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_devlink.h" 11 #include "ice_vsi_vlan_ops.h" 12 13 /** 14 * ice_vsi_type_str - maps VSI type enum to string equivalents 15 * @vsi_type: VSI type enum 16 */ 17 const char *ice_vsi_type_str(enum ice_vsi_type vsi_type) 18 { 19 switch (vsi_type) { 20 case ICE_VSI_PF: 21 return "ICE_VSI_PF"; 22 case ICE_VSI_VF: 23 return "ICE_VSI_VF"; 24 case ICE_VSI_CTRL: 25 return "ICE_VSI_CTRL"; 26 case ICE_VSI_CHNL: 27 return "ICE_VSI_CHNL"; 28 case ICE_VSI_LB: 29 return "ICE_VSI_LB"; 30 case ICE_VSI_SWITCHDEV_CTRL: 31 return "ICE_VSI_SWITCHDEV_CTRL"; 32 default: 33 return "unknown"; 34 } 35 } 36 37 /** 38 * ice_vsi_ctrl_all_rx_rings - Start or stop a VSI's Rx rings 39 * @vsi: the VSI being configured 40 * @ena: start or stop the Rx rings 41 * 42 * First enable/disable all of the Rx rings, flush any remaining writes, and 43 * then verify that they have all been enabled/disabled successfully. This will 44 * let all of the register writes complete when enabling/disabling the Rx rings 45 * before waiting for the change in hardware to complete. 46 */ 47 static int ice_vsi_ctrl_all_rx_rings(struct ice_vsi *vsi, bool ena) 48 { 49 int ret = 0; 50 u16 i; 51 52 ice_for_each_rxq(vsi, i) 53 ice_vsi_ctrl_one_rx_ring(vsi, ena, i, false); 54 55 ice_flush(&vsi->back->hw); 56 57 ice_for_each_rxq(vsi, i) { 58 ret = ice_vsi_wait_one_rx_ring(vsi, ena, i); 59 if (ret) 60 break; 61 } 62 63 return ret; 64 } 65 66 /** 67 * ice_vsi_alloc_arrays - Allocate queue and vector pointer arrays for the VSI 68 * @vsi: VSI pointer 69 * 70 * On error: returns error code (negative) 71 * On success: returns 0 72 */ 73 static int ice_vsi_alloc_arrays(struct ice_vsi *vsi) 74 { 75 struct ice_pf *pf = vsi->back; 76 struct device *dev; 77 78 dev = ice_pf_to_dev(pf); 79 if (vsi->type == ICE_VSI_CHNL) 80 return 0; 81 82 /* allocate memory for both Tx and Rx ring pointers */ 83 vsi->tx_rings = devm_kcalloc(dev, vsi->alloc_txq, 84 sizeof(*vsi->tx_rings), GFP_KERNEL); 85 if (!vsi->tx_rings) 86 return -ENOMEM; 87 88 vsi->rx_rings = devm_kcalloc(dev, vsi->alloc_rxq, 89 sizeof(*vsi->rx_rings), GFP_KERNEL); 90 if (!vsi->rx_rings) 91 goto err_rings; 92 93 /* txq_map needs to have enough space to track both Tx (stack) rings 94 * and XDP rings; at this point vsi->num_xdp_txq might not be set, 95 * so use num_possible_cpus() as we want to always provide XDP ring 96 * per CPU, regardless of queue count settings from user that might 97 * have come from ethtool's set_channels() callback; 98 */ 99 vsi->txq_map = devm_kcalloc(dev, (vsi->alloc_txq + num_possible_cpus()), 100 sizeof(*vsi->txq_map), GFP_KERNEL); 101 102 if (!vsi->txq_map) 103 goto err_txq_map; 104 105 vsi->rxq_map = devm_kcalloc(dev, vsi->alloc_rxq, 106 sizeof(*vsi->rxq_map), GFP_KERNEL); 107 if (!vsi->rxq_map) 108 goto err_rxq_map; 109 110 /* There is no need to allocate q_vectors for a loopback VSI. */ 111 if (vsi->type == ICE_VSI_LB) 112 return 0; 113 114 /* allocate memory for q_vector pointers */ 115 vsi->q_vectors = devm_kcalloc(dev, vsi->num_q_vectors, 116 sizeof(*vsi->q_vectors), GFP_KERNEL); 117 if (!vsi->q_vectors) 118 goto err_vectors; 119 120 vsi->af_xdp_zc_qps = bitmap_zalloc(max_t(int, vsi->alloc_txq, vsi->alloc_rxq), GFP_KERNEL); 121 if (!vsi->af_xdp_zc_qps) 122 goto err_zc_qps; 123 124 return 0; 125 126 err_zc_qps: 127 devm_kfree(dev, vsi->q_vectors); 128 err_vectors: 129 devm_kfree(dev, vsi->rxq_map); 130 err_rxq_map: 131 devm_kfree(dev, vsi->txq_map); 132 err_txq_map: 133 devm_kfree(dev, vsi->rx_rings); 134 err_rings: 135 devm_kfree(dev, vsi->tx_rings); 136 return -ENOMEM; 137 } 138 139 /** 140 * ice_vsi_set_num_desc - Set number of descriptors for queues on this VSI 141 * @vsi: the VSI being configured 142 */ 143 static void ice_vsi_set_num_desc(struct ice_vsi *vsi) 144 { 145 switch (vsi->type) { 146 case ICE_VSI_PF: 147 case ICE_VSI_SWITCHDEV_CTRL: 148 case ICE_VSI_CTRL: 149 case ICE_VSI_LB: 150 /* a user could change the values of num_[tr]x_desc using 151 * ethtool -G so we should keep those values instead of 152 * overwriting them with the defaults. 153 */ 154 if (!vsi->num_rx_desc) 155 vsi->num_rx_desc = ICE_DFLT_NUM_RX_DESC; 156 if (!vsi->num_tx_desc) 157 vsi->num_tx_desc = ICE_DFLT_NUM_TX_DESC; 158 break; 159 default: 160 dev_dbg(ice_pf_to_dev(vsi->back), "Not setting number of Tx/Rx descriptors for VSI type %d\n", 161 vsi->type); 162 break; 163 } 164 } 165 166 /** 167 * ice_vsi_set_num_qs - Set number of queues, descriptors and vectors for a VSI 168 * @vsi: the VSI being configured 169 * 170 * Return 0 on success and a negative value on error 171 */ 172 static void ice_vsi_set_num_qs(struct ice_vsi *vsi) 173 { 174 enum ice_vsi_type vsi_type = vsi->type; 175 struct ice_pf *pf = vsi->back; 176 struct ice_vf *vf = vsi->vf; 177 178 if (WARN_ON(vsi_type == ICE_VSI_VF && !vf)) 179 return; 180 181 switch (vsi_type) { 182 case ICE_VSI_PF: 183 if (vsi->req_txq) { 184 vsi->alloc_txq = vsi->req_txq; 185 vsi->num_txq = vsi->req_txq; 186 } else { 187 vsi->alloc_txq = min3(pf->num_lan_msix, 188 ice_get_avail_txq_count(pf), 189 (u16)num_online_cpus()); 190 } 191 192 pf->num_lan_tx = vsi->alloc_txq; 193 194 /* only 1 Rx queue unless RSS is enabled */ 195 if (!test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 196 vsi->alloc_rxq = 1; 197 } else { 198 if (vsi->req_rxq) { 199 vsi->alloc_rxq = vsi->req_rxq; 200 vsi->num_rxq = vsi->req_rxq; 201 } else { 202 vsi->alloc_rxq = min3(pf->num_lan_msix, 203 ice_get_avail_rxq_count(pf), 204 (u16)num_online_cpus()); 205 } 206 } 207 208 pf->num_lan_rx = vsi->alloc_rxq; 209 210 vsi->num_q_vectors = min_t(int, pf->num_lan_msix, 211 max_t(int, vsi->alloc_rxq, 212 vsi->alloc_txq)); 213 break; 214 case ICE_VSI_SWITCHDEV_CTRL: 215 /* The number of queues for ctrl VSI is equal to number of PRs 216 * Each ring is associated to the corresponding VF_PR netdev. 217 * Tx and Rx rings are always equal 218 */ 219 if (vsi->req_txq && vsi->req_rxq) { 220 vsi->alloc_txq = vsi->req_txq; 221 vsi->alloc_rxq = vsi->req_rxq; 222 } else { 223 vsi->alloc_txq = 1; 224 vsi->alloc_rxq = 1; 225 } 226 227 vsi->num_q_vectors = 1; 228 break; 229 case ICE_VSI_VF: 230 if (vf->num_req_qs) 231 vf->num_vf_qs = vf->num_req_qs; 232 vsi->alloc_txq = vf->num_vf_qs; 233 vsi->alloc_rxq = vf->num_vf_qs; 234 /* pf->vfs.num_msix_per includes (VF miscellaneous vector + 235 * data queue interrupts). Since vsi->num_q_vectors is number 236 * of queues vectors, subtract 1 (ICE_NONQ_VECS_VF) from the 237 * original vector count 238 */ 239 vsi->num_q_vectors = vf->num_msix - ICE_NONQ_VECS_VF; 240 break; 241 case ICE_VSI_CTRL: 242 vsi->alloc_txq = 1; 243 vsi->alloc_rxq = 1; 244 vsi->num_q_vectors = 1; 245 break; 246 case ICE_VSI_CHNL: 247 vsi->alloc_txq = 0; 248 vsi->alloc_rxq = 0; 249 break; 250 case ICE_VSI_LB: 251 vsi->alloc_txq = 1; 252 vsi->alloc_rxq = 1; 253 break; 254 default: 255 dev_warn(ice_pf_to_dev(pf), "Unknown VSI type %d\n", vsi_type); 256 break; 257 } 258 259 ice_vsi_set_num_desc(vsi); 260 } 261 262 /** 263 * ice_get_free_slot - get the next non-NULL location index in array 264 * @array: array to search 265 * @size: size of the array 266 * @curr: last known occupied index to be used as a search hint 267 * 268 * void * is being used to keep the functionality generic. This lets us use this 269 * function on any array of pointers. 270 */ 271 static int ice_get_free_slot(void *array, int size, int curr) 272 { 273 int **tmp_array = (int **)array; 274 int next; 275 276 if (curr < (size - 1) && !tmp_array[curr + 1]) { 277 next = curr + 1; 278 } else { 279 int i = 0; 280 281 while ((i < size) && (tmp_array[i])) 282 i++; 283 if (i == size) 284 next = ICE_NO_VSI; 285 else 286 next = i; 287 } 288 return next; 289 } 290 291 /** 292 * ice_vsi_delete_from_hw - delete a VSI from the switch 293 * @vsi: pointer to VSI being removed 294 */ 295 static void ice_vsi_delete_from_hw(struct ice_vsi *vsi) 296 { 297 struct ice_pf *pf = vsi->back; 298 struct ice_vsi_ctx *ctxt; 299 int status; 300 301 ice_fltr_remove_all(vsi); 302 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); 303 if (!ctxt) 304 return; 305 306 if (vsi->type == ICE_VSI_VF) 307 ctxt->vf_num = vsi->vf->vf_id; 308 ctxt->vsi_num = vsi->vsi_num; 309 310 memcpy(&ctxt->info, &vsi->info, sizeof(ctxt->info)); 311 312 status = ice_free_vsi(&pf->hw, vsi->idx, ctxt, false, NULL); 313 if (status) 314 dev_err(ice_pf_to_dev(pf), "Failed to delete VSI %i in FW - error: %d\n", 315 vsi->vsi_num, status); 316 317 kfree(ctxt); 318 } 319 320 /** 321 * ice_vsi_free_arrays - De-allocate queue and vector pointer arrays for the VSI 322 * @vsi: pointer to VSI being cleared 323 */ 324 static void ice_vsi_free_arrays(struct ice_vsi *vsi) 325 { 326 struct ice_pf *pf = vsi->back; 327 struct device *dev; 328 329 dev = ice_pf_to_dev(pf); 330 331 bitmap_free(vsi->af_xdp_zc_qps); 332 vsi->af_xdp_zc_qps = NULL; 333 /* free the ring and vector containers */ 334 devm_kfree(dev, vsi->q_vectors); 335 vsi->q_vectors = NULL; 336 devm_kfree(dev, vsi->tx_rings); 337 vsi->tx_rings = NULL; 338 devm_kfree(dev, vsi->rx_rings); 339 vsi->rx_rings = NULL; 340 devm_kfree(dev, vsi->txq_map); 341 vsi->txq_map = NULL; 342 devm_kfree(dev, vsi->rxq_map); 343 vsi->rxq_map = NULL; 344 } 345 346 /** 347 * ice_vsi_free_stats - Free the ring statistics structures 348 * @vsi: VSI pointer 349 */ 350 static void ice_vsi_free_stats(struct ice_vsi *vsi) 351 { 352 struct ice_vsi_stats *vsi_stat; 353 struct ice_pf *pf = vsi->back; 354 int i; 355 356 if (vsi->type == ICE_VSI_CHNL) 357 return; 358 if (!pf->vsi_stats) 359 return; 360 361 vsi_stat = pf->vsi_stats[vsi->idx]; 362 if (!vsi_stat) 363 return; 364 365 ice_for_each_alloc_txq(vsi, i) { 366 if (vsi_stat->tx_ring_stats[i]) { 367 kfree_rcu(vsi_stat->tx_ring_stats[i], rcu); 368 WRITE_ONCE(vsi_stat->tx_ring_stats[i], NULL); 369 } 370 } 371 372 ice_for_each_alloc_rxq(vsi, i) { 373 if (vsi_stat->rx_ring_stats[i]) { 374 kfree_rcu(vsi_stat->rx_ring_stats[i], rcu); 375 WRITE_ONCE(vsi_stat->rx_ring_stats[i], NULL); 376 } 377 } 378 379 kfree(vsi_stat->tx_ring_stats); 380 kfree(vsi_stat->rx_ring_stats); 381 kfree(vsi_stat); 382 pf->vsi_stats[vsi->idx] = NULL; 383 } 384 385 /** 386 * ice_vsi_alloc_ring_stats - Allocates Tx and Rx ring stats for the VSI 387 * @vsi: VSI which is having stats allocated 388 */ 389 static int ice_vsi_alloc_ring_stats(struct ice_vsi *vsi) 390 { 391 struct ice_ring_stats **tx_ring_stats; 392 struct ice_ring_stats **rx_ring_stats; 393 struct ice_vsi_stats *vsi_stats; 394 struct ice_pf *pf = vsi->back; 395 u16 i; 396 397 vsi_stats = pf->vsi_stats[vsi->idx]; 398 tx_ring_stats = vsi_stats->tx_ring_stats; 399 rx_ring_stats = vsi_stats->rx_ring_stats; 400 401 /* Allocate Tx ring stats */ 402 ice_for_each_alloc_txq(vsi, i) { 403 struct ice_ring_stats *ring_stats; 404 struct ice_tx_ring *ring; 405 406 ring = vsi->tx_rings[i]; 407 ring_stats = tx_ring_stats[i]; 408 409 if (!ring_stats) { 410 ring_stats = kzalloc(sizeof(*ring_stats), GFP_KERNEL); 411 if (!ring_stats) 412 goto err_out; 413 414 WRITE_ONCE(tx_ring_stats[i], ring_stats); 415 } 416 417 ring->ring_stats = ring_stats; 418 } 419 420 /* Allocate Rx ring stats */ 421 ice_for_each_alloc_rxq(vsi, i) { 422 struct ice_ring_stats *ring_stats; 423 struct ice_rx_ring *ring; 424 425 ring = vsi->rx_rings[i]; 426 ring_stats = rx_ring_stats[i]; 427 428 if (!ring_stats) { 429 ring_stats = kzalloc(sizeof(*ring_stats), GFP_KERNEL); 430 if (!ring_stats) 431 goto err_out; 432 433 WRITE_ONCE(rx_ring_stats[i], ring_stats); 434 } 435 436 ring->ring_stats = ring_stats; 437 } 438 439 return 0; 440 441 err_out: 442 ice_vsi_free_stats(vsi); 443 return -ENOMEM; 444 } 445 446 /** 447 * ice_vsi_free - clean up and deallocate the provided VSI 448 * @vsi: pointer to VSI being cleared 449 * 450 * This deallocates the VSI's queue resources, removes it from the PF's 451 * VSI array if necessary, and deallocates the VSI 452 */ 453 static void ice_vsi_free(struct ice_vsi *vsi) 454 { 455 struct ice_pf *pf = NULL; 456 struct device *dev; 457 458 if (!vsi || !vsi->back) 459 return; 460 461 pf = vsi->back; 462 dev = ice_pf_to_dev(pf); 463 464 if (!pf->vsi[vsi->idx] || pf->vsi[vsi->idx] != vsi) { 465 dev_dbg(dev, "vsi does not exist at pf->vsi[%d]\n", vsi->idx); 466 return; 467 } 468 469 mutex_lock(&pf->sw_mutex); 470 /* updates the PF for this cleared VSI */ 471 472 pf->vsi[vsi->idx] = NULL; 473 pf->next_vsi = vsi->idx; 474 475 ice_vsi_free_stats(vsi); 476 ice_vsi_free_arrays(vsi); 477 mutex_unlock(&pf->sw_mutex); 478 devm_kfree(dev, vsi); 479 } 480 481 void ice_vsi_delete(struct ice_vsi *vsi) 482 { 483 ice_vsi_delete_from_hw(vsi); 484 ice_vsi_free(vsi); 485 } 486 487 /** 488 * ice_msix_clean_ctrl_vsi - MSIX mode interrupt handler for ctrl VSI 489 * @irq: interrupt number 490 * @data: pointer to a q_vector 491 */ 492 static irqreturn_t ice_msix_clean_ctrl_vsi(int __always_unused irq, void *data) 493 { 494 struct ice_q_vector *q_vector = (struct ice_q_vector *)data; 495 496 if (!q_vector->tx.tx_ring) 497 return IRQ_HANDLED; 498 499 #define FDIR_RX_DESC_CLEAN_BUDGET 64 500 ice_clean_rx_irq(q_vector->rx.rx_ring, FDIR_RX_DESC_CLEAN_BUDGET); 501 