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 8d5cbedd7SAya Levin int mlx5e_health_fmsg_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 23d5cbedd7SAya Levin int mlx5e_health_fmsg_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 38d5cbedd7SAya Levin int mlx5e_health_cq_diag_fmsg(struct mlx5e_cq *cq, struct devlink_fmsg *fmsg) 392bf09e60SAya Levin { 402bf09e60SAya Levin u32 out[MLX5_ST_SZ_DW(query_cq_out)] = {}; 412bf09e60SAya Levin u8 hw_status; 422bf09e60SAya Levin void *cqc; 432bf09e60SAya Levin int err; 442bf09e60SAya Levin 454d0b7ef9SAya Levin err = mlx5_core_query_cq(cq->mdev, &cq->mcq, out); 462bf09e60SAya Levin if (err) 472bf09e60SAya Levin return err; 482bf09e60SAya Levin 492bf09e60SAya Levin cqc = MLX5_ADDR_OF(query_cq_out, out, cq_context); 502bf09e60SAya Levin hw_status = MLX5_GET(cqc, cqc, status); 512bf09e60SAya Levin 52d5cbedd7SAya Levin err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "CQ"); 532bf09e60SAya Levin if (err) 542bf09e60SAya Levin return err; 552bf09e60SAya Levin 562bf09e60SAya Levin err = devlink_fmsg_u32_pair_put(fmsg, "cqn", cq->mcq.cqn); 572bf09e60SAya Levin if (err) 582bf09e60SAya Levin return err; 592bf09e60SAya Levin 602bf09e60SAya Levin err = devlink_fmsg_u8_pair_put(fmsg, "HW status", hw_status); 612bf09e60SAya Levin if (err) 622bf09e60SAya Levin return err; 632bf09e60SAya Levin 643c9d1699SAya Levin err = devlink_fmsg_u32_pair_put(fmsg, "ci", mlx5_cqwq_get_ci(&cq->wq)); 653c9d1699SAya Levin if (err) 663c9d1699SAya Levin return err; 673c9d1699SAya Levin 683c9d1699SAya Levin err = devlink_fmsg_u32_pair_put(fmsg, "size", mlx5_cqwq_get_size(&cq->wq)); 693c9d1699SAya Levin if (err) 703c9d1699SAya Levin return err; 713c9d1699SAya Levin 72d5cbedd7SAya Levin err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 732bf09e60SAya Levin if (err) 742bf09e60SAya Levin return err; 752bf09e60SAya Levin 762bf09e60SAya Levin return 0; 772bf09e60SAya Levin } 782bf09e60SAya Levin 79d5cbedd7SAya Levin int mlx5e_health_cq_common_diag_fmsg(struct mlx5e_cq *cq, struct devlink_fmsg *fmsg) 802bf09e60SAya Levin { 812bf09e60SAya Levin u8 cq_log_stride; 822bf09e60SAya Levin u32 cq_sz; 832bf09e60SAya Levin int err; 842bf09e60SAya Levin 852bf09e60SAya Levin cq_sz = mlx5_cqwq_get_size(&cq->wq); 862bf09e60SAya Levin cq_log_stride = mlx5_cqwq_get_log_stride_size(&cq->wq); 872bf09e60SAya Levin 88d5cbedd7SAya Levin err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "CQ"); 892bf09e60SAya Levin if (err) 902bf09e60SAya Levin return err; 912bf09e60SAya Levin 922bf09e60SAya Levin err = devlink_fmsg_u64_pair_put(fmsg, "stride size", BIT(cq_log_stride)); 932bf09e60SAya Levin if (err) 942bf09e60SAya Levin return err; 952bf09e60SAya Levin 962bf09e60SAya Levin err = devlink_fmsg_u32_pair_put(fmsg, "size", cq_sz); 972bf09e60SAya Levin if (err) 982bf09e60SAya Levin return err; 992bf09e60SAya Levin 100d5cbedd7SAya Levin err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 1012bf09e60SAya Levin if (err) 1022bf09e60SAya Levin return err; 1032bf09e60SAya Levin 1042bf09e60SAya Levin return 0; 1052bf09e60SAya Levin } 1062bf09e60SAya Levin 10756837c2aSAya Levin int mlx5e_health_eq_diag_fmsg(struct mlx5_eq_comp *eq, struct devlink_fmsg *fmsg) 10856837c2aSAya Levin { 10956837c2aSAya Levin int err; 11056837c2aSAya