1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright (c) 2018, Intel Corporation. */ 3 4 #include "ice.h" 5 #include "ice_base.h" 6 #include "ice_flow.h" 7 #include "ice_lib.h" 8 #include "ice_fltr.h" 9 #include "ice_dcb_lib.h" 10 #include "ice_vsi_vlan_ops.h" 11 12 /** 13 * ice_vsi_type_str - maps VSI type enum to string equivalents 14 * @vsi_type: VSI type enum 15 */ 16 const char *ice_vsi_type_str(enum ice_vsi_type vsi_type) 17 { 18 switch (vsi_type) { 19 case ICE_VSI_PF: 20 return "ICE_VSI_PF"; 21 case ICE_VSI_VF: 22 return "ICE_VSI_VF"; 23 case ICE_VSI_CTRL: 24 return "ICE_VSI_CTRL"; 25 case ICE_VSI_CHNL: 26 return "ICE_VSI_CHNL"; 27 case ICE_VSI_LB: 28 return "ICE_VSI_LB"; 29 default: 30 return "unknown"; 31 } 32 } 33 34 /** 35 * ice_vsi_ctrl_all_rx_rings - Start or stop a VSI's Rx rings 36 * @vsi: the VSI being configured 37 * @ena: start or stop the Rx rings 38 * 39 * First enable/disable all of the Rx rings, flush any remaining writes, and 40 * then verify that they have all been enabled/disabled successfully. This will 41 * let all of the register writes complete when enabling/disabling the Rx rings 42 * before waiting for the change in hardware to complete. 43 */ 44 static int ice_vsi_ctrl_all_rx_rings(struct ice_vsi *vsi, bool ena) 45 { 46 int ret = 0; 47 u16 i; 48 49 ice_for_each_rxq(vsi, i) 50 ice_vsi_ctrl_one_rx_ring(vsi, ena, i, false); 51 52 ice_flush(&vsi->back->hw); 53 54 ice_for_each_rxq(vsi, i) { 55 ret = ice_vsi_wait_one_rx_ring(vsi, ena, i); 56 if (ret) 57 break; 58 } 59 60 return ret; 61 } 62 63 /** 64 * ice_vsi_alloc_arrays - Allocate queue and vector pointer arrays for the VSI 65 * @vsi: VSI pointer 66 * 67 * On error: returns error code (negative) 68 * On success: returns 0 69 */ 70 static int ice_vsi_alloc_arrays(struct ice_vsi *vsi) 71 { 72 struct ice_pf *pf = vsi->back; 73 struct device *dev; 74 75 dev = ice_pf_to_dev(pf); 76 if (vsi->type == ICE_VSI_CHNL) 77 return 0; 78 79 /* allocate memory for both Tx and Rx ring pointers */ 80 vsi->tx_rings = devm_kcalloc(dev, vsi->alloc_txq, 81 sizeof(*vsi->tx_rings), GFP_KERNEL); 82 if (!vsi->tx_rings) 83 return -ENOMEM; 84 85 vsi->rx_rings = devm_kcalloc(dev, vsi->alloc_rxq, 86 sizeof(*vsi->rx_rings), GFP_KERNEL); 87 if (!vsi->rx_rings) 88 goto err_rings; 89 90 /* txq_map needs to have enough space to track both Tx (stack) rings 91 * and XDP rings; at this point vsi->num_xdp_txq might not be set, 92 * so use num_possible_cpus() as we want to always provide XDP ring 93 * per CPU, regardless of queue count settings from user that might 94 * have come from ethtool's set_channels() callback; 95 */ 96 vsi->txq_map = devm_kcalloc(dev, (vsi->alloc_txq + num_possible_cpus()), 97 sizeof(*vsi->txq_map), GFP_KERNEL); 98 99 if (!vsi->txq_map) 100 goto err_txq_map; 101 102 vsi->rxq_map = devm_kcalloc(dev, vsi->alloc_rxq, 103 sizeof(*vsi->rxq_map), GFP_KERNEL); 104 if (!vsi->rxq_map) 105 goto err_rxq_map; 106 107 /* There is no need to allocate q_vectors for a loopback VSI. */ 108 if (vsi->type == ICE_VSI_LB) 109 return 0; 110 111 /* allocate memory for q_vector pointers */ 112 vsi->q_vectors = devm_kcalloc(dev, vsi->num_q_vectors, 113 sizeof(*vsi->q_vectors), GFP_KERNEL); 114 if (!vsi->q_vectors) 115 goto err_vectors; 116 117 return 0; 118 119 err_vectors: 120 devm_kfree(dev, vsi->rxq_map); 121 err_rxq_map: 122 devm_kfree(dev, vsi->txq_map); 123 err_txq_map: 124 devm_kfree(dev, vsi->rx_rings); 125 err_rings: 126 devm_kfree(dev, vsi->tx_rings); 127 return -ENOMEM; 128 } 129 130 /** 131 * ice_vsi_set_num_desc - Set number of descriptors for queues on this VSI 132 * @vsi: the VSI being configured 133 */ 134 static void ice_vsi_set_num_desc(struct ice_vsi *vsi) 135 { 136 switch (vsi->type) { 137 case ICE_VSI_PF: 138 case ICE_VSI_CTRL: 139 case ICE_VSI_LB: 140 /* a user could change the values of num_[tr]x_desc using 141 * ethtool -G so we should keep those values instead of 142 * overwriting them with the defaults. 143 */ 144 if (!vsi->num_rx_desc) 145 vsi->num_rx_desc = ICE_DFLT_NUM_RX_DESC; 146 if (!vsi->num_tx_desc) 147 vsi->num_tx_desc = ICE_DFLT_NUM_TX_DESC; 148 break; 149 default: 150 dev_dbg(ice_pf_to_dev(vsi->back), "Not setting number of Tx/Rx descriptors for VSI type %d\n", 151 vsi->type); 152 break; 153 } 154 } 155 156 /** 157 * ice_vsi_set_num_qs - Set number of queues, descriptors and vectors for a VSI 158 * @vsi: the VSI being configured 159 * 160 * Return 0 on success and a negative value on error 161 */ 162 static void ice_vsi_set_num_qs(struct ice_vsi *vsi) 163 { 164 enum ice_vsi_type vsi_type = vsi->type; 165 struct ice_pf *pf = vsi->back; 166 struct ice_vf *vf = vsi->vf; 167 168 if (WARN_ON(vsi_type == ICE_VSI_VF && !vf)) 169 return; 170 171 switch (vsi_type) { 172 case ICE_VSI_PF: 173 if (vsi->req_txq) { 174 vsi->alloc_txq = vsi->req_txq; 175 vsi->num_txq = vsi->req_txq; 176 } else { 177 vsi->alloc_txq = min3(pf->num_lan_msix, 178 ice_get_avail_txq_count(pf), 179 (u16)num_online_cpus()); 180 } 181 182 pf->num_lan_tx = vsi->alloc_txq; 183 184 /* only 1 Rx queue unless RSS is enabled */ 185 if (!test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 186 vsi->alloc_rxq = 1; 187 } else { 188 if (vsi->req_rxq) { 189 vsi->alloc_rxq = vsi->req_rxq; 190 vsi->num_rxq = vsi->req_rxq; 191 } else { 192 vsi->alloc_rxq = min3(pf->num_lan_msix, 193 ice_get_avail_rxq_count(pf), 194 (u16)num_online_cpus()); 195 } 196 } 197 198 pf->num_lan_rx = vsi->alloc_rxq; 199 200 vsi->num_q_vectors = min_t(int, pf->num_lan_msix, 201 max_t(int, vsi->alloc_rxq, 202 vsi->alloc_txq)); 203 break; 204 case ICE_VSI_VF: 205 if (vf->num_req_qs) 206 vf->num_vf_qs = vf->num_req_qs; 207 vsi->alloc_txq = vf->num_vf_qs; 208 vsi->alloc_rxq = vf->num_vf_qs; 209 /* pf->vfs.num_msix_per includes (VF miscellaneous vector + 210 * data queue interrupts). Since vsi->num_q_vectors is number 211 * of queues vectors, subtract 1 (ICE_NONQ_VECS_VF) from the 212 * original vector count 213 */ 214 vsi->num_q_vectors = vf->num_msix - ICE_NONQ_VECS_VF; 215 break; 216 case ICE_VSI_CTRL: 217 vsi->alloc_txq = 1; 218 vsi->alloc_rxq = 1; 219 vsi->num_q_vectors = 1; 220 break; 221 case ICE_VSI_CHNL: 222 vsi->alloc_txq = 0; 223 vsi->alloc_rxq = 0; 224 break; 225 case ICE_VSI_LB: 226 vsi->alloc_txq = 1; 227 vsi->alloc_rxq = 1; 228 break; 229 default: 230 dev_warn(ice_pf_to_dev(pf), "Unknown VSI type %d\n", vsi_type); 231 break; 232 } 233 234 ice_vsi_set_num_desc(vsi); 235 } 236 237 /** 238 * ice_get_free_slot - get the next non-NULL location index in array 239 * @array: array to search 240 * @size: size of the array 241 * @curr: last known occupied index to be used as a search hint 242 * 243 * void * is being used to keep the functionality generic. This lets us use this 244 * function on any array of pointers. 245 */ 246 static int ice_get_free_slot(void *array, int size, int curr) 247 { 248 int **tmp_array = (int **)array; 249 int next; 250 251 if (curr < (size - 1) && !tmp_array[curr + 1]) { 252 next = curr + 1; 253 } else { 254 int i = 0; 255 256 while ((i < size) && (tmp_array[i])) 257 i++; 258 if (i == size) 259 next = ICE_NO_VSI; 260 else 261 next = i; 262 } 263 return next; 264 } 265 266 /** 267 * ice_vsi_delete_from_hw - delete a VSI from the switch 268 * @vsi: pointer to VSI being removed 269 */ 270 static void ice_vsi_delete_from_hw(struct ice_vsi *vsi) 271 { 272 struct ice_pf *pf = vsi->back; 273 struct ice_vsi_ctx *ctxt; 274 int status; 275 276 ice_fltr_remove_all(vsi); 277 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); 278 if (!ctxt) 279 return; 280 281 if (vsi->type == ICE_VSI_VF) 282 ctxt->vf_num = vsi->vf->vf_id; 283 ctxt->vsi_num = vsi->vsi_num; 284 285 memcpy(&ctxt->info, &vsi->info, sizeof(ctxt->info)); 286 287 status = ice_free_vsi(&pf->hw, vsi->idx, ctxt, false, NULL); 288 if (status) 289 dev_err(ice_pf_to_dev(pf), "Failed to delete VSI %i in FW - error: %d\n", 290 vsi->vsi_num, status); 291 292 kfree(ctxt); 293 } 294 295 /** 296 * ice_vsi_free_arrays - De-allocate queue and vector pointer arrays for the VSI 297 * @vsi: pointer to VSI being cleared 298 */ 299 static void ice_vsi_free_arrays(struct ice_vsi *vsi) 300 { 301 struct ice_pf *pf = vsi->back; 302 struct device *dev; 303 304 dev = ice_pf_to_dev(pf); 305 306 /* free the ring and vector containers */ 307 devm_kfree(dev, vsi->q_vectors); 308 vsi->q_vectors = NULL; 309 devm_kfree(dev, vsi->tx_rings); 310 vsi->tx_rings = NULL; 311 devm_kfree(dev, vsi->rx_rings); 312 vsi->rx_rings = NULL; 313 devm_kfree(dev, vsi->txq_map); 314 vsi->txq_map = NULL; 315 devm_kfree(dev, vsi->rxq_map); 316 vsi->rxq_map = NULL; 317 } 318 319 /** 320 * ice_vsi_free_stats - Free the ring statistics structures 321 * @vsi: VSI pointer 322 */ 323 static void ice_vsi_free_stats(struct ice_vsi *vsi) 324 { 325 struct ice_vsi_stats *vsi_stat; 326 struct ice_pf *pf = vsi->back; 327 int i; 328 329 if (vsi->type == ICE_VSI_CHNL) 330 return; 331 if (!pf->vsi_stats) 332 return; 333 334 vsi_stat = pf->vsi_stats[vsi->idx]; 335 if (!vsi_stat) 336 return; 337 338 ice_for_each_alloc_txq(vsi, i) { 339 if (vsi_stat->tx_ring_stats[i]) { 340 kfree_rcu(vsi_stat->tx_ring_stats[i], rcu); 341 WRITE_ONCE(vsi_stat->tx_ring_stats[i], NULL); 342 } 343 } 344 345 ice_for_each_alloc_rxq(vsi, i) { 346 if (vsi_stat->rx_ring_stats[i]) { 347 kfree_rcu(vsi_stat->rx_ring_stats[i], rcu); 348 WRITE_ONCE(vsi_stat->rx_ring_stats[i], NULL); 349 } 350 } 351 352 kfree(vsi_stat->tx_ring_stats); 353 kfree(vsi_stat->rx_ring_stats); 354 kfree(vsi_stat); 355 pf->vsi_stats[vsi->idx] = NULL; 356 } 357 358 /** 359 * ice_vsi_alloc_ring_stats - Allocates Tx and Rx ring stats for the VSI 360 * @vsi: VSI which is having stats allocated 361 */ 362 static int ice_vsi_alloc_ring_stats(struct ice_vsi *vsi) 363 { 364 struct ice_ring_stats **tx_ring_stats; 365 struct ice_ring_stats **rx_ring_stats; 366 struct ice_vsi_stats *vsi_stats; 367 struct ice_pf *pf = vsi->back; 368 u16 i; 369 370 vsi_stats = pf->vsi_stats[vsi->idx]; 371 tx_ring_stats = vsi_stats->tx_ring_stats; 372 rx_ring_stats = vsi_stats->rx_ring_stats; 373 374 /* Allocate Tx ring stats */ 375 ice_for_each_alloc_txq(vsi, i) { 376 struct ice_ring_stats *ring_stats; 377 struct ice_tx_ring *ring; 378 379 ring = vsi->tx_rings[i]; 380 ring_stats = tx_ring_stats[i]; 381 382 if (!ring_stats) { 383 ring_stats = kzalloc(sizeof(*ring_stats), GFP_KERNEL); 384 if (!ring_stats) 385 goto err_out; 386 387 WRITE_ONCE(tx_ring_stats[i], ring_stats); 388 } 389 390 ring->ring_stats = ring_stats; 391 } 392 393 /* Allocate Rx ring stats */ 394 ice_for_each_alloc_rxq(vsi, i) { 395 struct ice_ring_stats *ring_stats; 396 struct ice_rx_ring *ring; 397 398 ring = vsi->rx_rings[i]; 399 ring_stats = rx_ring_stats[i]; 400 401 if (!ring_stats) { 402 ring_stats = kzalloc(sizeof(*ring_stats), GFP_KERNEL); 403 if (!ring_stats) 404 goto err_out; 405 406 WRITE_ONCE(rx_ring_stats[i], ring_stats); 407 } 408 409 ring->ring_stats = ring_stats; 410 } 411 412 return 0; 413 414 err_out: 415 ice_vsi_free_stats(vsi); 416 return -ENOMEM; 417 } 418 419 /** 420 * ice_vsi_free - clean up and deallocate the provided VSI 421 * @vsi: pointer to VSI being cleared 422 * 423 * This deallocates the VSI's queue resources, removes it from the PF's 424 * VSI array if necessary, and deallocates the VSI 425 */ 426 static void ice_vsi_free(struct ice_vsi *vsi) 427 { 428 struct ice_pf *pf = NULL; 429 struct device *dev; 430 431 if (!vsi || !vsi->back) 432 return; 433 434 pf = vsi->back; 435 dev = ice_pf_to_dev(pf); 436 437 if (!pf->vsi[vsi->idx] || pf->vsi[vsi->idx] != vsi) { 438 dev_dbg(dev, "vsi does not exist at pf->vsi[%d]\n", vsi->idx); 439 return; 440 } 441 442 mutex_lock(&pf->sw_mutex); 443 /* updates the PF for this cleared VSI */ 444 445 pf->vsi[vsi->idx] = NULL; 446 pf->next_vsi = vsi->idx; 447 448 ice_vsi_free_stats(vsi); 449 ice_vsi_free_arrays(vsi); 450 mutex_unlock(&pf->sw_mutex); 451 devm_kfree(dev, vsi); 452 } 453 454 void ice_vsi_delete(struct ice_vsi *vsi) 455 { 456 ice_vsi_delete_from_hw(vsi); 457 ice_vsi_free(vsi); 458 } 459 460 /** 461 * ice_msix_clean_ctrl_vsi - MSIX mode interrupt handler for ctrl VSI 462 * @irq: interrupt number 463 * @data: pointer to a q_vector 464 */ 465 static irqreturn_t ice_msix_clean_ctrl_vsi(int __always_unused irq, void *data) 466 { 467 struct ice_q_vector *q_vector = (struct ice_q_vector *)data; 468 469 if (!q_vector->tx.tx_ring) 470 return IRQ_HANDLED; 471 472 #define FDIR_RX_DESC_CLEAN_BUDGET 64 473 ice_clean_rx_irq(q_vector->rx.rx_ring, FDIR_RX_DESC_CLEAN_BUDGET); 474 ice_clean_ctrl_tx_irq(q_vector->tx.tx_ring); 475 476 return IRQ_HANDLED; 477 } 478 479 /** 480 * ice_msix_clean_rings - MSIX mode Interrupt Handler 481 * @irq: interrupt number 482 * @data: pointer to a q_vector 483 */ 484 static irqreturn_t ice_msix_clean_rings(int __always_unused irq, void *data) 485 { 486 struct ice_q_vector *q_vector = (struct ice_q_vector *)data; 487 488 if (!q_vector->tx.tx_ring && !q_vector->rx.rx_ring) 489 return IRQ_HANDLED; 490 491 q_vector->total_events++; 492 493 napi_schedule(&q_vector->napi); 494 495 return IRQ_HANDLED; 496 } 497 498 /** 499 * ice_vsi_alloc_stat_arrays - Allocate statistics arrays 500 * @vsi: VSI pointer 501 */ 502 static int ice_vsi_alloc_stat_arrays(struct ice_vsi *vsi) 503 { 504 struct ice_vsi_stats *vsi_stat; 505 struct ice_pf *pf = vsi->back; 506 507 if (vsi->type == ICE_VSI_CHNL) 508 return 0; 509 if (!pf->vsi_stats) 510 return -ENOENT; 511 512 if (pf->vsi_stats[vsi->idx]) 513 /* realloc will happen in rebuild path */ 514 return 0; 515 516 vsi_stat = kzalloc(sizeof(*vsi_stat), GFP_KERNEL); 517 if (!vsi_stat) 518 return -ENOMEM; 519 520 vsi_stat->tx_ring_stats = 521 kcalloc(vsi->alloc_txq, sizeof(*vsi_stat->tx_ring_stats), 522 GFP_KERNEL); 523 if (!vsi_stat->tx_ring_stats) 524 goto err_alloc_tx; 525 526 vsi_stat->rx_ring_stats = 527 kcalloc(vsi->alloc_rxq, sizeof(*vsi_stat->rx_ring_stats), 528 GFP_KERNEL); 529 if (!vsi_stat->rx_ring_stats) 530 goto err_alloc_rx; 531 532 pf->vsi_stats[vsi->idx] = vsi_stat; 533 534 return 0; 535 536 err_alloc_rx: 537 kfree(vsi_stat->rx_ring_stats); 538 err_alloc_tx: 539 kfree(vsi_stat->tx_ring_stats); 540 kfree(vsi_stat); 541 pf->vsi_stats[vsi->idx] = NULL; 542 return -ENOMEM; 543 } 544 545 /** 546 * ice_vsi_alloc_def - set default values for already allocated VSI 547 * @vsi: ptr to VSI 548 * @ch: ptr to channel 549 */ 550 static int 551 ice_vsi_alloc_def(struct ice_vsi *vsi, struct ice_channel *ch) 552 { 553 if (vsi->type != ICE_VSI_CHNL) { 554 ice_vsi_set_num_qs(vsi); 555 if (ice_vsi_alloc_arrays(vsi)) 556 return -ENOMEM; 557 } 558 559 switch (vsi->type) { 560 case ICE_VSI_PF: 561 /* Setup default MSIX irq handler for VSI */ 562 vsi->irq_handler = ice_msix_clean_rings; 563 break; 564 case ICE_VSI_CTRL: 565 /* Setup ctrl VSI MSIX irq handler */ 566 vsi->irq_handler = ice_msix_clean_ctrl_vsi; 567 break; 568 case ICE_VSI_CHNL: 569 if (!ch) 570 return -EINVAL; 571 572 vsi->num_rxq = ch->num_rxq; 573 vsi->num_txq = ch->num_txq; 574 vsi->next_base_q = ch->base_q; 575 break; 576 case ICE_VSI_VF: 577 case ICE_VSI_LB: 578 break; 579 default: 580 ice_vsi_free_arrays(vsi); 581 return -EINVAL; 582 } 583 584 return 0; 585 } 586 587 /** 588 * ice_vsi_alloc - Allocates the next available struct VSI in the PF 589 * @pf: board private structure 590 * 591 * Reserves a VSI index from the PF and allocates an empty VSI structure 592 * without a type. The VSI structure must later be initialized by calling 593 * ice_vsi_cfg(). 