ice_clean_ctrl_tx_irq(q_vector->tx.tx_ring); 502 503 return IRQ_HANDLED; 504 } 505 506 /** 507 * ice_msix_clean_rings - MSIX mode Interrupt Handler 508 * @irq: interrupt number 509 * @data: pointer to a q_vector 510 */ 511 static irqreturn_t ice_msix_clean_rings(int __always_unused irq, void *data) 512 { 513 struct ice_q_vector *q_vector = (struct ice_q_vector *)data; 514 515 if (!q_vector->tx.tx_ring && !q_vector->rx.rx_ring) 516 return IRQ_HANDLED; 517 518 q_vector->total_events++; 519 520 napi_schedule(&q_vector->napi); 521 522 return IRQ_HANDLED; 523 } 524 525 static irqreturn_t ice_eswitch_msix_clean_rings(int __always_unused irq, void *data) 526 { 527 struct ice_q_vector *q_vector = (struct ice_q_vector *)data; 528 struct ice_pf *pf = q_vector->vsi->back; 529 struct ice_repr *repr; 530 unsigned long id; 531 532 if (!q_vector->tx.tx_ring && !q_vector->rx.rx_ring) 533 return IRQ_HANDLED; 534 535 xa_for_each(&pf->eswitch.reprs, id, repr) 536 napi_schedule(&repr->q_vector->napi); 537 538 return IRQ_HANDLED; 539 } 540 541 /** 542 * ice_vsi_alloc_stat_arrays - Allocate statistics arrays 543 * @vsi: VSI pointer 544 */ 545 static int ice_vsi_alloc_stat_arrays(struct ice_vsi *vsi) 546 { 547 struct ice_vsi_stats *vsi_stat; 548 struct ice_pf *pf = vsi->back; 549 550 if (vsi->type == ICE_VSI_CHNL) 551 return 0; 552 if (!pf->vsi_stats) 553 return -ENOENT; 554 555 if (pf->vsi_stats[vsi->idx]) 556 /* realloc will happen in rebuild path */ 557 return 0; 558 559 vsi_stat = kzalloc(sizeof(*vsi_stat), GFP_KERNEL); 560 if (!vsi_stat) 561 return -ENOMEM; 562 563 vsi_stat->tx_ring_stats = 564 kcalloc(vsi->alloc_txq, sizeof(*vsi_stat->tx_ring_stats), 565 GFP_KERNEL); 566 if (!vsi_stat->tx_ring_stats) 567 goto err_alloc_tx; 568 569 vsi_stat->rx_ring_stats = 570 kcalloc(vsi->alloc_rxq, sizeof(*vsi_stat->rx_ring_stats), 571 GFP_KERNEL); 572 if (!vsi_stat->rx_ring_stats) 573 goto err_alloc_rx; 574 575 pf->vsi_stats[vsi->idx] = vsi_stat; 576 577 return 0; 578 579 err_alloc_rx: 580 kfree(vsi_stat->rx_ring_stats); 581 err_alloc_tx: 582 kfree(vsi_stat->tx_ring_stats); 583 kfree(vsi_stat); 584 pf->vsi_stats[vsi->idx] = NULL; 585 return -ENOMEM; 586 } 587 588 /** 589 * ice_vsi_alloc_def - set default values for already allocated VSI 590 * @vsi: ptr to VSI 591 * @ch: ptr to channel 592 */ 593 static int 594 ice_vsi_alloc_def(struct ice_vsi *vsi, struct ice_channel *ch) 595 { 596 if (vsi->type != ICE_VSI_CHNL) { 597 ice_vsi_set_num_qs(vsi); 598 if (ice_vsi_alloc_arrays(vsi)) 599 return -ENOMEM; 600 } 601 602 switch (vsi->type) { 603 case ICE_VSI_SWITCHDEV_CTRL: 604 /* Setup eswitch MSIX irq handler for VSI */ 605 vsi->irq_handler = ice_eswitch_msix_clean_rings; 606 break; 607 case ICE_VSI_PF: 608 /* Setup default MSIX irq handler for VSI */ 609 vsi->irq_handler = ice_msix_clean_rings; 610 break; 611 case ICE_VSI_CTRL: 612 /* Setup ctrl VSI MSIX irq handler */ 613 vsi->irq_handler = ice_msix_clean_ctrl_vsi; 614 break; 615 case ICE_VSI_CHNL: 616 if (!ch) 617 return -EINVAL; 618 619 vsi->num_rxq = ch->num_rxq; 620 vsi->num_txq = ch->num_txq; 621 vsi->next_base_q = ch->base_q; 622 break; 623 case ICE_VSI_VF: 624 case ICE_VSI_LB: 625 break; 626 default: 627 ice_vsi_free_arrays(vsi); 628 return -EINVAL; 629 } 630 631 return 0; 632 } 633 634 /** 635 * ice_vsi_alloc - Allocates the next available struct VSI in the PF 636 * @pf: board private structure 637 * 638 * Reserves a VSI index from the PF and allocates an empty VSI structure 639 * without a type. The VSI structure must later be initialized by calling 640 * ice_vsi_cfg(). 641 * 642 * returns a pointer to a VSI on success, NULL on failure. 643 */ 644 static struct ice_vsi *ice_vsi_alloc(struct ice_pf *pf) 645 { 646 struct device *dev = ice_pf_to_dev(pf); 647 struct ice_vsi *vsi = NULL; 648 649 /* Need to protect the allocation of the VSIs at the PF level */ 650 mutex_lock(&pf->sw_mutex); 651 652 /* If we have already allocated our maximum number of VSIs, 653 * pf->next_vsi will be ICE_NO_VSI. If not, pf->next_vsi index 654 * is available to be populated 655 */ 656 if (pf->next_vsi == ICE_NO_VSI) { 657 dev_dbg(dev, "out of VSI slots!\n"); 658 goto unlock_pf; 659 } 660 661 vsi = devm_kzalloc(dev, sizeof(*vsi), GFP_KERNEL); 662 if (!vsi) 663 goto unlock_pf; 664 665 vsi->back = pf; 666 set_bit(ICE_VSI_DOWN, vsi->state); 667 668 /* fill slot and make note of the index */ 669 vsi->idx = pf->next_vsi; 670 pf->vsi[pf->next_vsi] = vsi; 671 672 /* prepare pf->next_vsi for next use */ 673 pf->next_vsi = ice_get_free_slot(pf->vsi, pf->num_alloc_vsi, 674 pf->next_vsi); 675 676 unlock_pf: 677 mutex_unlock(&pf->sw_mutex); 678 return vsi; 679 } 680 681 /** 682 * ice_alloc_fd_res - Allocate FD resource for a VSI 683 * @vsi: pointer to the ice_vsi 684 * 685 * This allocates the FD resources 686 * 687 * Returns 0 on success, -EPERM on no-op or -EIO on failure 688 */ 689 static int ice_alloc_fd_res(struct ice_vsi *vsi) 690 { 691 struct ice_pf *pf = vsi->back; 692 u32 g_val, b_val; 693 694 /* Flow Director filters are only allocated/assigned to the PF VSI or 695 * CHNL VSI which passes the traffic. The CTRL VSI is only used to 696 * add/delete filters so resources are not allocated to it 697 */ 698 if (!test_bit(ICE_FLAG_FD_ENA, pf->flags)) 699 return -EPERM; 700 701 if (!(vsi->type == ICE_VSI_PF || vsi->type == ICE_VSI_VF || 702 vsi->type == ICE_VSI_CHNL)) 703 return -EPERM; 704 705 /* FD filters from guaranteed pool per VSI */ 706 g_val = pf->hw.func_caps.fd_fltr_guar; 707 if (!g_val) 708 return -EPERM; 709 710 /* FD filters from best effort pool */ 711 b_val = pf->hw.func_caps.fd_fltr_best_effort; 712 if (!b_val) 713 return -EPERM; 714 715 /* PF main VSI gets only 64 FD resources from guaranteed pool 716 * when ADQ is configured. 717 */ 718 #define ICE_PF_VSI_GFLTR 64 719 720 /* determine FD filter resources per VSI from shared(best effort) and 721 * dedicated pool 722 */ 723 if (vsi->type == ICE_VSI_PF) { 724 vsi->num_gfltr = g_val; 725 /* if MQPRIO is configured, main VSI doesn't get all FD 726 * resources from guaranteed pool. PF VSI gets 64 FD resources 727 */ 728 if (test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) { 729 if (g_val < ICE_PF_VSI_GFLTR) 730 return -EPERM; 731 /* allow bare minimum entries for PF VSI */ 732 vsi->num_gfltr = ICE_PF_VSI_GFLTR; 733 } 734 735 /* each VSI gets same "best_effort" quota */ 736 vsi->num_bfltr = b_val; 737 } else if (vsi->type == ICE_VSI_VF) { 738 vsi->num_gfltr = 0; 739 740 /* each VSI gets same "best_effort" quota */ 741 vsi->num_bfltr = b_val; 742 } else { 743 struct ice_vsi *main_vsi; 744 int numtc; 745 746 main_vsi = ice_get_main_vsi(pf); 747 if (!main_vsi) 748 return -EPERM; 749 750 if (!main_vsi->all_numtc) 751 return -EINVAL; 752 753 /* figure out ADQ numtc */ 754 numtc = main_vsi->all_numtc - ICE_CHNL_START_TC; 755 756 /* only one TC but still asking resources for channels, 757 * invalid config 758 */ 759 if (numtc < ICE_CHNL_START_TC) 760 return -EPERM; 761 762 g_val -= ICE_PF_VSI_GFLTR; 763 /* channel VSIs gets equal share from guaranteed pool */ 764 vsi->num_gfltr = g_val / numtc; 765 766 /* each VSI gets same "best_effort" quota */ 767 vsi->num_bfltr = b_val; 768 } 769 770 return 0; 771 } 772 773 /** 774 * ice_vsi_get_qs - Assign queues from PF to VSI 775 * @vsi: the VSI to assign queues to 776 * 777 * Returns 0 on success and a negative value on error 778 */ 779 static int ice_vsi_get_qs(struct ice_vsi *vsi) 780 { 781 struct ice_pf *pf = vsi->back; 782 struct ice_qs_cfg tx_qs_cfg = { 783 .qs_mutex = &pf->avail_q_mutex, 784 .pf_map = pf->avail_txqs, 785 .pf_map_size = pf->max_pf_txqs, 786 .q_count = vsi->alloc_txq, 787 .scatter_count = ICE_MAX_SCATTER_TXQS, 788 .vsi_map = vsi->txq_map, 789 .vsi_map_offset = 0, 790 .mapping_mode = ICE_VSI_MAP_CONTIG 791 }; 792 struct ice_qs_cfg rx_qs_cfg = { 793 .qs_mutex = &pf->avail_q_mutex, 794 .pf_map = pf->avail_rxqs, 795 .pf_map_size = pf->max_pf_rxqs, 796 .q_count = vsi->alloc_rxq, 797 .scatter_count = ICE_MAX_SCATTER_RXQS, 798 .vsi_map = vsi->rxq_map, 799 .vsi_map_offset = 0, 800 .mapping_mode = ICE_VSI_MAP_CONTIG 801 }; 802 int ret; 803 804 if (vsi->type == ICE_VSI_CHNL) 805 return 0; 806 807 ret = __ice_vsi_get_qs(&tx_qs_cfg); 808 if (ret) 809 return ret; 810 vsi->tx_mapping_mode = tx_qs_cfg.mapping_mode; 811 812 ret = __ice_vsi_get_qs(&rx_qs_cfg); 813 if (ret) 814 return ret; 815 vsi->rx_mapping_mode = rx_qs_cfg.mapping_mode; 816 817 return 0; 818 } 819 820 /** 821 * ice_vsi_put_qs - Release queues from VSI to PF 822 * @vsi: the VSI that is going to release queues 823 */ 824 static void ice_vsi_put_qs(struct ice_vsi *vsi) 825 { 826 struct ice_pf *pf = vsi->back; 827 int i; 828 829 mutex_lock(&pf->avail_q_mutex); 830 831 ice_for_each_alloc_txq(vsi, i) { 832 clear_bit(vsi->txq_map[i], pf->avail_txqs); 833 vsi->txq_map[i] = ICE_INVAL_Q_INDEX; 834 } 835 836 ice_for_each_alloc_rxq(vsi, i) { 837 clear_bit(vsi->rxq_map[i], pf->avail_rxqs); 838 vsi->rxq_map[i] = ICE_INVAL_Q_INDEX; 839 } 840 841 mutex_unlock(&pf->avail_q_mutex); 842 } 843 844 /** 845 * ice_is_safe_mode 846 * @pf: pointer to the PF struct 847 * 848 * returns true if driver is in safe mode, false otherwise 849 */ 850 bool ice_is_safe_mode(struct ice_pf *pf) 851 { 852 return !test_bit(ICE_FLAG_ADV_FEATURES, pf->flags); 853 } 854 855 /** 856 * ice_is_rdma_ena 857 * @pf: pointer to the PF struct 858 * 859 * returns true if RDMA is currently supported, false otherwise 860 */ 861 bool ice_is_rdma_ena(struct ice_pf *pf) 862 { 863 return test_bit(ICE_FLAG_RDMA_ENA, pf->flags); 864 } 865 866 /** 867 * ice_vsi_clean_rss_flow_fld - Delete RSS configuration 868 * @vsi: the VSI being cleaned up 869 * 870 * This function deletes RSS input set for all flows that were configured 871 * for this VSI 872 */ 873 static void ice_vsi_clean_rss_flow_fld(struct ice_vsi *vsi) 874 { 875 struct ice_pf *pf = vsi->back; 876 int status; 877 878 if (ice_is_safe_mode(pf)) 879 return; 880 881 status = ice_rem_vsi_rss_cfg(&pf->hw, vsi->idx); 882 if (status) 883 dev_dbg(ice_pf_to_dev(pf), "ice_rem_vsi_rss_cfg failed for vsi = %d, error = %d\n", 884 vsi->vsi_num, status); 885 } 886 887 /** 888 * ice_rss_clean - Delete RSS related VSI structures and configuration 889 * @vsi: the VSI being removed 890 */ 891 static void ice_rss_clean(struct ice_vsi *vsi) 892 { 893 struct ice_pf *pf = vsi->back; 894 struct device *dev; 895 896 dev = ice_pf_to_dev(pf); 897 898 devm_kfree(dev, vsi->rss_hkey_user); 899 devm_kfree(dev, vsi->rss_lut_user); 900 901 ice_vsi_clean_rss_flow_fld(vsi); 902 /* remove RSS replay list */ 903 if (!ice_is_safe_mode(pf)) 904 ice_rem_vsi_rss_list(&pf->hw, vsi->idx); 905 } 906 907 /** 908 * ice_vsi_set_rss_params - Setup RSS capabilities per VSI type 909 * @vsi: the VSI being configured 910 */ 911 static void ice_vsi_set_rss_params(struct ice_vsi *vsi) 912 { 913 struct ice_hw_common_caps *cap; 914 struct ice_pf *pf = vsi->back; 915 u16 max_rss_size; 916 917 if (!test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 918 vsi->rss_size = 1; 919 return; 920 } 921 922 cap = &pf->hw.func_caps.common_cap; 923 max_rss_size = BIT(cap->rss_table_entry_width); 924 switch (vsi->type) { 925 case ICE_VSI_CHNL: 926 case ICE_VSI_PF: 927 /* PF VSI will inherit RSS instance of PF */ 928 vsi->rss_table_size = (u16)cap->rss_table_size; 929 if (vsi->type == ICE_VSI_CHNL) 930 vsi->rss_size = min_t(u16, vsi->num_rxq, max_rss_size); 931 else 932 vsi->rss_size = min_t(u16, num_online_cpus(), 933 max_rss_size); 934 vsi->rss_lut_type = ICE_LUT_PF; 935 break; 936 case ICE_VSI_SWITCHDEV_CTRL: 937 vsi->rss_table_size = ICE_LUT_VSI_SIZE; 938 vsi->rss_size = min_t(u16, num_online_cpus(), max_rss_size); 939 vsi->rss_lut_type = ICE_LUT_VSI; 940 break; 941 case ICE_VSI_VF: 942 /* VF VSI will get a small RSS table. 943 * For VSI_LUT, LUT size should be set to 64 bytes. 944 */ 945 vsi->rss_table_size = ICE_LUT_VSI_SIZE; 946 vsi->rss_size = ICE_MAX_RSS_QS_PER_VF; 947 vsi->rss_lut_type = ICE_LUT_VSI; 948 break; 949 case ICE_VSI_LB: 950 break; 951 default: 952 dev_dbg(ice_pf_to_dev(pf), "Unsupported VSI type %s\n", 953 ice_vsi_type_str(vsi->type)); 954 break; 955 } 956 } 957 958 /** 959 * ice_set_dflt_vsi_ctx - Set default VSI context before adding a VSI 960 * @hw: HW structure used to determine the VLAN mode of the device 961 * @ctxt: the VSI context being set 962 * 963 * This initializes a default VSI context for all sections except the Queues. 964 */ 965 static void ice_set_dflt_vsi_ctx(struct ice_hw *hw, struct ice_vsi_ctx *ctxt) 966 { 967 u32 table = 0; 968 969 memset(&ctxt->info, 0, sizeof(ctxt->info)); 970 /* VSI's should be allocated from shared pool */ 971 ctxt->alloc_from_pool = true; 972 /* Src pruning enabled by default */ 973 ctxt->info.sw_flags = ICE_AQ_VSI_SW_FLAG_SRC_PRUNE; 974 /* Traffic from VSI can be sent to LAN */ 975 ctxt->info.sw_flags2 = ICE_AQ_VSI_SW_FLAG_LAN_ENA; 976 /* allow all untagged/tagged packets by default on Tx */ 977 ctxt->info.inner_vlan_flags = ((ICE_AQ_VSI_INNER_VLAN_TX_MODE_ALL & 978 ICE_AQ_VSI_INNER_VLAN_TX_MODE_M) >> 979 ICE_AQ_VSI_INNER_VLAN_TX_MODE_S); 980 /* SVM - by default bits 3 and 4 in inner_vlan_flags are 0's which 981 * results in legacy behavior (show VLAN, DEI, and UP) in descriptor. 