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_reporter_icosq_diagnose(struct mlx5e_icosq *icosq, u8 hw_state, 181 struct devlink_fmsg *fmsg) 182 { 183 int err; 184 185 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "ICOSQ"); 186 if (err) 187 return err; 188 189 err = devlink_fmsg_u32_pair_put(fmsg, "sqn", icosq->sqn); 190 if (err) 191 return err; 192 193 err = devlink_fmsg_u8_pair_put(fmsg, "HW state", hw_state); 194 if (err) 195 return err; 196 197 err = devlink_fmsg_u32_pair_put(fmsg, "cc", icosq->cc); 198 if (err) 199 return err; 200 201 err = devlink_fmsg_u32_pair_put(fmsg, "pc", icosq->pc); 202 if (err) 203 return err; 204 205 err = devlink_fmsg_u32_pair_put(fmsg, "WQE size", 206 mlx5_wq_cyc_get_size(&icosq->wq)); 207 if (err) 208 return err; 209 210 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "CQ"); 211 if (err) 212 return err; 213 214 err = devlink_fmsg_u32_pair_put(fmsg, "cqn", icosq->cq.mcq.cqn); 215 if (err) 216 return err; 217 218 err = devlink_fmsg_u32_pair_put(fmsg, "cc", icosq->cq.wq.cc); 219 if (err) 220 return err; 221 222 err = devlink_fmsg_u32_pair_put(fmsg, "size", mlx5_cqwq_get_size(&icosq->cq.wq)); 223 if (err) 224 return err; 225 226 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 227 if (err) 228 return err; 229 230 return mlx5e_health_fmsg_named_obj_nest_end(fmsg); 231 } 232 233 static int mlx5e_rx_reporter_build_diagnose_output(struct mlx5e_rq *rq, 234 struct devlink_fmsg *fmsg) 235 { 236 struct mlx5e_priv *priv = rq->channel->priv; 237 struct mlx5e_icosq *icosq; 238 u8 icosq_hw_state; 239 u16 wqe_counter; 240 int wqes_sz; 241 u8 hw_state; 242 u16 wq_head; 243 int err; 244 245 icosq = &rq->channel->icosq; 246 err = mlx5e_query_rq_state(priv->mdev, rq->rqn, &hw_state); 247 if (err) 248 return err; 249 250 err = mlx5_core_query_sq_state(priv->mdev, icosq->sqn, &icosq_hw_state); 251 if (err) 252 return err; 253 254 wqes_sz = mlx5e_rqwq_get_cur_sz(rq); 255 wq_head = mlx5e_rqwq_get_head(rq); 256 wqe_counter = mlx5e_rqwq_get_wqe_counter(rq); 257 258 err = devlink_fmsg_obj_nest_start(fmsg); 259 if (err) 260 return err; 261 262 err = devlink_fmsg_u32_pair_put(fmsg, "channel ix", rq->channel->ix); 263 if (err) 264 return err; 265 266 err = devlink_fmsg_u32_pair_put(fmsg, "rqn", rq->rqn); 267 if (err) 268 return err; 269 270 err = devlink_fmsg_u8_pair_put(fmsg, "HW state", hw_state); 271 if (err) 272 return err; 273 274 err = devlink_fmsg_u8_pair_put(fmsg, "SW state", rq->state); 275 if (err) 276 return err; 277 278 err = devlink_fmsg_u32_pair_put(fmsg, "WQE counter", wqe_counter); 279 if (err) 280 return err; 281 282 err = devlink_fmsg_u32_pair_put(fmsg, "posted WQEs", wqes_sz); 283 if (err) 284 return err; 285 286 err = devlink_fmsg_u32_pair_put(fmsg, "cc", wq_head); 287 if (err) 288 return err; 289 290 err = mlx5e_health_cq_diag_fmsg(&rq->cq, fmsg); 291 if (err) 292 return err; 293 294 err = mlx5e_health_eq_diag_fmsg(rq->cq.mcq.eq, fmsg); 295 if (err) 296 return err; 297 298 err = mlx5e_reporter_icosq_diagnose(icosq, icosq_hw_state, fmsg); 299 if (err) 300 return err; 301 302 err = devlink_fmsg_obj_nest_end(fmsg); 303 if (err) 304 return err; 305 306 return 0; 307 } 308 309 static int mlx5e_rx_reporter_diagnose(struct devlink_health_reporter *reporter, 310 struct devlink_fmsg *fmsg, 311 struct netlink_ext_ack *extack) 312 { 313 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter); 314 struct mlx5e_params *params = &priv->channels.params; 315 struct mlx5e_rq *generic_rq; 316 u32 rq_stride, rq_sz; 317 int i, err = 0; 318 319 mutex_lock(&priv->state_lock); 320 321 if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) 322 goto unlock; 323 324 generic_rq = &priv->channels.c[0]->rq; 325 rq_sz = mlx5e_rqwq_get_size(generic_rq); 326 rq_stride = BIT(mlx5e_mpwqe_get_log_stride_size(priv->mdev, params, NULL)); 327 328 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "Common config"); 329 if (err) 330 goto unlock; 331 332 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RQ"); 333 if (err) 334 goto unlock; 335 336 err = devlink_fmsg_u8_pair_put(fmsg, "type", params->rq_wq_type); 