594 * 595 * returns a pointer to a VSI on success, NULL on failure. 596 */ 597 static struct ice_vsi *ice_vsi_alloc(struct ice_pf *pf) 598 { 599 struct device *dev = ice_pf_to_dev(pf); 600 struct ice_vsi *vsi = NULL; 601 602 /* Need to protect the allocation of the VSIs at the PF level */ 603 mutex_lock(&pf->sw_mutex); 604 605 /* If we have already allocated our maximum number of VSIs, 606 * pf->next_vsi will be ICE_NO_VSI. If not, pf->next_vsi index 607 * is available to be populated 608 */ 609 if (pf->next_vsi == ICE_NO_VSI) { 610 dev_dbg(dev, "out of VSI slots!\n"); 611 goto unlock_pf; 612 } 613 614 vsi = devm_kzalloc(dev, sizeof(*vsi), GFP_KERNEL); 615 if (!vsi) 616 goto unlock_pf; 617 618 vsi->back = pf; 619 set_bit(ICE_VSI_DOWN, vsi->state); 620 621 /* fill slot and make note of the index */ 622 vsi->idx = pf->next_vsi; 623 pf->vsi[pf->next_vsi] = vsi; 624 625 /* prepare pf->next_vsi for next use */ 626 pf->next_vsi = ice_get_free_slot(pf->vsi, pf->num_alloc_vsi, 627 pf->next_vsi); 628 629 unlock_pf: 630 mutex_unlock(&pf->sw_mutex); 631 return vsi; 632 } 633 634 /** 635 * ice_alloc_fd_res - Allocate FD resource for a VSI 636 * @vsi: pointer to the ice_vsi 637 * 638 * This allocates the FD resources 639 * 640 * Returns 0 on success, -EPERM on no-op or -EIO on failure 641 */ 642 static int ice_alloc_fd_res(struct ice_vsi *vsi) 643 { 644 struct ice_pf *pf = vsi->back; 645 u32 g_val, b_val; 646 647 /* Flow Director filters are only allocated/assigned to the PF VSI or 648 * CHNL VSI which passes the traffic. The CTRL VSI is only used to 649 * add/delete filters so resources are not allocated to it 650 */ 651 if (!test_bit(ICE_FLAG_FD_ENA, pf->flags)) 652 return -EPERM; 653 654 if (!(vsi->type == ICE_VSI_PF || vsi->type == ICE_VSI_VF || 655 vsi->type == ICE_VSI_CHNL)) 656 return -EPERM; 657 658 /* FD filters from guaranteed pool per VSI */ 659 g_val = pf->hw.func_caps.fd_fltr_guar; 660 if (!g_val) 661 return -EPERM; 662 663 /* FD filters from best effort pool */ 664 b_val = pf->hw.func_caps.fd_fltr_best_effort; 665 if (!b_val) 666 return -EPERM; 667 668 /* PF main VSI gets only 64 FD resources from guaranteed pool 669 * when ADQ is configured. 670 */ 671 #define ICE_PF_VSI_GFLTR 64 672 673 /* determine FD filter resources per VSI from shared(best effort) and 674 * dedicated pool 675 */ 676 if (vsi->type == ICE_VSI_PF) { 677 vsi->num_gfltr = g_val; 678 /* if MQPRIO is configured, main VSI doesn't get all FD 679 * resources from guaranteed pool. PF VSI gets 64 FD resources 680 */ 681 if (test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) { 682 if (g_val < ICE_PF_VSI_GFLTR) 683 return -EPERM; 684 /* allow bare minimum entries for PF VSI */ 685 vsi->num_gfltr = ICE_PF_VSI_GFLTR; 686 } 687 688 /* each VSI gets same "best_effort" quota */ 689 vsi->num_bfltr = b_val; 690 } else if (vsi->type == ICE_VSI_VF) { 691 vsi->num_gfltr = 0; 692 693 /* each VSI gets same "best_effort" quota */ 694 vsi->num_bfltr = b_val; 695 } else { 696 struct ice_vsi *main_vsi; 697 int numtc; 698 699 main_vsi = ice_get_main_vsi(pf); 700 if (!main_vsi) 701 return -EPERM; 702 703 if (!main_vsi->all_numtc) 704 return -EINVAL; 705 706 /* figure out ADQ numtc */ 707 numtc = main_vsi->all_numtc - ICE_CHNL_START_TC; 708 709 /* only one TC but still asking resources for channels, 710 * invalid config 711 */ 712 if (numtc < ICE_CHNL_START_TC) 713 return -EPERM; 714 715 g_val -= ICE_PF_VSI_GFLTR; 716 /* channel VSIs gets equal share from guaranteed pool */ 717 vsi->num_gfltr = g_val / numtc; 718 719 /* each VSI gets same "best_effort" quota */ 720 vsi->num_bfltr = b_val; 721 } 722 723 return 0; 724 } 725 726 /** 727 * ice_vsi_get_qs - Assign queues from PF to VSI 728 * @vsi: the VSI to assign queues to 729 * 730 * Returns 0 on success and a negative value on error 731 */ 732 static int ice_vsi_get_qs(struct ice_vsi *vsi) 733 { 734 struct ice_pf *pf = vsi->back; 735 struct ice_qs_cfg tx_qs_cfg = { 736 .qs_mutex = &pf->avail_q_mutex, 737 .pf_map = pf->avail_txqs, 738 .pf_map_size = pf->max_pf_txqs, 739 .q_count = vsi->alloc_txq, 740 .scatter_count = ICE_MAX_SCATTER_TXQS, 741 .vsi_map = vsi->txq_map, 742 .vsi_map_offset = 0, 743 .mapping_mode = ICE_VSI_MAP_CONTIG 744 }; 745 struct ice_qs_cfg rx_qs_cfg = { 746 .qs_mutex = &pf->avail_q_mutex, 747 .pf_map = pf->avail_rxqs, 748 .pf_map_size = pf->max_pf_rxqs, 749 .q_count = vsi->alloc_rxq, 750 .scatter_count = ICE_MAX_SCATTER_RXQS, 751 .vsi_map = vsi->rxq_map, 752 .vsi_map_offset = 0, 753 .mapping_mode = ICE_VSI_MAP_CONTIG 754 }; 755 int ret; 756 757 if (vsi->type == ICE_VSI_CHNL) 758 return 0; 759 760 ret = __ice_vsi_get_qs(&tx_qs_cfg); 761 if (ret) 762 return ret; 763 vsi->tx_mapping_mode = tx_qs_cfg.mapping_mode; 764 765 ret = __ice_vsi_get_qs(&rx_qs_cfg); 766 if (ret) 767 return ret; 768 vsi->rx_mapping_mode = rx_qs_cfg.mapping_mode; 769 770 return 0; 771 } 772 773 /** 774 * ice_vsi_put_qs - Release queues from VSI to PF 775 * @vsi: the VSI that is going to release queues 776 */ 777 static void ice_vsi_put_qs(struct ice_vsi *vsi) 778 { 779 struct ice_pf *pf = vsi->back; 780 int i; 781 782 mutex_lock(&pf->avail_q_mutex); 783 784 ice_for_each_alloc_txq(vsi, i) { 785 clear_bit(vsi->txq_map[i], pf->avail_txqs); 786 vsi->txq_map[i] = ICE_INVAL_Q_INDEX; 787 } 788 789 ice_for_each_alloc_rxq(vsi, i) { 790 clear_bit(vsi->rxq_map[i], pf->avail_rxqs); 791 vsi->rxq_map[i] = ICE_INVAL_Q_INDEX; 792 } 793 794 mutex_unlock(&pf->avail_q_mutex); 795 } 796 797 /** 798 * ice_is_safe_mode 799 * @pf: pointer to the PF struct 800 * 801 * returns true if driver is in safe mode, false otherwise 802 */ 803 bool ice_is_safe_mode(struct ice_pf *pf) 804 { 805 return !test_bit(ICE_FLAG_ADV_FEATURES, pf->flags); 806 } 807 808 /** 809 * ice_is_rdma_ena 810 * @pf: pointer to the PF struct 811 * 812 * returns true if RDMA is currently supported, false otherwise 813 */ 814 bool ice_is_rdma_ena(struct ice_pf *pf) 815 { 816 return test_bit(ICE_FLAG_RDMA_ENA, pf->flags); 817 } 818 819 /** 820 * ice_vsi_clean_rss_flow_fld - Delete RSS configuration 821 * @vsi: the VSI being cleaned up 822 * 823 * This function deletes RSS input set for all flows that were configured 824 * for this VSI 825 */ 826 static void ice_vsi_clean_rss_flow_fld(struct ice_vsi *vsi) 827 { 828 struct ice_pf *pf = vsi->back; 829 int status; 830 831 if (ice_is_safe_mode(pf)) 832 return; 833 834 status = ice_rem_vsi_rss_cfg(&pf->hw, vsi->idx); 835 if (status) 836 dev_dbg(ice_pf_to_dev(pf), "ice_rem_vsi_rss_cfg failed for vsi = %d, error = %d\n", 837 vsi->vsi_num, status); 838 } 839 840 /** 841 * ice_rss_clean - Delete RSS related VSI structures and configuration 842 * @vsi: the VSI being removed 843 */ 844 static void ice_rss_clean(struct ice_vsi *vsi) 845 { 846 struct ice_pf *pf = vsi->back; 847 struct device *dev; 848 849 dev = ice_pf_to_dev(pf); 850 851 devm_kfree(dev, vsi->rss_hkey_user); 852 devm_kfree(dev, vsi->rss_lut_user); 853 854 ice_vsi_clean_rss_flow_fld(vsi); 855 /* remove RSS replay list */ 856 if (!ice_is_safe_mode(pf)) 857 ice_rem_vsi_rss_list(&pf->hw, vsi->idx); 858 } 859 860 /** 861 * ice_vsi_set_rss_params - Setup RSS capabilities per VSI type 862 * @vsi: the VSI being configured 863 */ 864 static void ice_vsi_set_rss_params(struct ice_vsi *vsi) 865 { 866 struct ice_hw_common_caps *cap; 867 struct ice_pf *pf = vsi->back; 868 u16 max_rss_size; 869 870 if (!test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 871 vsi->rss_size = 1; 872 return; 873 } 874 875 cap = &pf->hw.func_caps.common_cap; 876 max_rss_size = BIT(cap->rss_table_entry_width); 877 switch (vsi->type) { 878 case ICE_VSI_CHNL: 879 case ICE_VSI_PF: 880 /* PF VSI will inherit RSS instance of PF */ 881 vsi->rss_table_size = (u16)cap->rss_table_size; 882 if (vsi->type == ICE_VSI_CHNL) 883 vsi->rss_size = min_t(u16, vsi->num_rxq, max_rss_size); 884 else 885 vsi->rss_size = min_t(u16, num_online_cpus(), 886 max_rss_size); 887 vsi->rss_lut_type = ICE_LUT_PF; 888 break; 889 case ICE_VSI_VF: 890 /* VF VSI will get a small RSS table. 891 * For VSI_LUT, LUT size should be set to 64 bytes. 892 */ 893 vsi->rss_table_size = ICE_LUT_VSI_SIZE; 894 vsi->rss_size = ICE_MAX_RSS_QS_PER_VF; 895 vsi->rss_lut_type = ICE_LUT_VSI; 896 break; 897 case ICE_VSI_LB: 898 break; 899 default: 900 dev_dbg(ice_pf_to_dev(pf), "Unsupported VSI type %s\n", 901 ice_vsi_type_str(vsi->type)); 902 break; 903 } 904 } 905 906 /** 907 * ice_set_dflt_vsi_ctx - Set default VSI context before adding a VSI 908 * @hw: HW structure used to determine the VLAN mode of the device 909 * @ctxt: the VSI context being set 910 * 911 * This initializes a default VSI context for all sections except the Queues. 912 */ 913 static void ice_set_dflt_vsi_ctx(struct ice_hw *hw, struct ice_vsi_ctx *ctxt) 914 { 915 u32 table = 0; 916 917 memset(&ctxt->info, 0, sizeof(ctxt->info)); 918 /* VSI's should be allocated from shared pool */ 919 ctxt->alloc_from_pool = true; 920 /* Src pruning enabled by default */ 921 ctxt->info.sw_flags = ICE_AQ_VSI_SW_FLAG_SRC_PRUNE; 922 /* Traffic from VSI can be sent to LAN */ 923 ctxt->info.sw_flags2 = ICE_AQ_VSI_SW_FLAG_LAN_ENA; 924 /* allow all untagged/tagged packets by default on Tx */ 925 ctxt->info.inner_vlan_flags = FIELD_PREP(ICE_AQ_VSI_INNER_VLAN_TX_MODE_M, 926 ICE_AQ_VSI_INNER_VLAN_TX_MODE_ALL); 927 /* SVM - by default bits 3 and 4 in inner_vlan_flags are 0's which 928 * results in legacy behavior (show VLAN, DEI, and UP) in descriptor. 929 * 930 * DVM - leave inner VLAN in packet by default 931 */ 932 if (ice_is_dvm_ena(hw)) { 933 ctxt->info.inner_vlan_flags |= 934 FIELD_PREP(ICE_AQ_VSI_INNER_VLAN_EMODE_M, 935 ICE_AQ_VSI_INNER_VLAN_EMODE_NOTHING); 936 ctxt->info.outer_vlan_flags = 937 FIELD_PREP(ICE_AQ_VSI_OUTER_VLAN_TX_MODE_M, 938 ICE_AQ_VSI_OUTER_VLAN_TX_MODE_ALL); 939 ctxt->info.outer_vlan_flags |= 940 FIELD_PREP(ICE_AQ_VSI_OUTER_TAG_TYPE_M, 941 ICE_AQ_VSI_OUTER_TAG_VLAN_8100); 942 ctxt->info.outer_vlan_flags |= 943 FIELD_PREP(ICE_AQ_VSI_OUTER_VLAN_EMODE_M, 944 ICE_AQ_VSI_OUTER_VLAN_EMODE_NOTHING); 945 } 946 /* Have 1:1 UP mapping for both ingress/egress tables */ 947 table |= ICE_UP_TABLE_TRANSLATE(0, 0); 948 table |= ICE_UP_TABLE_TRANSLATE(1, 1); 949 table |= ICE_UP_TABLE_TRANSLATE(2, 2); 950 table |= ICE_UP_TABLE_TRANSLATE(3, 3); 951 table |= ICE_UP_TABLE_TRANSLATE(4, 4); 952 table |= ICE_UP_TABLE_TRANSLATE(5, 5); 953 table |= ICE_UP_TABLE_TRANSLATE(6, 6); 954 table |= ICE_UP_TABLE_TRANSLATE(7, 7); 955 ctxt->info.ingress_table = cpu_to_le32(table); 956 ctxt->info.egress_table = cpu_to_le32(table); 957 /* Have 1:1 UP mapping for outer to inner UP table */ 958 ctxt->info.outer_up_table = cpu_to_le32(table); 959 /* No Outer tag support outer_tag_flags remains to zero */ 960 } 961 962 /** 963 * ice_vsi_setup_q_map - Setup a VSI queue map 964 * @vsi: the VSI being configured 965 * @ctxt: VSI context structure 966 */ 967 static int ice_vsi_setup_q_map(struct ice_vsi *vsi, struct ice_vsi_ctx *ctxt) 968 { 969 u16 offset = 0, qmap = 0, tx_count = 0, rx_count = 0, pow = 0; 970 u16 num_txq_per_tc, num_rxq_per_tc; 971 u16 qcount_tx = vsi->alloc_txq; 972 u16 qcount_rx = vsi->alloc_rxq; 973 u8 netdev_tc = 0; 974 int i; 975 976 if (!vsi->tc_cfg.numtc) { 977 /* at least TC0 should be enabled by default */ 978 vsi->tc_cfg.numtc = 1; 979 vsi->tc_cfg.ena_tc = 1; 980 } 981 982 num_rxq_per_tc = min_t(u16, qcount_rx / vsi->tc_cfg.numtc, ICE_MAX_RXQS_PER_TC); 983 if (!num_rxq_per_tc) 984 num_rxq_per_tc = 1; 985 num_txq_per_tc = qcount_tx / vsi->tc_cfg.numtc; 986 if (!num_txq_per_tc) 987 num_txq_per_tc = 1; 988 989 /* find the (rounded up) power-of-2 of qcount */ 990 pow = (u16)order_base_2(num_rxq_per_tc); 991 992 /* TC mapping is a function of the number of Rx queues assigned to the 993 * VSI for each traffic class and the offset of these queues. 994 * The first 10 bits are for queue offset for TC0, next 4 bits for no:of 995 * queues allocated to TC0. No:of queues is a power-of-2. 996 * 997 * If TC is not enabled, the queue offset is set to 0, and allocate one 998 * queue, this way, traffic for the given TC will be sent to the default 999 * queue. 1000 * 1001 * Setup number and offset of Rx queues for all TCs for the VSI 1002 */ 1003 ice_for_each_traffic_class(i) { 1004 if (!(vsi->tc_cfg.ena_tc & BIT(i))) { 1005 /* TC is not enabled */ 1006 vsi->tc_cfg.tc_info[i].qoffset = 0; 1007 vsi->tc_cfg.tc_info[i].qcount_rx = 1; 1008 vsi->tc_cfg.tc_info[i].qcount_tx = 1; 1009 vsi->tc_cfg.tc_info[i].netdev_tc = 0; 1010 ctxt->info.tc_mapping[i] = 0; 1011 continue; 1012 } 1013 1014 /* TC is enabled */ 1015 vsi->tc_cfg.tc_info[i].qoffset = offset; 1016 vsi->tc_cfg.tc_info[i].qcount_rx = num_rxq_per_tc; 1017 vsi->tc_cfg.tc_info[i].qcount_tx = num_txq_per_tc; 1018 vsi->tc_cfg.tc_info[i].netdev_tc = netdev_tc++; 1019 1020 qmap = FIELD_PREP(ICE_AQ_VSI_TC_Q_OFFSET_M, offset); 1021 qmap |= FIELD_PREP(ICE_AQ_VSI_TC_Q_NUM_M, pow); 1022 offset += num_rxq_per_tc; 1023 tx_count += num_txq_per_tc; 1024 ctxt->info.tc_mapping[i] = cpu_to_le16(qmap); 1025 } 1026 1027 /* if offset is non-zero, means it is calculated correctly based on 1028 * enabled TCs for a given VSI otherwise qcount_rx will always 1029 * be correct and non-zero because it is based off - VSI's 1030 * allocated Rx queues which is at least 1 (hence qcount_tx will be 1031 * at least 1) 1032 */ 1033 if (offset) 1034 rx_count = offset; 1035 else 1036 rx_count = num_rxq_per_tc; 1037 1038 if (rx_count > vsi->alloc_rxq) { 1039 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Rx queues (%u), than were allocated (%u)!\n", 1040 rx_count, vsi->alloc_rxq); 1041 return -EINVAL; 1042 } 1043 1044 if (tx_count > vsi->alloc_txq) { 1045 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Tx queues (%u), than were allocated (%u)!