982 * 983 * DVM - leave inner VLAN in packet by default 984 */ 985 if (ice_is_dvm_ena(hw)) { 986 ctxt->info.inner_vlan_flags |= 987 ICE_AQ_VSI_INNER_VLAN_EMODE_NOTHING; 988 ctxt->info.outer_vlan_flags = 989 (ICE_AQ_VSI_OUTER_VLAN_TX_MODE_ALL << 990 ICE_AQ_VSI_OUTER_VLAN_TX_MODE_S) & 991 ICE_AQ_VSI_OUTER_VLAN_TX_MODE_M; 992 ctxt->info.outer_vlan_flags |= 993 (ICE_AQ_VSI_OUTER_TAG_VLAN_8100 << 994 ICE_AQ_VSI_OUTER_TAG_TYPE_S) & 995 ICE_AQ_VSI_OUTER_TAG_TYPE_M; 996 ctxt->info.outer_vlan_flags |= 997 FIELD_PREP(ICE_AQ_VSI_OUTER_VLAN_EMODE_M, 998 ICE_AQ_VSI_OUTER_VLAN_EMODE_NOTHING); 999 } 1000 /* Have 1:1 UP mapping for both ingress/egress tables */ 1001 table |= ICE_UP_TABLE_TRANSLATE(0, 0); 1002 table |= ICE_UP_TABLE_TRANSLATE(1, 1); 1003 table |= ICE_UP_TABLE_TRANSLATE(2, 2); 1004 table |= ICE_UP_TABLE_TRANSLATE(3, 3); 1005 table |= ICE_UP_TABLE_TRANSLATE(4, 4); 1006 table |= ICE_UP_TABLE_TRANSLATE(5, 5); 1007 table |= ICE_UP_TABLE_TRANSLATE(6, 6); 1008 table |= ICE_UP_TABLE_TRANSLATE(7, 7); 1009 ctxt->info.ingress_table = cpu_to_le32(table); 1010 ctxt->info.egress_table = cpu_to_le32(table); 1011 /* Have 1:1 UP mapping for outer to inner UP table */ 1012 ctxt->info.outer_up_table = cpu_to_le32(table); 1013 /* No Outer tag support outer_tag_flags remains to zero */ 1014 } 1015 1016 /** 1017 * ice_vsi_setup_q_map - Setup a VSI queue map 1018 * @vsi: the VSI being configured 1019 * @ctxt: VSI context structure 1020 */ 1021 static int ice_vsi_setup_q_map(struct ice_vsi *vsi, struct ice_vsi_ctx *ctxt) 1022 { 1023 u16 offset = 0, qmap = 0, tx_count = 0, rx_count = 0, pow = 0; 1024 u16 num_txq_per_tc, num_rxq_per_tc; 1025 u16 qcount_tx = vsi->alloc_txq; 1026 u16 qcount_rx = vsi->alloc_rxq; 1027 u8 netdev_tc = 0; 1028 int i; 1029 1030 if (!vsi->tc_cfg.numtc) { 1031 /* at least TC0 should be enabled by default */ 1032 vsi->tc_cfg.numtc = 1; 1033 vsi->tc_cfg.ena_tc = 1; 1034 } 1035 1036 num_rxq_per_tc = min_t(u16, qcount_rx / vsi->tc_cfg.numtc, ICE_MAX_RXQS_PER_TC); 1037 if (!num_rxq_per_tc) 1038 num_rxq_per_tc = 1; 1039 num_txq_per_tc = qcount_tx / vsi->tc_cfg.numtc; 1040 if (!num_txq_per_tc) 1041 num_txq_per_tc = 1; 1042 1043 /* find the (rounded up) power-of-2 of qcount */ 1044 pow = (u16)order_base_2(num_rxq_per_tc); 1045 1046 /* TC mapping is a function of the number of Rx queues assigned to the 1047 * VSI for each traffic class and the offset of these queues. 1048 * The first 10 bits are for queue offset for TC0, next 4 bits for no:of 1049 * queues allocated to TC0. No:of queues is a power-of-2. 1050 * 1051 * If TC is not enabled, the queue offset is set to 0, and allocate one 1052 * queue, this way, traffic for the given TC will be sent to the default 1053 * queue. 1054 * 1055 * Setup number and offset of Rx queues for all TCs for the VSI 1056 */ 1057 ice_for_each_traffic_class(i) { 1058 if (!(vsi->tc_cfg.ena_tc & BIT(i))) { 1059 /* TC is not enabled */ 1060 vsi->tc_cfg.tc_info[i].qoffset = 0; 1061 vsi->tc_cfg.tc_info[i].qcount_rx = 1; 1062 vsi->tc_cfg.tc_info[i].qcount_tx = 1; 1063 vsi->tc_cfg.tc_info[i].netdev_tc = 0; 1064 ctxt->info.tc_mapping[i] = 0; 1065 continue; 1066 } 1067 1068 /* TC is enabled */ 1069 vsi->tc_cfg.tc_info[i].qoffset = offset; 1070 vsi->tc_cfg.tc_info[i].qcount_rx = num_rxq_per_tc; 1071 vsi->tc_cfg.tc_info[i].qcount_tx = num_txq_per_tc; 1072 vsi->tc_cfg.tc_info[i].netdev_tc = netdev_tc++; 1073 1074 qmap = ((offset << ICE_AQ_VSI_TC_Q_OFFSET_S) & 1075 ICE_AQ_VSI_TC_Q_OFFSET_M) | 1076 ((pow << ICE_AQ_VSI_TC_Q_NUM_S) & 1077 ICE_AQ_VSI_TC_Q_NUM_M); 1078 offset += num_rxq_per_tc; 1079 tx_count += num_txq_per_tc; 1080 ctxt->info.tc_mapping[i] = cpu_to_le16(qmap); 1081 } 1082 1083 /* if offset is non-zero, means it is calculated correctly based on 1084 * enabled TCs for a given VSI otherwise qcount_rx will always 1085 * be correct and non-zero because it is based off - VSI's 1086 * allocated Rx queues which is at least 1 (hence qcount_tx will be 1087 * at least 1) 1088 */ 1089 if (offset) 1090 rx_count = offset; 1091 else 1092 rx_count = num_rxq_per_tc; 1093 1094 if (rx_count > vsi->alloc_rxq) { 1095 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Rx queues (%u), than were allocated (%u)!\n", 1096 rx_count, vsi->alloc_rxq); 1097 return -EINVAL; 1098 } 1099 1100 if (tx_count > vsi->alloc_txq) { 1101 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Tx queues (%u), than were allocated (%u)!\n", 1102 tx_count, vsi->alloc_txq); 1103 return -EINVAL; 1104 } 1105 1106 vsi->num_txq = tx_count; 1107 vsi->num_rxq = rx_count; 1108 1109 if (vsi->type == ICE_VSI_VF && vsi->num_txq != vsi->num_rxq) { 1110 dev_dbg(ice_pf_to_dev(vsi->back), "VF VSI should have same number of Tx and Rx queues. Hence making them equal\n"); 1111 /* since there is a chance that num_rxq could have been changed 1112 * in the above for loop, make num_txq equal to num_rxq. 1113 */ 1114 vsi->num_txq = vsi->num_rxq; 1115 } 1116 1117 /* Rx queue mapping */ 1118 ctxt->info.mapping_flags |= cpu_to_le16(ICE_AQ_VSI_Q_MAP_CONTIG); 1119 /* q_mapping buffer holds the info for the first queue allocated for 1120 * this VSI in the PF space and also the number of queues associated 1121 * with this VSI. 1122 */ 1123 ctxt->info.q_mapping[0] = cpu_to_le16(vsi->rxq_map[0]); 1124 ctxt->info.q_mapping[1] = cpu_to_le16(vsi->num_rxq); 1125 1126 return 0; 1127 } 1128 1129 /** 1130 * ice_set_fd_vsi_ctx - Set FD VSI context before adding a VSI 1131 * @ctxt: the VSI context being set 1132 * @vsi: the VSI being configured 1133 */ 1134 static void ice_set_fd_vsi_ctx(struct ice_vsi_ctx *ctxt, struct ice_vsi *vsi) 1135 { 1136 u8 dflt_q_group, dflt_q_prio; 1137 u16 dflt_q, report_q, val; 1138 1139 if (vsi->type != ICE_VSI_PF && vsi->type != ICE_VSI_CTRL && 1140 vsi->type != ICE_VSI_VF && vsi->type != ICE_VSI_CHNL) 1141 return; 1142 1143 val = ICE_AQ_VSI_PROP_FLOW_DIR_VALID; 1144 ctxt->info.valid_sections |= cpu_to_le16(val); 1145 dflt_q = 0; 1146 dflt_q_group = 0; 1147 report_q = 0; 1148 dflt_q_prio = 0; 1149 1150 /* enable flow director filtering/programming */ 1151 val = ICE_AQ_VSI_FD_ENABLE | ICE_AQ_VSI_FD_PROG_ENABLE; 1152 ctxt->info.fd_options = cpu_to_le16(val); 1153 /* max of allocated flow director filters */ 1154 ctxt->info.max_fd_fltr_dedicated = 1155 cpu_to_le16(vsi->num_gfltr); 1156 /* max of shared flow director filters any VSI may program */ 1157 ctxt->info.max_fd_fltr_shared = 1158 cpu_to_le16(vsi->num_bfltr); 1159 /* default queue index within the VSI of the default FD */ 1160 val = ((dflt_q << ICE_AQ_VSI_FD_DEF_Q_S) & 1161 ICE_AQ_VSI_FD_DEF_Q_M); 1162 /* target queue or queue group to the FD filter */ 1163 val |= ((dflt_q_group << ICE_AQ_VSI_FD_DEF_GRP_S) & 1164 ICE_AQ_VSI_FD_DEF_GRP_M); 1165 ctxt->info.fd_def_q = cpu_to_le16(val); 1166 /* queue index on which FD filter completion is reported */ 1167 val = ((report_q << ICE_AQ_VSI_FD_REPORT_Q_S) & 1168 ICE_AQ_VSI_FD_REPORT_Q_M); 1169 /* priority of the default qindex action */ 1170 val |= ((dflt_q_prio << ICE_AQ_VSI_FD_DEF_PRIORITY_S) & 1171 ICE_AQ_VSI_FD_DEF_PRIORITY_M); 1172 ctxt->info.fd_report_opt = cpu_to_le16(val); 1173 } 1174 1175 /** 1176 * ice_set_rss_vsi_ctx - Set RSS VSI context before adding a VSI 1177 * @ctxt: the VSI context being set 1178 * @vsi: the VSI being configured 1179 */ 1180 static void ice_set_rss_vsi_ctx(struct ice_vsi_ctx *ctxt, struct ice_vsi *vsi) 1181 { 1182 u8 lut_type, hash_type; 1183 struct device *dev; 1184 struct ice_pf *pf; 1185 1186 pf = vsi->back; 1187 dev = ice_pf_to_dev(pf); 1188 1189 switch (vsi->type) { 1190 case ICE_VSI_CHNL: 1191 case ICE_VSI_PF: 1192 /* PF VSI will inherit RSS instance of PF */ 1193 lut_type = ICE_AQ_VSI_Q_OPT_RSS_LUT_PF; 1194 hash_type = ICE_AQ_VSI_Q_OPT_RSS_TPLZ; 1195 break; 1196 case ICE_VSI_VF: 1197 /* VF VSI will gets a small RSS table which is a VSI LUT type */ 1198 lut_type = ICE_AQ_VSI_Q_OPT_RSS_LUT_VSI; 1199 hash_type = ICE_AQ_VSI_Q_OPT_RSS_TPLZ; 1200 break; 1201 default: 1202 dev_dbg(dev, "Unsupported VSI type %s\n", 1203 ice_vsi_type_str(vsi->type)); 1204 return; 1205 } 1206 1207 ctxt->info.q_opt_rss = ((lut_type << ICE_AQ_VSI_Q_OPT_RSS_LUT_S) & 1208 ICE_AQ_VSI_Q_OPT_RSS_LUT_M) | 1209 (hash_type & ICE_AQ_VSI_Q_OPT_RSS_HASH_M); 1210 } 1211 1212 static void 1213 ice_chnl_vsi_setup_q_map(struct ice_vsi *vsi, struct ice_vsi_ctx *ctxt) 1214 { 1215 struct ice_pf *pf = vsi->back; 1216 u16 qcount, qmap; 1217 u8 offset = 0; 1218 int pow; 1219 1220 qcount = min_t(int, vsi->num_rxq, pf->num_lan_msix); 1221 1222 pow = order_base_2(qcount); 1223 qmap = ((offset << ICE_AQ_VSI_TC_Q_OFFSET_S) & 1224 ICE_AQ_VSI_TC_Q_OFFSET_M) | 1225 ((pow << ICE_AQ_VSI_TC_Q_NUM_S) & 1226 ICE_AQ_VSI_TC_Q_NUM_M); 1227 1228 ctxt->info.tc_mapping[0] = cpu_to_le16(qmap); 1229 ctxt->info.mapping_flags |= cpu_to_le16(ICE_AQ_VSI_Q_MAP_CONTIG); 1230 ctxt->info.q_mapping[0] = cpu_to_le16(vsi->next_base_q); 1231 ctxt->info.q_mapping[1] = cpu_to_le16(qcount); 1232 } 1233 1234 /** 1235 * ice_vsi_is_vlan_pruning_ena - check if VLAN pruning is enabled or not 1236 * @vsi: VSI to check whether or not VLAN pruning is enabled. 1237 * 1238 * returns true if Rx VLAN pruning is enabled and false otherwise. 1239 */ 1240 static bool ice_vsi_is_vlan_pruning_ena(struct ice_vsi *vsi) 1241 { 1242 return vsi->info.sw_flags2 & ICE_AQ_VSI_SW_FLAG_RX_VLAN_PRUNE_ENA; 1243 } 1244 1245 /** 1246 * ice_vsi_init - Create and initialize a VSI 1247 * @vsi: the VSI being configured 1248 * @vsi_flags: VSI configuration flags 1249 * 1250 * Set ICE_FLAG_VSI_INIT to initialize a new VSI context, clear it to 1251 * reconfigure an existing context. 1252 * 1253 * This initializes a VSI context depending on the VSI type to be added and 1254 * passes it down to the add_vsi aq command to create a new VSI. 1255 */ 1256 static int ice_vsi_init(struct ice_vsi *vsi, u32 vsi_flags) 1257 { 1258 struct ice_pf *pf = vsi->back; 1259 struct ice_hw *hw = &pf->hw; 1260 struct ice_vsi_ctx *ctxt; 1261 struct device *dev; 1262 int ret = 0; 1263 1264 dev = ice_pf_to_dev(pf); 1265 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); 1266 if (!ctxt) 1267 return -ENOMEM; 1268 1269 switch (vsi->type) { 1270 case ICE_VSI_CTRL: 1271 case ICE_VSI_LB: 1272 case ICE_VSI_PF: 1273 ctxt->flags = ICE_AQ_VSI_TYPE_PF; 1274 break; 1275 case ICE_VSI_SWITCHDEV_CTRL: 1276 case ICE_VSI_CHNL: 1277 ctxt->flags = ICE_AQ_VSI_TYPE_VMDQ2; 1278 break; 1279 case ICE_VSI_VF: 1280 ctxt->flags = ICE_AQ_VSI_TYPE_VF; 1281 /* VF number here is the absolute VF number (0-255) */ 1282 ctxt->vf_num = vsi->vf->vf_id + hw->func_caps.vf_base_id; 1283 break; 1284 default: 1285 ret = -ENODEV; 1286 goto out; 1287 } 1288 1289 /* Handle VLAN pruning for channel VSI if main VSI has VLAN 1290 * prune enabled 1291 */ 1292 if (vsi->type == ICE_VSI_CHNL) { 1293 struct ice_vsi *main_vsi; 1294 1295 main_vsi = ice_get_main_vsi(pf); 1296 if (main_vsi && ice_vsi_is_vlan_pruning_ena(main_vsi)) 1297 ctxt->info.sw_flags2 |= 1298 ICE_AQ_VSI_SW_FLAG_RX_VLAN_PRUNE_ENA; 1299 else 1300 ctxt->info.sw_flags2 &= 1301 ~ICE_AQ_VSI_SW_FLAG_RX_VLAN_PRUNE_ENA; 1302 } 1303 1304 ice_set_dflt_vsi_ctx(hw, ctxt); 1305 if (test_bit(ICE_FLAG_FD_ENA, pf->flags)) 1306 ice_set_fd_vsi_ctx(ctxt, vsi); 1307 /* if the switch is in VEB mode, allow VSI loopback */ 1308 if (vsi->vsw->bridge_mode == BRIDGE_MODE_VEB) 1309 ctxt->info.sw_flags |= ICE_AQ_VSI_SW_FLAG_ALLOW_LB; 1310 1311 /* Set LUT type and HASH type if RSS is enabled */ 1312 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags) && 1313 vsi->type != ICE_VSI_CTRL) { 1314 ice_set_rss_vsi_ctx(ctxt, vsi); 1315 /* if updating VSI context, make sure to set valid_section: 1316 * to indicate which section of VSI context being updated 1317 */ 1318 if (!(vsi_flags & ICE_VSI_FLAG_INIT)) 1319 ctxt->info.valid_sections |= 1320 cpu_to_le16(ICE_AQ_VSI_PROP_Q_OPT_VALID); 1321 } 1322 1323 ctxt->info.sw_id = vsi->port_info->sw_id; 1324 if (vsi->type == ICE_VSI_CHNL) { 1325 ice_chnl_vsi_setup_q_map(vsi, ctxt); 1326 } else { 1327 ret = ice_vsi_setup_q_map(vsi, ctxt); 1328 if (ret) 1329 goto out; 1330 1331 if (!