1 // SPDX-License-Identifier: GPL-2.0
2 // Copyright (c) 2019 Mellanox Technologies.
11 static int mlx5e_query_rq_state(struct mlx5_core_dev *dev, u32 rqn, u8 *state)
13 int outlen = MLX5_ST_SZ_BYTES(query_rq_out);
18 out = kvzalloc(outlen, GFP_KERNEL);
22 err = mlx5_core_query_rq(dev, rqn, out);
26 rqc = MLX5_ADDR_OF(query_rq_out, out, rq_context);
27 *state = MLX5_GET(rqc, rqc, state);
34 static int mlx5e_wait_for_icosq_flush(struct mlx5e_icosq *icosq)
36 struct mlx5_core_dev *dev = icosq->channel->mdev;
37 unsigned long exp_time;
39 exp_time = jiffies + msecs_to_jiffies(mlx5_tout_ms(dev, FLUSH_ON_ERROR));
41 while (time_before(jiffies, exp_time)) {
42 if (icosq->cc == icosq->pc)
48 netdev_err(icosq->channel->netdev,
49 "Wait for ICOSQ 0x%x flush timeout (cc = 0x%x, pc = 0x%x)\n",
50 icosq->sqn, icosq->cc, icosq->pc);
55 static void mlx5e_reset_icosq_cc_pc(struct mlx5e_icosq *icosq)
57 WARN_ONCE(icosq->cc != icosq->pc, "ICOSQ 0x%x: cc (0x%x) != pc (0x%x)\n",
58 icosq->sqn, icosq->cc, icosq->pc);
63 static int mlx5e_rx_reporter_err_icosq_cqe_recover(void *ctx)
65 struct mlx5e_rq *xskrq = NULL;
66 struct mlx5_core_dev *mdev;
67 struct mlx5e_icosq *icosq;
68 struct net_device *dev;
75 mutex_lock(&icosq->channel->icosq_recovery_lock);
77 /* mlx5e_close_rq cancels this work before RQ and ICOSQ are killed. */
78 rq = &icosq->channel->rq;
79 if (test_bit(MLX5E_RQ_STATE_ENABLED, &icosq->channel->xskrq.state))
80 xskrq = &icosq->channel->xskrq;
81 mdev = icosq->channel->mdev;
82 dev = icosq->channel->netdev;
83 err = mlx5_core_query_sq_state(mdev, icosq->sqn, &state);
85 netdev_err(dev, "Failed to query ICOSQ 0x%x state. err = %d\n",
90 if (state != MLX5_SQC_STATE_ERR)
93 mlx5e_deactivate_rq(rq);
95 mlx5e_deactivate_rq(xskrq);
97 err = mlx5e_wait_for_icosq_flush(icosq);
101 mlx5e_deactivate_icosq(icosq);
103 /* At this point, both the rq and the icosq are disabled */
105 err = mlx5e_health_sq_to_ready(mdev, dev, icosq->sqn);
109 mlx5e_reset_icosq_cc_pc(icosq);
111 mlx5e_free_rx_in_progress_descs(rq);
113 mlx5e_free_rx_in_progress_descs(xskrq);
115 clear_bit(MLX5E_SQ_STATE_RECOVERING, &icosq->state);
116 mlx5e_activate_icosq(icosq);
118 mlx5e_activate_rq(rq);
119 rq->stats->recover++;
122 mlx5e_activate_rq(xskrq);
123 xskrq->stats->recover++;
126 mutex_unlock(&icosq->channel->icosq_recovery_lock);
130 clear_bit(MLX5E_SQ_STATE_RECOVERING, &icosq->state);
131 mutex_unlock(&icosq->channel->icosq_recovery_lock);
135 static int mlx5e_rq_to_ready(struct mlx5e_rq *rq, int curr_state)
137 struct net_device *dev = rq->netdev;
140 err = mlx5e_modify_rq_state(rq, curr_state, MLX5_RQC_STATE_RST);
142 netdev_err(dev, "Failed to move rq 0x%x to reset\n", rq->rqn);
145 err = mlx5e_modify_rq_state(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY);
147 netdev_err(dev, "Failed to move rq 0x%x to ready\n", rq->rqn);
154 static int mlx5e_rx_reporter_err_rq_cqe_recover(void *ctx)
156 struct mlx5e_rq *rq = ctx;
159 mlx5e_deactivate_rq(rq);
160 mlx5e_free_rx_descs(rq);
162 err = mlx5e_rq_to_ready(rq, MLX5_RQC_STATE_ERR);
166 clear_bit(MLX5E_RQ_STATE_RECOVERING, &rq->state);
167 mlx5e_activate_rq(rq);