\n", 1046 tx_count, vsi->alloc_txq); 1047 return -EINVAL; 1048 } 1049 1050 vsi->num_txq = tx_count; 1051 vsi->num_rxq = rx_count; 1052 1053 if (vsi->type == ICE_VSI_VF && vsi->num_txq != vsi->num_rxq) { 1054 dev_dbg(ice_pf_to_dev(vsi->back), "VF VSI should have same number of Tx and Rx queues. Hence making them equal\n"); 1055 /* since there is a chance that num_rxq could have been changed 1056 * in the above for loop, make num_txq equal to num_rxq. 1057 */ 1058 vsi->num_txq = vsi->num_rxq; 1059 } 1060 1061 /* Rx queue mapping */ 1062 ctxt->info.mapping_flags |= cpu_to_le16(ICE_AQ_VSI_Q_MAP_CONTIG); 1063 /* q_mapping buffer holds the info for the first queue allocated for 1064 * this VSI in the PF space and also the number of queues associated 1065 * with this VSI. 1066 */ 1067 ctxt->info.q_mapping[0] = cpu_to_le16(vsi->rxq_map[0]); 1068 ctxt->info.q_mapping[1] = cpu_to_le16(vsi->num_rxq); 1069 1070 return 0; 1071 } 1072 1073 /** 1074 * ice_set_fd_vsi_ctx - Set FD VSI context before adding a VSI 1075 * @ctxt: the VSI context being set 1076 * @vsi: the VSI being configured 1077 */ 1078 static void ice_set_fd_vsi_ctx(struct ice_vsi_ctx *ctxt, struct ice_vsi *vsi) 1079 { 1080 u8 dflt_q_group, dflt_q_prio; 1081 u16 dflt_q, report_q, val; 1082 1083 if (vsi->type != ICE_VSI_PF && vsi->type != ICE_VSI_CTRL && 1084 vsi->type != ICE_VSI_VF && vsi->type != ICE_VSI_CHNL) 1085 return; 1086 1087 val = ICE_AQ_VSI_PROP_FLOW_DIR_VALID; 1088 ctxt->info.valid_sections |= cpu_to_le16(val); 1089 dflt_q = 0; 1090 dflt_q_group = 0; 1091 report_q = 0; 1092 dflt_q_prio = 0; 1093 1094 /* enable flow director filtering/programming */ 1095 val = ICE_AQ_VSI_FD_ENABLE | ICE_AQ_VSI_FD_PROG_ENABLE; 1096 ctxt->info.fd_options = cpu_to_le16(val); 1097 /* max of allocated flow director filters */ 1098 ctxt->info.max_fd_fltr_dedicated = 1099 cpu_to_le16(vsi->num_gfltr); 1100 /* max of shared flow director filters any VSI may program */ 1101 ctxt->info.max_fd_fltr_shared = 1102 cpu_to_le16(vsi->num_bfltr); 1103 /* default queue index within the VSI of the default FD */ 1104 val = FIELD_PREP(ICE_AQ_VSI_FD_DEF_Q_M, dflt_q); 1105 /* target queue or queue group to the FD filter */ 1106 val |= FIELD_PREP(ICE_AQ_VSI_FD_DEF_GRP_M, dflt_q_group); 1107 ctxt->info.fd_def_q = cpu_to_le16(val); 1108 /* queue index on which FD filter completion is reported */ 1109 val = FIELD_PREP(ICE_AQ_VSI_FD_REPORT_Q_M, report_q); 1110 /* priority of the default qindex action */ 1111 val |= FIELD_PREP(ICE_AQ_VSI_FD_DEF_PRIORITY_M, dflt_q_prio); 1112 ctxt->info.fd_report_opt = cpu_to_le16(val); 1113 } 1114 1115 /** 1116 * ice_set_rss_vsi_ctx - Set RSS VSI context before adding a VSI 1117 * @ctxt: the VSI context being set 1118 * @vsi: the VSI being configured 1119 */ 1120 static void ice_set_rss_vsi_ctx(struct ice_vsi_ctx *ctxt, struct ice_vsi *vsi) 1121 { 1122 u8 lut_type, hash_type; 1123 struct device *dev; 1124 struct ice_pf *pf; 1125 1126 pf = vsi->back; 1127 dev = ice_pf_to_dev(pf); 1128 1129 switch (vsi->type) { 1130 case ICE_VSI_CHNL: 1131 case ICE_VSI_PF: 1132 /* PF VSI will inherit RSS instance of PF */ 1133 lut_type = ICE_AQ_VSI_Q_OPT_RSS_LUT_PF; 1134 break; 1135 case ICE_VSI_VF: 1136 /* VF VSI will gets a small RSS table which is a VSI LUT type */ 1137 lut_type = ICE_AQ_VSI_Q_OPT_RSS_LUT_VSI; 1138 break; 1139 default: 1140 dev_dbg(dev, "Unsupported VSI type %s\n", 1141 ice_vsi_type_str(vsi->type)); 1142 return; 1143 } 1144 1145 hash_type = ICE_AQ_VSI_Q_OPT_RSS_HASH_TPLZ; 1146 vsi->rss_hfunc = hash_type; 1147 1148 ctxt->info.q_opt_rss = 1149 FIELD_PREP(ICE_AQ_VSI_Q_OPT_RSS_LUT_M, lut_type) | 1150 FIELD_PREP(ICE_AQ_VSI_Q_OPT_RSS_HASH_M, hash_type); 1151 } 1152 1153 static void 1154 ice_chnl_vsi_setup_q_map(struct ice_vsi *vsi, struct ice_vsi_ctx *ctxt) 1155 { 1156 struct ice_pf *pf = vsi->back; 1157 u16 qcount, qmap; 1158 u8 offset = 0; 1159 int pow; 1160 1161 qcount = min_t(int, vsi->num_rxq, pf->num_lan_msix); 1162 1163 pow = order_base_2(qcount); 1164 qmap = FIELD_PREP(ICE_AQ_VSI_TC_Q_OFFSET_M, offset); 1165 qmap |= FIELD_PREP(ICE_AQ_VSI_TC_Q_NUM_M, pow); 1166 1167 ctxt->info.tc_mapping[0] = cpu_to_le16(qmap); 1168 ctxt->info.mapping_flags |= cpu_to_le16(ICE_AQ_VSI_Q_MAP_CONTIG); 1169 ctxt->info.q_mapping[0] = cpu_to_le16(vsi->next_base_q); 1170 ctxt->info.q_mapping[1] = cpu_to_le16(qcount); 1171 } 1172 1173 /** 1174 * ice_vsi_is_vlan_pruning_ena - check if VLAN pruning is enabled or not 1175 * @vsi: VSI to check whether or not VLAN pruning is enabled. 1176 * 1177 * returns true if Rx VLAN pruning is enabled and false otherwise. 1178 */ 1179 static bool ice_vsi_is_vlan_pruning_ena(struct ice_vsi *vsi) 1180 { 1181 return vsi->info.sw_flags2 & ICE_AQ_VSI_SW_FLAG_RX_VLAN_PRUNE_ENA; 1182 } 1183 1184 /** 1185 * ice_vsi_init - Create and initialize a VSI 1186 * @vsi: the VSI being configured 1187 * @vsi_flags: VSI configuration flags 1188 * 1189 * Set ICE_FLAG_VSI_INIT to initialize a new VSI context, clear it to 1190 * reconfigure an existing context. 1191 * 1192 * This initializes a VSI context depending on the VSI type to be added and 1193 * passes it down to the add_vsi aq command to create a new VSI. 1194 */ 1195 static int ice_vsi_init(struct ice_vsi *vsi, u32 vsi_flags) 1196 { 1197 struct ice_pf *pf = vsi->back; 1198 struct ice_hw *hw = &pf->hw; 1199 struct ice_vsi_ctx *ctxt; 1200 struct device *dev; 1201 int ret = 0; 1202 1203 dev = ice_pf_to_dev(pf); 1204 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); 1205 if (!ctxt) 1206 return -ENOMEM; 1207 1208 switch (vsi->type) { 1209 case ICE_VSI_CTRL: 1210 case ICE_VSI_LB: 1211 case ICE_VSI_PF: 1212 ctxt->flags = ICE_AQ_VSI_TYPE_PF; 1213 break; 1214 case ICE_VSI_CHNL: 1215 ctxt->flags = ICE_AQ_VSI_TYPE_VMDQ2; 1216 break; 1217 case ICE_VSI_VF: 1218 ctxt->flags = ICE_AQ_VSI_TYPE_VF; 1219 /* VF number here is the absolute VF number (0-255) */ 1220 ctxt->vf_num = vsi->vf->vf_id + hw->func_caps.vf_base_id; 1221 break; 1222 default: 1223 ret = -ENODEV; 1224 goto out; 1225 } 1226 1227 /* Handle VLAN pruning for channel VSI if main VSI has VLAN 1228 * prune enabled 1229 */ 1230 if (vsi->type == ICE_VSI_CHNL) { 1231 struct ice_vsi *main_vsi; 1232 1233 main_vsi = ice_get_main_vsi(pf); 1234 if (main_vsi && ice_vsi_is_vlan_pruning_ena(main_vsi)) 1235 ctxt->info.sw_flags2 |= 1236 ICE_AQ_VSI_SW_FLAG_RX_VLAN_PRUNE_ENA; 1237 else 1238 ctxt->info.sw_flags2 &= 1239 ~ICE_AQ_VSI_SW_FLAG_RX_VLAN_PRUNE_ENA; 1240 } 1241 1242 ice_set_dflt_vsi_ctx(hw, ctxt); 1243 if (test_bit(ICE_FLAG_FD_ENA, pf->flags)) 1244 ice_set_fd_vsi_ctx(ctxt, vsi); 1245 /* if the switch is in VEB mode, allow VSI loopback */ 1246 if (vsi->vsw->bridge_mode == BRIDGE_MODE_VEB) 1247 ctxt->info.sw_flags |= ICE_AQ_VSI_SW_FLAG_ALLOW_LB; 1248 1249 /* Set LUT type and HASH type if RSS is enabled */ 1250 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags) && 1251 vsi->type != ICE_VSI_CTRL) { 1252 ice_set_rss_vsi_ctx(ctxt, vsi); 1253 /* if updating VSI context, make sure to set valid_section: 1254 * to indicate which section of VSI context being updated 1255 */ 1256 if (!(vsi_flags & ICE_VSI_FLAG_INIT)) 1257 ctxt->info.valid_sections |= 1258 cpu_to_le16(ICE_AQ_VSI_PROP_Q_OPT_VALID); 1259 } 1260 1261 ctxt->info.sw_id = vsi->port_info->sw_id; 1262 if (vsi->type == ICE_VSI_CHNL) { 1263 ice_chnl_vsi_setup_q_map(vsi, ctxt); 1264 } else { 1265 ret = ice_vsi_setup_q_map(vsi, ctxt); 1266 if (ret) 1267 goto out; 1268 1269 if (!(vsi_flags & ICE_VSI_FLAG_INIT)) 1270 /* means VSI being updated */ 1271 /* must to indicate which section of VSI context are 1272 * being modified 1273 */ 1274 ctxt->info.valid_sections |= 1275 cpu_to_le16(ICE_AQ_VSI_PROP_RXQ_MAP_VALID); 1276 } 1277 1278 /* Allow control frames out of main VSI */ 1279 if (vsi->type == ICE_VSI_PF) { 1280 ctxt->info.sec_flags |= ICE_AQ_VSI_SEC_FLAG_ALLOW_DEST_OVRD; 1281 ctxt->info.valid_sections |= 1282 cpu_to_le16(ICE_AQ_VSI_PROP_SECURITY_VALID); 1283 } 1284 1285 if (vsi_flags & ICE_VSI_FLAG_INIT) { 1286 ret = ice_add_vsi(hw, vsi->idx, ctxt, NULL); 1287 if (ret) { 1288 dev_err(dev, "Add VSI failed, err %d\n", ret); 1289 ret = -EIO; 1290 goto out; 1291 } 1292 } else { 1293 ret = ice_update_vsi(hw, vsi->idx, ctxt, NULL); 1294 if (ret) { 1295 dev_err(dev, "Update VSI failed, err %d\n", ret); 1296 ret = -EIO; 1297 goto out; 1298 } 1299 } 1300 1301 /* keep context for update VSI operations */ 1302 vsi->info = ctxt->info; 1303 1304 /* record VSI number returned */ 1305 vsi->vsi_num = ctxt->vsi_num; 1306 1307 out: 1308 kfree(ctxt); 1309 return ret; 1310 } 1311 1312 /** 1313 * ice_vsi_clear_rings - Deallocates the Tx and Rx rings for VSI 1314 * @vsi: the VSI having rings deallocated 1315 */ 1316 static void ice_vsi_clear_rings(struct ice_vsi *vsi) 1317 { 1318 int i; 1319 1320 /* Avoid stale references by clearing map from vector to ring */ 1321 if (vsi->q_vectors) { 1322 ice_for_each_q_vector(vsi, i) { 1323 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 1324 1325 if (q_vector) { 1326 q_vector->tx.tx_ring = NULL; 1327 q_vector->rx.rx_ring = NULL; 1328 } 1329 } 1330 } 1331 1332 if (vsi->tx_rings) { 1333 ice_for_each_alloc_txq(vsi, i) { 1334 if (vsi->tx_rings[i]) { 1335 kfree_rcu(vsi->tx_rings[i], rcu); 1336 WRITE_ONCE(vsi->tx_rings[i], NULL); 1337 } 1338 } 1339 } 1340 if (vsi->rx_rings) { 1341 ice_for_each_alloc_rxq(vsi, i) { 1342 if (vsi->rx_rings[i]) { 1343 kfree_rcu(vsi->rx_rings[i], rcu); 1344 WRITE_ONCE(vsi->rx_rings[i], NULL); 1345 } 1346 } 1347 } 1348 } 1349 1350 /** 1351 * ice_vsi_alloc_rings - Allocates Tx and Rx rings for the VSI 1352 * @vsi: VSI which is having rings allocated 1353 */ 1354 static int ice_vsi_alloc_rings(struct ice_vsi *vsi) 1355 { 1356 bool dvm_ena = ice_is_dvm_ena(&vsi->back->hw); 1357 struct ice_pf *pf = vsi->back; 1358 struct device *dev; 1359 u16 i; 1360 1361 dev = ice_pf_to_dev(pf); 1362 /* Allocate Tx rings */ 1363 ice_for_each_alloc_txq(vsi, i) { 1364 struct ice_tx_ring *ring; 1365 1366 /* allocate with kzalloc(), free with kfree_rcu() */ 1367 ring = kzalloc(sizeof(*ring), GFP_KERNEL); 1368 1369 if (!ring) 1370 goto err_out; 1371 1372 ring->q_index = i; 1373 ring->reg_idx = vsi->txq_map[i]; 1374 ring->vsi = vsi; 1375 ring->tx_tstamps = &pf->ptp.port.tx; 1376 ring->dev = dev; 1377 ring->count = vsi->num_tx_desc; 1378 ring->txq_teid = ICE_INVAL_TEID; 1379 if (dvm_ena) 1380 ring->flags |= ICE_TX_FLAGS_RING_VLAN_L2TAG2; 1381 else 1382 ring->flags |= ICE_TX_FLAGS_RING_VLAN_L2TAG1; 1383 WRITE_ONCE(vsi->tx_rings[i], ring); 1384 } 1385 1386 /* Allocate Rx rings */ 1387 ice_for_each_alloc_rxq(vsi, i) { 1388 struct ice_rx_ring *ring; 1389 1390 /* allocate with kzalloc(), free with kfree_rcu() */ 1391 ring = kzalloc(sizeof(*ring), GFP_KERNEL); 1392 if (!ring) 1393 goto err_out; 1394 1395 ring->q_index = i; 1396 ring->reg_idx = vsi->rxq_map[i]; 1397 ring->vsi = vsi; 1398 ring->netdev = vsi->netdev; 1399 ring->dev = dev; 1400 ring->count = vsi->num_rx_desc; 1401 ring->cached_phctime = pf->ptp.cached_phc_time; 1402 WRITE_ONCE(vsi->rx_rings[i], ring); 1403 } 1404 1405 return 0; 1406 1407 err_out: 1408 ice_vsi_clear_rings(vsi); 1409 return -ENOMEM; 1410 } 1411 1412 /** 1413 * ice_vsi_manage_rss_lut - disable/enable RSS 1414 * @vsi: the VSI being changed 1415 * @ena: boolean value indicating if this is an enable or disable request 1416 * 1417 * In the event of disable request for RSS, this function will zero out RSS 1418 * LUT, while in the event of enable request for RSS, it will reconfigure RSS 1419 * LUT. 1420 */ 1421 void ice_vsi_manage_rss_lut(struct ice_vsi *vsi, bool ena) 1422 { 1423 u8 *lut; 1424 1425 lut = kzalloc(vsi->rss_table_size, GFP_KERNEL); 1426 if (!lut) 1427 return; 1428 1429 if (ena) { 1430 if (vsi->rss_lut_user) 1431 memcpy(lut, vsi->rss_lut_user, vsi->rss_table_size); 1432 else 1433 ice_fill_rss_lut(lut, vsi->rss_table_size, 1434 vsi->rss_size); 1435 } 1436 1437 ice_set_rss_lut(vsi, lut, vsi->rss_table_size); 1438 kfree(lut); 1439 } 1440 1441 /** 1442 * ice_vsi_cfg_crc_strip - Configure CRC stripping for a VSI 1443 * @vsi: VSI to be configured 1444 * @disable: set to true to have FCS / CRC in the frame data 1445 */ 1446 void ice_vsi_cfg_crc_strip(struct ice_vsi *vsi, bool disable) 1447 { 1448 int i; 1449 1450 ice_for_each_rxq(vsi, i) 1451 if (disable) 1452 vsi->rx_rings[i]->flags |= ICE_RX_FLAGS_CRC_STRIP_DIS; 1453 else 1454 vsi->rx_rings[i]->flags &= ~ICE_RX_FLAGS_CRC_STRIP_DIS; 1455 } 1456 1457 /** 1458 * ice_vsi_cfg_rss_lut_key - Configure RSS params for a VSI 1459 * @vsi: VSI to be configured 1460 */ 1461 int ice_vsi_cfg_rss_lut_key(struct ice_vsi *vsi) 1462 { 1463 struct ice_pf *pf = vsi->back; 1464 struct device *dev; 1465 u8 *lut, *key; 1466 int err; 1467 1468 dev = ice_pf_to_dev(pf); 1469 if (vsi->type == ICE_VSI_PF && vsi->ch_rss_size && 1470 (test_bit(ICE_FLAG_TC_MQPRIO, pf->flags))) { 1471 vsi->rss_size = min_t(u16, vsi->rss_size, vsi->ch_rss_size); 1472 } else { 1473 vsi->rss_size = min_t(u16, vsi->rss_size, vsi->num_rxq); 1474 1475 /* If orig_rss_size is valid and it is less than determined 1476 * main VSI's rss_size, update main VSI's rss_size to be 1477 * orig_rss_size so that when tc-qdisc is deleted, main VSI 1478 * RSS table gets programmed to be correct (whatever it was 1479 * to begin with (prior to setup-tc for ADQ config) 1480 */ 1481 if (vsi->orig_rss_size && vsi->rss_size < vsi->orig_rss_size && 1482 vsi->orig_rss_size <= vsi->num_rxq) { 1483 vsi->rss_size = vsi->orig_rss_size; 1484 /* now orig_rss_size is used, reset it to zero */ 1485 vsi->orig_rss_size = 0; 1486 } 1487 } 1488 1489 lut = kzalloc(vsi->rss_table_size, GFP_KERNEL); 1490 if (!lut) 1491 return -ENOMEM; 1492 1493 if (vsi->rss_lut_user) 1494 memcpy(lut, vsi->rss_lut_user, vsi->rss_table_size); 1495 else 1496 ice_fill_rss_lut(lut, vsi->rss_table_size, vsi->rss_size); 1497 1498 err = ice_set_rss_lut(vsi, lut, vsi->rss_table_size); 1499 if (err) { 1500 dev_err(dev, "set_rss_lut failed, error %d\n", err); 1501 goto ice_vsi_cfg_rss_exit; 1502 } 1503 1504 key = kzalloc(ICE_GET_SET_RSS_KEY_EXTEND_KEY_SIZE, GFP_KERNEL); 1505 if (!key) { 1506 err = -ENOMEM; 1507 goto ice_vsi_cfg_rss_exit; 1508 } 1509 1510 if (vsi->rss_hkey_user) 1511 memcpy(key, vsi->rss_hkey_user, ICE_GET_SET_RSS_KEY_EXTEND_KEY_SIZE); 1512 else 1513 netdev_rss_key_fill((void *)key, ICE_GET_SET_RSS_KEY_EXTEND_KEY_SIZE); 1514 1515 err = ice_set_rss_key(vsi, key); 1516 if (err) 1517 dev_err(dev, "set_rss_key failed, error %d\n", err); 1518 1519 kfree(key); 1520 ice_vsi_cfg_rss_exit: 1521 kfree(lut); 1522 return err; 1523 } 1524 1525 /** 1526 * ice_vsi_set_vf_rss_flow_fld - Sets VF VSI RSS input set for different flows 1527 * @vsi: VSI to be configured 1528 * 1529 * This function will only be called during the VF VSI setup. Upon successful 1530 * completion of package download, this function will configure default RSS 1531 * input sets for VF VSI. 1532 */ 1533 static void ice_vsi_set_vf_rss_flow_fld(struct ice_vsi *vsi) 1534 { 1535 struct ice_pf *pf = vsi->back; 1536 struct device *dev; 1537 int status; 1538 1539 dev = ice_pf_to_dev(pf); 1540 if (ice_is_safe_mode(pf)) { 1541 dev_dbg(dev, "Advanced RSS disabled. Package download failed, vsi num = %d\n", 1542 vsi->vsi_num); 1543 return; 1544 } 1545 1546 status = ice_add_avf_rss_cfg(&pf->hw, vsi, ICE_DEFAULT_RSS_HENA); 1547 if (status) 1548 dev_dbg(dev, "ice_add_avf_rss_cfg failed for vsi = %d, error = %d\n", 1549 vsi->vsi_num, status); 1550 } 1551 1552 static const struct ice_rss_hash_cfg default_rss_cfgs[] = { 1553 /* configure RSS for IPv4 with input set IP src/dst */ 1554 {ICE_FLOW_SEG_HDR_IPV4, ICE_FLOW_HASH_IPV4, ICE_RSS_ANY_HEADERS, false}, 1555 /* configure RSS for IPv6 with input set IPv6 src/dst */ 1556 {ICE_FLOW_SEG_HDR_IPV6, ICE_FLOW_HASH_IPV6, ICE_RSS_ANY_HEADERS, false}, 1557 /* configure RSS for tcp4 with input set IP src/dst, TCP