1c50de4afSAya Levin // SPDX-License-Identifier: GPL-2.0 2c50de4afSAya Levin // Copyright (c) 2019 Mellanox Technologies. 3c50de4afSAya Levin 4c50de4afSAya Levin #include "health.h" 5c50de4afSAya Levin #include "lib/eq.h" 65f29458bSAya Levin #include "lib/mlx5.h" 7c50de4afSAya Levin 82d708887SAya Levin int mlx5e_reporter_named_obj_nest_start(struct devlink_fmsg *fmsg, char *name) 92d708887SAya Levin { 102d708887SAya Levin int err; 112d708887SAya Levin 122d708887SAya Levin err = devlink_fmsg_pair_nest_start(fmsg, name); 132d708887SAya Levin if (err) 142d708887SAya Levin return err; 152d708887SAya Levin 162d708887SAya Levin err = devlink_fmsg_obj_nest_start(fmsg); 172d708887SAya Levin if (err) 182d708887SAya Levin return err; 192d708887SAya Levin 202d708887SAya Levin return 0; 212d708887SAya Levin } 222d708887SAya Levin 232d708887SAya Levin int mlx5e_reporter_named_obj_nest_end(struct devlink_fmsg *fmsg) 242d708887SAya Levin { 252d708887SAya Levin int err; 262d708887SAya Levin 272d708887SAya Levin err = devlink_fmsg_obj_nest_end(fmsg); 282d708887SAya Levin if (err) 292d708887SAya Levin return err; 302d708887SAya Levin 312d708887SAya Levin err = devlink_fmsg_pair_nest_end(fmsg); 322d708887SAya Levin if (err) 332d708887SAya Levin return err; 342d708887SAya Levin 352d708887SAya Levin return 0; 362d708887SAya Levin } 372d708887SAya Levin 382bf09e60SAya Levin int mlx5e_reporter_cq_diagnose(struct mlx5e_cq *cq, struct devlink_fmsg *fmsg) 392bf09e60SAya Levin { 402bf09e60SAya Levin struct mlx5e_priv *priv = cq->channel->priv; 412bf09e60SAya Levin u32 out[MLX5_ST_SZ_DW(query_cq_out)] = {}; 422bf09e60SAya Levin u8 hw_status; 432bf09e60SAya Levin void *cqc; 442bf09e60SAya Levin int err; 452bf09e60SAya Levin 462bf09e60SAya Levin err = mlx5_core_query_cq(priv->mdev, &cq->mcq, out, sizeof(out)); 472bf09e60SAya Levin if (err) 482bf09e60SAya Levin return err; 492bf09e60SAya Levin 502bf09e60SAya Levin cqc = MLX5_ADDR_OF(query_cq_out, out, cq_context); 512bf09e60SAya Levin hw_status = MLX5_GET(cqc, cqc, status); 522bf09e60SAya Levin 532bf09e60SAya Levin err = mlx5e_reporter_named_obj_nest_start(fmsg, "CQ"); 542bf09e60SAya Levin if (err) 552bf09e60SAya Levin return err; 562bf09e60SAya Levin 572bf09e60SAya Levin err = devlink_fmsg_u32_pair_put(fmsg, "cqn", cq->mcq.cqn); 582bf09e60SAya Levin if (err) 592bf09e60SAya Levin return err; 602bf09e60SAya Levin 612bf09e60SAya Levin err = devlink_fmsg_u8_pair_put(fmsg, "HW status", hw_status); 622bf09e60SAya Levin if (err) 632bf09e60SAya Levin return err; 642bf09e60SAya Levin 652bf09e60SAya Levin err = mlx5e_reporter_named_obj_nest_end(fmsg); 662bf09e60SAya Levin if (err) 672bf09e60SAya Levin return err; 682bf09e60SAya Levin 692bf09e60SAya Levin return 0; 702bf09e60SAya Levin } 712bf09e60SAya Levin 722bf09e60SAya Levin int mlx5e_reporter_cq_common_diagnose(struct mlx5e_cq *cq, struct devlink_fmsg *fmsg) 732bf09e60SAya Levin { 742bf09e60SAya Levin u8 cq_log_stride; 752bf09e60SAya Levin u32 cq_sz; 762bf09e60SAya Levin int err; 772bf09e60SAya Levin 782bf09e60SAya Levin cq_sz = mlx5_cqwq_get_size(&cq->wq); 792bf09e60SAya Levin cq_log_stride = mlx5_cqwq_get_log_stride_size(&cq->wq); 802bf09e60SAya Levin 812bf09e60SAya Levin err = mlx5e_reporter_named_obj_nest_start(fmsg, "CQ"); 