1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* Copyright (c) 2019 Mellanox Technologies. */ 3 4 #include <net/devlink.h> 5 #include "reporter.h" 6 #include "lib/eq.h" 7 8 #define MLX5E_TX_REPORTER_PER_SQ_MAX_LEN 256 9 10 struct mlx5e_tx_err_ctx { 11 int (*recover)(struct mlx5e_txqsq *sq); 12 struct mlx5e_txqsq *sq; 13 }; 14 15 static int mlx5e_wait_for_sq_flush(struct mlx5e_txqsq *sq) 16 { 17 unsigned long exp_time = jiffies + msecs_to_jiffies(2000); 18 19 while (time_before(jiffies, exp_time)) { 20 if (sq->cc == sq->pc) 21 return 0; 22 23 msleep(20); 24 } 25 26 netdev_err(sq->channel->netdev, 27 "Wait for SQ 0x%x flush timeout (sq cc = 0x%x, sq pc = 0x%x)\n", 28 sq->sqn, sq->cc, sq->pc); 29 30 return -ETIMEDOUT; 31 } 32 33 static void mlx5e_reset_txqsq_cc_pc(struct mlx5e_txqsq *sq) 34 { 35 WARN_ONCE(sq->cc != sq->pc, 36 "SQ 0x%x: cc (0x%x) != pc (0x%x)\n", 37 sq->sqn, sq->cc, sq->pc); 38 sq->cc = 0; 39 sq->dma_fifo_cc = 0; 40 sq->pc = 0; 41 } 42 43 static int mlx5e_sq_to_ready(struct mlx5e_txqsq *sq, int curr_state) 44 { 45 struct mlx5_core_dev *mdev = sq->channel->mdev; 46 struct net_device *dev = sq->channel->netdev; 47 struct mlx5e_modify_sq_param msp = {0}; 48 int err; 49 50 msp.curr_state = curr_state; 51 msp.next_state = MLX5_SQC_STATE_RST; 52 53 err = mlx5e_modify_sq(mdev, sq->sqn, &msp); 54 if (err) { 55 netdev_err(dev, "Failed to move sq 0x%x to reset\n", sq->sqn); 56 return err; 57 } 58 59 memset(&msp, 0, sizeof(msp)); 60 msp.curr_state = MLX5_SQC_STATE_RST; 61 msp.next_state = MLX5_SQC_STATE_RDY; 62 63 err = mlx5e_modify_sq(mdev, sq->sqn, &msp); 64 if (err) { 65 netdev_err(dev, "Failed to move sq 0x%x to ready\n", sq->sqn); 66 return err; 67 } 68 69 return 0; 70 } 71 72 static int mlx5e_tx_reporter_err_cqe_recover(struct mlx5e_txqsq *sq) 73 { 74 struct mlx5_core_dev *mdev = sq->channel->mdev; 75 struct net_device *dev = sq->channel->netdev; 76 u8 state; 77 int err; 78 79 if (!test_bit(MLX5E_SQ_STATE_RECOVERING, &sq->state)) 80 return 0; 81 82 err = mlx5_core_query_sq_state(mdev, sq->sqn, &state); 83 if (err) { 84 netdev_err(dev, "Failed to query SQ 0x%x state. err = %d\n", 85 sq->sqn, err); 86 return err; 87 } 88 89 if (state != MLX5_SQC_STATE_ERR) { 90 netdev_err(dev, "SQ 0x%x not in ERROR state\n", sq->sqn); 91 return -EINVAL; 92 } 93 94 mlx5e_tx_disable_queue(sq->txq); 95 96 err = mlx5e_wait_for_sq_flush(sq); 97 if (err) 98 return err; 99 100 /* At this point, no new packets will arrive from the stack as TXQ is 101 * marked with QUEUE_STATE_DRV_XOFF. In addition, NAPI cleared all 102 * pending WQEs. SQ can safely reset the SQ. 103 */ 104 105 err = mlx5e_sq_to_ready(sq, state); 106 if (err) 107 return err; 108 109 mlx5e_reset_txqsq_cc_pc(sq); 110 sq->stats->recover++; 