src/dst */ 1558 {ICE_FLOW_SEG_HDR_TCP | ICE_FLOW_SEG_HDR_IPV4, 1559 ICE_HASH_TCP_IPV4, ICE_RSS_ANY_HEADERS, false}, 1560 /* configure RSS for udp4 with input set IP src/dst, UDP src/dst */ 1561 {ICE_FLOW_SEG_HDR_UDP | ICE_FLOW_SEG_HDR_IPV4, 1562 ICE_HASH_UDP_IPV4, ICE_RSS_ANY_HEADERS, false}, 1563 /* configure RSS for sctp4 with input set IP src/dst - only support 1564 * RSS on SCTPv4 on outer headers (non-tunneled) 1565 */ 1566 {ICE_FLOW_SEG_HDR_SCTP | ICE_FLOW_SEG_HDR_IPV4, 1567 ICE_HASH_SCTP_IPV4, ICE_RSS_OUTER_HEADERS, false}, 1568 /* configure RSS for gtpc4 with input set IPv4 src/dst */ 1569 {ICE_FLOW_SEG_HDR_GTPC | ICE_FLOW_SEG_HDR_IPV4, 1570 ICE_FLOW_HASH_IPV4, ICE_RSS_OUTER_HEADERS, false}, 1571 /* configure RSS for gtpc4t with input set IPv4 src/dst */ 1572 {ICE_FLOW_SEG_HDR_GTPC_TEID | ICE_FLOW_SEG_HDR_IPV4, 1573 ICE_FLOW_HASH_GTP_C_IPV4_TEID, ICE_RSS_OUTER_HEADERS, false}, 1574 /* configure RSS for gtpu4 with input set IPv4 src/dst */ 1575 {ICE_FLOW_SEG_HDR_GTPU_IP | ICE_FLOW_SEG_HDR_IPV4, 1576 ICE_FLOW_HASH_GTP_U_IPV4_TEID, ICE_RSS_OUTER_HEADERS, false}, 1577 /* configure RSS for gtpu4e with input set IPv4 src/dst */ 1578 {ICE_FLOW_SEG_HDR_GTPU_EH | ICE_FLOW_SEG_HDR_IPV4, 1579 ICE_FLOW_HASH_GTP_U_IPV4_EH, ICE_RSS_OUTER_HEADERS, false}, 1580 /* configure RSS for gtpu4u with input set IPv4 src/dst */ 1581 { ICE_FLOW_SEG_HDR_GTPU_UP | ICE_FLOW_SEG_HDR_IPV4, 1582 ICE_FLOW_HASH_GTP_U_IPV4_UP, ICE_RSS_OUTER_HEADERS, false}, 1583 /* configure RSS for gtpu4d with input set IPv4 src/dst */ 1584 {ICE_FLOW_SEG_HDR_GTPU_DWN | ICE_FLOW_SEG_HDR_IPV4, 1585 ICE_FLOW_HASH_GTP_U_IPV4_DWN, ICE_RSS_OUTER_HEADERS, false}, 1586 1587 /* configure RSS for tcp6 with input set IPv6 src/dst, TCP src/dst */ 1588 {ICE_FLOW_SEG_HDR_TCP | ICE_FLOW_SEG_HDR_IPV6, 1589 ICE_HASH_TCP_IPV6, ICE_RSS_ANY_HEADERS, false}, 1590 /* configure RSS for udp6 with input set IPv6 src/dst, UDP src/dst */ 1591 {ICE_FLOW_SEG_HDR_UDP | ICE_FLOW_SEG_HDR_IPV6, 1592 ICE_HASH_UDP_IPV6, ICE_RSS_ANY_HEADERS, false}, 1593 /* configure RSS for sctp6 with input set IPv6 src/dst - only support 1594 * RSS on SCTPv6 on outer headers (non-tunneled) 1595 */ 1596 {ICE_FLOW_SEG_HDR_SCTP | ICE_FLOW_SEG_HDR_IPV6, 1597 ICE_HASH_SCTP_IPV6, ICE_RSS_OUTER_HEADERS, false}, 1598 /* configure RSS for IPSEC ESP SPI with input set MAC_IPV4_SPI */ 1599 {ICE_FLOW_SEG_HDR_ESP, 1600 ICE_FLOW_HASH_ESP_SPI, ICE_RSS_OUTER_HEADERS, false}, 1601 /* configure RSS for gtpc6 with input set IPv6 src/dst */ 1602 {ICE_FLOW_SEG_HDR_GTPC | ICE_FLOW_SEG_HDR_IPV6, 1603 ICE_FLOW_HASH_IPV6, ICE_RSS_OUTER_HEADERS, false}, 1604 /* configure RSS for gtpc6t with input set IPv6 src/dst */ 1605 {ICE_FLOW_SEG_HDR_GTPC_TEID | ICE_FLOW_SEG_HDR_IPV6, 1606 ICE_FLOW_HASH_GTP_C_IPV6_TEID, ICE_RSS_OUTER_HEADERS, false}, 1607 /* configure RSS for gtpu6 with input set IPv6 src/dst */ 1608 {ICE_FLOW_SEG_HDR_GTPU_IP | ICE_FLOW_SEG_HDR_IPV6, 1609 ICE_FLOW_HASH_GTP_U_IPV6_TEID, ICE_RSS_OUTER_HEADERS, false}, 1610 /* configure RSS for gtpu6e with input set IPv6 src/dst */ 1611 {ICE_FLOW_SEG_HDR_GTPU_EH | ICE_FLOW_SEG_HDR_IPV6, 1612 ICE_FLOW_HASH_GTP_U_IPV6_EH, ICE_RSS_OUTER_HEADERS, false}, 1613 /* configure RSS for gtpu6u with input set IPv6 src/dst */ 1614 { ICE_FLOW_SEG_HDR_GTPU_UP | ICE_FLOW_SEG_HDR_IPV6, 1615 ICE_FLOW_HASH_GTP_U_IPV6_UP, ICE_RSS_OUTER_HEADERS, false}, 1616 /* configure RSS for gtpu6d with input set IPv6 src/dst */ 1617 {ICE_FLOW_SEG_HDR_GTPU_DWN | ICE_FLOW_SEG_HDR_IPV6, 1618 ICE_FLOW_HASH_GTP_U_IPV6_DWN, ICE_RSS_OUTER_HEADERS, false}, 1619 }; 1620 1621 /** 1622 * ice_vsi_set_rss_flow_fld - Sets RSS input set for different flows 1623 * @vsi: VSI to be configured 1624 * 1625 * This function will only be called after successful download package call 1626 * during initialization of PF. Since the downloaded package will erase the 1627 * RSS section, this function will configure RSS input sets for different 1628 * flow types. The last profile added has the highest priority, therefore 2 1629 * tuple profiles (i.e. IPv4 src/dst) are added before 4 tuple profiles 1630 * (i.e. IPv4 src/dst TCP src/dst port). 1631 */ 1632 static void ice_vsi_set_rss_flow_fld(struct ice_vsi *vsi) 1633 { 1634 u16 vsi_num = vsi->vsi_num; 1635 struct ice_pf *pf = vsi->back; 1636 struct ice_hw *hw = &pf->hw; 1637 struct device *dev; 1638 int status; 1639 u32 i; 1640 1641 dev = ice_pf_to_dev(pf); 1642 if (ice_is_safe_mode(pf)) { 1643 dev_dbg(dev, "Advanced RSS disabled. Package download failed, vsi num = %d\n", 1644 vsi_num); 1645 return; 1646 } 1647 for (i = 0; i < ARRAY_SIZE(default_rss_cfgs); i++) { 1648 const struct ice_rss_hash_cfg *cfg = &default_rss_cfgs[i]; 1649 1650 status = ice_add_rss_cfg(hw, vsi, cfg); 1651 if (status) 1652 dev_dbg(dev, "ice_add_rss_cfg failed, addl_hdrs = %x, hash_flds = %llx, hdr_type = %d, symm = %d\n", 1653 cfg->addl_hdrs, cfg->hash_flds, 1654 cfg->hdr_type, cfg->symm); 1655 } 1656 } 1657 1658 /** 1659 * ice_pf_state_is_nominal - checks the PF for nominal state 1660 * @pf: pointer to PF to check 1661 * 1662 * Check the PF's state for a collection of bits that would indicate 1663 * the PF is in a state that would inhibit normal operation for 1664 * driver functionality. 1665 * 1666 * Returns true if PF is in a nominal state, false otherwise 1667 */ 1668 bool ice_pf_state_is_nominal(struct ice_pf *pf) 1669 { 1670 DECLARE_BITMAP(check_bits, ICE_STATE_NBITS) = { 0 }; 1671 1672 if (!pf) 1673 return false; 1674 1675 bitmap_set(check_bits, 0, ICE_STATE_NOMINAL_CHECK_BITS); 1676 if (bitmap_intersects(pf->state, check_bits, ICE_STATE_NBITS)) 1677 return false; 1678 1679 return true; 1680 } 1681 1682 /** 1683 * ice_update_eth_stats - Update VSI-specific ethernet statistics counters 1684 * @vsi: the VSI to be updated 1685 */ 1686 void ice_update_eth_stats(struct ice_vsi *vsi) 1687 { 1688 struct ice_eth_stats *prev_es, *cur_es; 1689 struct ice_hw *hw = &vsi->back->hw; 1690 struct ice_pf *pf = vsi->back; 1691 u16 vsi_num = vsi->vsi_num; /* HW absolute index of a VSI */ 1692 1693 prev_es = &vsi->eth_stats_prev; 1694 cur_es = &vsi->eth_stats; 1695 1696 if (ice_is_reset_in_progress(pf->state)) 1697 vsi->stat_offsets_loaded = false; 1698 1699 ice_stat_update40(hw, GLV_GORCL(vsi_num), vsi->stat_offsets_loaded, 1700 &prev_es->rx_bytes, &cur_es->rx_bytes); 1701 1702 ice_stat_update40(hw, GLV_UPRCL(vsi_num), vsi->stat_offsets_loaded, 1703 &prev_es->rx_unicast, &cur_es->rx_unicast); 1704 1705 ice_stat_update40(hw, GLV_MPRCL(vsi_num), vsi->stat_offsets_loaded, 1706 &prev_es->rx_multicast, &cur_es->rx_multicast); 1707 1708 ice_stat_update40(hw, GLV_BPRCL(vsi_num), vsi->stat_offsets_loaded, 1709 &prev_es->rx_broadcast, &cur_es->rx_broadcast); 1710 1711 ice_stat_update32(hw, GLV_RDPC(vsi_num), vsi->stat_offsets_loaded, 1712 &prev_es->rx_discards, &cur_es->rx_discards); 1713 1714 ice_stat_update40(hw, GLV_GOTCL(vsi_num), vsi->stat_offsets_loaded, 1715 &prev_es->tx_bytes, &cur_es->tx_bytes); 1716 1717 ice_stat_update40(hw, GLV_UPTCL(vsi_num), vsi->stat_offsets_loaded, 1718 &prev_es->tx_unicast, &cur_es->tx_unicast); 1719 1720 ice_stat_update40(hw, GLV_MPTCL(vsi_num), vsi->stat_offsets_loaded, 1721 &prev_es->tx_multicast, &cur_es->tx_multicast); 1722 1723 ice_stat_update40(hw, GLV_BPTCL(vsi_num), vsi->stat_offsets_loaded, 1724 &prev_es->tx_broadcast, &cur_es->tx_broadcast); 1725 1726 ice_stat_update32(hw, GLV_TEPC(vsi_num), vsi->stat_offsets_loaded, 1727 &prev_es->tx_errors, &cur_es->tx_errors); 1728 1729 vsi->stat_offsets_loaded = true; 1730 } 1731 1732 /** 1733 * ice_write_qrxflxp_cntxt - write/configure QRXFLXP_CNTXT register 1734 * @hw: HW pointer 1735 * @pf_q: index of the Rx queue in the PF's queue space 1736 * @rxdid: flexible descriptor RXDID 1737 * @prio: priority for the RXDID for this queue 1738 * @ena_ts: true to enable timestamp and false to disable timestamp 1739 */ 1740 void 1741 ice_write_qrxflxp_cntxt(struct ice_hw *hw, u16 pf_q, u32 rxdid, u32 prio, 1742 bool ena_ts) 1743 { 1744 int regval = rd32(hw, QRXFLXP_CNTXT(pf_q)); 1745 1746 /* clear any previous values */ 1747 regval &= ~(QRXFLXP_CNTXT_RXDID_IDX_M | 1748 QRXFLXP_CNTXT_RXDID_PRIO_M | 1749 QRXFLXP_CNTXT_TS_M); 1750 1751 regval |= FIELD_PREP(QRXFLXP_CNTXT_RXDID_IDX_M, rxdid); 1752 regval |= FIELD_PREP(QRXFLXP_CNTXT_RXDID_PRIO_M, prio); 1753 1754 if (ena_ts) 1755 /* Enable TimeSync on this queue */ 1756 regval |= QRXFLXP_CNTXT_TS_M; 1757 1758 wr32(hw, QRXFLXP_CNTXT(pf_q), regval); 1759 } 1760 1761 /** 1762 * ice_intrl_usec_to_reg - convert interrupt rate limit to register value 1763 * @intrl: interrupt rate limit in usecs 1764 * @gran: interrupt rate limit granularity in usecs 1765 * 1766 * This function converts a decimal interrupt rate limit in usecs to the format 1767 * expected by firmware. 1768 */ 1769 static u32 ice_intrl_usec_to_reg(u8 intrl, u8 gran) 1770 { 1771 u32 val = intrl / gran; 1772 1773 if (val) 1774 return val | GLINT_RATE_INTRL_ENA_M; 1775 return 0; 1776 } 1777 1778 /** 1779 * ice_write_intrl - write throttle rate limit to interrupt specific register 1780 * @q_vector: pointer to interrupt specific structure 1781 * @intrl: throttle rate limit in microseconds to write 1782 */ 1783 void ice_write_intrl(struct ice_q_vector *q_vector, u8 intrl) 1784 { 1785 struct ice_hw *hw = &q_vector->vsi->back->hw; 1786 1787 wr32(hw, GLINT_RATE(q_vector->reg_idx), 1788 ice_intrl_usec_to_reg(intrl, ICE_INTRL_GRAN_ABOVE_25)); 1789 } 1790 1791 static struct ice_q_vector *ice_pull_qvec_from_rc(struct ice_ring_container *rc) 1792 { 1793 switch (rc->type) { 1794 case ICE_RX_CONTAINER: 1795 if (rc->rx_ring) 1796 return rc->rx_ring->q_vector; 1797 break; 1798 case ICE_TX_CONTAINER: 1799 if (rc->tx_ring) 1800 return rc->tx_ring->q_vector; 1801 break; 1802 default: 1803 break; 1804 } 1805 1806 return NULL; 1807 } 1808 1809 /** 1810 * __ice_write_itr - write throttle rate to register 1811 * @q_vector: pointer to interrupt data structure 1812 * @rc: pointer to ring container 1813 * @itr: throttle rate in microseconds to write 1814 */ 1815 static void __ice_write_itr(struct ice_q_vector *q_vector, 1816 struct ice_ring_container *rc, u16 itr) 1817 { 1818 struct ice_hw *hw = &q_vector->vsi->back->hw; 1819 1820 wr32(hw, GLINT_ITR(rc->itr_idx, q_vector->reg_idx), 1821 ITR_REG_ALIGN(itr) >> ICE_ITR_GRAN_S); 1822 } 1823 1824 /** 1825 * ice_write_itr - write throttle rate to queue specific register 1826 * @rc: pointer to ring container 1827 * @itr: throttle rate in microseconds to write 1828 */ 1829 void ice_write_itr(struct ice_ring_container *rc, u16 itr) 1830 { 1831 struct ice_q_vector *q_vector; 1832 1833 q_vector = ice_pull_qvec_from_rc(rc); 1834 if (!q_vector) 1835 return; 1836 1837 __ice_write_itr(q_vector, rc, itr); 1838 } 1839 1840 /** 1841 * ice_set_q_vector_intrl - set up interrupt rate limiting 1842 * @q_vector: the vector to be configured 1843 * 1844 * Interrupt rate limiting is local to the vector, not per-queue so we must 1845 * detect if either ring container has dynamic moderation enabled to decide 1846 * what to set the interrupt rate limit to via INTRL settings. In the case that 1847 * dynamic moderation is disabled on both, write the value with the cached 1848 * setting to make sure INTRL register matches the user visible value. 1849 */ 1850 void ice_set_q_vector_intrl(struct ice_q_vector *q_vector) 1851 { 1852 if (ITR_IS_DYNAMIC(&q_vector->tx) || ITR_IS_DYNAMIC(&q_vector->rx)) { 1853 /* in the case of dynamic enabled, cap each vector to no more 1854 * than (4 us) 250,000 ints/sec, which allows low latency 1855 * but still less than 500,000 interrupts per second, which 1856 * reduces CPU a bit in the case of the lowest latency 1857 * setting. The 4 here is a value in microseconds. 1858 */ 1859 ice_write_intrl(q_vector, 4); 1860 } else { 1861 ice_write_intrl(q_vector, q_vector->intrl); 1862 } 1863 } 1864 1865 /** 1866 * ice_vsi_cfg_msix - MSIX mode Interrupt Config in the HW 1867 * @vsi: the VSI being configured 1868 * 1869 * This configures MSIX mode interrupts for the PF VSI, and should not be used 1870 * for the VF VSI. 1871 */ 1872 void ice_vsi_cfg_msix(struct ice_vsi *vsi) 1873 { 1874 struct ice_pf *pf = vsi->back; 1875 struct ice_hw *hw = &pf->hw; 1876 u16 txq = 0, rxq = 0; 1877 int i, q; 1878 1879 ice_for_each_q_vector(vsi, i) { 1880 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 1881 u16 reg_idx = q_vector->reg_idx; 1882 1883 ice_cfg_itr(hw, q_vector); 1884 1885 /* Both Transmit Queue Interrupt Cause Control register 1886 * and Receive Queue Interrupt Cause control register 1887 * expects MSIX_INDX field to be the vector index 1888 * within the function space and not the absolute 1889 * vector index across PF or across device. 1890 * For SR-IOV VF VSIs queue vector index always starts 1891 * with 1 since first vector index(0) is used for OICR 1892 * in VF space. Since VMDq and other PF VSIs are within 1893 * the PF function space, use the vector index that is 1894 * tracked for this PF. 1895 */ 1896 for (q = 0; q < q_vector->num_ring_tx; q++) { 1897 ice_cfg_txq_interrupt(vsi, txq, reg_idx, 1898 q_vector->tx.itr_idx); 1899 txq++; 1900 } 1901 1902 for (q = 0; q < q_vector->num_ring_rx; q++) { 1903 ice_cfg_rxq_interrupt(vsi, rxq, reg_idx, 1904 q_vector->rx.itr_idx); 1905 rxq++; 1906 } 1907 } 1908 } 1909 1910 /** 1911 * ice_vsi_start_all_rx_rings - start/enable all of a VSI's Rx rings 1912 * @vsi: the VSI whose rings are to be enabled 1913 * 1914 * Returns 0 on success and a negative value on error 1915 */ 1916 int ice_vsi_start_all_rx_rings(struct ice_vsi *vsi) 1917 { 1918 return ice_vsi_ctrl_all_rx_rings(vsi, true); 1919 } 1920 1921 /** 1922 * ice_vsi_stop_all_rx_rings - stop/disable all of a VSI's Rx rings 1923 * @vsi: the VSI whose rings are to be disabled 1924 * 1925 * Returns 0 on success and a negative value on error 1926 */ 1927 int ice_vsi_stop_all_rx_rings(struct ice_vsi *vsi) 1928 { 1929 return ice_vsi_ctrl_all_rx_rings(vsi, false); 1930 } 1931 1932 /** 1933 * ice_vsi_stop_tx_rings - Disable Tx rings 1934 * @vsi: the VSI being configured 1935 * @rst_src: reset source 1936 * @rel_vmvf_num: Relative ID of VF/VM 1937 * @rings: Tx ring array to be stopped 1938 * @count: number of Tx ring array elements 1939 */ 1940 static int 1941 ice_vsi_stop_tx_rings(struct ice_vsi *vsi, enum ice_disq_rst_src rst_src, 1942 u16 rel_vmvf_num, struct ice_tx_ring **rings, u16 count) 1943 { 1944 u16 q_idx; 1945 1946 if (vsi->num_txq > ICE_LAN_TXQ_MAX_QDIS) 1947 return -EINVAL; 1948 1949 for (q_idx = 0; q_idx < count; q_idx++) { 1950 struct ice_txq_meta txq_meta = { }; 1951 int status; 1952 1953 if (!rings || !rings[q_idx]) 1954 return -EINVAL; 1955 1956 ice_fill_txq_meta(vsi, rings[q_idx], &txq_meta); 1957 status = ice_vsi_stop_tx_ring(vsi, rst_src, rel_vmvf_num, 1958 rings[q_idx], &txq_meta); 1959 1960 if (status) 1961 return status; 1962 } 1963 1964 return 0; 1965 } 1966 1967 /** 1968 * ice_vsi_stop_lan_tx_rings - Disable LAN Tx rings 1969 * @vsi: the VSI being configured 1970 * @rst_src: reset source 1971 * @rel_vmvf_num: Relative ID of VF/VM 1972 */ 1973 int 1974 ice_vsi_stop_lan_tx_rings(struct ice_vsi *vsi, enum ice_disq_rst_src rst_src, 1975 u16 rel_vmvf_num) 1976 { 1977 return ice_vsi_stop_tx_rings(vsi, rst_src, rel_vmvf_num, vsi->tx_rings, vsi->num_txq); 1978 } 1979 1980 /** 1981 * ice_vsi_stop_xdp_tx_rings - Disable XDP Tx rings 1982 * @vsi: the VSI being configured 1983 */ 1984 int ice_vsi_stop_xdp_tx_rings(struct ice_vsi *vsi) 1985 { 1986 return ice_vsi_stop_tx_rings(vsi, ICE_NO_RESET, 0, vsi->xdp_rings, vsi->num_xdp_txq); 1987 } 1988 1989 /** 1990 * ice_vsi_is_rx_queue_active 1991 * @vsi: the VSI being configured 1992 * 1993 * Return true if at least one queue is active. 