Levin 11156837c2aSAya Levin err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "EQ"); 11256837c2aSAya Levin if (err) 11356837c2aSAya Levin return err; 11456837c2aSAya Levin 11556837c2aSAya Levin err = devlink_fmsg_u8_pair_put(fmsg, "eqn", eq->core.eqn); 11656837c2aSAya Levin if (err) 11756837c2aSAya Levin return err; 11856837c2aSAya Levin 11956837c2aSAya Levin err = devlink_fmsg_u32_pair_put(fmsg, "irqn", eq->core.irqn); 12056837c2aSAya Levin if (err) 12156837c2aSAya Levin return err; 12256837c2aSAya Levin 12356837c2aSAya Levin err = devlink_fmsg_u32_pair_put(fmsg, "vecidx", eq->core.vecidx); 12456837c2aSAya Levin if (err) 12556837c2aSAya Levin return err; 12656837c2aSAya Levin 12756837c2aSAya Levin err = devlink_fmsg_u32_pair_put(fmsg, "ci", eq->core.cons_index); 12856837c2aSAya Levin if (err) 12956837c2aSAya Levin return err; 13056837c2aSAya Levin 13156837c2aSAya Levin err = devlink_fmsg_u32_pair_put(fmsg, "size", eq->core.nent); 13256837c2aSAya Levin if (err) 13356837c2aSAya Levin return err; 13456837c2aSAya Levin 13556837c2aSAya Levin return mlx5e_health_fmsg_named_obj_nest_end(fmsg); 13656837c2aSAya Levin } 13756837c2aSAya Levin 138b3ea4c4fSEran Ben Elisha void mlx5e_health_create_reporters(struct mlx5e_priv *priv) 13911af6a6dSAya Levin { 140b3ea4c4fSEran Ben Elisha mlx5e_reporter_tx_create(priv); 141b3ea4c4fSEran Ben Elisha mlx5e_reporter_rx_create(priv); 14211af6a6dSAya Levin } 14311af6a6dSAya Levin 14411af6a6dSAya Levin void mlx5e_health_destroy_reporters(struct mlx5e_priv *priv) 14511af6a6dSAya Levin { 1469032e719SAya Levin mlx5e_reporter_rx_destroy(priv); 14711af6a6dSAya Levin mlx5e_reporter_tx_destroy(priv); 14811af6a6dSAya Levin } 14911af6a6dSAya Levin 15011af6a6dSAya Levin void mlx5e_health_channels_update(struct mlx5e_priv *priv) 15111af6a6dSAya Levin { 15211af6a6dSAya Levin if (priv->tx_reporter) 15311af6a6dSAya Levin devlink_health_reporter_state_update(priv->tx_reporter, 15411af6a6dSAya Levin DEVLINK_HEALTH_REPORTER_STATE_HEALTHY); 1559032e719SAya Levin if (priv->rx_reporter) 1569032e719SAya Levin devlink_health_reporter_state_update(priv->rx_reporter, 1579032e719SAya Levin DEVLINK_HEALTH_REPORTER_STATE_HEALTHY); 15811af6a6dSAya Levin } 15911af6a6dSAya Levin 160*4ad40d8eSEran Ben Elisha int mlx5e_health_sq_to_ready(struct mlx5_core_dev *mdev, struct net_device *dev, u32 sqn) 161c50de4afSAya Levin { 162c50de4afSAya Levin struct mlx5e_modify_sq_param msp = {}; 163c50de4afSAya Levin int err; 164c50de4afSAya Levin 165c50de4afSAya Levin msp.curr_state = MLX5_SQC_STATE_ERR; 166c50de4afSAya Levin msp.next_state = MLX5_SQC_STATE_RST; 167c50de4afSAya Levin 168c50de4afSAya Levin err = mlx5e_modify_sq(mdev, sqn, &msp); 169c50de4afSAya Levin if (err) { 170c50de4afSAya Levin netdev_err(dev, "Failed to move sq 0x%x to reset\n", sqn); 171c50de4afSAya Levin return err; 172c50de4afSAya Levin } 173c50de4afSAya Levin 174c50de4afSAya Levin memset(&msp, 0, sizeof(msp)); 175c50de4afSAya Levin msp.curr_state = MLX5_SQC_STATE_RST; 176c50de4afSAya Levin msp.next_state = MLX5_SQC_STATE_RDY; 177c50de4afSAya Levin 178c50de4afSAya Levin err = mlx5e_modify_sq(mdev, sqn, &msp); 179c50de4afSAya Levin if (err) { 180c50de4afSAya Levin netdev_err(dev, "Failed to move sq 0x%x to ready\n", sqn); 181c50de4afSAya Levin return err; 182c50de4afSAya