1 // SPDX-License-Identifier: GPL-2.0 2 // Copyright (c) 2019 Mellanox Technologies. 3 4 #include "health.h" 5 #include "params.h" 6 #include "txrx.h" 7 8 static int mlx5e_query_rq_state(struct mlx5_core_dev *dev, u32 rqn, u8 *state) 9 { 10 int outlen = MLX5_ST_SZ_BYTES(query_rq_out); 11 void *out; 12 void *rqc; 13 int err; 14 15 out = kvzalloc(outlen, GFP_KERNEL); 16 if (!out) 17 return -ENOMEM; 18 19 err = mlx5_core_query_rq(dev, rqn, out); 20 if (err) 21 goto out; 22 23 rqc = MLX5_ADDR_OF(query_rq_out, out, rq_context); 24 *state = MLX5_GET(rqc, rqc, state); 25 26 out: 27 kvfree(out); 28 return err; 29 } 30 31 static int mlx5e_wait_for_icosq_flush(struct mlx5e_icosq *icosq) 32 { 33 unsigned long exp_time = jiffies + 34 msecs_to_jiffies(MLX5E_REPORTER_FLUSH_TIMEOUT_MSEC); 35 36 while (time_before(jiffies, exp_time)) { 37 if (icosq->cc == icosq->pc) 38 return 0; 39 40 msleep(20); 41 } 42 43 netdev_err(icosq->channel->netdev, 44 "Wait for ICOSQ 0x%x flush timeout (cc = 0x%x, pc = 0x%x)\n", 45 icosq->sqn, icosq->cc, icosq->pc); 46 47 return -ETIMEDOUT; 48 } 49 50 static void mlx5e_reset_icosq_cc_pc(struct mlx5e_icosq *icosq) 51 { 52 WARN_ONCE(icosq->cc != icosq->pc, "ICOSQ 0x%x: cc (0x%x) != pc (0x%x)\n", 53 icosq->sqn, icosq->cc, icosq->pc); 54 icosq->cc = 0; 55 icosq->pc = 0; 56 } 57 58 static int mlx5e_rx_reporter_err_icosq_cqe_recover(void *ctx) 59 { 60 struct mlx5_core_dev *mdev; 61 struct mlx5e_icosq *icosq; 62 struct net_device *dev; 63 struct mlx5e_rq *rq; 64 u8 state; 65 int err; 66 67 icosq = ctx; 68 rq = &icosq->channel->rq; 69 mdev = icosq->channel->mdev; 70 dev = icosq->channel->netdev; 71 err = mlx5_core_query_sq_state(mdev, icosq->sqn, &state); 72 if (err) { 73 netdev_err(dev, "Failed to query ICOSQ 0x%x state. err = %d\n", 74 icosq->sqn, err); 75 goto out; 76 } 77 78 if (state != MLX5_SQC_STATE_ERR) 79 goto out; 80 81 mlx5e_deactivate_rq(rq); 82 err = mlx5e_wait_for_icosq_flush(icosq); 83 if (err) 84 goto out; 85 86 mlx5e_deactivate_icosq(icosq); 87 88 /* At this point, both the rq and the icosq are disabled */ 89 90 err = mlx5e_health_sq_to_ready(icosq->channel, icosq->sqn); 91 if (err) 92 goto out; 93 94 mlx5e_reset_icosq_cc_pc(icosq); 95 mlx5e_free_rx_in_progress_descs(rq); 96 clear_bit(MLX5E_SQ_STATE_RECOVERING, &icosq->state); 97 mlx5e_activate_icosq(icosq); 98 mlx5e_activate_rq(rq); 99 100 rq->stats->recover++; 101 return 0; 102 out: 103 clear_bit(MLX5E_SQ_STATE_RECOVERING, &icosq->state); 104 return err; 105 } 106 107 static int mlx5e_rq_to_ready(struct mlx5e_rq *rq, int curr_state) 108 { 109 struct net_device *dev = rq->netdev; 110 int err; 111 112 err = mlx5e_modify_rq_state(rq, curr_state, MLX5_RQC_STATE_RST); 113 if (err) { 114 netdev_err(dev, "Failed to move rq 0x%x to reset\n", rq->rqn); 115 return err; 116 } 117 err = mlx5e_modify_rq_state(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY); 118 if (err) { 119 netdev_err(dev, "Failed to move rq 0x%x to ready\n", rq->rqn); 120 return err; 121 } 122 123 return 0; 124 } 125 126 static int mlx5e_rx_reporter_err_rq_cqe_recover(void *ctx) 127 { 128 struct mlx5e_rq *rq = ctx; 129 int err; 130 131 mlx5e_deactivate_rq(rq); 132 mlx5e_free_rx_descs(rq); 133 134 err = mlx5e_rq_to_ready(rq, MLX5_RQC_STATE_ERR); 135 if (err) 136 goto out; 137 138 clear_bit(MLX5E_RQ_STATE_RECOVERING, &rq->state); 139 mlx5e_activate_rq(rq); 140 rq->stats->recover++; 141 return 0; 142 out: 143 clear_bit(MLX5E_RQ_STATE_RECOVERING, &rq->state); 144 return err; 145 } 146 147 static int mlx5e_rx_reporter_timeout_recover(void *ctx) 148 { 149 struct mlx5e_icosq *icosq; 150 struct mlx5_eq_comp *eq; 151 struct mlx5e_rq *rq; 152 int err; 153 154 rq = ctx; 155 icosq = &rq->channel->icosq; 156 eq = rq->cq.mcq.eq; 157 err = mlx5e_health_channel_eq_recover(eq, rq->channel); 158 if (err) 159 