168 rq->stats->recover++;
171 clear_bit(MLX5E_RQ_STATE_RECOVERING, &rq->state);
175 static int mlx5e_rx_reporter_timeout_recover(void *ctx)
177 struct mlx5_eq_comp *eq;
184 err = mlx5e_health_channel_eq_recover(rq->netdev, eq, rq->cq.ch_stats);
185 if (err && rq->icosq)
186 clear_bit(MLX5E_SQ_STATE_ENABLED, &rq->icosq->state);
191 static int mlx5e_rx_reporter_recover_from_ctx(struct mlx5e_err_ctx *err_ctx)
193 return err_ctx->recover(err_ctx->ctx);
196 static int mlx5e_rx_reporter_recover(struct devlink_health_reporter *reporter,
198 struct netlink_ext_ack *extack)
200 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
201 struct mlx5e_err_ctx *err_ctx = context;
203 return err_ctx ? mlx5e_rx_reporter_recover_from_ctx(err_ctx) :
204 mlx5e_health_recover_channels(priv);
207 static int mlx5e_reporter_icosq_diagnose(struct mlx5e_icosq *icosq, u8 hw_state,
208 struct devlink_fmsg *fmsg)
212 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "ICOSQ");
216 err = devlink_fmsg_u32_pair_put(fmsg, "sqn", icosq->sqn);
220 err = devlink_fmsg_u8_pair_put(fmsg, "HW state", hw_state);
224 err = devlink_fmsg_u32_pair_put(fmsg, "cc", icosq->cc);
228 err = devlink_fmsg_u32_pair_put(fmsg, "pc", icosq->pc);
232 err = devlink_fmsg_u32_pair_put(fmsg, "WQE size",
233 mlx5_wq_cyc_get_size(&icosq->wq));
237 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "CQ");
241 err = devlink_fmsg_u32_pair_put(fmsg, "cqn", icosq->cq.mcq.cqn);
245 err = devlink_fmsg_u32_pair_put(fmsg, "cc", icosq->cq.wq.cc);
249 err = devlink_fmsg_u32_pair_put(fmsg, "size", mlx5_cqwq_get_size(&icosq->cq.wq));
253 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
257 return mlx5e_health_fmsg_named_obj_nest_end(fmsg);
261 mlx5e_rx_reporter_build_diagnose_output_rq_common(struct mlx5e_rq *rq,
262 struct devlink_fmsg *fmsg)
270 err = mlx5e_query_rq_state(rq->mdev, rq->rqn, &hw_state);
274 wqes_sz = mlx5e_rqwq_get_cur_sz(rq);
275 wq_head = mlx5e_rqwq_get_head(rq);
276 wqe_counter = mlx5e_rqwq_get_wqe_counter(rq);
278 err = devlink_fmsg_u32_pair_put(fmsg, "rqn", rq->rqn);
282 err = devlink_fmsg_u8_pair_put(fmsg, "HW state", hw_state);
286 err = devlink_fmsg_u8_pair_put(fmsg, "SW state", rq->state);
290 err = devlink_fmsg_u32_pair_put(fmsg, "WQE counter", wqe_counter);
294 err = devlink_fmsg_u32_pair_put(fmsg, "posted WQEs", wqes_sz);
298 err = devlink_fmsg_u32_pair_put(fmsg, "cc", wq_head);
302 err = mlx5e_health_cq_diag_fmsg(&rq->cq, fmsg);
306 err = mlx5e_health_eq_diag_fmsg(rq->cq.mcq.eq, fmsg);
311 struct mlx5e_icosq *icosq = rq->icosq;
314 err = mlx5_core_query_sq_state(rq->mdev, icosq->sqn, &icosq_hw_state);
318 err = mlx5e_reporter_icosq_diagnose(icosq, icosq_hw_state, fmsg);
326 static int mlx5e_rx_reporter_build_diagnose_output(struct mlx5e_rq *rq,
327 struct devlink_fmsg *fmsg)
331 err = devlink_fmsg_obj_nest_start(fmsg);
335 err = devlink_fmsg_u32_pair_put(fmsg, "channel ix", rq->ix);
339 err = mlx5e_rx_reporter_build_diagnose_output_rq_common(rq, fmsg);
343 return devlink_fmsg_obj_nest_end(fmsg);
346 static int mlx5e_rx_reporter_diagnose_generic_rq(struct mlx5e_rq *rq,
347 struct devlink_fmsg *fmsg)