(vsi_flags & ICE_VSI_FLAG_INIT)) 1332 /* means VSI being updated */ 1333 /* must to indicate which section of VSI context are 1334 * being modified 1335 */ 1336 ctxt->info.valid_sections |= 1337 cpu_to_le16(ICE_AQ_VSI_PROP_RXQ_MAP_VALID); 1338 } 1339 1340 /* Allow control frames out of main VSI */ 1341 if (vsi->type == ICE_VSI_PF) { 1342 ctxt->info.sec_flags |= ICE_AQ_VSI_SEC_FLAG_ALLOW_DEST_OVRD; 1343 ctxt->info.valid_sections |= 1344 cpu_to_le16(ICE_AQ_VSI_PROP_SECURITY_VALID); 1345 } 1346 1347 if (vsi_flags & ICE_VSI_FLAG_INIT) { 1348 ret = ice_add_vsi(hw, vsi->idx, ctxt, NULL); 1349 if (ret) { 1350 dev_err(dev, "Add VSI failed, err %d\n", ret); 1351 ret = -EIO; 1352 goto out; 1353 } 1354 } else { 1355 ret = ice_update_vsi(hw, vsi->idx, ctxt, NULL); 1356 if (ret) { 1357 dev_err(dev, "Update VSI failed, err %d\n", ret); 1358 ret = -EIO; 1359 goto out; 1360 } 1361 } 1362 1363 /* keep context for update VSI operations */ 1364 vsi->info = ctxt->info; 1365 1366 /* record VSI number returned */ 1367 vsi->vsi_num = ctxt->vsi_num; 1368 1369 out: 1370 kfree(ctxt); 1371 return ret; 1372 } 1373 1374 /** 1375 * ice_vsi_clear_rings - Deallocates the Tx and Rx rings for VSI 1376 * @vsi: the VSI having rings deallocated 1377 */ 1378 static void ice_vsi_clear_rings(struct ice_vsi *vsi) 1379 { 1380 int i; 1381 1382 /* Avoid stale references by clearing map from vector to ring */ 1383 if (vsi->q_vectors) { 1384 ice_for_each_q_vector(vsi, i) { 1385 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 1386 1387 if (q_vector) { 1388 q_vector->tx.tx_ring = NULL; 1389 q_vector->rx.rx_ring = NULL; 1390 } 1391 } 1392 } 1393 1394 if (vsi->tx_rings) { 1395 ice_for_each_alloc_txq(vsi, i) { 1396 if (vsi->tx_rings[i]) { 1397 kfree_rcu(vsi->tx_rings[i], rcu); 1398 WRITE_ONCE(vsi->tx_rings[i], NULL); 1399 } 1400 } 1401 } 1402 if (vsi->rx_rings) { 1403 ice_for_each_alloc_rxq(vsi, i) { 1404 if (vsi->rx_rings[i]) { 1405 kfree_rcu(vsi->rx_rings[i], rcu); 1406 WRITE_ONCE(vsi->rx_rings[i], NULL); 1407 } 1408 } 1409 } 1410 } 1411 1412 /** 1413 * ice_vsi_alloc_rings - Allocates Tx and Rx rings for the VSI 1414 * @vsi: VSI which is having rings allocated 1415 */ 1416 static int ice_vsi_alloc_rings(struct ice_vsi *vsi) 1417 { 1418 bool dvm_ena = ice_is_dvm_ena(&vsi->back->hw); 1419 struct ice_pf *pf = vsi->back; 1420 struct device *dev; 1421 u16 i; 1422 1423 dev = ice_pf_to_dev(pf); 1424 /* Allocate Tx rings */ 1425 ice_for_each_alloc_txq(vsi, i) { 1426 struct ice_tx_ring *ring; 1427 1428 /* allocate with kzalloc(), free with kfree_rcu() */ 1429 ring = kzalloc(sizeof(*ring), GFP_KERNEL); 1430 1431 if (!ring) 1432 goto err_out; 1433 1434 ring->q_index = i; 1435 ring->reg_idx = vsi->txq_map[i]; 1436 ring->vsi = vsi; 1437 ring->tx_tstamps = &pf->ptp.port.tx; 1438 ring->dev = dev; 1439 ring->count = vsi->num_tx_desc; 1440 ring->txq_teid = ICE_INVAL_TEID; 1441 if (dvm_ena) 1442 ring->flags |= ICE_TX_FLAGS_RING_VLAN_L2TAG2; 1443 else 1444 ring->flags |= ICE_TX_FLAGS_RING_VLAN_L2TAG1; 1445 WRITE_ONCE(vsi->tx_rings[i], ring); 1446 } 1447 1448 /* Allocate Rx rings */ 1449 ice_for_each_alloc_rxq(vsi, i) { 1450 struct ice_rx_ring *ring; 1451 1452 /* allocate with kzalloc(), free with kfree_rcu() */ 1453 ring = kzalloc(sizeof(*ring), GFP_KERNEL); 1454 if (!ring) 1455 goto err_out; 1456 1457 ring->q_index = i; 1458 ring->reg_idx = vsi->rxq_map[i]; 1459 ring->vsi = vsi; 1460 ring->netdev = vsi->netdev; 1461 ring->dev = dev; 1462 ring->count = vsi->num_rx_desc; 1463 ring->cached_phctime = pf->ptp.cached_phc_time; 1464 WRITE_ONCE(vsi->rx_rings[i], ring); 1465 } 1466 1467 return 0; 1468 1469 err_out: 1470 ice_vsi_clear_rings(vsi); 1471 return -ENOMEM; 1472 } 1473 1474 /** 1475 * ice_vsi_manage_rss_lut - disable/enable RSS 1476 * @vsi: the VSI being changed 1477 * @ena: boolean value indicating if this is an enable or disable request 1478 * 1479 * In the event of disable request for RSS, this function will zero out RSS 1480 * LUT, while in the event of enable request for RSS, it will reconfigure RSS 1481 * LUT. 1482 */ 1483 void ice_vsi_manage_rss_lut(struct ice_vsi *vsi, bool ena) 1484 { 1485 u8 *lut; 1486 1487 lut = kzalloc(vsi->rss_table_size, GFP_KERNEL); 1488 if (!lut) 1489 return; 1490 1491 if (ena) { 1492 if (vsi->rss_lut_user) 1493 memcpy(lut, vsi->rss_lut_user, vsi->rss_table_size); 1494 else 1495 ice_fill_rss_lut(lut, vsi->rss_table_size, 1496 vsi->rss_size); 1497 } 1498 1499 ice_set_rss_lut(vsi, lut, vsi->rss_table_size); 1500 kfree(lut); 1501 } 1502 1503 /** 1504 * ice_vsi_cfg_crc_strip - Configure CRC stripping for a VSI 1505 * @vsi: VSI to be configured 1506 * @disable: set to true to have FCS / CRC in the frame data 1507 */ 1508 void ice_vsi_cfg_crc_strip(struct ice_vsi *vsi, bool disable) 1509 { 1510 int i; 1511 1512 ice_for_each_rxq(vsi, i) 1513 if (disable) 1514 vsi->rx_rings[i]->flags |= ICE_RX_FLAGS_CRC_STRIP_DIS; 1515 else 1516 vsi->rx_rings[i]->flags &= ~ICE_RX_FLAGS_CRC_STRIP_DIS; 1517 } 1518 1519 /** 1520 * ice_vsi_cfg_rss_lut_key - Configure RSS params for a VSI 1521 * @vsi: VSI to be configured 1522 */ 1523 int ice_vsi_cfg_rss_lut_key(struct ice_vsi *vsi) 1524 { 1525 struct ice_pf *pf = vsi->back; 1526 struct device *dev; 1527 u8 *lut, *key; 1528 int err; 1529 1530 dev = ice_pf_to_dev(pf); 1531 if (vsi->type == ICE_VSI_PF && vsi->ch_rss_size && 1532 (test_bit(ICE_FLAG_TC_MQPRIO, pf->flags))) { 1533 vsi->rss_size = min_t(u16, vsi->rss_size, vsi->ch_rss_size); 1534 } else { 1535 vsi->rss_size = min_t(u16, vsi->rss_size, vsi->num_rxq); 1536 1537 /* If orig_rss_size is valid and it is less than determined 1538 * main VSI's rss_size, update main VSI's rss_size to be 1539 * orig_rss_size so that when tc-qdisc is deleted, main VSI 1540 * RSS table gets programmed to be correct (whatever it was 1541 * to begin with (prior to setup-tc for ADQ config) 1542 */ 1543 if (vsi->orig_rss_size && vsi->rss_size < vsi->orig_rss_size && 1544 vsi->orig_rss_size <= vsi->num_rxq) { 1545 vsi->rss_size = vsi->orig_rss_size; 1546 /* now orig_rss_size is used, reset it to zero */ 1547 vsi->orig_rss_size = 0; 1548 } 1549 } 1550 1551 lut = kzalloc(vsi->rss_table_size, GFP_KERNEL); 1552 if (!lut) 1553 return -ENOMEM; 1554 1555 if (vsi->rss_lut_user) 1556 memcpy(lut, vsi->rss_lut_user, vsi->rss_table_size); 1557 else 1558 ice_fill_rss_lut(lut, vsi->rss_table_size, vsi->rss_size); 1559 1560 err = ice_set_rss_lut(vsi, lut, vsi->rss_table_size); 1561 if (err) { 1562 dev_err(dev, "set_rss_lut failed, error %d\n", err); 1563 goto ice_vsi_cfg_rss_exit; 1564 } 1565 1566 key = kzalloc(ICE_GET_SET_RSS_KEY_EXTEND_KEY_SIZE, GFP_KERNEL); 1567 if (!key) { 1568 err = -ENOMEM; 1569 goto ice_vsi_cfg_rss_exit; 1570 } 1571 1572 if (vsi->rss_hkey_user) 1573 memcpy(key, vsi->rss_hkey_user, ICE_GET_SET_RSS_KEY_EXTEND_KEY_SIZE); 1574 else 1575 netdev_rss_key_fill((void *)key, ICE_GET_SET_RSS_KEY_EXTEND_KEY_SIZE); 1576 1577 err = ice_set_rss_key(vsi, key); 1578 if (err) 1579 dev_err(dev, "set_rss_key failed, error %d\n", err); 1580 1581 kfree(key); 1582 ice_vsi_cfg_rss_exit: 1583 kfree(lut); 1584 return err; 1585 } 1586 1587 /** 1588 * ice_vsi_set_vf_rss_flow_fld - Sets VF VSI RSS input set for different flows 1589 * @vsi: VSI to be configured 1590 * 1591 * This function will only be called during the VF VSI setup. Upon successful 1592 * completion of package download, this function will configure default RSS 1593 * input sets for VF VSI. 1594 */ 1595 static void ice_vsi_set_vf_rss_flow_fld(struct ice_vsi *vsi) 1596 { 1597 struct ice_pf *pf = vsi->back; 1598 struct device *dev; 1599 int status; 1600 1601 dev = ice_pf_to_dev(pf); 1602 if (ice_is_safe_mode(pf)) { 1603 dev_dbg(dev, "Advanced RSS disabled. Package download failed, vsi num = %d\n", 1604 vsi->vsi_num); 1605 return; 1606 } 1607 1608 status = ice_add_avf_rss_cfg(&pf->hw, vsi->idx, ICE_DEFAULT_RSS_HENA); 1609 if (status) 1610 dev_dbg(dev, "ice_add_avf_rss_cfg failed for vsi = %d, error = %d\n", 1611 vsi->vsi_num, status); 1612 } 1613 1614 /** 1615 * ice_vsi_set_rss_flow_fld - Sets RSS input set for different flows 1616 * @vsi: VSI to be configured 1617 * 1618 * This function will only be called after successful download package call 1619 * during initialization of PF. Since the downloaded package will erase the 1620 * RSS section, this function will configure RSS input sets for different 1621 * flow types. The last profile added has the highest priority, therefore 2 1622 * tuple profiles (i.e. IPv4 src/dst) are added before 4 tuple profiles 1623 * (i.e. IPv4 src/dst TCP src/dst port). 1624 */ 1625 static void ice_vsi_set_rss_flow_fld(struct ice_vsi *vsi) 1626 { 1627 u16 vsi_handle = vsi->idx, vsi_num = vsi->vsi_num; 1628 struct ice_pf *pf = vsi->back; 1629 struct ice_hw *hw = &pf->hw; 1630 struct device *dev; 1631 int status; 1632 1633 dev = ice_pf_to_dev(pf); 1634 if (ice_is_safe_mode(pf)) { 1635 dev_dbg(dev, "Advanced RSS disabled. Package download failed, vsi num = %d\n", 1636 vsi_num); 1637 return; 1638 } 1639 /* configure RSS for IPv4 with input set IP src/dst */ 1640 status = ice_add_rss_cfg(hw, vsi_handle, ICE_FLOW_HASH_IPV4, 1641 ICE_FLOW_SEG_HDR_IPV4); 1642 if (status) 1643 dev_dbg(dev, "ice_add_rss_cfg failed for ipv4 flow, vsi = %d, error = %d\n", 1644 vsi_num, status); 1645 1646 /* configure RSS for IPv6 with input set IPv6 src/dst */ 1647 status = ice_add_rss_cfg(hw, vsi_handle, ICE_FLOW_HASH_IPV6, 1648 ICE_FLOW_SEG_HDR_IPV6); 1649 if (status) 1650 dev_dbg(dev, "ice_add_rss_cfg failed for ipv6 flow, vsi = %d, error = %d\n", 1651 vsi_num, status); 1652 1653 /* configure RSS for tcp4 with input set IP src/dst, TCP src/dst */ 1654 status = ice_add_rss_cfg(hw, vsi_handle, ICE_HASH_TCP_IPV4, 1655 ICE_FLOW_SEG_HDR_TCP | ICE_FLOW_SEG_HDR_IPV4); 1656 if (status) 1657 dev_dbg(dev, "ice_add_rss_cfg failed for tcp4 flow, vsi = %d, error = %d\n", 1658 vsi_num, status); 1659 1660 /* configure RSS for udp4 with input set IP src/dst, UDP src/dst */ 1661 status = ice_add_rss_cfg(hw, vsi_handle, ICE_HASH_UDP_IPV4, 1662 ICE_FLOW_SEG_HDR_UDP | ICE_FLOW_SEG_HDR_IPV4); 1663 if (status) 1664 dev_dbg(dev, "ice_add_rss_cfg failed for udp4 flow, vsi = %d, error = %d\n", 1665 vsi_num, status); 1666 1667 /* configure RSS for sctp4 with input set IP src/dst */ 1668 status = ice_add_rss_cfg(hw, vsi_handle, ICE_FLOW_HASH_IPV4, 1669 ICE_FLOW_SEG_HDR_SCTP | ICE_FLOW_SEG_HDR_IPV4); 1670 if (status) 1671 dev_dbg(dev, "ice_add_rss_cfg failed for sctp4 flow, vsi = %d, error = %d\n", 1672 vsi_num, status); 1673 1674 /* configure RSS for tcp6 with input set IPv6 src/dst, TCP src/dst */ 1675 status = ice_add_rss_cfg(hw, vsi_handle, ICE_HASH_TCP_IPV6, 1676 ICE_FLOW_SEG_HDR_TCP | ICE_FLOW_SEG_HDR_IPV6); 1677 if (status) 1678 dev_dbg(dev, "ice_add_rss_cfg failed for tcp6 flow, vsi = %d, error = %d\n", 1679 vsi_num, status); 1680 1681 /* configure RSS for udp6 with input set IPv6 src/dst, UDP src/dst */ 1682 status = ice_add_rss_cfg(hw, vsi_handle, ICE_HASH_UDP_IPV6, 1683 ICE_FLOW_SEG_HDR_UDP | ICE_FLOW_SEG_HDR_IPV6); 1684 if (status) 1685 dev_dbg(dev, "ice_add_rss_cfg failed for udp6 flow, vsi = %d, error = %d\n", 1686 vsi_num, status); 1687 1688 /* configure RSS for sctp6 with input set IPv6 src/dst */ 1689 status = ice_add_rss_cfg(hw, vsi_handle, ICE_FLOW_HASH_IPV6, 1690 ICE_FLOW_SEG_HDR_SCTP | ICE_FLOW_SEG_HDR_IPV6); 1691 if (status) 1692 dev_dbg(dev, "ice_add_rss_cfg failed for sctp6 flow, vsi = %d, error = %d\n", 1693 vsi_num, status); 1694 1695 status = ice_add_rss_cfg(hw, vsi_handle, ICE_FLOW_HASH_ESP_SPI, 1696 ICE_FLOW_SEG_HDR_ESP); 1697 if (status) 1698 dev_dbg(dev, "ice_add_rss_cfg failed for esp/spi flow, vsi = %d, error = %d\n", 1699 vsi_num, status); 1700 } 1701 1702 /** 1703 * ice_vsi_cfg_frame_size - setup max frame size and Rx buffer length 1704 * @vsi: VSI 1705 */ 1706 static void ice_vsi_cfg_frame_size(struct ice_vsi *vsi) 1707 { 1708 if (!vsi->netdev || test_bit(ICE_FLAG_LEGACY_RX, vsi->back->flags)) { 1709 vsi->max_frame = ICE_MAX_FRAME_LEGACY_RX; 1710 vsi->rx_buf_len = ICE_RXBUF_1664; 1711 #if (PAGE_SIZE < 8192) 1712 } else if (!ICE_2K_TOO_SMALL_WITH_PADDING && 1713 (vsi->netdev->mtu <= ETH_DATA_LEN)) { 1714 vsi->max_frame = ICE_RXBUF_1536 - NET_IP_ALIGN; 1715 vsi->rx_buf_len = ICE_RXBUF_1536 - NET_IP_ALIGN; 1716 #endif 1717 } else { 1718 vsi->max_frame = ICE_AQ_SET_MAC_FRAME_SIZE_MAX; 1719 vsi->rx_buf_len = ICE_RXBUF_3072; 1720 } 1721 } 1722 1723 /** 1724 * ice_pf_state_is_nominal - checks the PF for nominal state 1725 * @pf: pointer to PF to check 1726 * 1727 * Check the PF's state for a collection of bits that would indicate 1728 * the PF is in a state that would inhibit normal operation for 1729 * driver functionality. 