1994 */ 1995 bool ice_vsi_is_rx_queue_active(struct ice_vsi *vsi) 1996 { 1997 struct ice_pf *pf = vsi->back; 1998 struct ice_hw *hw = &pf->hw; 1999 int i; 2000 2001 ice_for_each_rxq(vsi, i) { 2002 u32 rx_reg; 2003 int pf_q; 2004 2005 pf_q = vsi->rxq_map[i]; 2006 rx_reg = rd32(hw, QRX_CTRL(pf_q)); 2007 if (rx_reg & QRX_CTRL_QENA_STAT_M) 2008 return true; 2009 } 2010 2011 return false; 2012 } 2013 2014 static void ice_vsi_set_tc_cfg(struct ice_vsi *vsi) 2015 { 2016 if (!test_bit(ICE_FLAG_DCB_ENA, vsi->back->flags)) { 2017 vsi->tc_cfg.ena_tc = ICE_DFLT_TRAFFIC_CLASS; 2018 vsi->tc_cfg.numtc = 1; 2019 return; 2020 } 2021 2022 /* set VSI TC information based on DCB config */ 2023 ice_vsi_set_dcb_tc_cfg(vsi); 2024 } 2025 2026 /** 2027 * ice_cfg_sw_lldp - Config switch rules for LLDP packet handling 2028 * @vsi: the VSI being configured 2029 * @tx: bool to determine Tx or Rx rule 2030 * @create: bool to determine create or remove Rule 2031 */ 2032 void ice_cfg_sw_lldp(struct ice_vsi *vsi, bool tx, bool create) 2033 { 2034 int (*eth_fltr)(struct ice_vsi *v, u16 type, u16 flag, 2035 enum ice_sw_fwd_act_type act); 2036 struct ice_pf *pf = vsi->back; 2037 struct device *dev; 2038 int status; 2039 2040 dev = ice_pf_to_dev(pf); 2041 eth_fltr = create ? ice_fltr_add_eth : ice_fltr_remove_eth; 2042 2043 if (tx) { 2044 status = eth_fltr(vsi, ETH_P_LLDP, ICE_FLTR_TX, 2045 ICE_DROP_PACKET); 2046 } else { 2047 if (ice_fw_supports_lldp_fltr_ctrl(&pf->hw)) { 2048 status = ice_lldp_fltr_add_remove(&pf->hw, vsi->vsi_num, 2049 create); 2050 } else { 2051 status = eth_fltr(vsi, ETH_P_LLDP, ICE_FLTR_RX, 2052 ICE_FWD_TO_VSI); 2053 } 2054 } 2055 2056 if (status) 2057 dev_dbg(dev, "Fail %s %s LLDP rule on VSI %i error: %d\n", 2058 create ? "adding" : "removing", tx ? "TX" : "RX", 2059 vsi->vsi_num, status); 2060 } 2061 2062 /** 2063 * ice_set_agg_vsi - sets up scheduler aggregator node and move VSI into it 2064 * @vsi: pointer to the VSI 2065 * 2066 * This function will allocate new scheduler aggregator now if needed and will 2067 * move specified VSI into it. 2068 */ 2069 static void ice_set_agg_vsi(struct ice_vsi *vsi) 2070 { 2071 struct device *dev = ice_pf_to_dev(vsi->back); 2072 struct ice_agg_node *agg_node_iter = NULL; 2073 u32 agg_id = ICE_INVALID_AGG_NODE_ID; 2074 struct ice_agg_node *agg_node = NULL; 2075 int node_offset, max_agg_nodes = 0; 2076 struct ice_port_info *port_info; 2077 struct ice_pf *pf = vsi->back; 2078 u32 agg_node_id_start = 0; 2079 int status; 2080 2081 /* create (as needed) scheduler aggregator node and move VSI into 2082 * corresponding aggregator node 2083 * - PF aggregator node to contains VSIs of type _PF and _CTRL 2084 * - VF aggregator nodes will contain VF VSI 2085 */ 2086 port_info = pf->hw.port_info; 2087 if (!port_info) 2088 return; 2089 2090 switch (vsi->type) { 2091 case ICE_VSI_CTRL: 2092 case ICE_VSI_CHNL: 2093 case ICE_VSI_LB: 2094 case ICE_VSI_PF: 2095 max_agg_nodes = ICE_MAX_PF_AGG_NODES; 2096 agg_node_id_start = ICE_PF_AGG_NODE_ID_START; 2097 agg_node_iter = &pf->pf_agg_node[0]; 2098 break; 2099 case ICE_VSI_VF: 2100 /* user can create 'n' VFs on a given PF, but since max children 2101 * per aggregator node can be only 64. Following code handles 2102 * aggregator(s) for VF VSIs, either selects a agg_node which 2103 * was already created provided num_vsis < 64, otherwise 2104 * select next available node, which will be created 2105 */ 2106 max_agg_nodes = ICE_MAX_VF_AGG_NODES; 2107 agg_node_id_start = ICE_VF_AGG_NODE_ID_START; 2108 agg_node_iter = &pf->vf_agg_node[0]; 2109 break; 2110 default: 2111 /* other VSI type, handle later if needed */ 2112 dev_dbg(dev, "unexpected VSI type %s\n", 2113 ice_vsi_type_str(vsi->type)); 2114 return; 2115 } 2116 2117 /* find the appropriate aggregator node */ 2118 for (node_offset = 0; node_offset < max_agg_nodes; node_offset++) { 2119 /* see if we can find space in previously created 2120 * node if num_vsis < 64, otherwise skip 2121 */ 2122 if (agg_node_iter->num_vsis && 2123 agg_node_iter->num_vsis == ICE_MAX_VSIS_IN_AGG_NODE) { 2124 agg_node_iter++; 2125 continue; 2126 } 2127 2128 if (agg_node_iter->valid && 2129 agg_node_iter->agg_id != ICE_INVALID_AGG_NODE_ID) { 2130 agg_id = agg_node_iter->agg_id; 2131 agg_node = agg_node_iter; 2132 break; 2133 } 2134 2135 /* find unclaimed agg_id */ 2136 if (agg_node_iter->agg_id == ICE_INVALID_AGG_NODE_ID) { 2137 agg_id = node_offset + agg_node_id_start; 2138 agg_node = agg_node_iter; 2139 break; 2140 } 2141 /* move to next agg_node */ 2142 agg_node_iter++; 2143 } 2144 2145 if (!agg_node) 2146 return; 2147 2148 /* if selected aggregator node was not created, create it */ 2149 if (!agg_node->valid) { 2150 status = ice_cfg_agg(port_info, agg_id, ICE_AGG_TYPE_AGG, 2151 (u8)vsi->tc_cfg.ena_tc); 2152 if (status) { 2153 dev_err(dev, "unable to create aggregator node with agg_id %u\n", 2154 agg_id); 2155 return; 2156 } 2157 /* aggregator node is created, store the needed info */ 2158 agg_node->valid = true; 2159 agg_node->agg_id = agg_id; 2160 } 2161 2162 /* move VSI to corresponding aggregator node */ 2163 status = ice_move_vsi_to_agg(port_info, agg_id, vsi->idx, 2164 (u8)vsi->tc_cfg.ena_tc); 2165 if (status) { 2166 dev_err(dev, "unable to move VSI idx %u into aggregator %u node", 2167 vsi->idx, agg_id); 2168 return; 2169 } 2170 2171 /* keep active children count for aggregator node */ 2172 agg_node->num_vsis++; 2173 2174 /* cache the 'agg_id' in VSI, so that after reset - VSI will be moved 2175 * to aggregator node 2176 */ 2177 vsi->agg_node = agg_node; 2178 dev_dbg(dev, "successfully moved VSI idx %u tc_bitmap 0x%x) into aggregator node %d which has num_vsis %u\n", 2179 vsi->idx, vsi->tc_cfg.ena_tc, vsi->agg_node->agg_id, 2180 vsi->agg_node->num_vsis); 2181 } 2182 2183 static int ice_vsi_cfg_tc_lan(struct ice_pf *pf, struct ice_vsi *vsi) 2184 { 2185 u16 max_txqs[ICE_MAX_TRAFFIC_CLASS] = { 0 }; 2186 struct device *dev = ice_pf_to_dev(pf); 2187 int ret, i; 2188 2189 /* configure VSI nodes based on number of queues and TC's */ 2190 ice_for_each_traffic_class(i) { 2191 if (!(vsi->tc_cfg.ena_tc & BIT(i))) 2192 continue; 2193 2194 if (vsi->type == ICE_VSI_CHNL) { 2195 if (!vsi->alloc_txq && vsi->num_txq) 2196 max_txqs[i] = vsi->num_txq; 2197 else 2198 max_txqs[i] = pf->num_lan_tx; 2199 } else { 2200 max_txqs[i] = vsi->alloc_txq; 2201 } 2202 2203 if (vsi->type == ICE_VSI_PF) 2204 max_txqs[i] += vsi->num_xdp_txq; 2205 } 2206 2207 dev_dbg(dev, "vsi->tc_cfg.ena_tc = %d\n", vsi->tc_cfg.ena_tc); 2208 ret = ice_cfg_vsi_lan(vsi->port_info, vsi->idx, vsi->tc_cfg.ena_tc, 2209 max_txqs); 2210 if (ret) { 2211 dev_err(dev, "VSI %d failed lan queue config, error %d\n", 2212 vsi->vsi_num, ret); 2213 return ret; 2214 } 2215 2216 return 0; 2217 } 2218 2219 /** 2220 * ice_vsi_cfg_def - configure default VSI based on the type 2221 * @vsi: pointer to VSI 2222 */ 2223 static int ice_vsi_cfg_def(struct ice_vsi *vsi) 2224 { 2225 struct device *dev = ice_pf_to_dev(vsi->back); 2226 struct ice_pf *pf = vsi->back; 2227 int ret; 2228 2229 vsi->vsw = pf->first_sw; 2230 2231 ret = ice_vsi_alloc_def(vsi, vsi->ch); 2232 if (ret) 2233 return ret; 2234 2235 /* allocate memory for Tx/Rx ring stat pointers */ 2236 ret = ice_vsi_alloc_stat_arrays(vsi); 2237 if (ret) 2238 goto unroll_vsi_alloc; 2239 2240 ice_alloc_fd_res(vsi); 2241 2242 ret = ice_vsi_get_qs(vsi); 2243 if (ret) { 2244 dev_err(dev, "Failed to allocate queues. vsi->idx = %d\n", 2245 vsi->idx); 2246 goto unroll_vsi_alloc_stat; 2247 } 2248 2249 /* set RSS capabilities */ 2250 ice_vsi_set_rss_params(vsi); 2251 2252 /* set TC configuration */ 2253 ice_vsi_set_tc_cfg(vsi); 2254 2255 /* create the VSI */ 2256 ret = ice_vsi_init(vsi, vsi->flags); 2257 if (ret) 2258 goto unroll_get_qs; 2259 2260 ice_vsi_init_vlan_ops(vsi); 2261 2262 switch (vsi->type) { 2263 case ICE_VSI_CTRL: 2264 case ICE_VSI_PF: 2265 ret = ice_vsi_alloc_q_vectors(vsi); 2266 if (ret) 2267 goto unroll_vsi_init; 2268 2269 ret = ice_vsi_alloc_rings(vsi); 2270 if (ret) 2271 goto unroll_vector_base; 2272 2273 ret = ice_vsi_alloc_ring_stats(vsi); 2274 if (ret) 2275 goto unroll_vector_base; 2276 2277 if (ice_is_xdp_ena_vsi(vsi)) { 2278 ret = ice_vsi_determine_xdp_res(vsi); 2279 if (ret) 2280 goto unroll_vector_base; 2281 ret = ice_prepare_xdp_rings(vsi, vsi->xdp_prog, 2282 ICE_XDP_CFG_PART); 2283 if (ret) 2284 goto unroll_vector_base; 2285 } 2286 2287 ice_vsi_map_rings_to_vectors(vsi); 2288 2289 /* Associate q_vector rings to napi */ 2290 ice_vsi_set_napi_queues(vsi); 2291 2292 vsi->stat_offsets_loaded = false; 2293 2294 /* ICE_VSI_CTRL does not need RSS so skip RSS processing */ 2295 if (vsi->type != ICE_VSI_CTRL) 2296 /* Do not exit if configuring RSS had an issue, at 2297 * least receive traffic on first queue. Hence no 2298 * need to capture return value 2299 */ 2300 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 2301 ice_vsi_cfg_rss_lut_key(vsi); 2302 ice_vsi_set_rss_flow_fld(vsi); 2303 } 2304 ice_init_arfs(vsi); 2305 break; 2306 case ICE_VSI_CHNL: 2307 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 2308 ice_vsi_cfg_rss_lut_key(vsi); 2309 ice_vsi_set_rss_flow_fld(vsi); 2310 } 2311 break; 2312 case ICE_VSI_VF: 2313 /* VF driver will take care of creating netdev for this type and 2314 * map queues to vectors through Virtchnl, PF driver only 2315 * creates a VSI and corresponding structures for bookkeeping 2316 * purpose 2317 */ 2318 ret = ice_vsi_alloc_q_vectors(vsi); 2319 if (ret) 2320 goto unroll_vsi_init; 2321 2322 ret = ice_vsi_alloc_rings(vsi); 2323 if (ret) 2324 goto unroll_alloc_q_vector; 2325 2326 ret = ice_vsi_alloc_ring_stats(vsi); 2327 if (ret) 2328 goto unroll_vector_base; 2329 2330 vsi->stat_offsets_loaded = false; 2331 2332 /* Do not exit if configuring RSS had an issue, at least 2333 * receive traffic on first queue. Hence no need to capture 2334 * return value 2335 */ 2336 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 2337 ice_vsi_cfg_rss_lut_key(vsi); 2338 ice_vsi_set_vf_rss_flow_fld(vsi); 2339 } 2340 break; 2341 case ICE_VSI_LB: 2342 ret = ice_vsi_alloc_rings(vsi); 2343 if (ret) 2344 goto unroll_vsi_init; 2345 2346 ret = ice_vsi_alloc_ring_stats(vsi); 2347 if (ret) 2348 goto unroll_vector_base; 2349 2350 break; 2351 default: 2352 /* clean up the resources and exit */ 2353 ret = -EINVAL; 2354 goto unroll_vsi_init; 2355 } 2356 2357 return 0; 2358 2359 unroll_vector_base: 2360 /* reclaim SW interrupts back to the common pool */ 2361 unroll_alloc_q_vector: 2362 ice_vsi_free_q_vectors(vsi); 2363 unroll_vsi_init: 2364 ice_vsi_delete_from_hw(vsi); 2365 unroll_get_qs: 2366 ice_vsi_put_qs(vsi); 2367 unroll_vsi_alloc_stat: 2368 ice_vsi_free_stats(vsi); 2369 unroll_vsi_alloc: 2370 ice_vsi_free_arrays(vsi); 2371 return ret; 2372 } 2373 2374 /** 2375 * ice_vsi_cfg - configure a previously allocated VSI 2376 * @vsi: pointer to VSI 2377 */ 2378 int ice_vsi_cfg(struct ice_vsi *vsi) 2379 { 2380 struct ice_pf *pf = vsi->back; 2381 int ret; 2382 2383 if (WARN_ON(vsi->type == ICE_VSI_VF && !vsi->vf)) 2384 return -EINVAL; 2385 2386 ret = ice_vsi_cfg_def(vsi); 2387 if (ret) 2388 return ret; 2389 2390 ret = ice_vsi_cfg_tc_lan(vsi->back, vsi); 2391 if (ret) 2392 ice_vsi_decfg(vsi); 2393 2394 if (vsi->type == ICE_VSI_CTRL) { 2395 if (vsi->vf) { 2396 WARN_ON(vsi->vf->ctrl_vsi_idx != ICE_NO_VSI); 2397 vsi->vf->ctrl_vsi_idx = vsi->idx; 2398 } else { 2399 WARN_ON(pf->ctrl_vsi_idx != ICE_NO_VSI); 2400 pf->ctrl_vsi_idx = vsi->idx; 2401 } 2402 } 2403 2404 return ret; 2405 } 2406 2407 /** 2408 * ice_vsi_decfg - remove all VSI configuration 2409 * @vsi: pointer to VSI 2410 */ 2411 void ice_vsi_decfg(struct ice_vsi *vsi) 2412 { 2413 struct ice_pf *pf = vsi->back; 2414 int err; 2415 2416 /* The Rx rule will only exist to remove if the LLDP FW 2417 * engine is currently stopped 2418 */ 2419 if (!ice_is_safe_mode(pf) && vsi->type == ICE_VSI_PF && 2420 !test_bit(ICE_FLAG_FW_LLDP_AGENT, pf->flags)) 2421 ice_cfg_sw_lldp(vsi, false, false); 2422 2423 ice_rm_vsi_lan_cfg(vsi->port_info, vsi->idx); 2424 err = ice_rm_vsi_rdma_cfg(vsi->port_info, vsi->idx); 2425 if (err) 2426 dev_err(ice_pf_to_dev(pf), "Failed to remove RDMA scheduler config for VSI %u, err %d\n", 2427 vsi->vsi_num, err); 2428 2429 if (ice_is_xdp_ena_vsi(vsi)) 2430 /* return value check can be skipped here, it always returns 2431 * 0 if reset is in progress 2432 */ 2433 ice_destroy_xdp_rings(vsi, ICE_XDP_CFG_PART); 2434 2435 ice_vsi_clear_rings(vsi); 2436 ice_vsi_free_q_vectors(vsi); 2437 ice_vsi_put_qs(vsi); 2438 ice_vsi_free_arrays(vsi); 2439 2440 /* SR-IOV determines needed MSIX resources all at once instead of per 2441 * VSI since when VFs are spawned we know how many VFs there are and how 2442 * many interrupts each VF needs. SR-IOV MSIX resources are also 2443 * cleared in the same manner. 2444 */ 2445 2446 if (vsi->type == ICE_VSI_VF && 2447 vsi->agg_node && vsi->agg_node->valid) 2448 vsi->agg_node->num_vsis--; 2449 } 2450 2451 /** 2452 * ice_vsi_setup - Set up a VSI by a given type 2453 * @pf: board private structure 2454 * @params: parameters to use when creating the VSI 2455 * 2456 * This allocates the sw VSI structure and its queue resources. 2457 * 2458 * Returns pointer to the successfully allocated and configured VSI sw struct on 2459 * success, NULL on failure. 2460 */ 2461 struct ice_vsi * 2462 ice_vsi_setup(struct ice_pf *pf, struct ice_vsi_cfg_params *params) 2463 { 2464 struct device *dev = ice_pf_to_dev(pf); 2465 struct ice_vsi *vsi; 2466 int ret; 2467 2468 /* ice_vsi_setup can only initialize a new VSI, and we must have 2469 * a port_info structure for it. 2470 */ 2471 if (WARN_ON(!(params->flags & ICE_VSI_FLAG_INIT)) || 2472 WARN_ON(!params->port_info)) 2473 return NULL; 2474 2475 vsi = ice_vsi_alloc(pf); 2476 if (!vsi) { 2477 dev_err(dev, "could not allocate VSI\n"); 2478 return NULL; 2479 } 2480 2481 vsi->params = *params; 2482 ret = ice_vsi_cfg(vsi); 2483 if (ret) 2484 goto err_vsi_cfg; 2485 2486 /* Add switch rule to drop all Tx Flow Control Frames, of look up 2487 * type ETHERTYPE from VSIs, and restrict malicious VF from sending 2488 * out PAUSE or PFC frames. If enabled, FW can still send FC frames. 2489 * The rule is added once for PF VSI in order to create appropriate 2490 * recipe, since VSI/VSI list is ignored with drop action... 