349 struct mlx5e_priv *priv = rq->priv;
350 struct mlx5e_params *params;
351 u32 rq_stride, rq_sz;
355 params = &priv->channels.params;
356 rq_sz = mlx5e_rqwq_get_size(rq);
357 real_time = mlx5_is_real_time_rq(priv->mdev);
358 rq_stride = BIT(mlx5e_mpwqe_get_log_stride_size(priv->mdev, params, NULL));
360 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RQ");
364 err = devlink_fmsg_u8_pair_put(fmsg, "type", params->rq_wq_type);
368 err = devlink_fmsg_u64_pair_put(fmsg, "stride size", rq_stride);
372 err = devlink_fmsg_u32_pair_put(fmsg, "size", rq_sz);
376 err = devlink_fmsg_string_pair_put(fmsg, "ts_format", real_time ? "RT" : "FRC");
380 err = mlx5e_health_cq_common_diag_fmsg(&rq->cq, fmsg);
384 return mlx5e_health_fmsg_named_obj_nest_end(fmsg);
388 mlx5e_rx_reporter_diagnose_common_ptp_config(struct mlx5e_priv *priv, struct mlx5e_ptp *ptp_ch,
389 struct devlink_fmsg *fmsg)
393 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "PTP");
397 err = devlink_fmsg_u32_pair_put(fmsg, "filter_type", priv->tstamp.rx_filter);
401 err = mlx5e_rx_reporter_diagnose_generic_rq(&ptp_ch->rq, fmsg);
405 return mlx5e_health_fmsg_named_obj_nest_end(fmsg);
409 mlx5e_rx_reporter_diagnose_common_config(struct devlink_health_reporter *reporter,
410 struct devlink_fmsg *fmsg)
412 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
413 struct mlx5e_rq *generic_rq = &priv->channels.c[0]->rq;
414 struct mlx5e_ptp *ptp_ch = priv->channels.ptp;
417 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "Common config");
421 err = mlx5e_rx_reporter_diagnose_generic_rq(generic_rq, fmsg);
425 if (ptp_ch && test_bit(MLX5E_PTP_STATE_RX, ptp_ch->state)) {
426 err = mlx5e_rx_reporter_diagnose_common_ptp_config(priv, ptp_ch, fmsg);
431 return mlx5e_health_fmsg_named_obj_nest_end(fmsg);
434 static int mlx5e_rx_reporter_build_diagnose_output_ptp_rq(struct mlx5e_rq *rq,
435 struct devlink_fmsg *fmsg)
439 err = devlink_fmsg_obj_nest_start(fmsg);
443 err = devlink_fmsg_string_pair_put(fmsg, "channel", "ptp");
447 err = mlx5e_rx_reporter_build_diagnose_output_rq_common(rq, fmsg);
451 err = devlink_fmsg_obj_nest_end(fmsg);
458 static int mlx5e_rx_reporter_diagnose(struct devlink_health_reporter *reporter,
459 struct devlink_fmsg *fmsg,
460 struct netlink_ext_ack *extack)
462 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
463 struct mlx5e_ptp *ptp_ch = priv->channels.ptp;
466 mutex_lock(&priv->state_lock);
468 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
471 err = mlx5e_rx_reporter_diagnose_common_config(reporter, fmsg);
475 err = devlink_fmsg_arr_pair_nest_start(fmsg, "RQs");
479 for (i = 0; i < priv->channels.num; i++) {
480 struct mlx5e_rq *rq = &priv->channels.c[i]->rq;
482 err = mlx5e_rx_reporter_build_diagnose_output(rq, fmsg);
486 if (ptp_ch && test_bit(MLX5E_PTP_STATE_RX, ptp_ch->state)) {
487 err = mlx5e_rx_reporter_build_diagnose_output_ptp_rq(&ptp_ch->rq, fmsg);
491 err = devlink_fmsg_arr_pair_nest_end(fmsg);
493 mutex_unlock(&priv->state_lock);
497 static int mlx5e_rx_reporter_dump_icosq(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg,
500 struct mlx5e_txqsq *icosq = ctx;