1730 * 1731 * Returns true if PF is in a nominal state, false otherwise 1732 */ 1733 bool ice_pf_state_is_nominal(struct ice_pf *pf) 1734 { 1735 DECLARE_BITMAP(check_bits, ICE_STATE_NBITS) = { 0 }; 1736 1737 if (!pf) 1738 return false; 1739 1740 bitmap_set(check_bits, 0, ICE_STATE_NOMINAL_CHECK_BITS); 1741 if (bitmap_intersects(pf->state, check_bits, ICE_STATE_NBITS)) 1742 return false; 1743 1744 return true; 1745 } 1746 1747 /** 1748 * ice_update_eth_stats - Update VSI-specific ethernet statistics counters 1749 * @vsi: the VSI to be updated 1750 */ 1751 void ice_update_eth_stats(struct ice_vsi *vsi) 1752 { 1753 struct ice_eth_stats *prev_es, *cur_es; 1754 struct ice_hw *hw = &vsi->back->hw; 1755 struct ice_pf *pf = vsi->back; 1756 u16 vsi_num = vsi->vsi_num; /* HW absolute index of a VSI */ 1757 1758 prev_es = &vsi->eth_stats_prev; 1759 cur_es = &vsi->eth_stats; 1760 1761 if (ice_is_reset_in_progress(pf->state)) 1762 vsi->stat_offsets_loaded = false; 1763 1764 ice_stat_update40(hw, GLV_GORCL(vsi_num), vsi->stat_offsets_loaded, 1765 &prev_es->rx_bytes, &cur_es->rx_bytes); 1766 1767 ice_stat_update40(hw, GLV_UPRCL(vsi_num), vsi->stat_offsets_loaded, 1768 &prev_es->rx_unicast, &cur_es->rx_unicast); 1769 1770 ice_stat_update40(hw, GLV_MPRCL(vsi_num), vsi->stat_offsets_loaded, 1771 &prev_es->rx_multicast, &cur_es->rx_multicast); 1772 1773 ice_stat_update40(hw, GLV_BPRCL(vsi_num), vsi->stat_offsets_loaded, 1774 &prev_es->rx_broadcast, &cur_es->rx_broadcast); 1775 1776 ice_stat_update32(hw, GLV_RDPC(vsi_num), vsi->stat_offsets_loaded, 1777 &prev_es->rx_discards, &cur_es->rx_discards); 1778 1779 ice_stat_update40(hw, GLV_GOTCL(vsi_num), vsi->stat_offsets_loaded, 1780 &prev_es->tx_bytes, &cur_es->tx_bytes); 1781 1782 ice_stat_update40(hw, GLV_UPTCL(vsi_num), vsi->stat_offsets_loaded, 1783 &prev_es->tx_unicast, &cur_es->tx_unicast); 1784 1785 ice_stat_update40(hw, GLV_MPTCL(vsi_num), vsi->stat_offsets_loaded, 1786 &prev_es->tx_multicast, &cur_es->tx_multicast); 1787 1788 ice_stat_update40(hw, GLV_BPTCL(vsi_num), vsi->stat_offsets_loaded, 1789 &prev_es->tx_broadcast, &cur_es->tx_broadcast); 1790 1791 ice_stat_update32(hw, GLV_TEPC(vsi_num), vsi->stat_offsets_loaded, 1792 &prev_es->tx_errors, &cur_es->tx_errors); 1793 1794 vsi->stat_offsets_loaded = true; 1795 } 1796 1797 /** 1798 * ice_write_qrxflxp_cntxt - write/configure QRXFLXP_CNTXT register 1799 * @hw: HW pointer 1800 * @pf_q: index of the Rx queue in the PF's queue space 1801 * @rxdid: flexible descriptor RXDID 1802 * @prio: priority for the RXDID for this queue 1803 * @ena_ts: true to enable timestamp and false to disable timestamp 1804 */ 1805 void 1806 ice_write_qrxflxp_cntxt(struct ice_hw *hw, u16 pf_q, u32 rxdid, u32 prio, 1807 bool ena_ts) 1808 { 1809 int regval = rd32(hw, QRXFLXP_CNTXT(pf_q)); 1810 1811 /* clear any previous values */ 1812 regval &= ~(QRXFLXP_CNTXT_RXDID_IDX_M | 1813 QRXFLXP_CNTXT_RXDID_PRIO_M | 1814 QRXFLXP_CNTXT_TS_M); 1815 1816 regval |= (rxdid << QRXFLXP_CNTXT_RXDID_IDX_S) & 1817 QRXFLXP_CNTXT_RXDID_IDX_M; 1818 1819 regval |= (prio << QRXFLXP_CNTXT_RXDID_PRIO_S) & 1820 QRXFLXP_CNTXT_RXDID_PRIO_M; 1821 1822 if (ena_ts) 1823 /* Enable TimeSync on this queue */ 1824 regval |= QRXFLXP_CNTXT_TS_M; 1825 1826 wr32(hw, QRXFLXP_CNTXT(pf_q), regval); 1827 } 1828 1829 int ice_vsi_cfg_single_rxq(struct ice_vsi *vsi, u16 q_idx) 1830 { 1831 if (q_idx >= vsi->num_rxq) 1832 return -EINVAL; 1833 1834 return ice_vsi_cfg_rxq(vsi->rx_rings[q_idx]); 1835 } 1836 1837 int ice_vsi_cfg_single_txq(struct ice_vsi *vsi, struct ice_tx_ring **tx_rings, u16 q_idx) 1838 { 1839 DEFINE_FLEX(struct ice_aqc_add_tx_qgrp, qg_buf, txqs, 1); 1840 1841 if (q_idx >= vsi->alloc_txq || !tx_rings || !tx_rings[q_idx]) 1842 return -EINVAL; 1843 1844 qg_buf->num_txqs = 1; 1845 1846 return ice_vsi_cfg_txq(vsi, tx_rings[q_idx], qg_buf); 1847 } 1848 1849 /** 1850 * ice_vsi_cfg_rxqs - Configure the VSI for Rx 1851 * @vsi: the VSI being configured 1852 * 1853 * Return 0 on success and a negative value on error 1854 * Configure the Rx VSI for operation. 1855 */ 1856 int ice_vsi_cfg_rxqs(struct ice_vsi *vsi) 1857 { 1858 u16 i; 1859 1860 if (vsi->type == ICE_VSI_VF) 1861 goto setup_rings; 1862 1863 ice_vsi_cfg_frame_size(vsi); 1864 setup_rings: 1865 /* set up individual rings */ 1866 ice_for_each_rxq(vsi, i) { 1867 int err = ice_vsi_cfg_rxq(vsi->rx_rings[i]); 1868 1869 if (err) 1870 return err; 1871 } 1872 1873 return 0; 1874 } 1875 1876 /** 1877 * ice_vsi_cfg_txqs - Configure the VSI for Tx 1878 * @vsi: the VSI being configured 1879 * @rings: Tx ring array to be configured 1880 * @count: number of Tx ring array elements 1881 * 1882 * Return 0 on success and a negative value on error 1883 * Configure the Tx VSI for operation. 1884 */ 1885 static int 1886 ice_vsi_cfg_txqs(struct ice_vsi *vsi, struct ice_tx_ring **rings, u16 count) 1887 { 1888 DEFINE_FLEX(struct ice_aqc_add_tx_qgrp, qg_buf, txqs, 1); 1889 int err = 0; 1890 u16 q_idx; 1891 1892 qg_buf->num_txqs = 1; 1893 1894 for (q_idx = 0; q_idx < count; q_idx++) { 1895 err = ice_vsi_cfg_txq(vsi, rings[q_idx], qg_buf); 1896 if (err) 1897 break; 1898 } 1899 1900 return err; 1901 } 1902 1903 /** 1904 * ice_vsi_cfg_lan_txqs - Configure the VSI for Tx 1905 * @vsi: the VSI being configured 1906 * 1907 * Return 0 on success and a negative value on error 1908 * Configure the Tx VSI for operation. 1909 */ 1910 int ice_vsi_cfg_lan_txqs(struct ice_vsi *vsi) 1911 { 1912 return ice_vsi_cfg_txqs(vsi, vsi->tx_rings, vsi->num_txq); 1913 } 1914 1915 /** 1916 * ice_vsi_cfg_xdp_txqs - Configure Tx queues dedicated for XDP in given VSI 1917 * @vsi: the VSI being configured 1918 * 1919 * Return 0 on success and a negative value on error 1920 * Configure the Tx queues dedicated for XDP in given VSI for operation. 1921 */ 1922 int ice_vsi_cfg_xdp_txqs(struct ice_vsi *vsi) 1923 { 1924 int ret; 1925 int i; 1926 1927 ret = ice_vsi_cfg_txqs(vsi, vsi->xdp_rings, vsi->num_xdp_txq); 1928 if (ret) 1929 return ret; 1930 1931 ice_for_each_rxq(vsi, i) 1932 ice_tx_xsk_pool(vsi, i); 1933 1934 return 0; 1935 } 1936 1937 /** 1938 * ice_intrl_usec_to_reg - convert interrupt rate limit to register value 1939 * @intrl: interrupt rate limit in usecs 1940 * @gran: interrupt rate limit granularity in usecs 1941 * 1942 * This function converts a decimal interrupt rate limit in usecs to the format 1943 * expected by firmware. 1944 */ 1945 static u32 ice_intrl_usec_to_reg(u8 intrl, u8 gran) 1946 { 1947 u32 val = intrl / gran; 1948 1949 if (val) 1950 return val | GLINT_RATE_INTRL_ENA_M; 1951 return 0; 1952 } 1953 1954 /** 1955 * ice_write_intrl - write throttle rate limit to interrupt specific register 1956 * @q_vector: pointer to interrupt specific structure 1957 * @intrl: throttle rate limit in microseconds to write 1958 */ 1959 void ice_write_intrl(struct ice_q_vector *q_vector, u8 intrl) 1960 { 1961 struct ice_hw *hw = &q_vector->vsi->back->hw; 1962 1963 wr32(hw, GLINT_RATE(q_vector->reg_idx), 1964 ice_intrl_usec_to_reg(intrl, ICE_INTRL_GRAN_ABOVE_25)); 1965 } 1966 1967 static struct ice_q_vector *ice_pull_qvec_from_rc(struct ice_ring_container *rc) 1968 { 1969 switch (rc->type) { 1970 case ICE_RX_CONTAINER: 1971 if (rc->rx_ring) 1972 return rc->rx_ring->q_vector; 1973 break; 1974 case ICE_TX_CONTAINER: 1975 if (rc->tx_ring) 1976 return rc->tx_ring->q_vector; 1977 break; 1978 default: 1979 break; 1980 } 1981 1982 return NULL; 1983 } 1984 1985 /** 1986 * __ice_write_itr - write throttle rate to register 1987 * @q_vector: pointer to interrupt data structure 1988 * @rc: pointer to ring container 1989 * @itr: throttle rate in microseconds to write 1990 */ 1991 static void __ice_write_itr(struct ice_q_vector *q_vector, 1992 struct ice_ring_container *rc, u16 itr) 1993 { 1994 struct ice_hw *hw = &q_vector->vsi->back->hw; 1995 1996 wr32(hw, GLINT_ITR(rc->itr_idx, q_vector->reg_idx), 1997 ITR_REG_ALIGN(itr) >> ICE_ITR_GRAN_S); 1998 } 1999 2000 /** 2001 * ice_write_itr - write throttle rate to queue specific register 2002 * @rc: pointer to ring container 2003 * @itr: throttle rate in microseconds to write 2004 */ 2005 void ice_write_itr(struct ice_ring_container *rc, u16 itr) 2006 { 2007 struct ice_q_vector *q_vector; 2008 2009 q_vector = ice_pull_qvec_from_rc(rc); 2010 if (!q_vector) 2011 return; 2012 2013 __ice_write_itr(q_vector, rc, itr); 2014 } 2015 2016 /** 2017 * ice_set_q_vector_intrl - set up interrupt rate limiting 2018 * @q_vector: the vector to be configured 2019 * 2020 * Interrupt rate limiting is local to the vector, not per-queue so we must 2021 * detect if either ring container has dynamic moderation enabled to decide 2022 * what to set the interrupt rate limit to via INTRL settings. In the case that 2023 * dynamic moderation is disabled on both, write the value with the cached 2024 * setting to make sure INTRL register matches the user visible value. 2025 */ 2026 void ice_set_q_vector_intrl(struct ice_q_vector *q_vector) 2027 { 2028 if (ITR_IS_DYNAMIC(&q_vector->tx) || ITR_IS_DYNAMIC(&q_vector->rx)) { 2029 /* in the case of dynamic enabled, cap each vector to no more 2030 * than (4 us) 250,000 ints/sec, which allows low latency 2031 * but still less than 500,000 interrupts per second, which 2032 * reduces CPU a bit in the case of the lowest latency 2033 * setting. The 4 here is a value in microseconds. 2034 */ 2035 ice_write_intrl(q_vector, 4); 2036 } else { 2037 ice_write_intrl(q_vector, q_vector->intrl); 2038 } 2039 } 2040 2041 /** 2042 * ice_vsi_cfg_msix - MSIX mode Interrupt Config in the HW 2043 * @vsi: the VSI being configured 2044 * 2045 * This configures MSIX mode interrupts for the PF VSI, and should not be used 2046 * for the VF VSI. 2047 */ 2048 void ice_vsi_cfg_msix(struct ice_vsi *vsi) 2049 { 2050 struct ice_pf *pf = vsi->back; 2051 struct ice_hw *hw = &pf->hw; 2052 u16 txq = 0, rxq = 0; 2053 int i, q; 2054 2055 ice_for_each_q_vector(vsi, i) { 2056 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 2057 u16 reg_idx = q_vector->reg_idx; 2058 2059 ice_cfg_itr(hw, q_vector); 2060 2061 /* Both Transmit Queue Interrupt Cause Control register 2062 * and Receive Queue Interrupt Cause control register 2063 * expects MSIX_INDX field to be the vector index 2064 * within the function space and not the absolute 2065 * vector index across PF or across device. 2066 * For SR-IOV VF VSIs queue vector index always starts 2067 * with 1 since first vector index(0) is used for OICR 2068 * in VF space. Since VMDq and other PF VSIs are within 2069 * the PF function space, use the vector index that is 2070 * tracked for this PF. 2071 */ 2072 for (q = 0; q < q_vector->num_ring_tx; q++) { 2073 ice_cfg_txq_interrupt(vsi, txq, reg_idx, 2074 q_vector->tx.itr_idx); 2075 txq++; 2076 } 2077 2078 for (q = 0; q < q_vector->num_ring_rx; q++) { 2079 ice_cfg_rxq_interrupt(vsi, rxq, reg_idx, 2080 q_vector->rx.itr_idx); 2081 rxq++; 2082 } 2083 } 2084 } 2085 2086 /** 2087 * ice_vsi_start_all_rx_rings - start/enable all of a VSI's Rx rings 2088 * @vsi: the VSI whose rings are to be enabled 2089 * 2090 * Returns 0 on success and a negative value on error 2091 */ 2092 int ice_vsi_start_all_rx_rings(struct ice_vsi *vsi) 2093 { 2094 return ice_vsi_ctrl_all_rx_rings(vsi, true); 2095 } 2096 2097 /** 2098 * ice_vsi_stop_all_rx_rings - stop/disable all of a VSI's Rx rings 2099 * @vsi: the VSI whose rings are to be disabled 2100 * 2101 * Returns 0 on success and a negative value on error 2102 */ 2103 int ice_vsi_stop_all_rx_rings(struct ice_vsi *vsi) 2104 { 2105 return ice_vsi_ctrl_all_rx_rings(vsi, false); 2106 } 2107 2108 /** 2109 * ice_vsi_stop_tx_rings - Disable Tx rings 2110 * @vsi: the VSI being configured 2111 * @rst_src: reset source 2112 * @rel_vmvf_num: Relative ID of VF/VM 2113 * @rings: Tx ring array to be stopped 2114 * @count: number of Tx ring array elements 2115 */ 2116 static int 2117 ice_vsi_stop_tx_rings(struct ice_vsi *vsi, enum ice_disq_rst_src rst_src, 2118 u16 rel_vmvf_num, struct ice_tx_ring **rings, u16 count) 2119 { 2120 u16 q_idx; 2121 2122 if (vsi->num_txq > ICE_LAN_TXQ_MAX_QDIS) 2123 return -EINVAL; 2124 2125 for (q_idx = 0; q_idx < count; q_idx++) { 2126 struct ice_txq_meta txq_meta = { }; 2127 int status; 2128 2129 if (!rings || !rings[q_idx]) 2130 return -EINVAL; 2131 2132 ice_fill_txq_meta(vsi, rings[q_idx], &txq_meta); 2133 status = ice_vsi_stop_tx_ring(vsi, rst_src, rel_vmvf_num, 2134 rings[q_idx], &txq_meta); 2135 2136 if (status) 2137 return status; 2138 } 2139 2140 return 0; 2141 } 2142 2143 /** 2144 * ice_vsi_stop_lan_tx_rings - Disable LAN Tx rings 2145 * @vsi: the VSI being configured 2146 * @rst_src: reset source 2147 * @rel_vmvf_num: Relative ID of VF/VM 2148 */ 2149 int 2150 ice_vsi_stop_lan_tx_rings(struct ice_vsi *vsi, enum ice_disq_rst_src rst_src, 2151 u16 rel_vmvf_num) 2152 { 2153 return ice_vsi_stop_tx_rings(vsi, rst_src, rel_vmvf_num, vsi->tx_rings, vsi->num_txq); 2154 } 2155 2156 /** 2157 * ice_vsi_stop_xdp_tx_rings - Disable XDP Tx rings 2158 * @vsi: the VSI being configured 2159 */ 2160 int ice_vsi_stop_xdp_tx_rings(struct ice_vsi *vsi) 2161 { 2162 return ice_vsi_stop_tx_rings(vsi, ICE_NO_RESET, 0, vsi->xdp_rings, vsi->num_xdp_txq); 2163 } 2164 2165 /** 2166 * ice_vsi_is_rx_queue_active 2167 * @vsi: the VSI being configured 2168 * 2169 * Return true if at least one queue is active. 