg, "posted WQEs", wqes_sz);
277 err = devlink_fmsg_u32_pair_put(fmsg, "cc", wq_head);
281 err = mlx5e_health_cq_diag_fmsg(&rq->cq, fmsg);
285 err = mlx5e_health_eq_diag_fmsg(rq->cq.mcq.eq, fmsg);
290 struct mlx5e_icosq *icosq = rq->icosq;
293 err = mlx5_core_query_sq_state(rq->mdev, icosq->sqn, &icosq_hw_state);
297 err = mlx5e_reporter_icosq_diagnose(icosq, icosq_hw_state, fmsg);
302 err = devlink_fmsg_obj_nest_end(fmsg);
309 static int mlx5e_rx_reporter_diagnose(struct devlink_health_reporter *reporter,
310 struct devlink_fmsg *fmsg,
311 struct netlink_ext_ack *extack)
313 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
314 struct mlx5e_params *params = &priv->channels.params;
315 struct mlx5e_rq *generic_rq;
316 u32 rq_stride, rq_sz;
319 mutex_lock(&priv->state_lock);
321 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
324 generic_rq = &priv->channels.c[0]->rq;
325 rq_sz = mlx5e_rqwq_get_size(generic_rq);
326 rq_stride = BIT(mlx5e_mpwqe_get_log_stride_size(priv->mdev, params, NULL));
328 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "Common config");
332 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RQ");
336 err = devlink_fmsg_u8_pair_put(fmsg, "type", params->rq_wq_type);
340 err = devlink_fmsg_u64_pair_put(fmsg, "stride size", rq_stride);
344 err = devlink_fmsg_u32_pair_put(fmsg, "size", rq_sz);
348 err = mlx5e_health_cq_common_diag_fmsg(&generic_rq->cq, fmsg);
352 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
356 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
360 err = devlink_fmsg_arr_pair_nest_start(fmsg, "RQs");
364 for (i = 0; i < priv->channels.num; i++) {
365 struct mlx5e_rq *rq = &priv->channels.c[i]->rq;
367 err = mlx5e_rx_reporter_build_diagnose_output(rq, fmsg);
371 err = devlink_fmsg_arr_pair_nest_end(fmsg);
375 mutex_unlock(&priv->state_lock);
379 static int mlx5e_rx_reporter_dump_icosq(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg,
382 struct mlx5e_txqsq *icosq = ctx;
383 struct mlx5_rsc_key key = {};
386 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
389 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "SX Slice");
393 key.size = PAGE_SIZE;
394 key.rsc = MLX5_SGMT_TYPE_SX_SLICE_ALL;
395 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
399 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
403 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "ICOSQ");
407 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "QPC");
411 key.rsc = MLX5_SGMT_TYPE_FULL_QPC;
412 key.index1 = icosq->sqn;
415 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
419 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
423 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "send_buff");
427 key.rsc = MLX5_SGMT_TYPE_SND_BUFF;
428 key.num_of_obj2 = MLX5_RSC_DUMP_ALL;
430 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
434 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
438 return mlx5e_health_fmsg_named_obj_nest_end(fmsg);
441 static int mlx5e_rx_reporter_dump_rq(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg,
444 struct mlx5_rsc_key key = {};
445 struct mlx5e_rq *rq = ctx;
448 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
451 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RX Slice");
455 key.size = PAGE_SIZE;
456 key.rsc = MLX5_SGMT_TYPE_RX_SLICE_ALL;
457 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
461 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
465 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RQ");
469 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "QPC");
473 key.rsc = MLX5_SGMT_TYPE_FULL_QPC;
474 key.index1 = rq->rqn;
477 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
481 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
485 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "receive_buff");
489 key.rsc = MLX5_SGMT_TYPE_RCV_BUFF;