501 struct mlx5_rsc_key key = {};
504 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
507 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "SX Slice");
511 key.size = PAGE_SIZE;
512 key.rsc = MLX5_SGMT_TYPE_SX_SLICE_ALL;
513 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
517 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
521 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "ICOSQ");
525 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "QPC");
529 key.rsc = MLX5_SGMT_TYPE_FULL_QPC;
530 key.index1 = icosq->sqn;
533 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
537 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
541 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "send_buff");
545 key.rsc = MLX5_SGMT_TYPE_SND_BUFF;
546 key.num_of_obj2 = MLX5_RSC_DUMP_ALL;
548 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
552 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
556 return mlx5e_health_fmsg_named_obj_nest_end(fmsg);
559 static int mlx5e_rx_reporter_dump_rq(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg,
562 struct mlx5_rsc_key key = {};
563 struct mlx5e_rq *rq = ctx;
566 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
569 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RX Slice");
573 key.size = PAGE_SIZE;
574 key.rsc = MLX5_SGMT_TYPE_RX_SLICE_ALL;
575 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
579 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
583 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RQ");
587 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "QPC");
591 key.rsc = MLX5_SGMT_TYPE_FULL_QPC;
592 key.index1 = rq->rqn;
595 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
599 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
603 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "receive_buff");
607 key.rsc = MLX5_SGMT_TYPE_RCV_BUFF;
608 key.num_of_obj2 = MLX5_RSC_DUMP_ALL;
609 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
613 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
617 return mlx5e_health_fmsg_named_obj_nest_end(fmsg);
620 static int mlx5e_rx_reporter_dump_all_rqs(struct mlx5e_priv *priv,
621 struct devlink_fmsg *fmsg)
623 struct mlx5e_ptp *ptp_ch = priv->channels.ptp;
624 struct mlx5_rsc_key key = {};
627 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
630 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RX Slice");
634 key.size = PAGE_SIZE;
635 key.rsc = MLX5_SGMT_TYPE_RX_SLICE_ALL;
636 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg);
640 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg);
644 err = devlink_fmsg_arr_pair_nest_start(fmsg, "RQs");
648 for (i = 0; i < priv->channels.num; i++) {
649 struct mlx5e_rq *rq = &priv->channels.c[i]->rq;
651 err = mlx5e_health_queue_dump(priv, fmsg, rq->rqn, "RQ");
656 if (ptp_ch && test_bit(MLX5E_PTP_STATE_RX, ptp_ch->state)) {
657 err = mlx5e_health_queue_dump(priv, fmsg, ptp_ch->rq.rqn, "PTP RQ");
662 return devlink_fmsg_arr_pair_nest_end(fmsg);
665 static int mlx5e_rx_reporter_dump_from_ctx(struct mlx5e_priv *priv,
666 struct mlx5e_err_ctx *err_ctx,