337 if (err) 338 goto unlock; 339 340 err = devlink_fmsg_u64_pair_put(fmsg, "stride size", rq_stride); 341 if (err) 342 goto unlock; 343 344 err = devlink_fmsg_u32_pair_put(fmsg, "size", rq_sz); 345 if (err) 346 goto unlock; 347 348 err = mlx5e_health_cq_common_diag_fmsg(&generic_rq->cq, fmsg); 349 if (err) 350 goto unlock; 351 352 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 353 if (err) 354 goto unlock; 355 356 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 357 if (err) 358 goto unlock; 359 360 err = devlink_fmsg_arr_pair_nest_start(fmsg, "RQs"); 361 if (err) 362 goto unlock; 363 364 for (i = 0; i < priv->channels.num; i++) { 365 struct mlx5e_rq *rq = &priv->channels.c[i]->rq; 366 367 err = mlx5e_rx_reporter_build_diagnose_output(rq, fmsg); 368 if (err) 369 goto unlock; 370 } 371 err = devlink_fmsg_arr_pair_nest_end(fmsg); 372 if (err) 373 goto unlock; 374 unlock: 375 mutex_unlock(&priv->state_lock); 376 return err; 377 } 378 379 static int mlx5e_rx_reporter_dump_icosq(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg, 380 void *ctx) 381 { 382 struct mlx5e_txqsq *icosq = ctx; 383 struct mlx5_rsc_key key = {}; 384 int err; 385 386 if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) 387 return 0; 388 389 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "SX Slice"); 390 if (err) 391 return err; 392 393 key.size = PAGE_SIZE; 394 key.rsc = MLX5_SGMT_TYPE_SX_SLICE_ALL; 395 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 396 if (err) 397 return err; 398 399 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 400 if (err) 401 return err; 402 403 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "ICOSQ"); 404 if (err) 405 return err; 406 407 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "QPC"); 408 if (err) 409 return err; 410 411 key.rsc = MLX5_SGMT_TYPE_FULL_QPC; 412 key.index1 = icosq->sqn; 413 key.num_of_obj1 = 1; 414 415 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 416 if (err) 417 return err; 418 419 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 420 if (err) 421 return err; 422 423 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "send_buff"); 424 if (err) 425 return err; 426 427 key.rsc = MLX5_SGMT_TYPE_SND_BUFF; 428 key.num_of_obj2 = MLX5_RSC_DUMP_ALL; 429 430 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 431 if (err) 432 return err; 433 434 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 435 if (err) 436 return err; 437 438 return mlx5e_health_fmsg_named_obj_nest_end(fmsg); 439 } 440 441 static int mlx5e_rx_reporter_dump_rq(struct mlx5e_priv *priv, struct devlink_fmsg *fmsg, 442 void *ctx) 443 { 444 struct mlx5_rsc_key key = {}; 445 struct mlx5e_rq *rq = ctx; 446 int err; 447 448 if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) 449 return 0; 450 451 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RX Slice"); 452 if (err) 453 return err; 454 455 key.size = PAGE_SIZE; 456 key.rsc = MLX5_SGMT_TYPE_RX_SLICE_ALL; 457 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 458 if (err) 459 return err; 460 461 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 462 if (err) 463 return err; 464 465 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RQ"); 466 if (err) 467 return err; 468 469 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "QPC"); 470 if (err) 471 return err; 472 473 key.rsc = MLX5_SGMT_TYPE_FULL_QPC; 474 key.index1 = rq->rqn; 475 key.num_of_obj1 = 1; 476 477 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 478 if (err) 479 return err; 480 481 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 482 if (err) 483 return err; 484 485 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "receive_buff"); 486 if (err) 487 return err; 488 489 key.rsc = MLX5_SGMT_TYPE_RCV_BUFF; 490 key.num_of_obj2 = MLX5_RSC_DUMP_ALL; 491 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 492 if (err) 493 return err; 494 495 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 496 if (err) 497 return err; 498 499 return mlx5e_health_fmsg_named_obj_nest_end(fmsg); 500 } 501 502 static int mlx5e_rx_reporter_dump_all_rqs(struct