822bf09e60SAya Levin if (err) 832bf09e60SAya Levin return err; 842bf09e60SAya Levin 852bf09e60SAya Levin err = devlink_fmsg_u64_pair_put(fmsg, "stride size", BIT(cq_log_stride)); 862bf09e60SAya Levin if (err) 872bf09e60SAya Levin return err; 882bf09e60SAya Levin 892bf09e60SAya Levin err = devlink_fmsg_u32_pair_put(fmsg, "size", cq_sz); 902bf09e60SAya Levin if (err) 912bf09e60SAya Levin return err; 922bf09e60SAya Levin 932bf09e60SAya Levin err = mlx5e_reporter_named_obj_nest_end(fmsg); 942bf09e60SAya Levin if (err) 952bf09e60SAya Levin return err; 962bf09e60SAya Levin 972bf09e60SAya Levin return 0; 982bf09e60SAya Levin } 992bf09e60SAya Levin 10011af6a6dSAya Levin int mlx5e_health_create_reporters(struct mlx5e_priv *priv) 10111af6a6dSAya Levin { 1029032e719SAya Levin int err; 1039032e719SAya Levin 1049032e719SAya Levin err = mlx5e_reporter_tx_create(priv); 1059032e719SAya Levin if (err) 1069032e719SAya Levin return err; 1079032e719SAya Levin 1089032e719SAya Levin err = mlx5e_reporter_rx_create(priv); 1099032e719SAya Levin if (err) 1109032e719SAya Levin return err; 1119032e719SAya Levin 1129032e719SAya Levin return 0; 11311af6a6dSAya Levin } 11411af6a6dSAya Levin 11511af6a6dSAya Levin void mlx5e_health_destroy_reporters(struct mlx5e_priv *priv) 11611af6a6dSAya Levin { 1179032e719SAya Levin mlx5e_reporter_rx_destroy(priv); 11811af6a6dSAya Levin mlx5e_reporter_tx_destroy(priv); 11911af6a6dSAya Levin } 12011af6a6dSAya Levin 12111af6a6dSAya Levin void mlx5e_health_channels_update(struct mlx5e_priv *priv) 12211af6a6dSAya Levin { 12311af6a6dSAya Levin if (priv->tx_reporter) 12411af6a6dSAya Levin devlink_health_reporter_state_update(priv->tx_reporter, 12511af6a6dSAya Levin DEVLINK_HEALTH_REPORTER_STATE_HEALTHY); 1269032e719SAya Levin if (priv->rx_reporter) 1279032e719SAya Levin devlink_health_reporter_state_update(priv->rx_reporter, 1289032e719SAya Levin DEVLINK_HEALTH_REPORTER_STATE_HEALTHY); 12911af6a6dSAya Levin } 13011af6a6dSAya Levin 131c50de4afSAya Levin int mlx5e_health_sq_to_ready(struct mlx5e_channel *channel, u32 sqn) 132c50de4afSAya Levin { 133c50de4afSAya Levin struct mlx5_core_dev *mdev = channel->mdev; 134c50de4afSAya Levin struct net_device *dev = channel->netdev; 135c50de4afSAya Levin struct mlx5e_modify_sq_param msp = {}; 136c50de4afSAya Levin int err; 137c50de4afSAya Levin 138c50de4afSAya Levin msp.curr_state = MLX5_SQC_STATE_ERR; 139c50de4afSAya Levin msp.next_state = MLX5_SQC_STATE_RST; 140c50de4afSAya Levin 141c50de4afSAya Levin err = mlx5e_modify_sq(mdev, sqn, &msp); 142c50de4afSAya Levin if (err) { 143c50de4afSAya Levin netdev_err(dev, "Failed to move sq 0x%x to reset\n", sqn); 144c50de4afSAya Levin return err; 145c50de4afSAya Levin } 146c50de4afSAya Levin 147c50de4afSAya Levin memset(&msp, 0, sizeof(msp)); 148c50de4afSAya Levin msp.curr_state = MLX5_SQC_STATE_RST; 149c50de4afSAya Levin msp.next_state = MLX5_SQC_STATE_RDY; 150c50de4afSAya Levin 151c50de4afSAya Levin err = mlx5e_modify_sq(mdev, sqn, &msp); 152c50de4afSAya Levin if (err) { 153c50de4afSAya Levin netdev_err(dev, "Failed to move sq 0x%x to ready\n", sqn); 154c50de4afSAya Levin return err; 155c50de4afSAya Levin } 156c50de4afSAya Levin 157c50de4afSAya Levin return 0; 158c50de4afSAya