Levin } 183c50de4afSAya Levin 184c50de4afSAya Levin return 0; 185c50de4afSAya Levin } 186c50de4afSAya Levin 187c50de4afSAya Levin int mlx5e_health_recover_channels(struct mlx5e_priv *priv) 188c50de4afSAya Levin { 189c50de4afSAya Levin int err = 0; 190c50de4afSAya Levin 191c50de4afSAya Levin rtnl_lock(); 192c50de4afSAya Levin mutex_lock(&priv->state_lock); 193c50de4afSAya Levin 194c50de4afSAya Levin if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) 195c50de4afSAya Levin goto out; 196c50de4afSAya Levin 197c50de4afSAya Levin err = mlx5e_safe_reopen_channels(priv); 198c50de4afSAya Levin 199c50de4afSAya Levin out: 200c50de4afSAya Levin mutex_unlock(&priv->state_lock); 201c50de4afSAya Levin rtnl_unlock(); 202c50de4afSAya Levin 203c50de4afSAya Levin return err; 204c50de4afSAya Levin } 205c50de4afSAya Levin 206521f31afSAya Levin int mlx5e_health_channel_eq_recover(struct net_device *dev, struct mlx5_eq_comp *eq, 207521f31afSAya Levin struct mlx5e_ch_stats *stats) 208c50de4afSAya Levin { 209c50de4afSAya Levin u32 eqe_count; 210c50de4afSAya Levin 211521f31afSAya Levin netdev_err(dev, "EQ 0x%x: Cons = 0x%x, irqn = 0x%x\n", 212c50de4afSAya Levin eq->core.eqn, eq->core.cons_index, eq->core.irqn); 213c50de4afSAya Levin 214c50de4afSAya Levin eqe_count = mlx5_eq_poll_irq_disabled(eq); 215c50de4afSAya Levin if (!eqe_count) 216c50de4afSAya Levin return -EIO; 217c50de4afSAya Levin 218521f31afSAya Levin netdev_err(dev, "Recovered %d eqes on EQ 0x%x\n", 219c50de4afSAya Levin eqe_count, eq->core.eqn); 220c50de4afSAya Levin 221521f31afSAya Levin stats->eq_rearm++; 222c50de4afSAya Levin return 0; 223c50de4afSAya Levin } 224c50de4afSAya Levin 225c50de4afSAya Levin int mlx5e_health_report(struct mlx5e_priv *priv, 226c50de4afSAya Levin struct devlink_health_reporter *reporter, char *err_str, 227c50de4afSAya Levin struct mlx5e_err_ctx *err_ctx) 228c50de4afSAya Levin { 229b21aef7eSJoe Perches netdev_err(priv->netdev, "%s\n", err_str); 23099cda454SEran Ben Elisha 23199cda454SEran Ben Elisha if (!reporter) 2321ad6c43cSAya Levin return err_ctx->recover(err_ctx->ctx); 23399cda454SEran Ben Elisha 234c50de4afSAya Levin return devlink_health_report(reporter, err_str, err_ctx); 235c50de4afSAya Levin } 2365f29458bSAya Levin 2375f29458bSAya Levin #define MLX5_HEALTH_DEVLINK_MAX_SIZE 1024 2385f29458bSAya Levin static int mlx5e_health_rsc_fmsg_binary(struct devlink_fmsg *fmsg, 2395f29458bSAya Levin const void *value, u32 value_len) 2405f29458bSAya Levin 2415f29458bSAya Levin { 2425f29458bSAya Levin u32 data_size; 24319f5b63bSMoshe Tal int err = 0; 2445f29458bSAya Levin u32 offset; 2455f29458bSAya Levin 2465f29458bSAya Levin for (offset = 0; offset < value_len; offset += data_size) { 2475f29458bSAya Levin data_size = value_len - offset; 2485f29458bSAya Levin if (data_size > MLX5_HEALTH_DEVLINK_MAX_SIZE) 2495f29458bSAya Levin data_size = MLX5_HEALTH_DEVLINK_MAX_SIZE; 2505f29458bSAya Levin err = devlink_fmsg_binary_put(fmsg, value + offset, data_size); 2515f29458bSAya Levin if (err) 2525f29458bSAya Levin break; 2535f29458bSAya Levin } 2545f29458bSAya Levin return err; 2555f29458bSAya Levin } 2565f29458bSAya Levin 2575f29458bSAya Levin int mlx5e_health_rsc_fmsg_dump(struct mlx5e_priv *priv, struct mlx5_rsc_key *key, 2585f29458bSAya Levin struct devlink_fmsg *fmsg) 