2491 * Also add rules to handle LLDP Tx packets. Tx LLDP packets need to 2492 * be dropped so that VFs cannot send LLDP packets to reconfig DCB 2493 * settings in the HW. 2494 */ 2495 if (!ice_is_safe_mode(pf) && vsi->type == ICE_VSI_PF) { 2496 ice_fltr_add_eth(vsi, ETH_P_PAUSE, ICE_FLTR_TX, 2497 ICE_DROP_PACKET); 2498 ice_cfg_sw_lldp(vsi, true, true); 2499 } 2500 2501 if (!vsi->agg_node) 2502 ice_set_agg_vsi(vsi); 2503 2504 return vsi; 2505 2506 err_vsi_cfg: 2507 ice_vsi_free(vsi); 2508 2509 return NULL; 2510 } 2511 2512 /** 2513 * ice_vsi_release_msix - Clear the queue to Interrupt mapping in HW 2514 * @vsi: the VSI being cleaned up 2515 */ 2516 static void ice_vsi_release_msix(struct ice_vsi *vsi) 2517 { 2518 struct ice_pf *pf = vsi->back; 2519 struct ice_hw *hw = &pf->hw; 2520 u32 txq = 0; 2521 u32 rxq = 0; 2522 int i, q; 2523 2524 ice_for_each_q_vector(vsi, i) { 2525 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 2526 2527 ice_write_intrl(q_vector, 0); 2528 for (q = 0; q < q_vector->num_ring_tx; q++) { 2529 ice_write_itr(&q_vector->tx, 0); 2530 wr32(hw, QINT_TQCTL(vsi->txq_map[txq]), 0); 2531 if (ice_is_xdp_ena_vsi(vsi)) { 2532 u32 xdp_txq = txq + vsi->num_xdp_txq; 2533 2534 wr32(hw, QINT_TQCTL(vsi->txq_map[xdp_txq]), 0); 2535 } 2536 txq++; 2537 } 2538 2539 for (q = 0; q < q_vector->num_ring_rx; q++) { 2540 ice_write_itr(&q_vector->rx, 0); 2541 wr32(hw, QINT_RQCTL(vsi->rxq_map[rxq]), 0); 2542 rxq++; 2543 } 2544 } 2545 2546 ice_flush(hw); 2547 } 2548 2549 /** 2550 * ice_vsi_free_irq - Free the IRQ association with the OS 2551 * @vsi: the VSI being configured 2552 */ 2553 void ice_vsi_free_irq(struct ice_vsi *vsi) 2554 { 2555 struct ice_pf *pf = vsi->back; 2556 int i; 2557 2558 if (!vsi->q_vectors || !vsi->irqs_ready) 2559 return; 2560 2561 ice_vsi_release_msix(vsi); 2562 if (vsi->type == ICE_VSI_VF) 2563 return; 2564 2565 vsi->irqs_ready = false; 2566 ice_free_cpu_rx_rmap(vsi); 2567 2568 ice_for_each_q_vector(vsi, i) { 2569 int irq_num; 2570 2571 irq_num = vsi->q_vectors[i]->irq.virq; 2572 2573 /* free only the irqs that were actually requested */ 2574 if (!vsi->q_vectors[i] || 2575 !(vsi->q_vectors[i]->num_ring_tx || 2576 vsi->q_vectors[i]->num_ring_rx)) 2577 continue; 2578 2579 /* clear the affinity notifier in the IRQ descriptor */ 2580 if (!IS_ENABLED(CONFIG_RFS_ACCEL)) 2581 irq_set_affinity_notifier(irq_num, NULL); 2582 2583 /* clear the affinity_hint in the IRQ descriptor */ 2584 irq_update_affinity_hint(irq_num, NULL); 2585 synchronize_irq(irq_num); 2586 devm_free_irq(ice_pf_to_dev(pf), irq_num, vsi->q_vectors[i]); 2587 } 2588 } 2589 2590 /** 2591 * ice_vsi_free_tx_rings - Free Tx resources for VSI queues 2592 * @vsi: the VSI having resources freed 2593 */ 2594 void ice_vsi_free_tx_rings(struct ice_vsi *vsi) 2595 { 2596 int i; 2597 2598 if (!vsi->tx_rings) 2599 return; 2600 2601 ice_for_each_txq(vsi, i) 2602 if (vsi->tx_rings[i] && vsi->tx_rings[i]->desc) 2603 ice_free_tx_ring(vsi->tx_rings[i]); 2604 } 2605 2606 /** 2607 * ice_vsi_free_rx_rings - Free Rx resources for VSI queues 2608 * @vsi: the VSI having resources freed 2609 */ 2610 void ice_vsi_free_rx_rings(struct ice_vsi *vsi) 2611 { 2612 int i; 2613 2614 if (!vsi->rx_rings) 2615 return; 2616 2617 ice_for_each_rxq(vsi, i) 2618 if (vsi->rx_rings[i] && vsi->rx_rings[i]->desc) 2619 ice_free_rx_ring(vsi->rx_rings[i]); 2620 } 2621 2622 /** 2623 * ice_vsi_close - Shut down a VSI 2624 * @vsi: the VSI being shut down 2625 */ 2626 void ice_vsi_close(struct ice_vsi *vsi) 2627 { 2628 if (!test_and_set_bit(ICE_VSI_DOWN, vsi->state)) 2629 ice_down(vsi); 2630 2631 ice_vsi_free_irq(vsi); 2632 ice_vsi_free_tx_rings(vsi); 2633 ice_vsi_free_rx_rings(vsi); 2634 } 2635 2636 /** 2637 * ice_ena_vsi - resume a VSI 2638 * @vsi: the VSI being resume 2639 * @locked: is the rtnl_lock already held 2640 */ 2641 int ice_ena_vsi(struct ice_vsi *vsi, bool locked) 2642 { 2643 int err = 0; 2644 2645 if (!test_bit(ICE_VSI_NEEDS_RESTART, vsi->state)) 2646 return 0; 2647 2648 clear_bit(ICE_VSI_NEEDS_RESTART, vsi->state); 2649 2650 if (vsi->netdev && vsi->type == ICE_VSI_PF) { 2651 if (netif_running(vsi->netdev)) { 2652 if (!locked) 2653 rtnl_lock(); 2654 2655 err = ice_open_internal(vsi->netdev); 2656 2657 if (!locked) 2658 rtnl_unlock(); 2659 } 2660 } else if (vsi->type == ICE_VSI_CTRL) { 2661 err = ice_vsi_open_ctrl(vsi); 2662 } 2663 2664 return err; 2665 } 2666 2667 /** 2668 * ice_dis_vsi - pause a VSI 2669 * @vsi: the VSI being paused 2670 * @locked: is the rtnl_lock already held 2671 */ 2672 void ice_dis_vsi(struct ice_vsi *vsi, bool locked) 2673 { 2674 if (test_bit(ICE_VSI_DOWN, vsi->state)) 2675 return; 2676 2677 set_bit(ICE_VSI_NEEDS_RESTART, vsi->state); 2678 2679 if (vsi->type == ICE_VSI_PF && vsi->netdev) { 2680 if (netif_running(vsi->netdev)) { 2681 if (!locked) 2682 rtnl_lock(); 2683 2684 ice_vsi_close(vsi); 2685 2686 if (!locked) 2687 rtnl_unlock(); 2688 } else { 2689 ice_vsi_close(vsi); 2690 } 2691 } else if (vsi->type == ICE_VSI_CTRL) { 2692 ice_vsi_close(vsi); 2693 } 2694 } 2695 2696 /** 2697 * __ice_queue_set_napi - Set the napi instance for the queue 2698 * @dev: device to which NAPI and queue belong 2699 * @queue_index: Index of queue 2700 * @type: queue type as RX or TX 2701 * @napi: NAPI context 2702 * @locked: is the rtnl_lock already held 2703 * 2704 * Set the napi instance for the queue. Caller indicates the lock status. 2705 */ 2706 static void 2707 __ice_queue_set_napi(struct net_device *dev, unsigned int queue_index, 2708 enum netdev_queue_type type, struct napi_struct *napi, 2709 bool locked) 2710 { 2711 if (!locked) 2712 rtnl_lock(); 2713 netif_queue_set_napi(dev, queue_index, type, napi); 2714 if (!locked) 2715 rtnl_unlock(); 2716 } 2717 2718 /** 2719 * ice_queue_set_napi - Set the napi instance for the queue 2720 * @vsi: VSI being configured 2721 * @queue_index: Index of queue 2722 * @type: queue type as RX or TX 2723 * @napi: NAPI context 2724 * 2725 * Set the napi instance for the queue. The rtnl lock state is derived from the 2726 * execution path. 2727 */ 2728 void 2729 ice_queue_set_napi(struct ice_vsi *vsi, unsigned int queue_index, 2730 enum netdev_queue_type type, struct napi_struct *napi) 2731 { 2732 struct ice_pf *pf = vsi->back; 2733 2734 if (!vsi->netdev) 2735 return; 2736 2737 if (current_work() == &pf->serv_task || 2738 test_bit(ICE_PREPARED_FOR_RESET, pf->state) || 2739 test_bit(ICE_DOWN, pf->state) || 2740 test_bit(ICE_SUSPENDED, pf->state)) 2741 __ice_queue_set_napi(vsi->netdev, queue_index, type, napi, 2742 false); 2743 else 2744 __ice_queue_set_napi(vsi->netdev, queue_index, type, napi, 2745 true); 2746 } 2747 2748 /** 2749 * __ice_q_vector_set_napi_queues - Map queue[s] associated with the napi 2750 * @q_vector: q_vector pointer 2751 * @locked: is the rtnl_lock already held 2752 * 2753 * Associate the q_vector napi with all the queue[s] on the vector. 2754 * Caller indicates the lock status. 2755 */ 2756 void __ice_q_vector_set_napi_queues(struct ice_q_vector *q_vector, bool locked) 2757 { 2758 struct ice_rx_ring *rx_ring; 2759 struct ice_tx_ring *tx_ring; 2760 2761 ice_for_each_rx_ring(rx_ring, q_vector->rx) 2762 __ice_queue_set_napi(q_vector->vsi->netdev, rx_ring->q_index, 2763 NETDEV_QUEUE_TYPE_RX, &q_vector->napi, 2764 locked); 2765 2766 ice_for_each_tx_ring(tx_ring, q_vector->tx) 2767 __ice_queue_set_napi(q_vector->vsi->netdev, tx_ring->q_index, 2768 NETDEV_QUEUE_TYPE_TX, &q_vector->napi, 2769 locked); 2770 /* Also set the interrupt number for the NAPI */ 2771 netif_napi_set_irq(&q_vector->napi, q_vector->irq.virq); 2772 } 2773 2774 /** 2775 * ice_q_vector_set_napi_queues - Map queue[s] associated with the napi 2776 * @q_vector: q_vector pointer 2777 * 2778 * Associate the q_vector napi with all the queue[s] on the vector 2779 */ 2780 void ice_q_vector_set_napi_queues(struct ice_q_vector *q_vector) 2781 { 2782 struct ice_rx_ring *rx_ring; 2783 struct ice_tx_ring *tx_ring; 2784 2785 ice_for_each_rx_ring(rx_ring, q_vector->rx) 2786 ice_queue_set_napi(q_vector->vsi, rx_ring->q_index, 2787 NETDEV_QUEUE_TYPE_RX, &q_vector->napi); 2788 2789 ice_for_each_tx_ring(tx_ring, q_vector->tx) 2790 ice_queue_set_napi(q_vector->vsi, tx_ring->q_index, 2791 NETDEV_QUEUE_TYPE_TX, &q_vector->napi); 2792 /* Also set the interrupt number for the NAPI */ 2793 netif_napi_set_irq(&q_vector->napi, q_vector->irq.virq); 2794 } 2795 2796 /** 2797 * ice_vsi_set_napi_queues 2798 * @vsi: VSI pointer 2799 * 2800 * Associate queue[s] with napi for all vectors 2801 */ 2802 void ice_vsi_set_napi_queues(struct ice_vsi *vsi) 2803 { 2804 int i; 2805 2806 if (!vsi->netdev) 2807 return; 2808 2809 ice_for_each_q_vector(vsi, i) 2810 ice_q_vector_set_napi_queues(vsi->q_vectors[i]); 2811 } 2812 2813 /** 2814 * ice_vsi_release - Delete a VSI and free its resources 2815 * @vsi: the VSI being removed 2816 * 2817 * Returns 0 on success or < 0 on error 2818 */ 2819 int ice_vsi_release(struct ice_vsi *vsi) 2820 { 2821 struct ice_pf *pf; 2822 2823 if (!vsi->back) 2824 return -ENODEV; 2825 pf = vsi->back; 2826 2827 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) 2828 ice_rss_clean(vsi); 2829 2830 ice_vsi_close(vsi); 2831 ice_vsi_decfg(vsi); 2832 2833 /* retain SW VSI data structure since it is needed to unregister and 2834 * free VSI netdev when PF is not in reset recovery pending state,\ 2835 * for ex: during rmmod. 2836 */ 2837 if (!ice_is_reset_in_progress(pf->state)) 2838 ice_vsi_delete(vsi); 2839 2840 return 0; 2841 } 2842 2843 /** 2844 * ice_vsi_rebuild_get_coalesce - get coalesce from all q_vectors 2845 * @vsi: VSI connected with q_vectors 2846 * @coalesce: array of struct with stored coalesce 2847 * 2848 * Returns array size. 2849 */ 2850 static int 2851 ice_vsi_rebuild_get_coalesce(struct ice_vsi *vsi, 2852 struct ice_coalesce_stored *coalesce) 2853 { 2854 int i; 2855 2856 ice_for_each_q_vector(vsi, i) { 2857 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 2858 2859 coalesce[i].itr_tx = q_vector->tx.itr_settings; 2860 coalesce[i].itr_rx = q_vector->rx.itr_settings; 2861 coalesce[i].intrl = q_vector->intrl; 2862 2863 if (i < vsi->num_txq) 2864 coalesce[i].tx_valid = true; 2865 if (i < vsi->num_rxq) 2866 coalesce[i].rx_valid = true; 2867 } 2868 2869 return vsi->num_q_vectors; 2870 } 2871 2872 /** 2873 * ice_vsi_rebuild_set_coalesce - set coalesce from earlier saved arrays 2874 * @vsi: VSI connected with q_vectors 2875 * @coalesce: pointer to array of struct with stored coalesce 2876 * @size: size of coalesce array 2877 * 2878 * Before this function, ice_vsi_rebuild_get_coalesce should be called to save 2879 * ITR params in arrays. If size is 0 or coalesce wasn't stored set coalesce 2880 * to default value. 2881 */ 2882 static void 2883 ice_vsi_rebuild_set_coalesce(struct ice_vsi *vsi, 2884 struct ice_coalesce_stored *coalesce, int size) 2885 { 2886 struct ice_ring_container *rc; 2887 int i; 2888 2889 if ((size && !coalesce) || !vsi) 2890 return; 2891 2892 /* There are a couple of cases that have to be handled here: 2893 * 1. The case where the number of queue vectors stays the same, but 2894 * the number of Tx or Rx rings changes (the first for loop) 2895 * 2. The case where the number of queue vectors increased (the 2896 * second for loop) 2897 */ 2898 for (i = 0; i < size && i < vsi->num_q_vectors; i++) { 2899 /* There are 2 cases to handle here and they are the same for 2900 * both Tx and Rx: 2901 * if the entry was valid previously (coalesce[i].[tr]x_valid 2902 * and the loop variable is less than the number of rings 2903 * allocated, then write the previous values 2904 * 2905 * if the entry was not valid previously, but the number of 2906 * rings is less than are allocated (this means the number of 2907 * rings increased from previously), then write out the 2908 * values in the first element 2909 * 2910 * Also, always write the ITR, even if in ITR_IS_DYNAMIC 2911 * as there is no harm because the dynamic algorithm 2912 * will just overwrite. 2913 */ 2914 if (i < vsi->alloc_rxq && coalesce[i].rx_valid) { 2915 rc = &vsi->q_vectors[i]->rx; 2916 rc->itr_settings = coalesce[i].itr_rx; 2917 ice_write_itr(rc, rc->itr_setting); 2918 } else if (i < vsi->alloc_rxq) { 2919 rc = &vsi->q_vectors[i]->rx; 2920 rc->itr_settings = coalesce[0].itr_rx; 2921 ice_write_itr(rc, rc->itr_setting); 2922 } 2923 2924 if (i < vsi->alloc_txq && coalesce[i].tx_valid) { 2925 rc = &vsi->q_vectors[i]->tx; 2926 rc->itr_settings = coalesce[i].itr_tx; 2927 ice_write_itr(rc, rc->itr_setting); 2928 } else if (i < vsi->alloc_txq) { 2929 rc = &vsi->q_vectors[i]->tx; 2930 rc->itr_settings = coalesce[0].itr_tx; 2931 ice_write_itr(rc, rc->itr_setting); 2932 } 2933 2934 vsi->q_vectors[i]->intrl = coalesce[i].intrl; 2935 ice_set_q_vector_intrl(vsi->q_vectors[i]); 2936 } 2937 2938 /* the number of queue vectors increased so write whatever is in 2939 * the first element 2940 */ 2941 for (; i < vsi->num_q_vectors; i++) { 2942 /* transmit */ 2943 rc = &vsi->q_vectors[i]->tx; 2944 rc->itr_settings = coalesce[0].itr_tx; 2945 ice_write_itr(rc, rc->itr_setting); 2946 2947 /* receive */ 2948 rc = &vsi->q_vectors[i]->rx; 2949 rc->itr_settings = coalesce[0].itr_rx; 2950 ice_write_itr(rc, rc->itr_setting); 2951 2952 vsi->q_vectors[i]->intrl = coalesce[0].intrl; 2953 ice_set_q_vector_intrl(vsi->q_vectors[i]); 2954 } 2955 } 2956 2957 /** 2958 * ice_vsi_realloc_stat_arrays - Frees unused stat structures or alloc new ones 2959 * @vsi: VSI pointer 2960 */ 2961 static int 2962 ice_vsi_realloc_stat_arrays(struct ice_vsi *vsi) 2963 { 2964 u16 req_txq = vsi->req_txq ? vsi->req_txq : vsi->alloc_txq; 2965 u16 req_rxq = vsi->req_rxq ? vsi->req_rxq : vsi->alloc_rxq; 2966 struct ice_ring_stats **tx_ring_stats; 2967 struct ice_ring_stats **rx_ring_stats; 2968 struct ice_vsi_stats *vsi_stat; 2969 struct ice_pf *pf = vsi->back; 2970 u16 prev_txq = vsi->alloc_txq; 2971 u16 prev_rxq = vsi->alloc_rxq; 2972 int i; 2973 2974 vsi_stat = pf->vsi_stats[vsi->idx]; 2975 2976 if (req_txq < prev_txq) { 2977 for (i = req_txq; i < prev_txq; i++) { 2978 if (vsi_stat->tx_ring_stats[i]) { 2979 kfree_rcu(vsi_stat->tx_ring_stats[i], rcu); 2980 WRITE_ONCE(vsi_stat->tx_ring_stats[i], NULL); 2981 } 2982 } 2983 } 2984 2985 tx_ring_stats = vsi_stat->tx_ring_stats; 2986 vsi_stat->tx_ring_stats = 2987 krealloc_array(vsi_stat->tx_ring_stats, req_txq, 2988 sizeof(*vsi_stat->tx_ring_stats), 2989 GFP_KERNEL | __GFP_ZERO); 2990 if (!vsi_stat->tx_ring_stats) { 2991 vsi_stat->tx_ring_stats = tx_ring_stats; 2992 return -ENOMEM; 2993 } 2994 2995 if (req_rxq < prev_rxq) { 2996 for (i = req_rxq; i < prev_rxq; i++) { 2997 if (vsi_stat->rx_ring_stats[i]) { 2998 kfree_rcu(vsi_stat->rx_ring_stats[i], rcu); 2999 WRITE_ONCE(vsi_stat->rx_ring_stats[i], NULL); 3000 } 3001 } 3002 } 3003 3004 rx_ring_stats = vsi_stat->rx_ring_stats; 3005 vsi_stat->rx_ring_stats = 3006 krealloc_array(vsi_stat->rx_ring_stats, req_rxq, 3007 sizeof(*vsi_stat->rx_ring_stats), 3008 GFP_KERNEL | __GFP_ZERO); 3009 if (!vsi_stat->rx_ring_stats) { 3010 vsi_stat->rx_ring_stats = rx_ring_stats; 3011 return -ENOMEM; 3012 } 3013 3014 return 0; 3015 } 3016 3017 /** 3018 * ice_vsi_rebuild - Rebuild VSI after reset 3019 * @vsi: VSI to be rebuild 3020 * @vsi_flags: flags used for VSI rebuild flow 3021 * 3022 * Set vsi_flags to ICE_VSI_FLAG_INIT to initialize a new VSI, or 3023 * ICE_VSI_FLAG_NO_INIT to rebuild an existing VSI in hardware. 