clear_bit(MLX5E_SQ_STATE_ENABLED, &icosq->state); 160 161 return err; 162 } 163 164 static int mlx5e_rx_reporter_recover_from_ctx(struct mlx5e_err_ctx *err_ctx) 165 { 166 return err_ctx->recover(err_ctx->ctx); 167 } 168 169 static int mlx5e_rx_reporter_recover(struct devlink_health_reporter *reporter, 170 void *context, 171 struct netlink_ext_ack *extack) 172 { 173 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter); 174 struct mlx5e_err_ctx *err_ctx = context; 175 176 return err_ctx ? mlx5e_rx_reporter_recover_from_ctx(err_ctx) : 177 mlx5e_health_recover_channels(priv); 178 } 179 180 static int mlx5e_rx_reporter_build_diagnose_output(struct mlx5e_rq *rq, 181 struct devlink_fmsg *fmsg) 182 { 183 struct mlx5e_priv *priv = rq->channel->priv; 184 struct mlx5e_icosq *icosq; 185 u8 icosq_hw_state; 186 u16 wqe_counter; 187 int wqes_sz; 188 u8 hw_state; 189 u16 wq_head; 190 int err; 191 192 icosq = &rq->channel->icosq; 193 err = mlx5e_query_rq_state(priv->mdev, rq->rqn, &hw_state); 194 if (err) 195 return err; 196 197 err = mlx5_core_query_sq_state(priv->mdev, icosq->sqn, &icosq_hw_state); 198 if (err) 199 return err; 200 201 wqes_sz = mlx5e_rqwq_get_cur_sz(rq); 202 wq_head = mlx5e_rqwq_get_head(rq); 203 wqe_counter = mlx5e_rqwq_get_wqe_counter(rq); 204 205 err = devlink_fmsg_obj_nest_start(fmsg); 206 if (err) 207 return err; 208 209 err = devlink_fmsg_u32_pair_put(fmsg, "channel ix", rq->channel->ix); 210 if (err) 211 return err; 212 213 err = devlink_fmsg_u32_pair_put(fmsg, "rqn", rq->rqn); 214 if (err) 215 return err; 216 217 err = devlink_fmsg_u8_pair_put(fmsg, "HW state", hw_state); 218 if (err) 219 return err; 220 221 err = devlink_fmsg_u8_pair_put(fmsg, "SW state", rq->state); 222 if (err) 223 return err; 224 225 err = devlink_fmsg_u32_pair_put(fmsg, "WQE counter", wqe_counter); 226 if (err) 227 return err; 228 229 err = devlink_fmsg_u32_pair_put(fmsg, "posted WQEs", wqes_sz); 230 if (err) 231 return err; 232 233 err = devlink_fmsg_u32_pair_put(fmsg, "cc", wq_head); 234 if (err) 235 return err; 236 237 err = devlink_fmsg_u8_pair_put(fmsg, "ICOSQ HW state", icosq_hw_state); 238 if (err) 239 return err; 240 241 err = mlx5e_health_cq_diag_fmsg(&rq->cq, fmsg); 242 if (err) 243 return err; 244 245 err = devlink_fmsg_obj_nest_end(fmsg); 246 if (err) 247 return err; 248 249 return 0; 250 } 251 252 static int mlx5e_rx_reporter_diagnose(struct devlink_health_reporter *reporter, 253 struct devlink_fmsg *fmsg, 254 struct netlink_ext_ack *extack) 255 { 256 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter); 257 struct mlx5e_params *params = &priv->channels.params; 258 struct mlx5e_rq *generic_rq; 259 u32 rq_stride, rq_sz; 260 int i, err = 0; 261 262 mutex_lock(&priv->state_lock); 263 264 if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) 265 goto unlock; 266 267 generic_rq = &priv->channels.c[0]->rq; 268 rq_sz = mlx5e_rqwq_get_size(generic_rq); 269 rq_stride = BIT(mlx5e_mpwqe_get_log_stride_size(priv->mdev, params, NULL)); 270 271 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "Common config"); 272 if (err) 273 goto unlock; 274 275 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RQ"); 276 if (err) 277 goto unlock; 278 279 err = devlink_fmsg_u8_pair_put(fmsg, "type", params->rq_wq_type); 280 if (err) 281 goto unlock; 282 283 err = devlink_fmsg_u64_pair_put(fmsg, "stride size", rq_stride); 284 if (err) 285 goto unlock; 286 287 err = devlink_fmsg_u32_pair_put(fmsg, "size", rq_sz); 288 if (err) 289 goto unlock; 290 291 err = mlx5e_health_cq_common_diag_fmsg(&generic_rq->cq, fmsg); 292 if (err) 293 goto unlock; 294 295 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 296 if (err) 297 goto unlock; 298 299 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 300 if (err) 301 goto unlock; 302 303 err = devlink_fmsg_arr_pair_nest_start(fmsg, "RQs"); 304 