2170 */ 2171 bool ice_vsi_is_rx_queue_active(struct ice_vsi *vsi) 2172 { 2173 struct ice_pf *pf = vsi->back; 2174 struct ice_hw *hw = &pf->hw; 2175 int i; 2176 2177 ice_for_each_rxq(vsi, i) { 2178 u32 rx_reg; 2179 int pf_q; 2180 2181 pf_q = vsi->rxq_map[i]; 2182 rx_reg = rd32(hw, QRX_CTRL(pf_q)); 2183 if (rx_reg & QRX_CTRL_QENA_STAT_M) 2184 return true; 2185 } 2186 2187 return false; 2188 } 2189 2190 static void ice_vsi_set_tc_cfg(struct ice_vsi *vsi) 2191 { 2192 if (!test_bit(ICE_FLAG_DCB_ENA, vsi->back->flags)) { 2193 vsi->tc_cfg.ena_tc = ICE_DFLT_TRAFFIC_CLASS; 2194 vsi->tc_cfg.numtc = 1; 2195 return; 2196 } 2197 2198 /* set VSI TC information based on DCB config */ 2199 ice_vsi_set_dcb_tc_cfg(vsi); 2200 } 2201 2202 /** 2203 * ice_cfg_sw_lldp - Config switch rules for LLDP packet handling 2204 * @vsi: the VSI being configured 2205 * @tx: bool to determine Tx or Rx rule 2206 * @create: bool to determine create or remove Rule 2207 */ 2208 void ice_cfg_sw_lldp(struct ice_vsi *vsi, bool tx, bool create) 2209 { 2210 int (*eth_fltr)(struct ice_vsi *v, u16 type, u16 flag, 2211 enum ice_sw_fwd_act_type act); 2212 struct ice_pf *pf = vsi->back; 2213 struct device *dev; 2214 int status; 2215 2216 dev = ice_pf_to_dev(pf); 2217 eth_fltr = create ? ice_fltr_add_eth : ice_fltr_remove_eth; 2218 2219 if (tx) { 2220 status = eth_fltr(vsi, ETH_P_LLDP, ICE_FLTR_TX, 2221 ICE_DROP_PACKET); 2222 } else { 2223 if (ice_fw_supports_lldp_fltr_ctrl(&pf->hw)) { 2224 status = ice_lldp_fltr_add_remove(&pf->hw, vsi->vsi_num, 2225 create); 2226 } else { 2227 status = eth_fltr(vsi, ETH_P_LLDP, ICE_FLTR_RX, 2228 ICE_FWD_TO_VSI); 2229 } 2230 } 2231 2232 if (status) 2233 dev_dbg(dev, "Fail %s %s LLDP rule on VSI %i error: %d\n", 2234 create ? "adding" : "removing", tx ? "TX" : "RX", 2235 vsi->vsi_num, status); 2236 } 2237 2238 /** 2239 * ice_set_agg_vsi - sets up scheduler aggregator node and move VSI into it 2240 * @vsi: pointer to the VSI 2241 * 2242 * This function will allocate new scheduler aggregator now if needed and will 2243 * move specified VSI into it. 2244 */ 2245 static void ice_set_agg_vsi(struct ice_vsi *vsi) 2246 { 2247 struct device *dev = ice_pf_to_dev(vsi->back); 2248 struct ice_agg_node *agg_node_iter = NULL; 2249 u32 agg_id = ICE_INVALID_AGG_NODE_ID; 2250 struct ice_agg_node *agg_node = NULL; 2251 int node_offset, max_agg_nodes = 0; 2252 struct ice_port_info *port_info; 2253 struct ice_pf *pf = vsi->back; 2254 u32 agg_node_id_start = 0; 2255 int status; 2256 2257 /* create (as needed) scheduler aggregator node and move VSI into 2258 * corresponding aggregator node 2259 * - PF aggregator node to contains VSIs of type _PF and _CTRL 2260 * - VF aggregator nodes will contain VF VSI 2261 */ 2262 port_info = pf->hw.port_info; 2263 if (!port_info) 2264 return; 2265 2266 switch (vsi->type) { 2267 case ICE_VSI_CTRL: 2268 case ICE_VSI_CHNL: 2269 case ICE_VSI_LB: 2270 case ICE_VSI_PF: 2271 case ICE_VSI_SWITCHDEV_CTRL: 2272 max_agg_nodes = ICE_MAX_PF_AGG_NODES; 2273 agg_node_id_start = ICE_PF_AGG_NODE_ID_START; 2274 agg_node_iter = &pf->pf_agg_node[0]; 2275 break; 2276 case ICE_VSI_VF: 2277 /* user can create 'n' VFs on a given PF, but since max children 2278 * per aggregator node can be only 64. Following code handles 2279 * aggregator(s) for VF VSIs, either selects a agg_node which 2280 * was already created provided num_vsis < 64, otherwise 2281 * select next available node, which will be created 2282 */ 2283 max_agg_nodes = ICE_MAX_VF_AGG_NODES; 2284 agg_node_id_start = ICE_VF_AGG_NODE_ID_START; 2285 agg_node_iter = &pf->vf_agg_node[0]; 2286 break; 2287 default: 2288 /* other VSI type, handle later if needed */ 2289 dev_dbg(dev, "unexpected VSI type %s\n", 2290 ice_vsi_type_str(vsi->type)); 2291 return; 2292 } 2293 2294 /* find the appropriate aggregator node */ 2295 for (node_offset = 0; node_offset < max_agg_nodes; node_offset++) { 2296 /* see if we can find space in previously created 2297 * node if num_vsis < 64, otherwise skip 2298 */ 2299 if (agg_node_iter->num_vsis && 2300 agg_node_iter->num_vsis == ICE_MAX_VSIS_IN_AGG_NODE) { 2301 agg_node_iter++; 2302 continue; 2303 } 2304 2305 if (agg_node_iter->valid && 2306 agg_node_iter->agg_id != ICE_INVALID_AGG_NODE_ID) { 2307 agg_id = agg_node_iter->agg_id; 2308 agg_node = agg_node_iter; 2309 break; 2310 } 2311 2312 /* find unclaimed agg_id */ 2313 if (agg_node_iter->agg_id == ICE_INVALID_AGG_NODE_ID) { 2314 agg_id = node_offset + agg_node_id_start; 2315 agg_node = agg_node_iter; 2316 break; 2317 } 2318 /* move to next agg_node */ 2319 agg_node_iter++; 2320 } 2321 2322 if (!agg_node) 2323 return; 2324 2325 /* if selected aggregator node was not created, create it */ 2326 if (!agg_node->valid) { 2327 status = ice_cfg_agg(port_info, agg_id, ICE_AGG_TYPE_AGG, 2328 (u8)vsi->tc_cfg.ena_tc); 2329 if (status) { 2330 dev_err(dev, "unable to create aggregator node with agg_id %u\n", 2331 agg_id); 2332 return; 2333 } 2334 /* aggregator node is created, store the needed info */ 2335 agg_node->valid = true; 2336 agg_node->agg_id = agg_id; 2337 } 2338 2339 /* move VSI to corresponding aggregator node */ 2340 status = ice_move_vsi_to_agg(port_info, agg_id, vsi->idx, 2341 (u8)vsi->tc_cfg.ena_tc); 2342 if (status) { 2343 dev_err(dev, "unable to move VSI idx %u into aggregator %u node", 2344 vsi->idx, agg_id); 2345 return; 2346 } 2347 2348 /* keep active children count for aggregator node */ 2349 agg_node->num_vsis++; 2350 2351 /* cache the 'agg_id' in VSI, so that after reset - VSI will be moved 2352 * to aggregator node 2353 */ 2354 vsi->agg_node = agg_node; 2355 dev_dbg(dev, "successfully moved VSI idx %u tc_bitmap 0x%x) into aggregator node %d which has num_vsis %u\n", 2356 vsi->idx, vsi->tc_cfg.ena_tc, vsi->agg_node->agg_id, 2357 vsi->agg_node->num_vsis); 2358 } 2359 2360 static int ice_vsi_cfg_tc_lan(struct ice_pf *pf, struct ice_vsi *vsi) 2361 { 2362 u16 max_txqs[ICE_MAX_TRAFFIC_CLASS] = { 0 }; 2363 struct device *dev = ice_pf_to_dev(pf); 2364 int ret, i; 2365 2366 /* configure VSI nodes based on number of queues and TC's */ 2367 ice_for_each_traffic_class(i) { 2368 if (!(vsi->tc_cfg.ena_tc & BIT(i))) 2369 continue; 2370 2371 if (vsi->type == ICE_VSI_CHNL) { 2372 if (!vsi->alloc_txq && vsi->num_txq) 2373 max_txqs[i] = vsi->num_txq; 2374 else 2375 max_txqs[i] = pf->num_lan_tx; 2376 } else { 2377 max_txqs[i] = vsi->alloc_txq; 2378 } 2379 } 2380 2381 dev_dbg(dev, "vsi->tc_cfg.ena_tc = %d\n", vsi->tc_cfg.ena_tc); 2382 ret = ice_cfg_vsi_lan(vsi->port_info, vsi->idx, vsi->tc_cfg.ena_tc, 2383 max_txqs); 2384 if (ret) { 2385 dev_err(dev, "VSI %d failed lan queue config, error %d\n", 2386 vsi->vsi_num, ret); 2387 return ret; 2388 } 2389 2390 return 0; 2391 } 2392 2393 /** 2394 * ice_vsi_cfg_def - configure default VSI based on the type 2395 * @vsi: pointer to VSI 2396 * @params: the parameters to configure this VSI with 2397 */ 2398 static int 2399 ice_vsi_cfg_def(struct ice_vsi *vsi, struct ice_vsi_cfg_params *params) 2400 { 2401 struct device *dev = ice_pf_to_dev(vsi->back); 2402 struct ice_pf *pf = vsi->back; 2403 int ret; 2404 2405 vsi->vsw = pf->first_sw; 2406 2407 ret = ice_vsi_alloc_def(vsi, params->ch); 2408 if (ret) 2409 return ret; 2410 2411 /* allocate memory for Tx/Rx ring stat pointers */ 2412 ret = ice_vsi_alloc_stat_arrays(vsi); 2413 if (ret) 2414 goto unroll_vsi_alloc; 2415 2416 ice_alloc_fd_res(vsi); 2417 2418 ret = ice_vsi_get_qs(vsi); 2419 if (ret) { 2420 dev_err(dev, "Failed to allocate queues. vsi->idx = %d\n", 2421 vsi->idx); 2422 goto unroll_vsi_alloc_stat; 2423 } 2424 2425 /* set RSS capabilities */ 2426 ice_vsi_set_rss_params(vsi); 2427 2428 /* set TC configuration */ 2429 ice_vsi_set_tc_cfg(vsi); 2430 2431 /* create the VSI */ 2432 ret = ice_vsi_init(vsi, params->flags); 2433 if (ret) 2434 goto unroll_get_qs; 2435 2436 ice_vsi_init_vlan_ops(vsi); 2437 2438 switch (vsi->type) { 2439 case ICE_VSI_CTRL: 2440 case ICE_VSI_SWITCHDEV_CTRL: 2441 case ICE_VSI_PF: 2442 ret = ice_vsi_alloc_q_vectors(vsi); 2443 if (ret) 2444 goto unroll_vsi_init; 2445 2446 ret = ice_vsi_alloc_rings(vsi); 2447 if (ret) 2448 goto unroll_vector_base; 2449 2450 ret = ice_vsi_alloc_ring_stats(vsi); 2451 if (ret) 2452 goto unroll_vector_base; 2453 2454 ice_vsi_map_rings_to_vectors(vsi); 2455 vsi->stat_offsets_loaded = false; 2456 2457 if (ice_is_xdp_ena_vsi(vsi)) { 2458 ret = ice_vsi_determine_xdp_res(vsi); 2459 if (ret) 2460 goto unroll_vector_base; 2461 ret = ice_prepare_xdp_rings(vsi, vsi->xdp_prog); 2462 if (ret) 2463 goto unroll_vector_base; 2464 } 2465 2466 /* ICE_VSI_CTRL does not need RSS so skip RSS processing */ 2467 if (vsi->type != ICE_VSI_CTRL) 2468 /* Do not exit if configuring RSS had an issue, at 2469 * least receive traffic on first queue. Hence no 2470 * need to capture return value 2471 */ 2472 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 2473 ice_vsi_cfg_rss_lut_key(vsi); 2474 ice_vsi_set_rss_flow_fld(vsi); 2475 } 2476 ice_init_arfs(vsi); 2477 break; 2478 case ICE_VSI_CHNL: 2479 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 2480 ice_vsi_cfg_rss_lut_key(vsi); 2481 ice_vsi_set_rss_flow_fld(vsi); 2482 } 2483 break; 2484 case ICE_VSI_VF: 2485 /* VF driver will take care of creating netdev for this type and 2486 * map queues to vectors through Virtchnl, PF driver only 2487 * creates a VSI and corresponding structures for bookkeeping 2488 * purpose 2489 */ 2490 ret = ice_vsi_alloc_q_vectors(vsi); 2491 if (ret) 2492 goto unroll_vsi_init; 2493 2494 ret = ice_vsi_alloc_rings(vsi); 2495 if (ret) 2496 goto unroll_alloc_q_vector; 2497 2498 ret = ice_vsi_alloc_ring_stats(vsi); 2499 if (ret) 2500 goto unroll_vector_base; 2501 2502 vsi->stat_offsets_loaded = false; 2503 2504 /* Do not exit if configuring RSS had an issue, at least 2505 * receive traffic on first queue. Hence no need to capture 2506 * return value 2507 */ 2508 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 2509 ice_vsi_cfg_rss_lut_key(vsi); 2510 ice_vsi_set_vf_rss_flow_fld(vsi); 2511 } 2512 break; 2513 case ICE_VSI_LB: 2514 ret = ice_vsi_alloc_rings(vsi); 2515 if (ret) 2516 goto unroll_vsi_init; 2517 2518 ret = ice_vsi_alloc_ring_stats(vsi); 2519 if (ret) 2520 goto unroll_vector_base; 2521 2522 break; 2523 default: 2524 /* clean up the resources and exit */ 2525 ret = -EINVAL; 2526 goto unroll_vsi_init; 2527 } 2528 2529 return 0; 2530 2531 unroll_vector_base: 2532 /* reclaim SW interrupts back to the common pool */ 2533 unroll_alloc_q_vector: 2534 ice_vsi_free_q_vectors(vsi); 2535 unroll_vsi_init: 2536 ice_vsi_delete_from_hw(vsi); 2537 unroll_get_qs: 2538 ice_vsi_put_qs(vsi); 2539 unroll_vsi_alloc_stat: 2540 ice_vsi_free_stats(vsi); 2541 unroll_vsi_alloc: 2542 ice_vsi_free_arrays(vsi); 2543 return ret; 2544 } 2545 2546 /** 2547 * ice_vsi_cfg - configure a previously allocated VSI 2548 * @vsi: pointer to VSI 2549 * @params: parameters used to configure this VSI 2550 */ 2551 int ice_vsi_cfg(struct ice_vsi *vsi, struct ice_vsi_cfg_params *params) 2552 { 2553 struct ice_pf *pf = vsi->back; 2554 int ret; 2555 2556 if (WARN_ON(params->type == ICE_VSI_VF && !params->vf)) 2557 return -EINVAL; 2558 2559 vsi->type = params->type; 2560 vsi->port_info = params->pi; 2561 2562 /* For VSIs which don't have a connected VF, this will be NULL */ 2563 vsi->vf = params->vf; 2564 2565 ret = ice_vsi_cfg_def(vsi, params); 2566 if (ret) 2567 return ret; 2568 2569 ret = ice_vsi_cfg_tc_lan(vsi->back, vsi); 2570 if (ret) 2571 ice_vsi_decfg(vsi); 2572 2573 if (vsi->type == ICE_VSI_CTRL) { 2574 if (vsi->vf) { 2575 WARN_ON(vsi->vf->ctrl_vsi_idx != ICE_NO_VSI); 2576 vsi->vf->ctrl_vsi_idx = vsi->idx; 2577 } else { 2578 WARN_ON(pf->ctrl_vsi_idx != ICE_NO_VSI); 2579 pf->ctrl_vsi_idx = vsi->idx; 2580 } 2581 } 2582 2583 return ret; 2584 } 2585 2586 /** 2587 * ice_vsi_decfg - remove all VSI configuration 2588 * @vsi: pointer to VSI 2589 */ 2590 void ice_vsi_decfg(struct ice_vsi *vsi) 2591 { 2592 struct ice_pf *pf = vsi->back; 2593 int err; 2594 2595 /* The Rx rule will only exist to remove if the LLDP FW 2596 * engine is currently stopped 2597 */ 2598 if (!ice_is_safe_mode(pf) && vsi->type == ICE_VSI_PF && 2599 !test_bit(ICE_FLAG_FW_LLDP_AGENT, pf->flags)) 2600 ice_cfg_sw_lldp(vsi, false, false); 2601 2602 ice_rm_vsi_lan_cfg(vsi->port_info, vsi->idx); 2603 err = ice_rm_vsi_rdma_cfg(vsi->port_info, vsi->idx); 2604 if (err) 2605 dev_err(ice_pf_to_dev(pf), "Failed to remove RDMA scheduler config for VSI %u, err %d\n", 2606 vsi->vsi_num, err); 2607 2608 if (ice_is_xdp_ena_vsi(vsi)) 2609 /* return value check can be skipped here, it always returns 2610 * 0 if reset is in progress 2611 */ 2612 ice_destroy_xdp_rings(vsi); 2613 2614 ice_vsi_clear_rings(vsi); 2615 ice_vsi_free_q_vectors(vsi); 2616 ice_vsi_put_qs(vsi); 2617 ice_vsi_free_arrays(vsi); 2618 2619 /* SR-IOV determines needed MSIX resources all at once instead of per 2620 * VSI since when VFs are spawned we know how many VFs there are and how 2621 * many interrupts each VF needs. SR-IOV MSIX resources are also 2622 * cleared in the same manner. 2623 */ 2624 2625 if (vsi->type == ICE_VSI_VF && 2626 vsi->agg_node && vsi->agg_node->valid) 2627 vsi->agg_node->num_vsis--; 2628 if (vsi->agg_node) { 2629 vsi->agg_node->valid = false; 2630 vsi->agg_node->agg_id = 0; 2631 } 2632 } 2633 2634 /** 2635 * ice_vsi_setup - Set up a VSI by a given type 2636 * @pf: board private structure 2637 * @params: parameters to use when creating the VSI 2638 * 2639 * This allocates the sw VSI structure and its queue resources. 2640 * 2641 * Returns pointer to the successfully allocated and configured VSI sw struct on 2642 * success, NULL on failure. 2643 */ 2644 struct ice_vsi * 2645 ice_vsi_setup(struct ice_pf *pf, struct ice_vsi_cfg_params *params) 2646 { 2647 struct device *dev = ice_pf_to_dev(pf); 2648 struct ice_vsi *vsi; 2649 int ret; 2650 2651 /* ice_vsi_setup can only initialize a new VSI, and we must have 2652 * a port_info structure for it. 2653 */ 2654 if (WARN_ON(!(params->flags & ICE_VSI_FLAG_INIT)) || 2655 WARN_ON(!params->pi)) 2656 return NULL; 2657 2658 vsi = ice_vsi_alloc(pf); 2659 if (!vsi) { 2660 dev_err(dev, "could not allocate VSI\n"); 2661 return NULL; 2662 } 2663 2664 ret = ice_vsi_cfg(vsi, params); 2665 if (ret) 2666 goto err_vsi_cfg; 2667 2668 /* Add switch rule to drop all Tx Flow Control Frames, of look up 2669 * type ETHERTYPE from VSIs, and restrict malicious VF from sending 2670 * out PAUSE or PFC frames. If enabled, FW can still send FC frames. 