mlx5e_priv *priv, 503 struct devlink_fmsg *fmsg) 504 { 505 struct mlx5_rsc_key key = {}; 506 int i, err; 507 508 if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) 509 return 0; 510 511 err = mlx5e_health_fmsg_named_obj_nest_start(fmsg, "RX Slice"); 512 if (err) 513 return err; 514 515 key.size = PAGE_SIZE; 516 key.rsc = MLX5_SGMT_TYPE_RX_SLICE_ALL; 517 err = mlx5e_health_rsc_fmsg_dump(priv, &key, fmsg); 518 if (err) 519 return err; 520 521 err = mlx5e_health_fmsg_named_obj_nest_end(fmsg); 522 if (err) 523 return err; 524 525 err = devlink_fmsg_arr_pair_nest_start(fmsg, "RQs"); 526 if (err) 527 return err; 528 529 for (i = 0; i < priv->channels.num; i++) { 530 struct mlx5e_rq *rq = &priv->channels.c[i]->rq; 531 532 err = mlx5e_health_queue_dump(priv, fmsg, rq->rqn, "RQ"); 533 if (err) 534 return err; 535 } 536 537 return devlink_fmsg_arr_pair_nest_end(fmsg); 538 } 539 540 static int mlx5e_rx_reporter_dump_from_ctx(struct mlx5e_priv *priv, 541 struct mlx5e_err_ctx *err_ctx, 542 struct devlink_fmsg *fmsg) 543 { 544 return err_ctx->dump(priv, fmsg, err_ctx->ctx); 545 } 546 547 static int mlx5e_rx_reporter_dump(struct devlink_health_reporter *reporter, 548 struct devlink_fmsg *fmsg, void *context, 549 struct netlink_ext_ack *extack) 550 { 551 struct mlx5e_priv *priv = devlink_health_reporter_priv(reporter); 552 struct mlx5e_err_ctx *err_ctx = context; 553 554 return err_ctx ? mlx5e_rx_reporter_dump_from_ctx(priv, err_ctx, fmsg) : 555 mlx5e_rx_reporter_dump_all_rqs(priv, fmsg); 556 } 557 558 void mlx5e_reporter_rx_timeout(struct mlx5e_rq *rq) 559 { 560 struct mlx5e_icosq *icosq = &rq->channel->icosq; 561 struct mlx5e_priv *priv = rq->channel->priv; 562 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN]; 563 struct mlx5e_err_ctx err_ctx = {}; 564 565 err_ctx.ctx = rq; 566 err_ctx.recover = mlx5e_rx_reporter_timeout_recover; 567 err_ctx.dump = mlx5e_rx_reporter_dump_rq; 568 snprintf(err_str, sizeof(err_str), 569 "RX timeout on channel: %d, ICOSQ: 0x%x RQ: 0x%x, CQ: 0x%x", 570 icosq->channel->ix, icosq->sqn, rq->rqn, rq->cq.mcq.cqn); 571 572 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx); 573 } 574 575 void mlx5e_reporter_rq_cqe_err(struct mlx5e_rq *rq) 576 { 577 struct mlx5e_priv *priv = rq->channel->priv; 578 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN]; 579 struct mlx5e_err_ctx err_ctx = {}; 580 581 err_ctx.ctx = rq; 582 err_ctx.recover = mlx5e_rx_reporter_err_rq_cqe_recover; 583 err_ctx.dump = mlx5e_rx_reporter_dump_rq; 584 snprintf(err_str, sizeof(err_str), "ERR CQE on RQ: 0x%x", rq->rqn); 585 586 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx); 587 } 588 589 void mlx5e_reporter_icosq_cqe_err(struct mlx5e_icosq *icosq) 590 { 591 struct mlx5e_priv *priv = icosq->channel->priv; 592 char err_str[MLX5E_REPORTER_PER_Q_MAX_LEN]; 593 struct mlx5e_err_ctx err_ctx = {}; 594 595 err_ctx.ctx = icosq; 596 err_ctx.recover = mlx5e_rx_reporter_err_icosq_cqe_recover; 597 err_ctx.dump = mlx5e_rx_reporter_dump_icosq; 598 snprintf(err_str, sizeof(err_str), "ERR CQE on ICOSQ: 0x%x", icosq->sqn); 599 600 mlx5e_health_report(priv, priv->rx_reporter, err_str, &err_ctx); 601 } 602 603 static const struct devlink_health_reporter_ops mlx5_rx_reporter_ops = { 604 .name = "rx", 605 .recover = mlx5e_rx_reporter_recover, 606 .diagnose = mlx5e_rx_reporter_diagnose, 607 .dump = mlx5e_rx_reporter_dump, 608 }; 609 610 #define MLX5E_REPORTER_RX_GRACEFUL_PERIOD 500 611 612 void mlx5e_reporter_rx_create(struct mlx5e_priv *priv) 613 { 614 struct devlink *devlink = priv_to_devlink(priv->mdev); 615 struct devlink_health_reporter *reporter; 616 617 reporter = devlink_health_reporter_create(devlink, 618 &mlx5_rx_reporter_ops, 619 MLX5E_REPORTER_RX_GRACEFUL_PERIOD, 620 priv); 621 if (IS_ERR(reporter)) { 622 netdev_warn(priv->netdev, "Failed to create rx reporter, err = %ld\n", 623 PTR_ERR(reporter)); 624 return; 625 } 626 priv->rx_reporter = reporter; 627 } 628 629 void mlx5e_reporter_rx_destroy(struct mlx5e_priv *priv) 630 { 631 if (!priv->rx_reporter) 632 return; 633 634 devlink_health_reporter_destroy(priv->rx_reporter); 635 } 636