3024 * 3025 * Returns 0 on success and negative value on failure 3026 */ 3027 int ice_vsi_rebuild(struct ice_vsi *vsi, u32 vsi_flags) 3028 { 3029 struct ice_coalesce_stored *coalesce; 3030 int prev_num_q_vectors; 3031 struct ice_pf *pf; 3032 int ret; 3033 3034 if (!vsi) 3035 return -EINVAL; 3036 3037 vsi->flags = vsi_flags; 3038 pf = vsi->back; 3039 if (WARN_ON(vsi->type == ICE_VSI_VF && !vsi->vf)) 3040 return -EINVAL; 3041 3042 ret = ice_vsi_realloc_stat_arrays(vsi); 3043 if (ret) 3044 goto err_vsi_cfg; 3045 3046 ice_vsi_decfg(vsi); 3047 ret = ice_vsi_cfg_def(vsi); 3048 if (ret) 3049 goto err_vsi_cfg; 3050 3051 coalesce = kcalloc(vsi->num_q_vectors, 3052 sizeof(struct ice_coalesce_stored), GFP_KERNEL); 3053 if (!coalesce) 3054 return -ENOMEM; 3055 3056 prev_num_q_vectors = ice_vsi_rebuild_get_coalesce(vsi, coalesce); 3057 3058 ret = ice_vsi_cfg_tc_lan(pf, vsi); 3059 if (ret) { 3060 if (vsi_flags & ICE_VSI_FLAG_INIT) { 3061 ret = -EIO; 3062 goto err_vsi_cfg_tc_lan; 3063 } 3064 3065 kfree(coalesce); 3066 return ice_schedule_reset(pf, ICE_RESET_PFR); 3067 } 3068 3069 ice_vsi_rebuild_set_coalesce(vsi, coalesce, prev_num_q_vectors); 3070 kfree(coalesce); 3071 3072 return 0; 3073 3074 err_vsi_cfg_tc_lan: 3075 ice_vsi_decfg(vsi); 3076 kfree(coalesce); 3077 err_vsi_cfg: 3078 return ret; 3079 } 3080 3081 /** 3082 * ice_is_reset_in_progress - check for a reset in progress 3083 * @state: PF state field 3084 */ 3085 bool ice_is_reset_in_progress(unsigned long *state) 3086 { 3087 return test_bit(ICE_RESET_OICR_RECV, state) || 3088 test_bit(ICE_PFR_REQ, state) || 3089 test_bit(ICE_CORER_REQ, state) || 3090 test_bit(ICE_GLOBR_REQ, state); 3091 } 3092 3093 /** 3094 * ice_wait_for_reset - Wait for driver to finish reset and rebuild 3095 * @pf: pointer to the PF structure 3096 * @timeout: length of time to wait, in jiffies 3097 * 3098 * Wait (sleep) for a short time until the driver finishes cleaning up from 3099 * a device reset. The caller must be able to sleep. Use this to delay 3100 * operations that could fail while the driver is cleaning up after a device 3101 * reset. 3102 * 3103 * Returns 0 on success, -EBUSY if the reset is not finished within the 3104 * timeout, and -ERESTARTSYS if the thread was interrupted. 3105 */ 3106 int ice_wait_for_reset(struct ice_pf *pf, unsigned long timeout) 3107 { 3108 long ret; 3109 3110 ret = wait_event_interruptible_timeout(pf->reset_wait_queue, 3111 !ice_is_reset_in_progress(pf->state), 3112 timeout); 3113 if (ret < 0) 3114 return ret; 3115 else if (!ret) 3116 return -EBUSY; 3117 else 3118 return 0; 3119 } 3120 3121 /** 3122 * ice_vsi_update_q_map - update our copy of the VSI info with new queue map 3123 * @vsi: VSI being configured 3124 * @ctx: the context buffer returned from AQ VSI update command 3125 */ 3126 static void ice_vsi_update_q_map(struct ice_vsi *vsi, struct ice_vsi_ctx *ctx) 3127 { 3128 vsi->info.mapping_flags = ctx->info.mapping_flags; 3129 memcpy(&vsi->info.q_mapping, &ctx->info.q_mapping, 3130 sizeof(vsi->info.q_mapping)); 3131 memcpy(&vsi->info.tc_mapping, ctx->info.tc_mapping, 3132 sizeof(vsi->info.tc_mapping)); 3133 } 3134 3135 /** 3136 * ice_vsi_cfg_netdev_tc - Setup the netdev TC configuration 3137 * @vsi: the VSI being configured 3138 * @ena_tc: TC map to be enabled 3139 */ 3140 void ice_vsi_cfg_netdev_tc(struct ice_vsi *vsi, u8 ena_tc) 3141 { 3142 struct net_device *netdev = vsi->netdev; 3143 struct ice_pf *pf = vsi->back; 3144 int numtc = vsi->tc_cfg.numtc; 3145 struct ice_dcbx_cfg *dcbcfg; 3146 u8 netdev_tc; 3147 int i; 3148 3149 if (!netdev) 3150 return; 3151 3152 /* CHNL VSI doesn't have it's own netdev, hence, no netdev_tc */ 3153 if (vsi->type == ICE_VSI_CHNL) 3154 return; 3155 3156 if (!ena_tc) { 3157 netdev_reset_tc(netdev); 3158 return; 3159 } 3160 3161 if (vsi->type == ICE_VSI_PF && ice_is_adq_active(pf)) 3162 numtc = vsi->all_numtc; 3163 3164 if (netdev_set_num_tc(netdev, numtc)) 3165 return; 3166 3167 dcbcfg = &pf->hw.port_info->qos_cfg.local_dcbx_cfg; 3168 3169 ice_for_each_traffic_class(i) 3170 if (vsi->tc_cfg.ena_tc & BIT(i)) 3171 netdev_set_tc_queue(netdev, 3172 vsi->tc_cfg.tc_info[i].netdev_tc, 3173 vsi->tc_cfg.tc_info[i].qcount_tx, 3174 vsi->tc_cfg.tc_info[i].qoffset); 3175 /* setup TC queue map for CHNL TCs */ 3176 ice_for_each_chnl_tc(i) { 3177 if (!(vsi->all_enatc & BIT(i))) 3178 break; 3179 if (!vsi->mqprio_qopt.qopt.count[i]) 3180 break; 3181 netdev_set_tc_queue(netdev, i, 3182 vsi->mqprio_qopt.qopt.count[i], 3183 vsi->mqprio_qopt.qopt.offset[i]); 3184 } 3185 3186 if (test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3187 return; 3188 3189 for (i = 0; i < ICE_MAX_USER_PRIORITY; i++) { 3190 u8 ets_tc = dcbcfg->etscfg.prio_table[i]; 3191 3192 /* Get the mapped netdev TC# for the UP */ 3193 netdev_tc = vsi->tc_cfg.tc_info[ets_tc].netdev_tc; 3194 netdev_set_prio_tc_map(netdev, i, netdev_tc); 3195 } 3196 } 3197 3198 /** 3199 * ice_vsi_setup_q_map_mqprio - Prepares mqprio based tc_config 3200 * @vsi: the VSI being configured, 3201 * @ctxt: VSI context structure 3202 * @ena_tc: number of traffic classes to enable 3203 * 3204 * Prepares VSI tc_config to have queue configurations based on MQPRIO options. 3205 */ 3206 static int 3207 ice_vsi_setup_q_map_mqprio(struct ice_vsi *vsi, struct ice_vsi_ctx *ctxt, 3208 u8 ena_tc) 3209 { 3210 u16 pow, offset = 0, qcount_tx = 0, qcount_rx = 0, qmap; 3211 u16 tc0_offset = vsi->mqprio_qopt.qopt.offset[0]; 3212 int tc0_qcount = vsi->mqprio_qopt.qopt.count[0]; 3213 u16 new_txq, new_rxq; 3214 u8 netdev_tc = 0; 3215 int i; 3216 3217 vsi->tc_cfg.ena_tc = ena_tc ? ena_tc : 1; 3218 3219 pow = order_base_2(tc0_qcount); 3220 qmap = FIELD_PREP(ICE_AQ_VSI_TC_Q_OFFSET_M, tc0_offset); 3221 qmap |= FIELD_PREP(ICE_AQ_VSI_TC_Q_NUM_M, pow); 3222 3223 ice_for_each_traffic_class(i) { 3224 if (!(vsi->tc_cfg.ena_tc & BIT(i))) { 3225 /* TC is not enabled */ 3226 vsi->tc_cfg.tc_info[i].qoffset = 0; 3227 vsi->tc_cfg.tc_info[i].qcount_rx = 1; 3228 vsi->tc_cfg.tc_info[i].qcount_tx = 1; 3229 vsi->tc_cfg.tc_info[i].netdev_tc = 0; 3230 ctxt->info.tc_mapping[i] = 0; 3231 continue; 3232 } 3233 3234 offset = vsi->mqprio_qopt.qopt.offset[i]; 3235 qcount_rx = vsi->mqprio_qopt.qopt.count[i]; 3236 qcount_tx = vsi->mqprio_qopt.qopt.count[i]; 3237 vsi->tc_cfg.tc_info[i].qoffset = offset; 3238 vsi->tc_cfg.tc_info[i].qcount_rx = qcount_rx; 3239 vsi->tc_cfg.tc_info[i].qcount_tx = qcount_tx; 3240 vsi->tc_cfg.tc_info[i].netdev_tc = netdev_tc++; 3241 } 3242 3243 if (vsi->all_numtc && vsi->all_numtc != vsi->tc_cfg.numtc) { 3244 ice_for_each_chnl_tc(i) { 3245 if (!(vsi->all_enatc & BIT(i))) 3246 continue; 3247 offset = vsi->mqprio_qopt.qopt.offset[i]; 3248 qcount_rx = vsi->mqprio_qopt.qopt.count[i]; 3249 qcount_tx = vsi->mqprio_qopt.qopt.count[i]; 3250 } 3251 } 3252 3253 new_txq = offset + qcount_tx; 3254 if (new_txq > vsi->alloc_txq) { 3255 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Tx queues (%u), than were allocated (%u)!\n", 3256 new_txq, vsi->alloc_txq); 3257 return -EINVAL; 3258 } 3259 3260 new_rxq = offset + qcount_rx; 3261 if (new_rxq > vsi->alloc_rxq) { 3262 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Rx queues (%u), than were allocated (%u)!\n", 3263 new_rxq, vsi->alloc_rxq); 3264 return -EINVAL; 3265 } 3266 3267 /* Set actual Tx/Rx queue pairs */ 3268 vsi->num_txq = new_txq; 3269 vsi->num_rxq = new_rxq; 3270 3271 /* Setup queue TC[0].qmap for given VSI context */ 3272 ctxt->info.tc_mapping[0] = cpu_to_le16(qmap); 3273 ctxt->info.q_mapping[0] = cpu_to_le16(vsi->rxq_map[0]); 3274 ctxt->info.q_mapping[1] = cpu_to_le16(tc0_qcount); 3275 3276 /* Find queue count available for channel VSIs and starting offset 3277 * for channel VSIs 3278 */ 3279 if (tc0_qcount && tc0_qcount < vsi->num_rxq) { 3280 vsi->cnt_q_avail = vsi->num_rxq - tc0_qcount; 3281 vsi->next_base_q = tc0_qcount; 3282 } 3283 dev_dbg(ice_pf_to_dev(vsi->back), "vsi->num_txq = %d\n", vsi->num_txq); 3284 dev_dbg(ice_pf_to_dev(vsi->back), "vsi->num_rxq = %d\n", vsi->num_rxq); 3285 dev_dbg(ice_pf_to_dev(vsi->back), "all_numtc %u, all_enatc: 0x%04x, tc_cfg.numtc %u\n", 3286 vsi->all_numtc, vsi->all_enatc, vsi->tc_cfg.numtc); 3287 3288 return 0; 3289 } 3290 3291 /** 3292 * ice_vsi_cfg_tc - Configure VSI Tx Sched for given TC map 3293 * @vsi: VSI to be configured 3294 * @ena_tc: TC bitmap 3295 * 3296 * VSI queues expected to be quiesced before calling this function 3297 */ 3298 int ice_vsi_cfg_tc(struct ice_vsi *vsi, u8 ena_tc) 3299 { 3300 u16 max_txqs[ICE_MAX_TRAFFIC_CLASS] = { 0 }; 3301 struct ice_pf *pf = vsi->back; 3302 struct ice_tc_cfg old_tc_cfg; 3303 struct ice_vsi_ctx *ctx; 3304 struct device *dev; 3305 int i, ret = 0; 3306 u8 num_tc = 0; 3307 3308 dev = ice_pf_to_dev(pf); 3309 if (vsi->tc_cfg.ena_tc == ena_tc && 3310 vsi->mqprio_qopt.mode != TC_MQPRIO_MODE_CHANNEL) 3311 return 0; 3312 3313 ice_for_each_traffic_class(i) { 3314 /* build bitmap of enabled TCs */ 3315 if (ena_tc & BIT(i)) 3316 num_tc++; 3317 /* populate max_txqs per TC */ 3318 max_txqs[i] = vsi->alloc_txq; 3319 /* Update max_txqs if it is CHNL VSI, because alloc_t[r]xq are 3320 * zero for CHNL VSI, hence use num_txq instead as max_txqs 3321 */ 3322 if (vsi->type == ICE_VSI_CHNL && 3323 test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3324 max_txqs[i] = vsi->num_txq; 3325 } 3326 3327 memcpy(&old_tc_cfg, &vsi->tc_cfg, sizeof(old_tc_cfg)); 3328 vsi->tc_cfg.ena_tc = ena_tc; 3329 vsi->tc_cfg.numtc = num_tc; 3330 3331 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); 3332 if (!ctx) 3333 return -ENOMEM; 3334 3335 ctx->vf_num = 0; 3336 ctx->info = vsi->info; 3337 3338 if (vsi->type == ICE_VSI_PF && 3339 test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3340 ret = ice_vsi_setup_q_map_mqprio(vsi, ctx, ena_tc); 3341 else 3342 ret = ice_vsi_setup_q_map(vsi, ctx); 3343 3344 if (ret) { 3345 memcpy(&vsi->tc_cfg, &old_tc_cfg, sizeof(vsi->tc_cfg)); 3346 goto out; 3347 } 3348 3349 /* must to indicate which section of VSI context are being modified */ 3350 ctx->info.valid_sections = cpu_to_le16(ICE_AQ_VSI_PROP_RXQ_MAP_VALID); 3351 ret = ice_update_vsi(&pf->hw, vsi->idx, ctx, NULL); 3352 if (ret) { 3353 dev_info(dev, "Failed VSI Update\n"); 3354 goto out; 3355 } 3356 3357 if (vsi->type == ICE_VSI_PF && 3358 test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3359 ret = ice_cfg_vsi_lan(vsi->port_info, vsi->idx, 1, max_txqs); 3360 else 3361 ret = ice_cfg_vsi_lan(vsi->port_info, vsi->idx, 3362 vsi->tc_cfg.ena_tc, max_txqs); 3363 3364 if (ret) { 3365 dev_err(dev, "VSI %d failed TC config, error %d\n", 3366 vsi->vsi_num, ret); 3367 goto out; 3368 } 3369 ice_vsi_update_q_map(vsi, ctx); 3370 vsi->info.valid_sections = 0; 3371 3372 ice_vsi_cfg_netdev_tc(vsi, ena_tc); 3373 out: 3374 kfree(ctx); 3375 return ret; 3376 } 3377 3378 /** 3379 * ice_update_ring_stats - Update ring statistics 3380 * @stats: stats to be updated 3381 * @pkts: number of processed packets 3382 * @bytes: number of processed bytes 3383 * 3384 * This function assumes that caller has acquired a u64_stats_sync lock. 3385 */ 3386 static void ice_update_ring_stats(struct ice_q_stats *stats, u64 pkts, u64 bytes) 3387 { 3388 stats->bytes += bytes; 3389 stats->pkts += pkts; 3390 } 3391 3392 /** 3393 * ice_update_tx_ring_stats - Update Tx ring specific counters 3394 * @tx_ring: ring to update 3395 * @pkts: number of processed packets 3396 * @bytes: number of processed bytes 3397 */ 3398 void ice_update_tx_ring_stats(struct ice_tx_ring *tx_ring, u64 pkts, u64 bytes) 3399 { 3400 u64_stats_update_begin(&tx_ring->ring_stats->syncp); 3401 ice_update_ring_stats(&tx_ring->ring_stats->stats, pkts, bytes); 3402 u64_stats_update_end(&tx_ring->ring_stats->syncp); 3403 } 3404 3405 /** 3406 * ice_update_rx_ring_stats - Update Rx ring specific counters 3407 * @rx_ring: ring to update 3408 * @pkts: number of processed packets 3409 * @bytes: number of processed bytes 3410 */ 3411 void ice_update_rx_ring_stats(struct ice_rx_ring *rx_ring, u64 pkts, u64 bytes) 3412 { 3413 u64_stats_update_begin(&rx_ring->ring_stats->syncp); 3414 ice_update_ring_stats(&rx_ring->ring_stats->stats, pkts, bytes); 3415 u64_stats_update_end(&rx_ring->ring_stats->syncp); 3416 } 3417 3418 /** 3419 * ice_is_dflt_vsi_in_use - check if the default forwarding VSI is being used 3420 * @pi: port info of the switch with default VSI 3421 * 3422 * Return true if the there is a single VSI in default forwarding VSI list 3423 */ 3424 bool ice_is_dflt_vsi_in_use(struct ice_port_info *pi) 3425 { 3426 bool exists = false; 3427 3428 ice_check_if_dflt_vsi(pi, 0, &exists); 3429 return exists; 3430 } 3431 3432 /** 3433 * ice_is_vsi_dflt_vsi - check if the VSI passed in is the default VSI 3434 * @vsi: VSI to compare against default forwarding VSI 3435 * 3436 * If this VSI passed in is the default forwarding VSI then return true, else 3437 * return false 3438 */ 3439 bool ice_is_vsi_dflt_vsi(struct ice_vsi *vsi) 3440 { 3441 return ice_check_if_dflt_vsi(vsi->port_info, vsi->idx, NULL); 3442 } 3443 3444 /** 3445 * ice_set_dflt_vsi - set the default forwarding VSI 3446 * @vsi: VSI getting set as the default forwarding VSI on the switch 3447 * 3448 * If the VSI passed in is already the default VSI and it's enabled just return 3449 * success. 3450 * 3451 * Otherwise try to set the VSI passed in as the switch's default VSI and 3452 * return the result. 3453 */ 3454 int ice_set_dflt_vsi(struct ice_vsi *vsi) 3455 { 3456 struct device *dev; 3457 int status; 3458 3459 if (!vsi) 3460 return -EINVAL; 3461 3462 dev = ice_pf_to_dev(vsi->back); 3463 3464 if (ice_lag_is_switchdev_running(vsi->back)) { 3465 dev_dbg(dev, "VSI %d passed is a part of LAG containing interfaces in switchdev mode, nothing to do\n", 3466 vsi->vsi_num); 3467 return 0; 3468 } 3469 3470 /* the VSI passed in is already the default VSI */ 3471 if (ice_is_vsi_dflt_vsi(vsi)) { 3472 dev_dbg(dev, "VSI %d passed in is already the default forwarding VSI, nothing to do\n", 3473 vsi->vsi_num); 3474 return 0; 3475 } 3476 3477 status = ice_cfg_dflt_vsi(vsi->port_info, vsi->idx, true, ICE_FLTR_RX); 3478 if (status) { 3479 dev_err(dev, "Failed to set VSI %d as the default forwarding VSI, error %d\n", 3480 vsi->vsi_num, status); 3481 return status; 3482 } 3483 3484 return 0; 3485 } 3486 3487 /** 3488 * ice_clear_dflt_vsi - clear the default forwarding VSI 3489 * @vsi: VSI to remove from filter list 3490 * 3491 * If the switch has no default VSI or it's not enabled then return error. 3492 * 3493 * Otherwise try to clear the default VSI and return the result. 