2671 * The rule is added once for PF VSI in order to create appropriate 2672 * recipe, since VSI/VSI list is ignored with drop action... 2673 * Also add rules to handle LLDP Tx packets. Tx LLDP packets need to 2674 * be dropped so that VFs cannot send LLDP packets to reconfig DCB 2675 * settings in the HW. 2676 */ 2677 if (!ice_is_safe_mode(pf) && vsi->type == ICE_VSI_PF) { 2678 ice_fltr_add_eth(vsi, ETH_P_PAUSE, ICE_FLTR_TX, 2679 ICE_DROP_PACKET); 2680 ice_cfg_sw_lldp(vsi, true, true); 2681 } 2682 2683 if (!vsi->agg_node) 2684 ice_set_agg_vsi(vsi); 2685 2686 return vsi; 2687 2688 err_vsi_cfg: 2689 ice_vsi_free(vsi); 2690 2691 return NULL; 2692 } 2693 2694 /** 2695 * ice_vsi_release_msix - Clear the queue to Interrupt mapping in HW 2696 * @vsi: the VSI being cleaned up 2697 */ 2698 static void ice_vsi_release_msix(struct ice_vsi *vsi) 2699 { 2700 struct ice_pf *pf = vsi->back; 2701 struct ice_hw *hw = &pf->hw; 2702 u32 txq = 0; 2703 u32 rxq = 0; 2704 int i, q; 2705 2706 ice_for_each_q_vector(vsi, i) { 2707 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 2708 2709 ice_write_intrl(q_vector, 0); 2710 for (q = 0; q < q_vector->num_ring_tx; q++) { 2711 ice_write_itr(&q_vector->tx, 0); 2712 wr32(hw, QINT_TQCTL(vsi->txq_map[txq]), 0); 2713 if (ice_is_xdp_ena_vsi(vsi)) { 2714 u32 xdp_txq = txq + vsi->num_xdp_txq; 2715 2716 wr32(hw, QINT_TQCTL(vsi->txq_map[xdp_txq]), 0); 2717 } 2718 txq++; 2719 } 2720 2721 for (q = 0; q < q_vector->num_ring_rx; q++) { 2722 ice_write_itr(&q_vector->rx, 0); 2723 wr32(hw, QINT_RQCTL(vsi->rxq_map[rxq]), 0); 2724 rxq++; 2725 } 2726 } 2727 2728 ice_flush(hw); 2729 } 2730 2731 /** 2732 * ice_vsi_free_irq - Free the IRQ association with the OS 2733 * @vsi: the VSI being configured 2734 */ 2735 void ice_vsi_free_irq(struct ice_vsi *vsi) 2736 { 2737 struct ice_pf *pf = vsi->back; 2738 int i; 2739 2740 if (!vsi->q_vectors || !vsi->irqs_ready) 2741 return; 2742 2743 ice_vsi_release_msix(vsi); 2744 if (vsi->type == ICE_VSI_VF) 2745 return; 2746 2747 vsi->irqs_ready = false; 2748 ice_free_cpu_rx_rmap(vsi); 2749 2750 ice_for_each_q_vector(vsi, i) { 2751 int irq_num; 2752 2753 irq_num = vsi->q_vectors[i]->irq.virq; 2754 2755 /* free only the irqs that were actually requested */ 2756 if (!vsi->q_vectors[i] || 2757 !(vsi->q_vectors[i]->num_ring_tx || 2758 vsi->q_vectors[i]->num_ring_rx)) 2759 continue; 2760 2761 /* clear the affinity notifier in the IRQ descriptor */ 2762 if (!IS_ENABLED(CONFIG_RFS_ACCEL)) 2763 irq_set_affinity_notifier(irq_num, NULL); 2764 2765 /* clear the affinity_mask in the IRQ descriptor */ 2766 irq_set_affinity_hint(irq_num, NULL); 2767 synchronize_irq(irq_num); 2768 devm_free_irq(ice_pf_to_dev(pf), irq_num, vsi->q_vectors[i]); 2769 } 2770 } 2771 2772 /** 2773 * ice_vsi_free_tx_rings - Free Tx resources for VSI queues 2774 * @vsi: the VSI having resources freed 2775 */ 2776 void ice_vsi_free_tx_rings(struct ice_vsi *vsi) 2777 { 2778 int i; 2779 2780 if (!vsi->tx_rings) 2781 return; 2782 2783 ice_for_each_txq(vsi, i) 2784 if (vsi->tx_rings[i] && vsi->tx_rings[i]->desc) 2785 ice_free_tx_ring(vsi->tx_rings[i]); 2786 } 2787 2788 /** 2789 * ice_vsi_free_rx_rings - Free Rx resources for VSI queues 2790 * @vsi: the VSI having resources freed 2791 */ 2792 void ice_vsi_free_rx_rings(struct ice_vsi *vsi) 2793 { 2794 int i; 2795 2796 if (!vsi->rx_rings) 2797 return; 2798 2799 ice_for_each_rxq(vsi, i) 2800 if (vsi->rx_rings[i] && vsi->rx_rings[i]->desc) 2801 ice_free_rx_ring(vsi->rx_rings[i]); 2802 } 2803 2804 /** 2805 * ice_vsi_close - Shut down a VSI 2806 * @vsi: the VSI being shut down 2807 */ 2808 void ice_vsi_close(struct ice_vsi *vsi) 2809 { 2810 if (!test_and_set_bit(ICE_VSI_DOWN, vsi->state)) 2811 ice_down(vsi); 2812 2813 ice_vsi_free_irq(vsi); 2814 ice_vsi_free_tx_rings(vsi); 2815 ice_vsi_free_rx_rings(vsi); 2816 } 2817 2818 /** 2819 * ice_ena_vsi - resume a VSI 2820 * @vsi: the VSI being resume 2821 * @locked: is the rtnl_lock already held 2822 */ 2823 int ice_ena_vsi(struct ice_vsi *vsi, bool locked) 2824 { 2825 int err = 0; 2826 2827 if (!test_bit(ICE_VSI_NEEDS_RESTART, vsi->state)) 2828 return 0; 2829 2830 clear_bit(ICE_VSI_NEEDS_RESTART, vsi->state); 2831 2832 if (vsi->netdev && vsi->type == ICE_VSI_PF) { 2833 if (netif_running(vsi->netdev)) { 2834 if (!locked) 2835 rtnl_lock(); 2836 2837 err = ice_open_internal(vsi->netdev); 2838 2839 if (!locked) 2840 rtnl_unlock(); 2841 } 2842 } else if (vsi->type == ICE_VSI_CTRL) { 2843 err = ice_vsi_open_ctrl(vsi); 2844 } 2845 2846 return err; 2847 } 2848 2849 /** 2850 * ice_dis_vsi - pause a VSI 2851 * @vsi: the VSI being paused 2852 * @locked: is the rtnl_lock already held 2853 */ 2854 void ice_dis_vsi(struct ice_vsi *vsi, bool locked) 2855 { 2856 if (test_bit(ICE_VSI_DOWN, vsi->state)) 2857 return; 2858 2859 set_bit(ICE_VSI_NEEDS_RESTART, vsi->state); 2860 2861 if (vsi->type == ICE_VSI_PF && vsi->netdev) { 2862 if (netif_running(vsi->netdev)) { 2863 if (!locked) 2864 rtnl_lock(); 2865 2866 ice_vsi_close(vsi); 2867 2868 if (!locked) 2869 rtnl_unlock(); 2870 } else { 2871 ice_vsi_close(vsi); 2872 } 2873 } else if (vsi->type == ICE_VSI_CTRL || 2874 vsi->type == ICE_VSI_SWITCHDEV_CTRL) { 2875 ice_vsi_close(vsi); 2876 } 2877 } 2878 2879 /** 2880 * ice_vsi_dis_irq - Mask off queue interrupt generation on the VSI 2881 * @vsi: the VSI being un-configured 2882 */ 2883 void ice_vsi_dis_irq(struct ice_vsi *vsi) 2884 { 2885 struct ice_pf *pf = vsi->back; 2886 struct ice_hw *hw = &pf->hw; 2887 u32 val; 2888 int i; 2889 2890 /* disable interrupt causation from each queue */ 2891 if (vsi->tx_rings) { 2892 ice_for_each_txq(vsi, i) { 2893 if (vsi->tx_rings[i]) { 2894 u16 reg; 2895 2896 reg = vsi->tx_rings[i]->reg_idx; 2897 val = rd32(hw, QINT_TQCTL(reg)); 2898 val &= ~QINT_TQCTL_CAUSE_ENA_M; 2899 wr32(hw, QINT_TQCTL(reg), val); 2900 } 2901 } 2902 } 2903 2904 if (vsi->rx_rings) { 2905 ice_for_each_rxq(vsi, i) { 2906 if (vsi->rx_rings[i]) { 2907 u16 reg; 2908 2909 reg = vsi->rx_rings[i]->reg_idx; 2910 val = rd32(hw, QINT_RQCTL(reg)); 2911 val &= ~QINT_RQCTL_CAUSE_ENA_M; 2912 wr32(hw, QINT_RQCTL(reg), val); 2913 } 2914 } 2915 } 2916 2917 /* disable each interrupt */ 2918 ice_for_each_q_vector(vsi, i) { 2919 if (!vsi->q_vectors[i]) 2920 continue; 2921 wr32(hw, GLINT_DYN_CTL(vsi->q_vectors[i]->reg_idx), 0); 2922 } 2923 2924 ice_flush(hw); 2925 2926 /* don't call synchronize_irq() for VF's from the host */ 2927 if (vsi->type == ICE_VSI_VF) 2928 return; 2929 2930 ice_for_each_q_vector(vsi, i) 2931 synchronize_irq(vsi->q_vectors[i]->irq.virq); 2932 } 2933 2934 /** 2935 * ice_vsi_release - Delete a VSI and free its resources 2936 * @vsi: the VSI being removed 2937 * 2938 * Returns 0 on success or < 0 on error 2939 */ 2940 int ice_vsi_release(struct ice_vsi *vsi) 2941 { 2942 struct ice_pf *pf; 2943 2944 if (!vsi->back) 2945 return -ENODEV; 2946 pf = vsi->back; 2947 2948 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) 2949 ice_rss_clean(vsi); 2950 2951 ice_vsi_close(vsi); 2952 ice_vsi_decfg(vsi); 2953 2954 /* retain SW VSI data structure since it is needed to unregister and 2955 * free VSI netdev when PF is not in reset recovery pending state,\ 2956 * for ex: during rmmod. 2957 */ 2958 if (!ice_is_reset_in_progress(pf->state)) 2959 ice_vsi_delete(vsi); 2960 2961 return 0; 2962 } 2963 2964 /** 2965 * ice_vsi_rebuild_get_coalesce - get coalesce from all q_vectors 2966 * @vsi: VSI connected with q_vectors 2967 * @coalesce: array of struct with stored coalesce 2968 * 2969 * Returns array size. 2970 */ 2971 static int 2972 ice_vsi_rebuild_get_coalesce(struct ice_vsi *vsi, 2973 struct ice_coalesce_stored *coalesce) 2974 { 2975 int i; 2976 2977 ice_for_each_q_vector(vsi, i) { 2978 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 2979 2980 coalesce[i].itr_tx = q_vector->tx.itr_settings; 2981 coalesce[i].itr_rx = q_vector->rx.itr_settings; 2982 coalesce[i].intrl = q_vector->intrl; 2983 2984 if (i < vsi->num_txq) 2985 coalesce[i].tx_valid = true; 2986 if (i < vsi->num_rxq) 2987 coalesce[i].rx_valid = true; 2988 } 2989 2990 return vsi->num_q_vectors; 2991 } 2992 2993 /** 2994 * ice_vsi_rebuild_set_coalesce - set coalesce from earlier saved arrays 2995 * @vsi: VSI connected with q_vectors 2996 * @coalesce: pointer to array of struct with stored coalesce 2997 * @size: size of coalesce array 2998 * 2999 * Before this function, ice_vsi_rebuild_get_coalesce should be called to save 3000 * ITR params in arrays. If size is 0 or coalesce wasn't stored set coalesce 3001 * to default value. 3002 */ 3003 static void 3004 ice_vsi_rebuild_set_coalesce(struct ice_vsi *vsi, 3005 struct ice_coalesce_stored *coalesce, int size) 3006 { 3007 struct ice_ring_container *rc; 3008 int i; 3009 3010 if ((size && !coalesce) || !vsi) 3011 return; 3012 3013 /* There are a couple of cases that have to be handled here: 3014 * 1. The case where the number of queue vectors stays the same, but 3015 * the number of Tx or Rx rings changes (the first for loop) 3016 * 2. The case where the number of queue vectors increased (the 3017 * second for loop) 3018 */ 3019 for (i = 0; i < size && i < vsi->num_q_vectors; i++) { 3020 /* There are 2 cases to handle here and they are the same for 3021 * both Tx and Rx: 3022 * if the entry was valid previously (coalesce[i].[tr]x_valid 3023 * and the loop variable is less than the number of rings 3024 * allocated, then write the previous values 3025 * 3026 * if the entry was not valid previously, but the number of 3027 * rings is less than are allocated (this means the number of 3028 * rings increased from previously), then write out the 3029 * values in the first element 3030 * 3031 * Also, always write the ITR, even if in ITR_IS_DYNAMIC 3032 * as there is no harm because the dynamic algorithm 3033 * will just overwrite. 3034 */ 3035 if (i < vsi->alloc_rxq && coalesce[i].rx_valid) { 3036 rc = &vsi->q_vectors[i]->rx; 3037 rc->itr_settings = coalesce[i].itr_rx; 3038 ice_write_itr(rc, rc->itr_setting); 3039 } else if (i < vsi->alloc_rxq) { 3040 rc = &vsi->q_vectors[i]->rx; 3041 rc->itr_settings = coalesce[0].itr_rx; 3042 ice_write_itr(rc, rc->itr_setting); 3043 } 3044 3045 if (i < vsi->alloc_txq && coalesce[i].tx_valid) { 3046 rc = &vsi->q_vectors[i]->tx; 3047 rc->itr_settings = coalesce[i].itr_tx; 3048 ice_write_itr(rc, rc->itr_setting); 3049 } else if (i < vsi->alloc_txq) { 3050 rc = &vsi->q_vectors[i]->tx; 3051 rc->itr_settings = coalesce[0].itr_tx; 3052 ice_write_itr(rc, rc->itr_setting); 3053 } 3054 3055 vsi->q_vectors[i]->intrl = coalesce[i].intrl; 3056 ice_set_q_vector_intrl(vsi->q_vectors[i]); 3057 } 3058 3059 /* the number of queue vectors increased so write whatever is in 3060 * the first element 3061 */ 3062 for (; i < vsi->num_q_vectors; i++) { 3063 /* transmit */ 3064 rc = &vsi->q_vectors[i]->tx; 3065 rc->itr_settings = coalesce[0].itr_tx; 3066 ice_write_itr(rc, rc->itr_setting); 3067 3068 /* receive */ 3069 rc = &vsi->q_vectors[i]->rx; 3070 rc->itr_settings = coalesce[0].itr_rx; 3071 ice_write_itr(rc, rc->itr_setting); 3072 3073 vsi->q_vectors[i]->intrl = coalesce[0].intrl; 3074 ice_set_q_vector_intrl(vsi->q_vectors[i]); 3075 } 3076 } 3077 3078 /** 3079 * ice_vsi_realloc_stat_arrays - Frees unused stat structures or alloc new ones 3080 * @vsi: VSI pointer 3081 */ 3082 static int 3083 ice_vsi_realloc_stat_arrays(struct ice_vsi *vsi) 3084 { 3085 u16 req_txq = vsi->req_txq ? vsi->req_txq : vsi->alloc_txq; 3086 u16 req_rxq = vsi->req_rxq ? vsi->req_rxq : vsi->alloc_rxq; 3087 struct ice_ring_stats **tx_ring_stats; 3088 struct ice_ring_stats **rx_ring_stats; 3089 struct ice_vsi_stats *vsi_stat; 3090 struct ice_pf *pf = vsi->back; 3091 u16 prev_txq = vsi->alloc_txq; 3092 u16 prev_rxq = vsi->alloc_rxq; 3093 int i; 3094 3095 vsi_stat = pf->vsi_stats[vsi->idx]; 3096 3097 if (req_txq < prev_txq) { 3098 for (i = req_txq; i < prev_txq; i++) { 3099 if (vsi_stat->tx_ring_stats[i]) { 3100 kfree_rcu(vsi_stat->tx_ring_stats[i], rcu); 3101 WRITE_ONCE(vsi_stat->tx_ring_stats[i], NULL); 3102 } 3103 } 3104 } 3105 3106 tx_ring_stats = vsi_stat->rx_ring_stats; 3107 vsi_stat->tx_ring_stats = 3108 krealloc_array(vsi_stat->tx_ring_stats, req_txq, 3109 sizeof(*vsi_stat->tx_ring_stats), 3110 GFP_KERNEL | __GFP_ZERO); 3111 if (!vsi_stat->tx_ring_stats) { 3112 vsi_stat->tx_ring_stats = tx_ring_stats; 3113 return -ENOMEM; 3114 } 3115 3116 if (req_rxq < prev_rxq) { 3117 for (i = req_rxq; i < prev_rxq; i++) { 3118 if (vsi_stat->rx_ring_stats[i]) { 3119 kfree_rcu(vsi_stat->rx_ring_stats[i], rcu); 3120 WRITE_ONCE(vsi_stat->rx_ring_stats[i], NULL); 3121 } 3122 } 3123 } 3124 3125 rx_ring_stats = vsi_stat->rx_ring_stats; 3126 vsi_stat->rx_ring_stats = 3127 krealloc_array(vsi_stat->rx_ring_stats, req_rxq, 3128 sizeof(*vsi_stat->rx_ring_stats), 3129 GFP_KERNEL | __GFP_ZERO); 3130 if (!vsi_stat->rx_ring_stats) { 3131 vsi_stat->rx_ring_stats = rx_ring_stats; 3132 return -ENOMEM; 3133 } 3134 3135 return 0; 3136 } 3137 3138 /** 3139 * ice_vsi_rebuild - Rebuild VSI after reset 3140 * @vsi: VSI to be rebuild 3141 * @vsi_flags: flags used for VSI rebuild flow 3142 * 3143 * Set vsi_flags to ICE_VSI_FLAG_INIT to initialize a new VSI, or 3144 * ICE_VSI_FLAG_NO_INIT to rebuild an existing VSI in hardware. 