1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* Copyright (c) 2019 Mellanox Technologies. */
4 #include <net/devlink.h>
8 #define MLX5E_TX_REPORTER_PER_SQ_MAX_LEN 256
10 struct mlx5e_tx_err_ctx {
11 int (*recover)(struct mlx5e_txqsq *sq);
12 struct mlx5e_txqsq *sq;
15 static int mlx5e_wait_for_sq_flush(struct mlx5e_txqsq *sq)
17 unsigned long exp_time = jiffies + msecs_to_jiffies(2000);
19 while (time_before(jiffies, exp_time)) {
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);
33 static void mlx5e_reset_txqsq_cc_pc(struct mlx5e_txqsq *sq)
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);
43 static int mlx5e_sq_to_ready(struct mlx5e_txqsq *sq, int curr_state)
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};
50 msp.curr_state = curr_state;
51 msp.next_state = MLX5_SQC_STATE_RST;
53 err = mlx5e_modify_sq(mdev, sq->sqn, &msp);
55 netdev_err(dev, "Failed to move sq 0x%x to reset\n", sq->sqn);
59 memset(&msp, 0, sizeof(msp));
60 msp.curr_state = MLX5_SQC_STATE_RST;
61 msp.next_state = MLX5_SQC_STATE_RDY;
63 err = mlx5e_modify_sq(mdev, sq->sqn, &msp);
65 netdev_err(dev, "Failed to move sq 0x%x to ready\n", sq->sqn);
72 static int mlx5e_tx_reporter_err_cqe_recover(struct mlx5e_txqsq *sq)
74 struct mlx5_core_dev *mdev = sq->channel->mdev;
75 struct net_device *dev = sq->channel->netdev;
79 if (!test_bit(MLX5E_SQ_STATE_RECOVERING, &sq->state))
82 err = mlx5_core_query_sq_state(mdev, sq->sqn, &state);
84 netdev_err(dev, "Failed to query SQ 0x%x state. err = %d\n",
89 if (state != MLX5_SQC_STATE_ERR) {
90 netdev_err(dev, "SQ 0x%x not in ERROR state\n", sq->sqn);
94 mlx5e_tx_disable_queue(sq->txq);
96 err = mlx5e_wait_for_sq_flush(sq);
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.
105 err = mlx5e_sq_to_ready(sq, state);
109 mlx5e_reset_txqsq_cc_pc(sq);
110 sq->stats->recover++;
111 mlx5e_activate_txqsq(sq);
116 static int mlx5_tx_health_report(struct devlink_health_reporter *tx_reporter,
118 struct mlx5e_tx_err_ctx *err_ctx)
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);
125 return devlink_health_report(tx_reporter, err_str, err_ctx);
128 void mlx5e_tx_reporter_err_cqe(struct mlx5e_txqsq *sq)
130 char err_str[MLX5E_TX_REPORTER_PER_SQ_MAX_LEN];
131 struct mlx5e_tx_err_ctx err_ctx = {0};
134 err_ctx.recover = mlx5e_tx_reporter_err_cqe_recover;
135 sprintf(err_str, "ERR CQE on SQ: 0x%x", sq->sqn);
137 mlx5_tx_health_report(sq->channel->priv->tx_reporter, err_str,
141 static int mlx5e_tx_reporter_timeout_recover(struct mlx5e_txqsq *sq)
143 struct mlx5_eq_comp *eq = sq->cq.mcq.eq;
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);
150 eqe_count = mlx5_eq_poll_irq_disabled(eq);
151 ret = eqe_count ? false : true;
153 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
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++;
163 int mlx5e_tx_reporter_timeout(struct mlx5e_txqsq *sq)
165 char err_str[MLX5E_TX_REPORTER_PER_SQ_MAX_LEN];
166 struct mlx5e_tx_err_ctx err_ctx;
169 err_ctx.recover = mlx5e_tx_reporter_timeout_recover;
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));
175 return mlx5_tx_health_report(sq->channel->priv->tx_reporter, err_str,
179 /* state lock cannot be grabbed within this function.
180 * It can cause a dead lock or a read-after-free.
182 static int mlx5e_tx_reporter_recover_from_ctx(struct mlx5e_tx_err_ctx *err_ctx)
184 return err_ctx->recover(err_ctx->sq);
187 static int mlx5e_tx_reporter_recover_all(struct mlx5e_priv *priv)
192 mutex_lock(&priv->state_lock);
194 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
197 err = mlx5e_safe_reopen_channels(priv);
200 mutex_unlock(&priv->state_lock);
206 static int mlx5e_tx_reporter_recover(struct devlink_health_reporter *reporter,
209 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
210 struct mlx5e_tx_err_ctx *err_ctx = context;
212 return err_ctx ? mlx5e_tx_reporter_recover_from_ctx(err_ctx) :
213 mlx5e_tx_reporter_recover_all(priv);
217 mlx5e_tx_reporter_build_diagnose_output(struct devlink_fmsg *fmsg,
218 u32 sqn, u8 state, bool stopped)
222 err = devlink_fmsg_obj_nest_start(fmsg);
226 err = devlink_fmsg_u32_pair_put(fmsg, "sqn", sqn);
230 err = devlink_fmsg_u8_pair_put(fmsg, "HW state", state);
234 err = devlink_fmsg_bool_pair_put(fmsg, "stopped", stopped);
238 err = devlink_fmsg_obj_nest_end(fmsg);
245 static int mlx5e_tx_reporter_diagnose(struct devlink_health_reporter *reporter,
246 struct devlink_fmsg *fmsg)
248 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
251 mutex_lock(&priv->state_lock);
253 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
256 err = devlink_fmsg_arr_pair_nest_start(fmsg, "SQs");
260 for (i = 0; i < priv->channels.num * priv->channels.params.num_tc;
262 struct mlx5e_txqsq *sq = priv->txq2sq[i];
265 err = mlx5_core_query_sq_state(priv->mdev, sq->sqn, &state);
269 err = mlx5e_tx_reporter_build_diagnose_output(fmsg, sq->sqn,
271 netif_xmit_stopped(sq->txq));
275 err = devlink_fmsg_arr_pair_nest_end(fmsg);
280 mutex_unlock(&priv->state_lock);
284 static const struct devlink_health_reporter_ops mlx5_tx_reporter_ops = {
286 .recover = mlx5e_tx_reporter_recover,
287 .diagnose = mlx5e_tx_reporter_diagnose,
290 #define MLX5_REPORTER_TX_GRACEFUL_PERIOD 500
292 int mlx5e_tx_reporter_create(struct mlx5e_priv *priv)
294 struct mlx5_core_dev *mdev = priv->mdev;
295 struct devlink *devlink = priv_to_devlink(mdev);
298 devlink_health_reporter_create(devlink, &mlx5_tx_reporter_ops,
299 MLX5_REPORTER_TX_GRACEFUL_PERIOD,
301 if (IS_ERR(priv->tx_reporter))
302 netdev_warn(priv->netdev,
303 "Failed to create tx reporter, err = %ld\n",
304 PTR_ERR(priv->tx_reporter));
305 return IS_ERR_OR_NULL(priv->tx_reporter);
308 void mlx5e_tx_reporter_destroy(struct mlx5e_priv *priv)
310 if (IS_ERR_OR_NULL(priv->tx_reporter))
313 devlink_health_reporter_destroy(priv->tx_reporter);