111 mlx5e_activate_txqsq(sq); 112 113 return 0; 114 } 115 116 static int mlx5_tx_health_report(struct devlink_health_reporter *tx_reporter, 117 char *err_str, 118 struct mlx5e_tx_err_ctx *err_ctx) 119 { 120 if (IS_ERR_OR_NULL(tx_reporter)) { 121 netdev_err(err_ctx->sq->channel->netdev, err_str); 122 return err_ctx->recover(err_ctx->sq); 123 } 124 125 return devlink_health_report(tx_reporter, err_str, err_ctx); 126 } 127 128 void mlx5e_tx_reporter_err_cqe(struct mlx5e_txqsq *sq) 129 { 130 char err_str[MLX5E_TX_REPORTER_PER_SQ_MAX_LEN]; 131 struct mlx5e_tx_err_ctx err_ctx = {0}; 132 133 err_ctx.sq = sq; 134 err_ctx.recover = mlx5e_tx_reporter_err_cqe_recover; 135 sprintf(err_str, "ERR CQE on SQ: 0x%x", sq->sqn); 136 137 mlx5_tx_health_report(sq->channel->priv->tx_reporter, err_str, 138 &err_ctx); 139 } 140 141 static int mlx5e_tx_reporter_timeout_recover(struct mlx5e_txqsq *sq) 142 { 143 struct mlx5_eq_comp *eq = sq->cq.mcq.eq; 144 u32 eqe_count; 145 int ret; 146 147 netdev_err(sq->channel->netdev, "EQ 0x%x: Cons = 0x%x, irqn = 0x%x\n", 148 eq->core.eqn, eq->core.cons_index, eq->core.irqn); 149 150 eqe_count = mlx5_eq_poll_irq_disabled(eq); 151 ret = eqe_count ? true : false; 152 if (!eqe_count) { 153 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state); 154 return ret; 155 } 156 157 netdev_err(sq->channel->netdev, "Recover %d eqes on EQ 0x%x\n", 158 eqe_count, eq->core.eqn); 159 sq->channel->stats->eq_rearm++; 160 return ret; 161 } 162 163 int mlx5e_tx_reporter_timeout(struct mlx5e_txqsq *sq) 164 { 165 char err_str[MLX5E_TX_REPORTER_PER_SQ_MAX_LEN]; 166 struct mlx5e_tx_err_ctx err_ctx; 167 168 err_ctx.sq = sq; 169 err_ctx.recover = mlx5e_tx_reporter_timeout_recover; 170 sprintf(err_str, 171 "TX timeout on queue: %d, SQ: 0x%x, CQ: 0x%x, SQ Cons: 0x%x SQ Prod: 0x%x, usecs since last trans: %u\n", 172 sq->channel->ix, sq->sqn, sq->cq.mcq.cqn, sq->cc, sq->pc, 173 jiffies_to_usecs(jiffies - sq->txq->trans_start)); 174 175 return mlx5_tx_health_report(sq->channel->priv->tx_reporter, err_str, 176 &err_ctx); 177 } 178 179 /* state lock cannot be grabbed within this function. 180 * It can cause a dead lock or a read-after-free. 181 */ 182 int mlx5e_tx_reporter_recover_from_ctx(struct mlx5e_tx_err_ctx *err_ctx) 183 { 184 return err_ctx->recover(err_ctx->sq); 185 } 186 187 static int mlx5e_tx_reporter_recover_all(struct mlx5e_priv *priv) 188 { 189 int err; 190 191 rtnl_lock(); 192 mutex_lock(&priv->state_lock); 193 mlx5e_close_locked(priv->netdev); 194 err = mlx5e_open_locked(priv->netdev); 195 mutex_unlock(&priv->state_lock); 196 rtnl_unlock(); 197 198 return err; 199 } 200 201 static int mlx5e_tx_reporter_recover(struct devlink_health_reporter *reporter, 202 void *context) 203 { 204 