Levin } 159c50de4afSAya Levin 160c50de4afSAya Levin int mlx5e_health_recover_channels(struct mlx5e_priv *priv) 161c50de4afSAya Levin { 162c50de4afSAya Levin int err = 0; 163c50de4afSAya Levin 164c50de4afSAya Levin rtnl_lock(); 165c50de4afSAya Levin mutex_lock(&priv->state_lock); 166c50de4afSAya Levin 167c50de4afSAya Levin if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) 168c50de4afSAya Levin goto out; 169c50de4afSAya Levin 170c50de4afSAya Levin err = mlx5e_safe_reopen_channels(priv); 171c50de4afSAya Levin 172c50de4afSAya Levin out: 173c50de4afSAya Levin mutex_unlock(&priv->state_lock); 174c50de4afSAya Levin rtnl_unlock(); 175c50de4afSAya Levin 176c50de4afSAya Levin return err; 177c50de4afSAya Levin } 178c50de4afSAya Levin 179c50de4afSAya Levin int mlx5e_health_channel_eq_recover(struct mlx5_eq_comp *eq, struct mlx5e_channel *channel) 180c50de4afSAya Levin { 181c50de4afSAya Levin u32 eqe_count; 182c50de4afSAya Levin 183c50de4afSAya Levin netdev_err(channel->netdev, "EQ 0x%x: Cons = 0x%x, irqn = 0x%x\n", 184c50de4afSAya Levin eq->core.eqn, eq->core.cons_index, eq->core.irqn); 185c50de4afSAya Levin 186c50de4afSAya Levin eqe_count = mlx5_eq_poll_irq_disabled(eq); 187c50de4afSAya Levin if (!eqe_count) 188c50de4afSAya Levin return -EIO; 189c50de4afSAya Levin 190c50de4afSAya Levin netdev_err(channel->netdev, "Recovered %d eqes on EQ 0x%x\n", 191c50de4afSAya Levin eqe_count, eq->core.eqn); 192c50de4afSAya Levin 193c50de4afSAya Levin channel->stats->eq_rearm++; 194c50de4afSAya Levin return 0; 195c50de4afSAya Levin } 196c50de4afSAya Levin 197c50de4afSAya Levin int mlx5e_health_report(struct mlx5e_priv *priv, 198c50de4afSAya Levin struct devlink_health_reporter *reporter, char *err_str, 199c50de4afSAya Levin struct mlx5e_err_ctx *err_ctx) 200c50de4afSAya Levin { 201c50de4afSAya Levin netdev_err(priv->netdev, err_str); 20299cda454SEran Ben Elisha 20399cda454SEran Ben Elisha if (!reporter) 204c50de4afSAya Levin return err_ctx->recover(&err_ctx->ctx); 20599cda454SEran Ben Elisha 206c50de4afSAya Levin return devlink_health_report(reporter, err_str, err_ctx); 207c50de4afSAya Levin } 2085f29458bSAya Levin 2095f29458bSAya Levin #define MLX5_HEALTH_DEVLINK_MAX_SIZE 1024 2105f29458bSAya Levin static int mlx5e_health_rsc_fmsg_binary(struct devlink_fmsg *fmsg, 2115f29458bSAya Levin const void *value, u32 value_len) 2125f29458bSAya Levin 2135f29458bSAya Levin { 2145f29458bSAya Levin u32 data_size; 2155f29458bSAya Levin u32 offset; 2165f29458bSAya Levin int err; 2175f29458bSAya Levin 2185f29458bSAya Levin for (offset = 0; offset < value_len; offset += data_size) { 2195f29458bSAya Levin data_size = value_len - offset; 2205f29458bSAya Levin if (data_size > MLX5_HEALTH_DEVLINK_MAX_SIZE) 2215f29458bSAya Levin data_size = MLX5_HEALTH_DEVLINK_MAX_SIZE; 2225f29458bSAya Levin err = devlink_fmsg_binary_put(fmsg, value + offset, data_size); 2235f29458bSAya Levin if (err) 2245f29458bSAya Levin break; 2255f29458bSAya Levin } 2265f29458bSAya Levin return err; 2275f29458bSAya Levin } 2285f29458bSAya Levin 2295f29458bSAya Levin int mlx5e_health_rsc_fmsg_dump(struct mlx5e_priv *priv, struct mlx5_rsc_key *key, 2305f29458bSAya Levin struct devlink_fmsg *fmsg) 2315f29458bSAya Levin { 2325f29458bSAya