2595f29458bSAya Levin { 2605f29458bSAya Levin struct mlx5_core_dev *mdev = priv->mdev; 2615f29458bSAya Levin struct mlx5_rsc_dump_cmd *cmd; 2625f29458bSAya Levin struct page *page; 2635f29458bSAya Levin int cmd_err, err; 2645f29458bSAya Levin int end_err; 2655f29458bSAya Levin int size; 2665f29458bSAya Levin 2675f29458bSAya Levin if (IS_ERR_OR_NULL(mdev->rsc_dump)) 2685f29458bSAya Levin return -EOPNOTSUPP; 2695f29458bSAya Levin 2705f29458bSAya Levin page = alloc_page(GFP_KERNEL); 2715f29458bSAya Levin if (!page) 2725f29458bSAya Levin return -ENOMEM; 2735f29458bSAya Levin 2745f29458bSAya Levin err = devlink_fmsg_binary_pair_nest_start(fmsg, "data"); 2755f29458bSAya Levin if (err) 2765f29458bSAya Levin return err; 2775f29458bSAya Levin 2785f29458bSAya Levin cmd = mlx5_rsc_dump_cmd_create(mdev, key); 2795f29458bSAya Levin if (IS_ERR(cmd)) { 2805f29458bSAya Levin err = PTR_ERR(cmd); 2815f29458bSAya Levin goto free_page; 2825f29458bSAya Levin } 2835f29458bSAya Levin 2845f29458bSAya Levin do { 2855f29458bSAya Levin cmd_err = mlx5_rsc_dump_next(mdev, cmd, page, &size); 2865f29458bSAya Levin if (cmd_err < 0) { 2875f29458bSAya Levin err = cmd_err; 2885f29458bSAya Levin goto destroy_cmd; 2895f29458bSAya Levin } 2905f29458bSAya Levin 2915f29458bSAya Levin err = mlx5e_health_rsc_fmsg_binary(fmsg, page_address(page), size); 2925f29458bSAya Levin if (err) 2935f29458bSAya Levin goto destroy_cmd; 2945f29458bSAya Levin 2955f29458bSAya Levin } while (cmd_err > 0); 2965f29458bSAya Levin 2975f29458bSAya Levin destroy_cmd: 2985f29458bSAya Levin mlx5_rsc_dump_cmd_destroy(cmd); 2995f29458bSAya Levin end_err = devlink_fmsg_binary_pair_nest_end(fmsg); 3005f29458bSAya Levin if (end_err) 3015f29458bSAya Levin err = end_err; 3025f29458bSAya Levin free_page: 3035f29458bSAya Levin __free_page(page); 3045f29458bSAya Levin return err; 3055f29458bSAya Levin } 3065f29458bSAya Levin 3075f29458bSAya Levin int mlx5e_health_queue_dump(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg, 3085f29458bSAya Levin int queue_idx, char *lbl) 3095f29458bSAya Levin { 3105f29458bSAya Levin struct mlx5_rsc_key key = {}; 3115f29458bSAya Levin int err; 3125f29458bSAya Levin 3135f29458bSAya Levin key.rsc = MLX5_SGMT_TYPE_FULL_QPC; 3145f29458bSAya Levin key.index1 = queue_idx; 3155f29458bSAya Levin key.size = PAGE_SIZE; 3165f29458bSAya Levin key.num_of_obj1 = 1; 3175f29458bSAya Levin 3185f29458bSAya Levin err = devlink_fmsg_obj_nest_start(fmsg); 3195f29458bSAya Levin if (err) 3205f29458bSAya Levin return err; 3215f29458bSAya Levin 322d5cbedd7SAya Levin err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, lbl); 3235f29458bSAya Levin if (err) 3245f29458bSAya Levin return err; 3255f29458bSAya Levin 3265f29458bSAya Levin err = devlink_fmsg_u32_pair_put(fmsg, "index", queue_idx); 3275f29458bSAya Levin if (err) 3285f29458bSAya Levin return err; 3295f29458bSAya Levin 3305f29458bSAya Levin err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 3315f29458bSAya Levin if (err) 3325f29458bSAya Levin return err; 3335f29458bSAya Levin 334d5cbedd7SAya Levin err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 3355f29458bSAya Levin if (err) 3365f29458bSAya Levin return err; 3375f29458bSAya Levin 3385f29458bSAya Levin return devlink_fmsg_obj_nest_end(fmsg); 3395f29458bSAya Levin } 340