if (err) 305 goto unlock; 306 307 for (i = 0; i < priv->channels.num; i++) { 308 struct mlx5e_rq *rq = &priv->channels.c[i]->rq; 309 310 err = mlx5e_rx_reporter_build_diagnose_output(rq, fmsg); 311 if (err) 312 goto unlock; 313 } 314 err = devlink_fmsg_arr_pair_nest_end(fmsg); 315 if (err) 316 goto unlock; 317 unlock: 318 mutex_unlock(&priv->state_lock); 319 return err; 320 } 321 322 static int mlx5e_rx_reporter_dump_icosq(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg, 323 void *ctx) 324 { 325 struct mlx5e_txqsq *icosq = ctx; 326 struct mlx5_rsc_key key = {}; 327 int err; 328 329 if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) 330 return 0; 331 332 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "SX Slice"); 333 if (err) 334 return err; 335 336 key.size = PAGE_SIZE; 337 key.rsc = MLX5_SGMT_TYPE_SX_SLICE_ALL; 338 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 339 if (err) 340 return err; 341 342 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 343 if (err) 344 return err; 345 346 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "ICOSQ"); 347 if (err) 348 return err; 349 350 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "QPC"); 351 if (err) 352 return err; 353 354 key.rsc = MLX5_SGMT_TYPE_FULL_QPC; 355 key.index1 = icosq->sqn; 356 key.num_of_obj1 = 1; 357 358 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 359 if (err) 360 return err; 361 362 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 363 if (err) 364 return err; 365 366 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "send_buff"); 367 if (err) 368 return err; 369 370 key.rsc = MLX5_SGMT_TYPE_SND_BUFF; 371 key.num_of_obj2 = MLX5_RSC_DUMP_ALL; 372 373 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 374 if (err) 375 return err; 376 377 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 378 if (err) 379 return err; 380 381 return mlx5e_health_fmsg_named_obj_nest_end(fmsg); 382 } 383 384 static int mlx5e_rx_reporter_dump_rq(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg, 385 void *ctx) 386 { 387 struct mlx5_rsc_key key = {}; 388 struct mlx5e_rq *rq = ctx; 389 int err; 390 391 if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) 392 return 0; 393 394 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RX Slice"); 395 if (err) 396 return err; 397 398 key.size = PAGE_SIZE; 399 key.rsc = MLX5_SGMT_TYPE_RX_SLICE_ALL; 400 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 401 if (err) 402 return err; 403 404 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 405 if (err) 406 return err; 407 408 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RQ"); 409 if (err) 410 return err; 411 412 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "QPC"); 413 if (err) 414 return err; 415 416 key.rsc = MLX5_SGMT_TYPE_FULL_QPC; 417 key.index1 = rq->rqn; 418 key.num_of_obj1 = 1; 419 420 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 421 if (err) 422 return err; 423 424 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 425 if (err) 426 return err; 427 428 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "receive_buff"); 429 if (err) 430 return err; 431 432 key.rsc = MLX5_SGMT_TYPE_RCV_BUFF; 433 key.num_of_obj2 = MLX5_RSC_DUMP_ALL; 434 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 435 if (err) 436 return err; 437 438 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 439 if (err) 440 return err; 441 442 return mlx5e_health_fmsg_named_obj_nest_end(fmsg); 443 } 444 445 static int mlx5e_rx_reporter_dump_all_rqs(struct mlx5e_priv *priv, 446 struct devlink_fmsg *fmsg) 447 { 448 struct mlx5_rsc_key key = {}; 449 int i, err; 450 451 if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) 452 return 0; 453 454 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RX Slice"); 455 if (err) 456 return err; 457 