Levin struct mlx5_core_dev *mdev = priv->mdev; 2335f29458bSAya Levin struct mlx5_rsc_dump_cmd *cmd; 2345f29458bSAya Levin struct page *page; 2355f29458bSAya Levin int cmd_err, err; 2365f29458bSAya Levin int end_err; 2375f29458bSAya Levin int size; 2385f29458bSAya Levin 2395f29458bSAya Levin if (IS_ERR_OR_NULL(mdev->rsc_dump)) 2405f29458bSAya Levin return -EOPNOTSUPP; 2415f29458bSAya Levin 2425f29458bSAya Levin page = alloc_page(GFP_KERNEL); 2435f29458bSAya Levin if (!page) 2445f29458bSAya Levin return -ENOMEM; 2455f29458bSAya Levin 2465f29458bSAya Levin err = devlink_fmsg_binary_pair_nest_start(fmsg, "data"); 2475f29458bSAya Levin if (err) 2485f29458bSAya Levin return err; 2495f29458bSAya Levin 2505f29458bSAya Levin cmd = mlx5_rsc_dump_cmd_create(mdev, key); 2515f29458bSAya Levin if (IS_ERR(cmd)) { 2525f29458bSAya Levin err = PTR_ERR(cmd); 2535f29458bSAya Levin goto free_page; 2545f29458bSAya Levin } 2555f29458bSAya Levin 2565f29458bSAya Levin do { 2575f29458bSAya Levin cmd_err = mlx5_rsc_dump_next(mdev, cmd, page, &size); 2585f29458bSAya Levin if (cmd_err < 0) { 2595f29458bSAya Levin err = cmd_err; 2605f29458bSAya Levin goto destroy_cmd; 2615f29458bSAya Levin } 2625f29458bSAya Levin 2635f29458bSAya Levin err = mlx5e_health_rsc_fmsg_binary(fmsg, page_address(page), size); 2645f29458bSAya Levin if (err) 2655f29458bSAya Levin goto destroy_cmd; 2665f29458bSAya Levin 2675f29458bSAya Levin } while (cmd_err > 0); 2685f29458bSAya Levin 2695f29458bSAya Levin destroy_cmd: 2705f29458bSAya Levin mlx5_rsc_dump_cmd_destroy(cmd); 2715f29458bSAya Levin end_err = devlink_fmsg_binary_pair_nest_end(fmsg); 2725f29458bSAya Levin if (end_err) 2735f29458bSAya Levin err = end_err; 2745f29458bSAya Levin free_page: 2755f29458bSAya Levin __free_page(page); 2765f29458bSAya Levin return err; 2775f29458bSAya Levin } 2785f29458bSAya Levin 2795f29458bSAya Levin int mlx5e_health_queue_dump(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg, 2805f29458bSAya Levin int queue_idx, char *lbl) 2815f29458bSAya Levin { 2825f29458bSAya Levin struct mlx5_rsc_key key = {}; 2835f29458bSAya Levin int err; 2845f29458bSAya Levin 2855f29458bSAya Levin key.rsc = MLX5_SGMT_TYPE_FULL_QPC; 2865f29458bSAya Levin key.index1 = queue_idx; 2875f29458bSAya Levin key.size = PAGE_SIZE; 2885f29458bSAya Levin key.num_of_obj1 = 1; 2895f29458bSAya Levin 2905f29458bSAya Levin err = devlink_fmsg_obj_nest_start(fmsg); 2915f29458bSAya Levin if (err) 2925f29458bSAya Levin return err; 2935f29458bSAya Levin 2945f29458bSAya Levin err = mlx5e_reporter_named_obj_nest_start(fmsg, lbl); 2955f29458bSAya Levin if (err) 2965f29458bSAya Levin return err; 2975f29458bSAya Levin 2985f29458bSAya Levin err = devlink_fmsg_u32_pair_put(fmsg, "index", queue_idx); 2995f29458bSAya Levin if (err) 3005f29458bSAya Levin return err; 3015f29458bSAya Levin 3025f29458bSAya Levin err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 3035f29458bSAya Levin if (err) 3045f29458bSAya Levin return err; 3055f29458bSAya Levin 3065f29458bSAya Levin err = mlx5e_reporter_named_obj_nest_end(fmsg); 3075f29458bSAya Levin if (err) 3085f29458bSAya Levin return err; 3095f29458bSAya Levin 3105f29458bSAya Levin return devlink_fmsg_obj_nest_end(fmsg); 3115f29458bSAya Levin } 312