1 /* SPDX-License-Identifier: GPL-2.0 */
2 /* Copyright (c) 2019 Mellanox Technologies. */
3
4 #include "health.h"
5
mlx5e_wait_for_sq_flush(struct mlx5e_txqsq * sq)6 static int mlx5e_wait_for_sq_flush(struct mlx5e_txqsq *sq)
7 {
8 unsigned long exp_time = jiffies + msecs_to_jiffies(2000);
9
10 while (time_before(jiffies, exp_time)) {
11 if (sq->cc == sq->pc)
12 return 0;
13
14 msleep(20);
15 }
16
17 netdev_err(sq->channel->netdev,
18 "Wait for SQ 0x%x flush timeout (sq cc = 0x%x, sq pc = 0x%x)\n",
19 sq->sqn, sq->cc, sq->pc);
20
21 return -ETIMEDOUT;
22 }
23
mlx5e_reset_txqsq_cc_pc(struct mlx5e_txqsq * sq)24 static void mlx5e_reset_txqsq_cc_pc(struct mlx5e_txqsq *sq)
25 {
26 WARN_ONCE(sq->cc != sq->pc,
27 "SQ 0x%x: cc (0x%x) != pc (0x%x)\n",
28 sq->sqn, sq->cc, sq->pc);
29 sq->cc = 0;
30 sq->dma_fifo_cc = 0;
31 sq->pc = 0;
32 }
33
mlx5e_tx_reporter_err_cqe_recover(void * ctx)34 static int mlx5e_tx_reporter_err_cqe_recover(void *ctx)
35 {
36 struct mlx5_core_dev *mdev;
37 struct net_device *dev;
38 struct mlx5e_txqsq *sq;
39 u8 state;
40 int err;
41
42 sq = ctx;
43 mdev = sq->channel->mdev;
44 dev = sq->channel->netdev;
45
46 if (!test_bit(MLX5E_SQ_STATE_RECOVERING, &sq->state))
47 return 0;
48
49 err = mlx5_core_query_sq_state(mdev, sq->sqn, &state);
50 if (err) {
51 netdev_err(dev, "Failed to query SQ 0x%x state. err = %d\n",
52 sq->sqn, err);
53 goto out;
54 }
55
56 if (state != MLX5_SQC_STATE_ERR)
57 goto out;
58
59 mlx5e_tx_disable_queue(sq->txq);
60
61 err = mlx5e_wait_for_sq_flush(sq);
62 if (err)
63 goto out;
64
65 /* At this point, no new packets will arrive from the stack as TXQ is
66 * marked with QUEUE_STATE_DRV_XOFF. In addition, NAPI cleared all
67 * pending WQEs. SQ can safely reset the SQ.
68 */
69
70 err = mlx5e_health_sq_to_ready(sq->channel, sq->sqn);
71 if (err)
72 goto out;
73
74 mlx5e_reset_txqsq_cc_pc(sq);
75 sq->stats->recover++;
76 clear_bit(MLX5E_SQ_STATE_RECOVERING, &sq->state);
77 mlx5e_activate_txqsq(sq);
78
79 return 0;
80 out:
81 clear_bit(MLX5E_SQ_STATE_RECOVERING, &sq->state);
82 return err;
83 }
84
mlx5e_reporter_tx_err_cqe(struct mlx5e_txqsq * sq)85 void mlx5e_reporter_tx_err_cqe(struct mlx5e_txqsq *sq)
86 {
87 struct mlx5e_priv *priv = sq->channel->priv;
88 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
89 struct mlx5e_err_ctx err_ctx = {0};
90
91 err_ctx.ctx = sq;
92 err_ctx.recover = mlx5e_tx_reporter_err_cqe_recover;
93 sprintf(err_str, "ERR CQE on SQ: 0x%x", sq->sqn);
94
95 mlx5e_health_report(priv, priv->tx_reporter, err_str, &err_ctx);
96 }
97
mlx5e_tx_reporter_timeout_recover(void * ctx)98 static int mlx5e_tx_reporter_timeout_recover(void *ctx)
99 {
100 struct mlx5_eq_comp *eq;
101 struct mlx5e_txqsq *sq;
102 int err;
103
104 sq = ctx;
105 eq = sq->cq.mcq.eq;
106 err = mlx5e_health_channel_eq_recover(eq, sq->channel);
107 if (err)
108 clear_bit(MLX5E_SQ_STATE_ENABLED, &sq->state);
109