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter); 205 struct mlx5e_tx_err_ctx *err_ctx = context; 206 207 return err_ctx ? mlx5e_tx_reporter_recover_from_ctx(err_ctx) : 208 mlx5e_tx_reporter_recover_all(priv); 209 } 210 211 static int 212 mlx5e_tx_reporter_build_diagnose_output(struct devlink_fmsg *fmsg, 213 u32 sqn, u8 state, bool stopped) 214 { 215 int err; 216 217 err = devlink_fmsg_obj_nest_start(fmsg); 218 if (err) 219 return err; 220 221 err = devlink_fmsg_u32_pair_put(fmsg, "sqn", sqn); 222 if (err) 223 return err; 224 225 err = devlink_fmsg_u8_pair_put(fmsg, "HW state", state); 226 if (err) 227 return err; 228 229 err = devlink_fmsg_bool_pair_put(fmsg, "stopped", stopped); 230 if (err) 231 return err; 232 233 err = devlink_fmsg_obj_nest_end(fmsg); 234 if (err) 235 return err; 236 237 return 0; 238 } 239 240 static int mlx5e_tx_reporter_diagnose(struct devlink_health_reporter *reporter, 241 struct devlink_fmsg *fmsg) 242 { 243 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter); 244 int i, err = 0; 245 246 mutex_lock(&priv->state_lock); 247 248 if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) 249 goto unlock; 250 251 err = devlink_fmsg_arr_pair_nest_start(fmsg, "SQs"); 252 if (err) 253 goto unlock; 254 255 for (i = 0; i < priv->channels.num * priv->channels.params.num_tc; 256 i++) { 257 struct mlx5e_txqsq *sq = priv->txq2sq[i]; 258 u8 state; 259 260 err = mlx5_core_query_sq_state(priv->mdev, sq->sqn, &state); 261 if (err) 262 break; 263 264 err = mlx5e_tx_reporter_build_diagnose_output(fmsg, sq->sqn, 265 state, 266 netif_xmit_stopped(sq->txq)); 267 if (err) 268 break; 269 } 270 err = devlink_fmsg_arr_pair_nest_end(fmsg); 271 if (err) 272 goto unlock; 273 274 unlock: 275 mutex_unlock(&priv->state_lock); 276 return err; 277 } 278 279 static const struct devlink_health_reporter_ops mlx5_tx_reporter_ops = { 280 .name = "tx", 281 .recover = mlx5e_tx_reporter_recover, 282 .diagnose = mlx5e_tx_reporter_diagnose, 283 }; 284 285 #define MLX5_REPORTER_TX_GRACEFUL_PERIOD 500 286 287 int mlx5e_tx_reporter_create(struct mlx5e_priv *priv) 288 { 289 struct mlx5_core_dev *mdev = priv->mdev; 290 struct devlink *devlink = priv_to_devlink(mdev); 291 292 priv->tx_reporter = 293 devlink_health_reporter_create(devlink, &mlx5_tx_reporter_ops, 294 MLX5_REPORTER_TX_GRACEFUL_PERIOD, 295 true, priv); 296 if (IS_ERR(priv->tx_reporter)) 297 netdev_warn(priv->netdev, 298 "Failed to create tx reporter, err = %ld\n", 299 PTR_ERR(priv->tx_reporter)); 300 return PTR_ERR_OR_ZERO(priv->tx_reporter); 301 } 302 303 void mlx5e_tx_reporter_destroy(struct mlx5e_priv *priv) 304 { 305 if (IS_ERR_OR_NULL(priv->tx_reporter)) 306 return; 307 308 devlink_health_reporter_destroy(priv->tx_reporter); 309 } 310