490 key.num_of_obj2 = MLX5_RSC_DUMP_ALL;
491 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
495 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
499 return mlx5e_health_fmsg_named_obj_nest_end(fmsg);
502 static int mlx5e_rx_reporter_dump_all_rqs(struct mlx5e_priv *priv,
503 struct devlink_fmsg *fmsg)
505 struct mlx5_rsc_key key = {};
508 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
511 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RX Slice");
515 key.size = PAGE_SIZE;
516 key.rsc = MLX5_SGMT_TYPE_RX_SLICE_ALL;
517 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
521 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
525 err = devlink_fmsg_arr_pair_nest_start(fmsg, "RQs");
529 for (i = 0; i < priv->channels.num; i++) {
530 struct mlx5e_rq *rq = &priv->channels.c[i]->rq;
532 err = mlx5e_health_queue_dump(priv, fmsg, rq->rqn, "RQ");
537 return devlink_fmsg_arr_pair_nest_end(fmsg);
540 static int mlx5e_rx_reporter_dump_from_ctx(struct mlx5e_priv *priv,
541 struct mlx5e_err_ctx *err_ctx,
542 struct devlink_fmsg *fmsg)
544 return err_ctx->dump(priv, fmsg, err_ctx->ctx);
547 static int mlx5e_rx_reporter_dump(struct devlink_health_reporter *reporter,
548 struct devlink_fmsg *fmsg, void *context,
549 struct netlink_ext_ack *extack)
551 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
552 struct mlx5e_err_ctx *err_ctx = context;
554 return err_ctx ? mlx5e_rx_reporter_dump_from_ctx(priv, err_ctx, fmsg) :
555 mlx5e_rx_reporter_dump_all_rqs(priv, fmsg);
558 void mlx5e_reporter_rx_timeout(struct mlx5e_rq *rq)
560 char icosq_str[MLX5E_REPORTER_PER_Q_MAX_LEN] = {};
561 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
562 struct mlx5e_icosq *icosq = rq->icosq;
563 struct mlx5e_priv *priv = rq->priv;
564 struct mlx5e_err_ctx err_ctx = {};
567 err_ctx.recover = mlx5e_rx_reporter_timeout_recover;
568 err_ctx.dump = mlx5e_rx_reporter_dump_rq;
571 snprintf(icosq_str, sizeof(icosq_str), "ICOSQ: 0x%x, ", icosq->sqn);
572 snprintf(err_str, sizeof(err_str),
573 "RX timeout on channel: %d, %sRQ: 0x%x, CQ: 0x%x",
574 rq->ix, icosq_str, rq->rqn, rq->cq.mcq.cqn);
576 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx);
579 void mlx5e_reporter_rq_cqe_err(struct mlx5e_rq *rq)
581 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
582 struct mlx5e_priv *priv = rq->priv;
583 struct mlx5e_err_ctx err_ctx = {};
586 err_ctx.recover = mlx5e_rx_reporter_err_rq_cqe_recover;
587 err_ctx.dump = mlx5e_rx_reporter_dump_rq;
588 snprintf(err_str, sizeof(err_str), "ERR CQE on RQ: 0x%x", rq->rqn);
590 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx);
593 void mlx5e_reporter_icosq_cqe_err(struct mlx5e_icosq *icosq)
595 struct mlx5e_priv *priv = icosq->channel->priv;
596 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
597 struct mlx5e_err_ctx err_ctx = {};
600 err_ctx.recover = mlx5e_rx_reporter_err_icosq_cqe_recover;
601 err_ctx.dump = mlx5e_rx_reporter_dump_icosq;
602 snprintf(err_str, sizeof(err_str), "ERR CQE on ICOSQ: 0x%x", icosq->sqn);
604 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx);
607 static const struct devlink_health_reporter_ops mlx5_rx_reporter_ops = {
609 .recover = mlx5e_rx_reporter_recover,
610 .diagnose = mlx5e_rx_reporter_diagnose,
611 .dump = mlx5e_rx_reporter_dump,
614 #define MLX5E_REPORTER_RX_GRACEFUL_PERIOD 500
616 void mlx5e_reporter_rx_create(struct mlx5e_priv *priv)
618 struct devlink_health_reporter *reporter;
620 reporter = devlink_port_health_reporter_create(&priv->dl_port, &mlx5_rx_reporter_ops,
621 MLX5E_REPORTER_RX_GRACEFUL_PERIOD, priv);
622 if (IS_ERR(reporter)) {
623 netdev_warn(priv->netdev, "Failed to create rx reporter, err = %ld\n",
627 priv->rx_reporter = reporter;
630 void mlx5e_reporter_rx_destroy(struct mlx5e_priv *priv)
632 if (!priv->rx_reporter)
635 devlink_port_health_reporter_destroy(priv->rx_reporter);