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" 6c50de4afSAya Levin 72d708887SAya Levin int mlx5e_reporter_named_obj_nest_start(struct devlink_fmsg *fmsg, char *name) 82d708887SAya Levin { 92d708887SAya Levin int err; 102d708887SAya Levin 112d708887SAya Levin err = devlink_fmsg_pair_nest_start(fmsg, name); 122d708887SAya Levin if (err) 132d708887SAya Levin return err; 142d708887SAya Levin 152d708887SAya Levin err = devlink_fmsg_obj_nest_start(fmsg); 162d708887SAya Levin if (err) 172d708887SAya Levin return err; 182d708887SAya Levin 192d708887SAya Levin return 0; 202d708887SAya Levin } 212d708887SAya Levin 222d708887SAya Levin int mlx5e_reporter_named_obj_nest_end(struct devlink_fmsg *fmsg) 232d708887SAya Levin { 242d708887SAya Levin int err; 252d708887SAya Levin 262d708887SAya Levin err = devlink_fmsg_obj_nest_end(fmsg); 272d708887SAya Levin if (err) 282d708887SAya Levin return err; 292d708887SAya Levin 302d708887SAya Levin err = devlink_fmsg_pair_nest_end(fmsg); 312d708887SAya Levin if (err) 322d708887SAya Levin return err; 332d708887SAya Levin 342d708887SAya Levin return 0; 352d708887SAya Levin } 362d708887SAya Levin 372bf09e60SAya Levin int mlx5e_reporter_cq_diagnose(struct mlx5e_cq *cq, struct devlink_fmsg *fmsg) 382bf09e60SAya Levin { 392bf09e60SAya Levin struct mlx5e_priv *priv = cq->channel->priv; 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 452bf09e60SAya Levin err = mlx5_core_query_cq(priv->mdev, &cq->mcq, out, sizeof(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 522bf09e60SAya Levin err = mlx5e_reporter_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 642bf09e60SAya Levin err = mlx5e_reporter_named_obj_nest_end(fmsg); 652bf09e60SAya Levin if (err) 662bf09e60SAya Levin return err; 672bf09e60SAya Levin 682bf09e60SAya Levin return 0; 692bf09e60SAya Levin } 702bf09e60SAya Levin 712bf09e60SAya Levin int mlx5e_reporter_cq_common_diagnose(struct mlx5e_cq *cq, struct devlink_fmsg *fmsg) 722bf09e60SAya Levin { 732bf09e60SAya Levin u8 cq_log_stride; 742bf09e60SAya Levin u32 cq_sz; 752bf09e60SAya Levin int err; 762bf09e60SAya Levin 772bf09e60SAya Levin cq_sz = mlx5_cqwq_get_size(&cq->wq); 782bf09e60SAya Levin cq_log_stride = mlx5_cqwq_get_log_stride_size(&cq->wq); 792bf09e60SAya Levin 802bf09e60SAya Levin err = mlx5e_reporter_named_obj_nest_start(fmsg, "CQ"); 812bf09e60SAya Levin if (err) 822bf09e60SAya Levin return err; 832bf09e60SAya Levin 842bf09e60SAya Levin err = devlink_fmsg_u64_pair_put(fmsg, "stride size", BIT(cq_log_stride)); 852bf09e60SAya Levin if (err) 862bf09e60SAya Levin return err; 872bf09e60SAya Levin 882bf09e60SAya Levin err = devlink_fmsg_u32_pair_put(fmsg, "size", cq_sz); 892bf09e60SAya Levin if (err) 902bf09e60SAya Levin return err; 912bf09e60SAya Levin 922bf09e60SAya Levin err = mlx5e_reporter_named_obj_nest_end(fmsg); 932bf09e60SAya Levin if (err) 942bf09e60SAya Levin return err; 952bf09e60SAya Levin 962bf09e60SAya Levin return 0; 972bf09e60SAya Levin } 982bf09e60SAya Levin 9911af6a6dSAya Levin int mlx5e_health_create_reporters(struct mlx5e_priv *priv) 10011af6a6dSAya Levin { 10111af6a6dSAya Levin return mlx5e_reporter_tx_create(priv); 10211af6a6dSAya Levin } 10311af6a6dSAya Levin 10411af6a6dSAya Levin void mlx5e_health_destroy_reporters(struct mlx5e_priv *priv) 10511af6a6dSAya Levin { 10611af6a6dSAya Levin mlx5e_reporter_tx_destroy(priv); 10711af6a6dSAya Levin } 10811af6a6dSAya Levin 10911af6a6dSAya Levin void mlx5e_health_channels_update(struct mlx5e_priv *priv) 11011af6a6dSAya Levin { 11111af6a6dSAya Levin if (priv->tx_reporter) 11211af6a6dSAya Levin devlink_health_reporter_state_update(priv->tx_reporter, 11311af6a6dSAya Levin DEVLINK_HEALTH_REPORTER_STATE_HEALTHY); 11411af6a6dSAya Levin } 11511af6a6dSAya Levin 116c50de4afSAya Levin int mlx5e_health_sq_to_ready(struct mlx5e_channel *channel, u32 sqn) 117c50de4afSAya Levin { 118c50de4afSAya Levin struct mlx5_core_dev *mdev = channel->mdev; 119c50de4afSAya Levin struct net_device *dev = channel->netdev; 120c50de4afSAya Levin struct mlx5e_modify_sq_param msp = {}; 121c50de4afSAya Levin int err; 122c50de4afSAya Levin 123c50de4afSAya Levin msp.curr_state = MLX5_SQC_STATE_ERR; 124c50de4afSAya Levin msp.next_state = MLX5_SQC_STATE_RST; 125c50de4afSAya Levin 126c50de4afSAya Levin err = mlx5e_modify_sq(mdev, sqn, &msp); 127c50de4afSAya Levin if (err) { 128c50de4afSAya Levin netdev_err(dev, "Failed to move sq 0x%x to reset\n", sqn); 129c50de4afSAya Levin return err; 130c50de4afSAya Levin } 131c50de4afSAya Levin 132c50de4afSAya Levin memset(&msp, 0, sizeof(msp)); 133c50de4afSAya Levin msp.curr_state = MLX5_SQC_STATE_RST; 134c50de4afSAya Levin msp.next_state = MLX5_SQC_STATE_RDY; 135c50de4afSAya Levin 136c50de4afSAya Levin err = mlx5e_modify_sq(mdev, sqn, &msp); 137c50de4afSAya Levin if (err) { 138c50de4afSAya Levin netdev_err(dev, "Failed to move sq 0x%x to ready\n", sqn); 139c50de4afSAya Levin return err; 140c50de4afSAya Levin } 141c50de4afSAya Levin 142c50de4afSAya Levin return 0; 143c50de4afSAya Levin } 144c50de4afSAya Levin 145c50de4afSAya Levin int mlx5e_health_recover_channels(struct mlx5e_priv *priv) 146c50de4afSAya Levin { 147c50de4afSAya Levin int err = 0; 148c50de4afSAya Levin 149c50de4afSAya Levin rtnl_lock(); 150c50de4afSAya Levin mutex_lock(&priv->state_lock); 151c50de4afSAya Levin 152c50de4afSAya Levin if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) 153c50de4afSAya Levin goto out; 154c50de4afSAya Levin 155c50de4afSAya Levin err = mlx5e_safe_reopen_channels(priv); 156c50de4afSAya Levin 157c50de4afSAya Levin out: 158c50de4afSAya Levin mutex_unlock(&priv->state_lock); 159c50de4afSAya Levin rtnl_unlock(); 160c50de4afSAya Levin 161c50de4afSAya Levin return err; 162c50de4afSAya Levin } 163c50de4afSAya Levin 164c50de4afSAya Levin int mlx5e_health_channel_eq_recover(struct mlx5_eq_comp *eq, struct mlx5e_channel *channel) 165c50de4afSAya Levin { 166c50de4afSAya Levin u32 eqe_count; 167c50de4afSAya Levin 168c50de4afSAya Levin netdev_err(channel->netdev, "EQ 0x%x: Cons = 0x%x, irqn = 0x%x\n", 169c50de4afSAya Levin eq->core.eqn, eq->core.cons_index, eq->core.irqn); 170c50de4afSAya Levin 171c50de4afSAya Levin eqe_count = mlx5_eq_poll_irq_disabled(eq); 172c50de4afSAya Levin if (!eqe_count) 173c50de4afSAya Levin return -EIO; 174c50de4afSAya Levin 175c50de4afSAya Levin netdev_err(channel->netdev, "Recovered %d eqes on EQ 0x%x\n", 176c50de4afSAya Levin eqe_count, eq->core.eqn); 177c50de4afSAya Levin 178c50de4afSAya Levin channel->stats->eq_rearm++; 179c50de4afSAya Levin return 0; 180c50de4afSAya Levin } 181c50de4afSAya Levin 182c50de4afSAya Levin int mlx5e_health_report(struct mlx5e_priv *priv, 183c50de4afSAya Levin struct devlink_health_reporter *reporter, char *err_str, 184c50de4afSAya Levin struct mlx5e_err_ctx *err_ctx) 185c50de4afSAya Levin { 186c50de4afSAya Levin if (!reporter) { 187c50de4afSAya Levin netdev_err(priv->netdev, err_str); 188c50de4afSAya Levin return err_ctx->recover(&err_ctx->ctx); 189c50de4afSAya Levin } 190c50de4afSAya Levin return devlink_health_report(reporter, err_str, err_ctx); 191c50de4afSAya Levin } 192