110 return err;
111 }
112
mlx5e_reporter_tx_timeout(struct mlx5e_txqsq * sq)113 int mlx5e_reporter_tx_timeout(struct mlx5e_txqsq *sq)
114 {
115 struct mlx5e_priv *priv = sq->channel->priv;
116 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN];
117 struct mlx5e_err_ctx err_ctx;
118
119 err_ctx.ctx = sq;
120 err_ctx.recover = mlx5e_tx_reporter_timeout_recover;
121 sprintf(err_str,
122 "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",
123 sq->channel->ix, sq->sqn, sq->cq.mcq.cqn, sq->cc, sq->pc,
124 jiffies_to_usecs(jiffies - sq->txq->trans_start));
125
126 return mlx5e_health_report(priv, priv->tx_reporter, err_str, &err_ctx);
127 }
128
129 /* state lock cannot be grabbed within this function.
130 * It can cause a dead lock or a read-after-free.
131 */
mlx5e_tx_reporter_recover_from_ctx(struct mlx5e_err_ctx * err_ctx)132 static int mlx5e_tx_reporter_recover_from_ctx(struct mlx5e_err_ctx *err_ctx)
133 {
134 return err_ctx->recover(err_ctx->ctx);
135 }
136
mlx5e_tx_reporter_recover(struct devlink_health_reporter * reporter,void * context)137 static int mlx5e_tx_reporter_recover(struct devlink_health_reporter *reporter,
138 void *context)
139 {
140 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
141 struct mlx5e_err_ctx *err_ctx = context;
142
143 return err_ctx ? mlx5e_tx_reporter_recover_from_ctx(err_ctx) :
144 mlx5e_health_recover_channels(priv);
145 }
146
147 static int
mlx5e_tx_reporter_build_diagnose_output(struct devlink_fmsg * fmsg,struct mlx5e_txqsq * sq,int tc)148 mlx5e_tx_reporter_build_diagnose_output(struct devlink_fmsg *fmsg,
149 struct mlx5e_txqsq *sq, int tc)
150 {
151 struct mlx5e_priv *priv = sq->channel->priv;
152 bool stopped = netif_xmit_stopped(sq->txq);
153 u8 state;
154 int err;
155
156 err = mlx5_core_query_sq_state(priv->mdev, sq->sqn, &state);
157 if (err)
158 return err;
159
160 err = devlink_fmsg_obj_nest_start(fmsg);
161 if (err)
162 return err;
163
164 err = devlink_fmsg_u32_pair_put(fmsg, "channel ix", sq->ch_ix);
165 if (err)
166 return err;
167
168 err = devlink_fmsg_u32_pair_put(fmsg, "tc", tc);
169 if (err)
170 return err;
171
172 err = devlink_fmsg_u32_pair_put(fmsg, "txq ix", sq->txq_ix);
173 if (err)
174 return err;
175
176 err = devlink_fmsg_u32_pair_put(fmsg, "sqn", sq->sqn);
177 if (err)
178 return err;
179
180 err = devlink_fmsg_u8_pair_put(fmsg, "HW state", state);
181 if (err)
182 return err;
183
184 err = devlink_fmsg_bool_pair_put(fmsg, "stopped", stopped);
185 if (err)
186 return err;
187
188 err = devlink_fmsg_u32_pair_put(fmsg, "cc", sq->cc);
189 if (err)
190 return err;
191
192 err = devlink_fmsg_u32_pair_put(fmsg, "pc", sq->pc);
193 if (err)
194 return err;
195
196 err = mlx5e_reporter_cq_diagnose(&sq->cq, fmsg);
197 if (err)
198 return err;
199
200 err = devlink_fmsg_obj_nest_end(fmsg);
201 if (err)
202 return err;
203
204 return 0;
205 }
206