458 key.size = PAGE_SIZE; 459 key.rsc = MLX5_SGMT_TYPE_RX_SLICE_ALL; 460 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 461 if (err) 462 return err; 463 464 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 465 if (err) 466 return err; 467 468 err = devlink_fmsg_arr_pair_nest_start(fmsg, "RQs"); 469 if (err) 470 return err; 471 472 for (i = 0; i < priv->channels.num; i++) { 473 struct mlx5e_rq *rq = &priv->channels.c[i]->rq; 474 475 err = mlx5e_health_queue_dump(priv, fmsg, rq->rqn, "RQ"); 476 if (err) 477 return err; 478 } 479 480 return devlink_fmsg_arr_pair_nest_end(fmsg); 481 } 482 483 static int mlx5e_rx_reporter_dump_from_ctx(struct mlx5e_priv *priv, 484 struct mlx5e_err_ctx *err_ctx, 485 struct devlink_fmsg *fmsg) 486 { 487 return err_ctx->dump(priv, fmsg, err_ctx->ctx); 488 } 489 490 static int mlx5e_rx_reporter_dump(struct devlink_health_reporter *reporter, 491 struct devlink_fmsg *fmsg, void *context, 492 struct netlink_ext_ack *extack) 493 { 494 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter); 495 struct mlx5e_err_ctx *err_ctx = context; 496 497 return err_ctx ? mlx5e_rx_reporter_dump_from_ctx(priv, err_ctx, fmsg) : 498 mlx5e_rx_reporter_dump_all_rqs(priv, fmsg); 499 } 500 501 void mlx5e_reporter_rx_timeout(struct mlx5e_rq *rq) 502 { 503 struct mlx5e_icosq *icosq = &rq->channel->icosq; 504 struct mlx5e_priv *priv = rq->channel->priv; 505 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN]; 506 struct mlx5e_err_ctx err_ctx = {}; 507 508 err_ctx.ctx = rq; 509 err_ctx.recover = mlx5e_rx_reporter_timeout_recover; 510 err_ctx.dump = mlx5e_rx_reporter_dump_rq; 511 snprintf(err_str, sizeof(err_str), 512 "RX timeout on channel: %d, ICOSQ: 0x%x RQ: 0x%x, CQ: 0x%x", 513 icosq->channel->ix, icosq->sqn, rq->rqn, rq->cq.mcq.cqn); 514 515 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx); 516 } 517 518 void mlx5e_reporter_rq_cqe_err(struct mlx5e_rq *rq) 519 { 520 struct mlx5e_priv *priv = rq->channel->priv; 521 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN]; 522 struct mlx5e_err_ctx err_ctx = {}; 523 524 err_ctx.ctx = rq; 525 err_ctx.recover = mlx5e_rx_reporter_err_rq_cqe_recover; 526 err_ctx.dump = mlx5e_rx_reporter_dump_rq; 527 snprintf(err_str, sizeof(err_str), "ERR CQE on RQ: 0x%x", rq->rqn); 528 529 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx); 530 } 531 532 void mlx5e_reporter_icosq_cqe_err(struct mlx5e_icosq *icosq) 533 { 534 struct mlx5e_priv *priv = icosq->channel->priv; 535 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN]; 536 struct mlx5e_err_ctx err_ctx = {}; 537 538 err_ctx.ctx = icosq; 539 err_ctx.recover = mlx5e_rx_reporter_err_icosq_cqe_recover; 540 err_ctx.dump = mlx5e_rx_reporter_dump_icosq; 541 snprintf(err_str, sizeof(err_str), "ERR CQE on ICOSQ: 0x%x", icosq->sqn); 542 543 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx); 544 } 545 546 static const struct devlink_health_reporter_ops mlx5_rx_reporter_ops = { 547 .name = "rx", 548 .recover = mlx5e_rx_reporter_recover, 549 .diagnose = mlx5e_rx_reporter_diagnose, 550 .dump = mlx5e_rx_reporter_dump, 551 }; 552 553 #define MLX5E_REPORTER_RX_GRACEFUL_PERIOD 500 554 555 void mlx5e_reporter_rx_create(struct mlx5e_priv *priv) 556 { 557 struct devlink *devlink = priv_to_devlink(priv->mdev); 558 struct devlink_health_reporter *reporter; 559 560 reporter = devlink_health_reporter_create(devlink, 561 &mlx5_rx_reporter_ops, 562 MLX5E_REPORTER_RX_GRACEFUL_PERIOD, 563 priv); 564 if (IS_ERR(reporter)) { 565 netdev_warn(priv->netdev, "Failed to create rx reporter, err = %ld\n", 566 PTR_ERR(reporter)); 567 return; 568 } 569 priv->rx_reporter = reporter; 570 } 571 572 void mlx5e_reporter_rx_destroy(struct mlx5e_priv *priv) 573 { 574 if (!priv->rx_reporter) 575 return; 576 577 devlink_health_reporter_destroy(priv->rx_reporter); 578 } 579