3494 */ 3495 int ice_clear_dflt_vsi(struct ice_vsi *vsi) 3496 { 3497 struct device *dev; 3498 int status; 3499 3500 if (!vsi) 3501 return -EINVAL; 3502 3503 dev = ice_pf_to_dev(vsi->back); 3504 3505 /* there is no default VSI configured */ 3506 if (!ice_is_dflt_vsi_in_use(vsi->port_info)) 3507 return -ENODEV; 3508 3509 status = ice_cfg_dflt_vsi(vsi->port_info, vsi->idx, false, 3510 ICE_FLTR_RX); 3511 if (status) { 3512 dev_err(dev, "Failed to clear the default forwarding VSI %d, error %d\n", 3513 vsi->vsi_num, status); 3514 return -EIO; 3515 } 3516 3517 return 0; 3518 } 3519 3520 /** 3521 * ice_get_link_speed_mbps - get link speed in Mbps 3522 * @vsi: the VSI whose link speed is being queried 3523 * 3524 * Return current VSI link speed and 0 if the speed is unknown. 3525 */ 3526 int ice_get_link_speed_mbps(struct ice_vsi *vsi) 3527 { 3528 unsigned int link_speed; 3529 3530 link_speed = vsi->port_info->phy.link_info.link_speed; 3531 3532 return (int)ice_get_link_speed(fls(link_speed) - 1); 3533 } 3534 3535 /** 3536 * ice_get_link_speed_kbps - get link speed in Kbps 3537 * @vsi: the VSI whose link speed is being queried 3538 * 3539 * Return current VSI link speed and 0 if the speed is unknown. 3540 */ 3541 int ice_get_link_speed_kbps(struct ice_vsi *vsi) 3542 { 3543 int speed_mbps; 3544 3545 speed_mbps = ice_get_link_speed_mbps(vsi); 3546 3547 return speed_mbps * 1000; 3548 } 3549 3550 /** 3551 * ice_set_min_bw_limit - setup minimum BW limit for Tx based on min_tx_rate 3552 * @vsi: VSI to be configured 3553 * @min_tx_rate: min Tx rate in Kbps to be configured as BW limit 3554 * 3555 * If the min_tx_rate is specified as 0 that means to clear the minimum BW limit 3556 * profile, otherwise a non-zero value will force a minimum BW limit for the VSI 3557 * on TC 0. 3558 */ 3559 int ice_set_min_bw_limit(struct ice_vsi *vsi, u64 min_tx_rate) 3560 { 3561 struct ice_pf *pf = vsi->back; 3562 struct device *dev; 3563 int status; 3564 int speed; 3565 3566 dev = ice_pf_to_dev(pf); 3567 if (!vsi->port_info) { 3568 dev_dbg(dev, "VSI %d, type %u specified doesn't have valid port_info\n", 3569 vsi->idx, vsi->type); 3570 return -EINVAL; 3571 } 3572 3573 speed = ice_get_link_speed_kbps(vsi); 3574 if (min_tx_rate > (u64)speed) { 3575 dev_err(dev, "invalid min Tx rate %llu Kbps specified for %s %d is greater than current link speed %u Kbps\n", 3576 min_tx_rate, ice_vsi_type_str(vsi->type), vsi->idx, 3577 speed); 3578 return -EINVAL; 3579 } 3580 3581 /* Configure min BW for VSI limit */ 3582 if (min_tx_rate) { 3583 status = ice_cfg_vsi_bw_lmt_per_tc(vsi->port_info, vsi->idx, 0, 3584 ICE_MIN_BW, min_tx_rate); 3585 if (status) { 3586 dev_err(dev, "failed to set min Tx rate(%llu Kbps) for %s %d\n", 3587 min_tx_rate, ice_vsi_type_str(vsi->type), 3588 vsi->idx); 3589 return status; 3590 } 3591 3592 dev_dbg(dev, "set min Tx rate(%llu Kbps) for %s\n", 3593 min_tx_rate, ice_vsi_type_str(vsi->type)); 3594 } else { 3595 status = ice_cfg_vsi_bw_dflt_lmt_per_tc(vsi->port_info, 3596 vsi->idx, 0, 3597 ICE_MIN_BW); 3598 if (status) { 3599 dev_err(dev, "failed to clear min Tx rate configuration for %s %d\n", 3600 ice_vsi_type_str(vsi->type), vsi->idx); 3601 return status; 3602 } 3603 3604 dev_dbg(dev, "cleared min Tx rate configuration for %s %d\n", 3605 ice_vsi_type_str(vsi->type), vsi->idx); 3606 } 3607 3608 return 0; 3609 } 3610 3611 /** 3612 * ice_set_max_bw_limit - setup maximum BW limit for Tx based on max_tx_rate 3613 * @vsi: VSI to be configured 3614 * @max_tx_rate: max Tx rate in Kbps to be configured as BW limit 3615 * 3616 * If the max_tx_rate is specified as 0 that means to clear the maximum BW limit 3617 * profile, otherwise a non-zero value will force a maximum BW limit for the VSI 3618 * on TC 0. 3619 */ 3620 int ice_set_max_bw_limit(struct ice_vsi *vsi, u64 max_tx_rate) 3621 { 3622 struct ice_pf *pf = vsi->back; 3623 struct device *dev; 3624 int status; 3625 int speed; 3626 3627 dev = ice_pf_to_dev(pf); 3628 if (!vsi->port_info) { 3629 dev_dbg(dev, "VSI %d, type %u specified doesn't have valid port_info\n", 3630 vsi->idx, vsi->type); 3631 return -EINVAL; 3632 } 3633 3634 speed = ice_get_link_speed_kbps(vsi); 3635 if (max_tx_rate > (u64)speed) { 3636 dev_err(dev, "invalid max Tx rate %llu Kbps specified for %s %d is greater than current link speed %u Kbps\n", 3637 max_tx_rate, ice_vsi_type_str(vsi->type), vsi->idx, 3638 speed); 3639 return -EINVAL; 3640 } 3641 3642 /* Configure max BW for VSI limit */ 3643 if (max_tx_rate) { 3644 status = ice_cfg_vsi_bw_lmt_per_tc(vsi->port_info, vsi->idx, 0, 3645 ICE_MAX_BW, max_tx_rate); 3646 if (status) { 3647 dev_err(dev, "failed setting max Tx rate(%llu Kbps) for %s %d\n", 3648 max_tx_rate, ice_vsi_type_str(vsi->type), 3649 vsi->idx); 3650 return status; 3651 } 3652 3653 dev_dbg(dev, "set max Tx rate(%llu Kbps) for %s %d\n", 3654 max_tx_rate, ice_vsi_type_str(vsi->type), vsi->idx); 3655 } else { 3656 status = ice_cfg_vsi_bw_dflt_lmt_per_tc(vsi->port_info, 3657 vsi->idx, 0, 3658 ICE_MAX_BW); 3659 if (status) { 3660 dev_err(dev, "failed clearing max Tx rate configuration for %s %d\n", 3661 ice_vsi_type_str(vsi->type), vsi->idx); 3662 return status; 3663 } 3664 3665 dev_dbg(dev, "cleared max Tx rate configuration for %s %d\n", 3666 ice_vsi_type_str(vsi->type), vsi->idx); 3667 } 3668 3669 return 0; 3670 } 3671 3672 /** 3673 * ice_set_link - turn on/off physical link 3674 * @vsi: VSI to modify physical link on 3675 * @ena: turn on/off physical link 3676 */ 3677 int ice_set_link(struct ice_vsi *vsi, bool ena) 3678 { 3679 struct device *dev = ice_pf_to_dev(vsi->back); 3680 struct ice_port_info *pi = vsi->port_info; 3681 struct ice_hw *hw = pi->hw; 3682 int status; 3683 3684 if (vsi->type != ICE_VSI_PF) 3685 return -EINVAL; 3686 3687 status = ice_aq_set_link_restart_an(pi, ena, NULL); 3688 3689 /* if link is owned by manageability, FW will return ICE_AQ_RC_EMODE. 3690 * this is not a fatal error, so print a warning message and return 3691 * a success code. Return an error if FW returns an error code other 3692 * than ICE_AQ_RC_EMODE 3693 */ 3694 if (status == -EIO) { 3695 if (hw->adminq.sq_last_status == ICE_AQ_RC_EMODE) 3696 dev_dbg(dev, "can't set link to %s, err %d aq_err %s. not fatal, continuing\n", 3697 (ena ? "ON" : "OFF"), status, 3698 ice_aq_str(hw->adminq.sq_last_status)); 3699 } else if (status) { 3700 dev_err(dev, "can't set link to %s, err %d aq_err %s\n", 3701 (ena ? "ON" : "OFF"), status, 3702 ice_aq_str(hw->adminq.sq_last_status)); 3703 return status; 3704 } 3705 3706 return 0; 3707 } 3708 3709 /** 3710 * ice_vsi_add_vlan_zero - add VLAN 0 filter(s) for this VSI 3711 * @vsi: VSI used to add VLAN filters 3712 * 3713 * In Single VLAN Mode (SVM), single VLAN filters via ICE_SW_LKUP_VLAN are based 3714 * on the inner VLAN ID, so the VLAN TPID (i.e. 0x8100 or 0x888a8) doesn't 3715 * matter. In Double VLAN Mode (DVM), outer/single VLAN filters via 3716 * ICE_SW_LKUP_VLAN are based on the outer/single VLAN ID + VLAN TPID. 3717 * 3718 * For both modes add a VLAN 0 + no VLAN TPID filter to handle untagged traffic 3719 * when VLAN pruning is enabled. Also, this handles VLAN 0 priority tagged 3720 * traffic in SVM, since the VLAN TPID isn't part of filtering. 3721 * 3722 * If DVM is enabled then an explicit VLAN 0 + VLAN TPID filter needs to be 3723 * added to allow VLAN 0 priority tagged traffic in DVM, since the VLAN TPID is 3724 * part of filtering. 3725 */ 3726 int ice_vsi_add_vlan_zero(struct ice_vsi *vsi) 3727 { 3728 struct ice_vsi_vlan_ops *vlan_ops = ice_get_compat_vsi_vlan_ops(vsi); 3729 struct ice_vlan vlan; 3730 int err; 3731 3732 vlan = ICE_VLAN(0, 0, 0); 3733 err = vlan_ops->add_vlan(vsi, &vlan); 3734 if (err && err != -EEXIST) 3735 return err; 3736 3737 /* in SVM both VLAN 0 filters are identical */ 3738 if (!ice_is_dvm_ena(&vsi->back->hw)) 3739 return 0; 3740 3741 vlan = ICE_VLAN(ETH_P_8021Q, 0, 0); 3742 err = vlan_ops->add_vlan(vsi, &vlan); 3743 if (err && err != -EEXIST) 3744 return err; 3745 3746 return 0; 3747 } 3748 3749 /** 3750 * ice_vsi_del_vlan_zero - delete VLAN 0 filter(s) for this VSI 3751 * @vsi: VSI used to add VLAN filters 3752 * 3753 * Delete the VLAN 0 filters in the same manner that they were added in 3754 * ice_vsi_add_vlan_zero. 3755 */ 3756 int ice_vsi_del_vlan_zero(struct ice_vsi *vsi) 3757 { 3758 struct ice_vsi_vlan_ops *vlan_ops = ice_get_compat_vsi_vlan_ops(vsi); 3759 struct ice_vlan vlan; 3760 int err; 3761 3762 vlan = ICE_VLAN(0, 0, 0); 3763 err = vlan_ops->del_vlan(vsi, &vlan); 3764 if (err && err != -EEXIST) 3765 return err; 3766 3767 /* in SVM both VLAN 0 filters are identical */ 3768 if (!ice_is_dvm_ena(&vsi->back->hw)) 3769 return 0; 3770 3771 vlan = ICE_VLAN(ETH_P_8021Q, 0, 0); 3772 err = vlan_ops->del_vlan(vsi, &vlan); 3773 if (err && err != -EEXIST) 3774 return err; 3775 3776 /* when deleting the last VLAN filter, make sure to disable the VLAN 3777 * promisc mode so the filter isn't left by accident 3778 */ 3779 return ice_clear_vsi_promisc(&vsi->back->hw, vsi->idx, 3780 ICE_MCAST_VLAN_PROMISC_BITS, 0); 3781 } 3782 3783 /** 3784 * ice_vsi_num_zero_vlans - get number of VLAN 0 filters based on VLAN mode 3785 * @vsi: VSI used to get the VLAN mode 3786 * 3787 * If DVM is enabled then 2 VLAN 0 filters are added, else if SVM is enabled 3788 * then 1 VLAN 0 filter is added. See ice_vsi_add_vlan_zero for more details. 3789 */ 3790 static u16 ice_vsi_num_zero_vlans(struct ice_vsi *vsi) 3791 { 3792 #define ICE_DVM_NUM_ZERO_VLAN_FLTRS 2 3793 #define ICE_SVM_NUM_ZERO_VLAN_FLTRS 1 3794 /* no VLAN 0 filter is created when a port VLAN is active */ 3795 if (vsi->type == ICE_VSI_VF) { 3796 if (WARN_ON(!vsi->vf)) 3797 return 0; 3798 3799 if (ice_vf_is_port_vlan_ena(vsi->vf)) 3800 return 0; 3801 } 3802 3803 if (ice_is_dvm_ena(&vsi->back->hw)) 3804 return ICE_DVM_NUM_ZERO_VLAN_FLTRS; 3805 else 3806 return ICE_SVM_NUM_ZERO_VLAN_FLTRS; 3807 } 3808 3809 /** 3810 * ice_vsi_has_non_zero_vlans - check if VSI has any non-zero VLANs 3811 * @vsi: VSI used to determine if any non-zero VLANs have been added 3812 */ 3813 bool ice_vsi_has_non_zero_vlans(struct ice_vsi *vsi) 3814 { 3815 return (vsi->num_vlan > ice_vsi_num_zero_vlans(vsi)); 3816 } 3817 3818 /** 3819 * ice_vsi_num_non_zero_vlans - get the number of non-zero VLANs for this VSI 3820 * @vsi: VSI used to get the number of non-zero VLANs added 3821 */ 3822 u16 ice_vsi_num_non_zero_vlans(struct ice_vsi *vsi) 3823 { 3824 return (vsi->num_vlan - ice_vsi_num_zero_vlans(vsi)); 3825 } 3826 3827 /** 3828 * ice_is_feature_supported 3829 * @pf: pointer to the struct ice_pf instance 3830 * @f: feature enum to be checked 3831 * 3832 * returns true if feature is supported, false otherwise 3833 */ 3834 bool ice_is_feature_supported(struct ice_pf *pf, enum ice_feature f) 3835 { 3836 if (f < 0 || f >= ICE_F_MAX) 3837 return false; 3838 3839 return test_bit(f, pf->features); 3840 } 3841 3842 /** 3843 * ice_set_feature_support 3844 * @pf: pointer to the struct ice_pf instance 3845 * @f: feature enum to set 3846 */ 3847 void ice_set_feature_support(struct ice_pf *pf, enum ice_feature f) 3848 { 3849 if (f < 0 || f >= ICE_F_MAX) 3850 return; 3851 3852 set_bit(f, pf->features); 3853 } 3854 3855 /** 3856 * ice_clear_feature_support 3857 * @pf: pointer to the struct ice_pf instance 3858 * @f: feature enum to clear 3859 */ 3860 void ice_clear_feature_support(struct ice_pf *pf, enum ice_feature f) 3861 { 3862 if (f < 0 || f >= ICE_F_MAX) 3863 return; 3864 3865 clear_bit(f, pf->features); 3866 } 3867 3868 /** 3869 * ice_init_feature_support 3870 * @pf: pointer to the struct ice_pf instance 3871 * 3872 * called during init to setup supported feature 3873 */ 3874 void ice_init_feature_support(struct ice_pf *pf) 3875 { 3876 switch (pf->hw.device_id) { 3877 case ICE_DEV_ID_E810C_BACKPLANE: 3878 case ICE_DEV_ID_E810C_QSFP: 3879 case ICE_DEV_ID_E810C_SFP: 3880 case ICE_DEV_ID_E810_XXV_BACKPLANE: 3881 case ICE_DEV_ID_E810_XXV_QSFP: 3882 case ICE_DEV_ID_E810_XXV_SFP: 3883 ice_set_feature_support(pf, ICE_F_DSCP); 3884 if (ice_is_phy_rclk_in_netlist(&pf->hw)) 3885 ice_set_feature_support(pf, ICE_F_PHY_RCLK); 3886 /* If we don't own the timer - don't enable other caps */ 3887 if (!ice_pf_src_tmr_owned(pf)) 3888 break; 3889 if (ice_is_cgu_in_netlist(&pf->hw)) 3890 ice_set_feature_support(pf, ICE_F_CGU); 3891 if (ice_is_clock_mux_in_netlist(&pf->hw)) 3892 ice_set_feature_support(pf, ICE_F_SMA_CTRL); 3893 if (ice_gnss_is_gps_present(&pf->hw)) 3894 ice_set_feature_support(pf, ICE_F_GNSS); 3895 break; 3896 default: 3897 break; 3898 } 3899 } 3900 3901 /** 3902 * ice_vsi_update_security - update security block in VSI 3903 * @vsi: pointer to VSI structure 3904 * @fill: function pointer to fill ctx 3905 */ 3906 int 3907 ice_vsi_update_security(struct ice_vsi *vsi, void (*fill)(struct ice_vsi_ctx *)) 3908 { 3909 struct ice_vsi_ctx ctx = { 0 }; 3910 3911 ctx.info = vsi->info; 3912 ctx.info.valid_sections = cpu_to_le16(ICE_AQ_VSI_PROP_SECURITY_VALID); 3913 fill(&ctx); 3914 3915 if (ice_update_vsi(&vsi->back->hw, vsi->idx, &ctx, NULL)) 3916 return -ENODEV; 3917 3918 vsi->info = ctx.info; 3919 return 0; 3920 } 3921 3922 /** 3923 * ice_vsi_ctx_set_antispoof - set antispoof function in VSI ctx 3924 * @ctx: pointer to VSI ctx structure 3925 */ 3926 void ice_vsi_ctx_set_antispoof(struct ice_vsi_ctx *ctx) 3927 { 3928 ctx->info.sec_flags |= ICE_AQ_VSI_SEC_FLAG_ENA_MAC_ANTI_SPOOF | 3929 (ICE_AQ_VSI_SEC_TX_VLAN_PRUNE_ENA << 3930 ICE_AQ_VSI_SEC_TX_PRUNE_ENA_S); 3931 } 3932 3933 /** 3934 * ice_vsi_ctx_clear_antispoof - clear antispoof function in VSI ctx 3935 * @ctx: pointer to VSI ctx structure 3936 */ 3937 void ice_vsi_ctx_clear_antispoof(struct ice_vsi_ctx *ctx) 3938 { 3939 ctx->info.sec_flags &= ~ICE_AQ_VSI_SEC_FLAG_ENA_MAC_ANTI_SPOOF & 3940 ~(ICE_AQ_VSI_SEC_TX_VLAN_PRUNE_ENA << 3941 ICE_AQ_VSI_SEC_TX_PRUNE_ENA_S); 3942 } 3943 3944 /** 3945 * ice_vsi_ctx_set_allow_override - allow destination override on VSI 3946 * @ctx: pointer to VSI ctx structure 3947 */ 3948 void ice_vsi_ctx_set_allow_override(struct ice_vsi_ctx *ctx) 3949 { 3950 ctx->info.sec_flags |= ICE_AQ_VSI_SEC_FLAG_ALLOW_DEST_OVRD; 3951 } 3952 3953 /** 3954 * ice_vsi_ctx_clear_allow_override - turn off destination override on VSI 3955 * @ctx: pointer to VSI ctx structure 3956 */ 3957 void ice_vsi_ctx_clear_allow_override(struct ice_vsi_ctx *ctx) 3958 { 3959 ctx->info.sec_flags &= ~ICE_AQ_VSI_SEC_FLAG_ALLOW_DEST_OVRD; 3960 } 3961 3962 /** 3963 * ice_vsi_update_local_lb - update sw block in VSI with local loopback bit 3964 * @vsi: pointer to VSI structure 3965 * @set: set or unset the bit 3966 */ 3967 int 3968 ice_vsi_update_local_lb(struct ice_vsi *vsi, bool set) 3969 { 3970 struct ice_vsi_ctx ctx = { 3971 .info = vsi->info, 3972 }; 3973 3974 ctx.info.valid_sections = cpu_to_le16(ICE_AQ_VSI_PROP_SW_VALID); 3975 if (set) 3976 ctx.info.sw_flags |= ICE_AQ_VSI_SW_FLAG_LOCAL_LB; 3977 else 3978 ctx.info.sw_flags &= ~ICE_AQ_VSI_SW_FLAG_LOCAL_LB; 3979 3980 if (ice_update_vsi(&vsi->back->hw, vsi->idx, &ctx, NULL)) 3981 return -ENODEV; 3982 3983 vsi->info = ctx.info; 3984 return 0; 3985 } 3986