mlx5e_tx_reporter_diagnose(struct devlink_health_reporter * reporter,struct devlink_fmsg * fmsg)207 static int mlx5e_tx_reporter_diagnose(struct devlink_health_reporter *reporter,
208 struct devlink_fmsg *fmsg)
209 {
210 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter);
211 struct mlx5e_txqsq *generic_sq = priv->txq2sq[0];
212 u32 sq_stride, sq_sz;
213
214 int i, tc, err = 0;
215
216 mutex_lock(&priv->state_lock);
217
218 if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
219 goto unlock;
220
221 sq_sz = mlx5_wq_cyc_get_size(&generic_sq->wq);
222 sq_stride = MLX5_SEND_WQE_BB;
223
224 err = mlx5e_reporter_named_obj_nest_start(fmsg, "Common Config");
225 if (err)
226 goto unlock;
227
228 err = mlx5e_reporter_named_obj_nest_start(fmsg, "SQ");
229 if (err)
230 goto unlock;
231
232 err = devlink_fmsg_u64_pair_put(fmsg, "stride size", sq_stride);
233 if (err)
234 goto unlock;
235
236 err = devlink_fmsg_u32_pair_put(fmsg, "size", sq_sz);
237 if (err)
238 goto unlock;
239
240 err = mlx5e_reporter_cq_common_diagnose(&generic_sq->cq, fmsg);
241 if (err)
242 goto unlock;
243
244 err = mlx5e_reporter_named_obj_nest_end(fmsg);
245 if (err)
246 goto unlock;
247
248 err = mlx5e_reporter_named_obj_nest_end(fmsg);
249 if (err)
250 goto unlock;
251
252 err = devlink_fmsg_arr_pair_nest_start(fmsg, "SQs");
253 if (err)
254 goto unlock;
255
256 for (i = 0; i < priv->channels.num; i++) {
257 struct mlx5e_channel *c = priv->channels.c[i];
258
259 for (tc = 0; tc < priv->channels.params.num_tc; tc++) {
260 struct mlx5e_txqsq *sq = &c->sq[tc];
261
262 err = mlx5e_tx_reporter_build_diagnose_output(fmsg, sq, tc);
263 if (err)
264 goto unlock;
265 }
266 }
267 err = devlink_fmsg_arr_pair_nest_end(fmsg);
268 if (err)
269 goto unlock;
270
271 unlock:
272 mutex_unlock(&priv->state_lock);
273 return err;
274 }
275
276 static const struct devlink_health_reporter_ops mlx5_tx_reporter_ops = {
277 .name = "tx",
278 .recover = mlx5e_tx_reporter_recover,
279 .diagnose = mlx5e_tx_reporter_diagnose,
280 };
281
282 #define MLX5_REPORTER_TX_GRACEFUL_PERIOD 500
283
mlx5e_reporter_tx_create(struct mlx5e_priv * priv)284 int mlx5e_reporter_tx_create(struct mlx5e_priv *priv)
285 {
286 struct devlink_health_reporter *reporter;
287 struct mlx5_core_dev *mdev = priv->mdev;
288 struct devlink *devlink;
289
290 devlink = priv_to_devlink(mdev);
291 reporter =
292 devlink_health_reporter_create(devlink, &mlx5_tx_reporter_ops,
293 MLX5_REPORTER_TX_GRACEFUL_PERIOD,
294 true, priv);
295 if (IS_ERR(reporter)) {
296 netdev_warn(priv->netdev,
297 "Failed to create tx reporter, err = %ld\n",
298 PTR_ERR(reporter));
299 return PTR_ERR(reporter);
300 }
301 priv->tx_reporter = reporter;
302 return 0;
303 }
304
mlx5e_reporter_tx_destroy(struct mlx5e_priv * priv)305 void mlx5e_reporter_tx_destroy(struct mlx5e_priv *priv)
306 {
307 if (!priv->tx_reporter)
308 return;
309
310 devlink_health_reporter_destroy(priv->tx_reporter);
311 }
312