667 struct devlink_fmsg *fmsg)
669 return err_ctx->dump(priv, fmsg, err_ctx->ctx);
672 static int mlx5e_rx_reporter_dump(struct devlink_health_reporter *reporter,
673 struct devlink_fmsg *fmsg, void *context,
674 struct netlink_ext_ack *extack)
676 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
677 struct mlx5e_err_ctx *err_ctx = context;
679 return err_ctx ? mlx5e_rx_reporter_dump_from_ctx(priv, err_ctx, fmsg) :
680 mlx5e_rx_reporter_dump_all_rqs(priv, fmsg);
683 void mlx5e_reporter_rx_timeout(struct mlx5e_rq *rq)
685 char icosq_str[MLX5E_REPORTER_PER_Q_MAX_LEN] = {};
686 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
687 struct mlx5e_icosq *icosq = rq->icosq;
688 struct mlx5e_priv *priv = rq->priv;
689 struct mlx5e_err_ctx err_ctx = {};
692 err_ctx.recover = mlx5e_rx_reporter_timeout_recover;
693 err_ctx.dump = mlx5e_rx_reporter_dump_rq;
696 snprintf(icosq_str, sizeof(icosq_str), "ICOSQ: 0x%x, ", icosq->sqn);
697 snprintf(err_str, sizeof(err_str),
698 "RX timeout on channel: %d, %sRQ: 0x%x, CQ: 0x%x",
699 rq->ix, icosq_str, rq->rqn, rq->cq.mcq.cqn);
701 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx);
704 void mlx5e_reporter_rq_cqe_err(struct mlx5e_rq *rq)
706 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
707 struct mlx5e_priv *priv = rq->priv;
708 struct mlx5e_err_ctx err_ctx = {};
711 err_ctx.recover = mlx5e_rx_reporter_err_rq_cqe_recover;
712 err_ctx.dump = mlx5e_rx_reporter_dump_rq;
713 snprintf(err_str, sizeof(err_str), "ERR CQE on RQ: 0x%x", rq->rqn);
715 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx);
718 void mlx5e_reporter_icosq_cqe_err(struct mlx5e_icosq *icosq)
720 struct mlx5e_priv *priv = icosq->channel->priv;
721 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
722 struct mlx5e_err_ctx err_ctx = {};
725 err_ctx.recover = mlx5e_rx_reporter_err_icosq_cqe_recover;
726 err_ctx.dump = mlx5e_rx_reporter_dump_icosq;
727 snprintf(err_str, sizeof(err_str), "ERR CQE on ICOSQ: 0x%x", icosq->sqn);
729 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx);
732 void mlx5e_reporter_icosq_suspend_recovery(struct mlx5e_channel *c)
734 mutex_lock(&c->icosq_recovery_lock);
737 void mlx5e_reporter_icosq_resume_recovery(struct mlx5e_channel *c)
739 mutex_unlock(&c->icosq_recovery_lock);
742 static const struct devlink_health_reporter_ops mlx5_rx_reporter_ops = {
744 .recover = mlx5e_rx_reporter_recover,
745 .diagnose = mlx5e_rx_reporter_diagnose,
746 .dump = mlx5e_rx_reporter_dump,
749 #define MLX5E_REPORTER_RX_GRACEFUL_PERIOD 500
751 void mlx5e_reporter_rx_create(struct mlx5e_priv *priv)
753 struct devlink_port *dl_port = mlx5e_devlink_get_dl_port(priv);
754 struct devlink_health_reporter *reporter;
756 reporter = devlink_port_health_reporter_create(dl_port, &mlx5_rx_reporter_ops,
757 MLX5E_REPORTER_RX_GRACEFUL_PERIOD, priv);
758 if (IS_ERR(reporter)) {
759 netdev_warn(priv->netdev, "Failed to create rx reporter, err = %ld\n",
763 priv->rx_reporter = reporter;
766 void mlx5e_reporter_rx_destroy(struct mlx5e_priv *priv)
768 if (!priv->rx_reporter)
771 devlink_port_health_reporter_destroy(priv->rx_reporter);
772 priv->rx_reporter = NULL;