3145 * 3146 * Returns 0 on success and negative value on failure 3147 */ 3148 int ice_vsi_rebuild(struct ice_vsi *vsi, u32 vsi_flags) 3149 { 3150 struct ice_vsi_cfg_params params = {}; 3151 struct ice_coalesce_stored *coalesce; 3152 int prev_num_q_vectors = 0; 3153 struct ice_pf *pf; 3154 int ret; 3155 3156 if (!vsi) 3157 return -EINVAL; 3158 3159 params = ice_vsi_to_params(vsi); 3160 params.flags = vsi_flags; 3161 3162 pf = vsi->back; 3163 if (WARN_ON(vsi->type == ICE_VSI_VF && !vsi->vf)) 3164 return -EINVAL; 3165 3166 coalesce = kcalloc(vsi->num_q_vectors, 3167 sizeof(struct ice_coalesce_stored), GFP_KERNEL); 3168 if (!coalesce) 3169 return -ENOMEM; 3170 3171 prev_num_q_vectors = ice_vsi_rebuild_get_coalesce(vsi, coalesce); 3172 3173 ret = ice_vsi_realloc_stat_arrays(vsi); 3174 if (ret) 3175 goto err_vsi_cfg; 3176 3177 ice_vsi_decfg(vsi); 3178 ret = ice_vsi_cfg_def(vsi, ¶ms); 3179 if (ret) 3180 goto err_vsi_cfg; 3181 3182 ret = ice_vsi_cfg_tc_lan(pf, vsi); 3183 if (ret) { 3184 if (vsi_flags & ICE_VSI_FLAG_INIT) { 3185 ret = -EIO; 3186 goto err_vsi_cfg_tc_lan; 3187 } 3188 3189 kfree(coalesce); 3190 return ice_schedule_reset(pf, ICE_RESET_PFR); 3191 } 3192 3193 ice_vsi_rebuild_set_coalesce(vsi, coalesce, prev_num_q_vectors); 3194 kfree(coalesce); 3195 3196 return 0; 3197 3198 err_vsi_cfg_tc_lan: 3199 ice_vsi_decfg(vsi); 3200 err_vsi_cfg: 3201 kfree(coalesce); 3202 return ret; 3203 } 3204 3205 /** 3206 * ice_is_reset_in_progress - check for a reset in progress 3207 * @state: PF state field 3208 */ 3209 bool ice_is_reset_in_progress(unsigned long *state) 3210 { 3211 return test_bit(ICE_RESET_OICR_RECV, state) || 3212 test_bit(ICE_PFR_REQ, state) || 3213 test_bit(ICE_CORER_REQ, state) || 3214 test_bit(ICE_GLOBR_REQ, state); 3215 } 3216 3217 /** 3218 * ice_wait_for_reset - Wait for driver to finish reset and rebuild 3219 * @pf: pointer to the PF structure 3220 * @timeout: length of time to wait, in jiffies 3221 * 3222 * Wait (sleep) for a short time until the driver finishes cleaning up from 3223 * a device reset. The caller must be able to sleep. Use this to delay 3224 * operations that could fail while the driver is cleaning up after a device 3225 * reset. 3226 * 3227 * Returns 0 on success, -EBUSY if the reset is not finished within the 3228 * timeout, and -ERESTARTSYS if the thread was interrupted. 3229 */ 3230 int ice_wait_for_reset(struct ice_pf *pf, unsigned long timeout) 3231 { 3232 long ret; 3233 3234 ret = wait_event_interruptible_timeout(pf->reset_wait_queue, 3235 !ice_is_reset_in_progress(pf->state), 3236 timeout); 3237 if (ret < 0) 3238 return ret; 3239 else if (!ret) 3240 return -EBUSY; 3241 else 3242 return 0; 3243 } 3244 3245 /** 3246 * ice_vsi_update_q_map - update our copy of the VSI info with new queue map 3247 * @vsi: VSI being configured 3248 * @ctx: the context buffer returned from AQ VSI update command 3249 */ 3250 static void ice_vsi_update_q_map(struct ice_vsi *vsi, struct ice_vsi_ctx *ctx) 3251 { 3252 vsi->info.mapping_flags = ctx->info.mapping_flags; 3253 memcpy(&vsi->info.q_mapping, &ctx->info.q_mapping, 3254 sizeof(vsi->info.q_mapping)); 3255 memcpy(&vsi->info.tc_mapping, ctx->info.tc_mapping, 3256 sizeof(vsi->info.tc_mapping)); 3257 } 3258 3259 /** 3260 * ice_vsi_cfg_netdev_tc - Setup the netdev TC configuration 3261 * @vsi: the VSI being configured 3262 * @ena_tc: TC map to be enabled 3263 */ 3264 void ice_vsi_cfg_netdev_tc(struct ice_vsi *vsi, u8 ena_tc) 3265 { 3266 struct net_device *netdev = vsi->netdev; 3267 struct ice_pf *pf = vsi->back; 3268 int numtc = vsi->tc_cfg.numtc; 3269 struct ice_dcbx_cfg *dcbcfg; 3270 u8 netdev_tc; 3271 int i; 3272 3273 if (!netdev) 3274 return; 3275 3276 /* CHNL VSI doesn't have it's own netdev, hence, no netdev_tc */ 3277 if (vsi->type == ICE_VSI_CHNL) 3278 return; 3279 3280 if (!ena_tc) { 3281 netdev_reset_tc(netdev); 3282 return; 3283 } 3284 3285 if (vsi->type == ICE_VSI_PF && ice_is_adq_active(pf)) 3286 numtc = vsi->all_numtc; 3287 3288 if (netdev_set_num_tc(netdev, numtc)) 3289 return; 3290 3291 dcbcfg = &pf->hw.port_info->qos_cfg.local_dcbx_cfg; 3292 3293 ice_for_each_traffic_class(i) 3294 if (vsi->tc_cfg.ena_tc & BIT(i)) 3295 netdev_set_tc_queue(netdev, 3296 vsi->tc_cfg.tc_info[i].netdev_tc, 3297 vsi->tc_cfg.tc_info[i].qcount_tx, 3298 vsi->tc_cfg.tc_info[i].qoffset); 3299 /* setup TC queue map for CHNL TCs */ 3300 ice_for_each_chnl_tc(i) { 3301 if (!(vsi->all_enatc & BIT(i))) 3302 break; 3303 if (!vsi->mqprio_qopt.qopt.count[i]) 3304 break; 3305 netdev_set_tc_queue(netdev, i, 3306 vsi->mqprio_qopt.qopt.count[i], 3307 vsi->mqprio_qopt.qopt.offset[i]); 3308 } 3309 3310 if (test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3311 return; 3312 3313 for (i = 0; i < ICE_MAX_USER_PRIORITY; i++) { 3314 u8 ets_tc = dcbcfg->etscfg.prio_table[i]; 3315 3316 /* Get the mapped netdev TC# for the UP */ 3317 netdev_tc = vsi->tc_cfg.tc_info[ets_tc].netdev_tc; 3318 netdev_set_prio_tc_map(netdev, i, netdev_tc); 3319 } 3320 } 3321 3322 /** 3323 * ice_vsi_setup_q_map_mqprio - Prepares mqprio based tc_config 3324 * @vsi: the VSI being configured, 3325 * @ctxt: VSI context structure 3326 * @ena_tc: number of traffic classes to enable 3327 * 3328 * Prepares VSI tc_config to have queue configurations based on MQPRIO options. 3329 */ 3330 static int 3331 ice_vsi_setup_q_map_mqprio(struct ice_vsi *vsi, struct ice_vsi_ctx *ctxt, 3332 u8 ena_tc) 3333 { 3334 u16 pow, offset = 0, qcount_tx = 0, qcount_rx = 0, qmap; 3335 u16 tc0_offset = vsi->mqprio_qopt.qopt.offset[0]; 3336 int tc0_qcount = vsi->mqprio_qopt.qopt.count[0]; 3337 u16 new_txq, new_rxq; 3338 u8 netdev_tc = 0; 3339 int i; 3340 3341 vsi->tc_cfg.ena_tc = ena_tc ? ena_tc : 1; 3342 3343 pow = order_base_2(tc0_qcount); 3344 qmap = ((tc0_offset << ICE_AQ_VSI_TC_Q_OFFSET_S) & 3345 ICE_AQ_VSI_TC_Q_OFFSET_M) | 3346 ((pow << ICE_AQ_VSI_TC_Q_NUM_S) & ICE_AQ_VSI_TC_Q_NUM_M); 3347 3348 ice_for_each_traffic_class(i) { 3349 if (!(vsi->tc_cfg.ena_tc & BIT(i))) { 3350 /* TC is not enabled */ 3351 vsi->tc_cfg.tc_info[i].qoffset = 0; 3352 vsi->tc_cfg.tc_info[i].qcount_rx = 1; 3353 vsi->tc_cfg.tc_info[i].qcount_tx = 1; 3354 vsi->tc_cfg.tc_info[i].netdev_tc = 0; 3355 ctxt->info.tc_mapping[i] = 0; 3356 continue; 3357 } 3358 3359 offset = vsi->mqprio_qopt.qopt.offset[i]; 3360 qcount_rx = vsi->mqprio_qopt.qopt.count[i]; 3361 qcount_tx = vsi->mqprio_qopt.qopt.count[i]; 3362 vsi->tc_cfg.tc_info[i].qoffset = offset; 3363 vsi->tc_cfg.tc_info[i].qcount_rx = qcount_rx; 3364 vsi->tc_cfg.tc_info[i].qcount_tx = qcount_tx; 3365 vsi->tc_cfg.tc_info[i].netdev_tc = netdev_tc++; 3366 } 3367 3368 if (vsi->all_numtc && vsi->all_numtc != vsi->tc_cfg.numtc) { 3369 ice_for_each_chnl_tc(i) { 3370 if (!(vsi->all_enatc & BIT(i))) 3371 continue; 3372 offset = vsi->mqprio_qopt.qopt.offset[i]; 3373 qcount_rx = vsi->mqprio_qopt.qopt.count[i]; 3374 qcount_tx = vsi->mqprio_qopt.qopt.count[i]; 3375 } 3376 } 3377 3378 new_txq = offset + qcount_tx; 3379 if (new_txq > vsi->alloc_txq) { 3380 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Tx queues (%u), than were allocated (%u)!\n", 3381 new_txq, vsi->alloc_txq); 3382 return -EINVAL; 3383 } 3384 3385 new_rxq = offset + qcount_rx; 3386 if (new_rxq > vsi->alloc_rxq) { 3387 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Rx queues (%u), than were allocated (%u)!\n", 3388 new_rxq, vsi->alloc_rxq); 3389 return -EINVAL; 3390 } 3391 3392 /* Set actual Tx/Rx queue pairs */ 3393 vsi->num_txq = new_txq; 3394 vsi->num_rxq = new_rxq; 3395 3396 /* Setup queue TC[0].qmap for given VSI context */ 3397 ctxt->info.tc_mapping[0] = cpu_to_le16(qmap); 3398 ctxt->info.q_mapping[0] = cpu_to_le16(vsi->rxq_map[0]); 3399 ctxt->info.q_mapping[1] = cpu_to_le16(tc0_qcount); 3400 3401 /* Find queue count available for channel VSIs and starting offset 3402 * for channel VSIs 3403 */ 3404 if (tc0_qcount && tc0_qcount < vsi->num_rxq) { 3405 vsi->cnt_q_avail = vsi->num_rxq - tc0_qcount; 3406 vsi->next_base_q = tc0_qcount; 3407 } 3408 dev_dbg(ice_pf_to_dev(vsi->back), "vsi->num_txq = %d\n", vsi->num_txq); 3409 dev_dbg(ice_pf_to_dev(vsi->back), "vsi->num_rxq = %d\n", vsi->num_rxq); 3410 dev_dbg(ice_pf_to_dev(vsi->back), "all_numtc %u, all_enatc: 0x%04x, tc_cfg.numtc %u\n", 3411 vsi->all_numtc, vsi->all_enatc, vsi->tc_cfg.numtc); 3412 3413 return 0; 3414 } 3415 3416 /** 3417 * ice_vsi_cfg_tc - Configure VSI Tx Sched for given TC map 3418 * @vsi: VSI to be configured 3419 * @ena_tc: TC bitmap 3420 * 3421 * VSI queues expected to be quiesced before calling this function 3422 */ 3423 int ice_vsi_cfg_tc(struct ice_vsi *vsi, u8 ena_tc) 3424 { 3425 u16 max_txqs[ICE_MAX_TRAFFIC_CLASS] = { 0 }; 3426 struct ice_pf *pf = vsi->back; 3427 struct ice_tc_cfg old_tc_cfg; 3428 struct ice_vsi_ctx *ctx; 3429 struct device *dev; 3430 int i, ret = 0; 3431 u8 num_tc = 0; 3432 3433 dev = ice_pf_to_dev(pf); 3434 if (vsi->tc_cfg.ena_tc == ena_tc && 3435 vsi->mqprio_qopt.mode != TC_MQPRIO_MODE_CHANNEL) 3436 return 0; 3437 3438 ice_for_each_traffic_class(i) { 3439 /* build bitmap of enabled TCs */ 3440 if (ena_tc & BIT(i)) 3441 num_tc++; 3442 /* populate max_txqs per TC */ 3443 max_txqs[i] = vsi->alloc_txq; 3444 /* Update max_txqs if it is CHNL VSI, because alloc_t[r]xq are 3445 * zero for CHNL VSI, hence use num_txq instead as max_txqs 3446 */ 3447 if (vsi->type == ICE_VSI_CHNL && 3448 test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3449 max_txqs[i] = vsi->num_txq; 3450 } 3451 3452 memcpy(&old_tc_cfg, &vsi->tc_cfg, sizeof(old_tc_cfg)); 3453 vsi->tc_cfg.ena_tc = ena_tc; 3454 vsi->tc_cfg.numtc = num_tc; 3455 3456 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); 3457 if (!ctx) 3458 return -ENOMEM; 3459 3460 ctx->vf_num = 0; 3461 ctx->info = vsi->info; 3462 3463 if (vsi->type == ICE_VSI_PF && 3464 test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3465 ret = ice_vsi_setup_q_map_mqprio(vsi, ctx, ena_tc); 3466 else 3467 ret = ice_vsi_setup_q_map(vsi, ctx); 3468 3469 if (ret) { 3470 memcpy(&vsi->tc_cfg, &old_tc_cfg, sizeof(vsi->tc_cfg)); 3471 goto out; 3472 } 3473 3474 /* must to indicate which section of VSI context are being modified */ 3475 ctx->info.valid_sections = cpu_to_le16(ICE_AQ_VSI_PROP_RXQ_MAP_VALID); 3476 ret = ice_update_vsi(&pf->hw, vsi->idx, ctx, NULL); 3477 if (ret) { 3478 dev_info(dev, "Failed VSI Update\n"); 3479 goto out; 3480 } 3481 3482 if (vsi->type == ICE_VSI_PF && 3483 test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3484 ret = ice_cfg_vsi_lan(vsi->port_info, vsi->idx, 1, max_txqs); 3485 else 3486 ret = ice_cfg_vsi_lan(vsi->port_info, vsi->idx, 3487 vsi->tc_cfg.ena_tc, max_txqs); 3488 3489 if (ret) { 3490 dev_err(dev, "VSI %d failed TC config, error %d\n", 3491 vsi->vsi_num, ret); 3492 goto out; 3493 } 3494 ice_vsi_update_q_map(vsi, ctx); 3495 vsi->info.valid_sections = 0; 3496 3497 ice_vsi_cfg_netdev_tc(vsi, ena_tc); 3498 out: 3499 kfree(ctx); 3500 return ret; 3501 } 3502 3503 /** 3504 * ice_update_ring_stats - Update ring statistics 3505 * @stats: stats to be updated 3506 * @pkts: number of processed packets 3507 * @bytes: number of processed bytes 3508 * 3509 * This function assumes that caller has acquired a u64_stats_sync lock. 3510 */ 3511 static void ice_update_ring_stats(struct ice_q_stats *stats, u64 pkts, u64 bytes) 3512 { 3513 stats->bytes += bytes; 3514 stats->pkts += pkts; 3515 } 3516 3517 /** 3518 * ice_update_tx_ring_stats - Update Tx ring specific counters 3519 * @tx_ring: ring to update 3520 * @pkts: number of processed packets 3521 * @bytes: number of processed bytes 3522 */ 3523 void ice_update_tx_ring_stats(struct ice_tx_ring *tx_ring, u64 pkts, u64 bytes) 3524 { 3525 u64_stats_update_begin(&tx_ring->ring_stats->syncp); 3526 ice_update_ring_stats(&tx_ring->ring_stats->stats, pkts, bytes); 3527 u64_stats_update_end(&tx_ring->ring_stats->syncp); 3528 } 3529 3530 /** 3531 * ice_update_rx_ring_stats - Update Rx ring specific counters 3532 * @rx_ring: ring to update 3533 * @pkts: number of processed packets 3534 * @bytes: number of processed bytes 3535 */ 3536 void ice_update_rx_ring_stats(struct ice_rx_ring *rx_ring, u64 pkts, u64 bytes) 3537 { 3538 u64_stats_update_begin(&rx_ring->ring_stats->syncp); 3539 ice_update_ring_stats(&rx_ring->ring_stats->stats, pkts, bytes); 3540 u64_stats_update_end(&rx_ring->ring_stats->syncp); 3541 } 3542 3543 /** 3544 * ice_is_dflt_vsi_in_use - check if the default forwarding VSI is being used 3545 * @pi: port info of the switch with default VSI 3546 * 3547 * Return true if the there is a single VSI in default forwarding VSI list 3548 */ 3549 bool ice_is_dflt_vsi_in_use(struct ice_port_info *pi) 3550 { 3551 bool exists = false; 3552 3553 ice_check_if_dflt_vsi(pi, 0, &exists); 3554 return exists; 3555 } 3556 3557 /** 3558 * ice_is_vsi_dflt_vsi - check if the VSI passed in is the default VSI 3559 * @vsi: VSI to compare against default forwarding VSI 3560 * 3561 * If this VSI passed in is the default forwarding VSI then return true, else 3562 * return false 3563 */ 3564 bool ice_is_vsi_dflt_vsi(struct ice_vsi *vsi) 3565 { 3566 return ice_check_if_dflt_vsi(vsi->port_info, vsi->idx, NULL); 3567 } 3568 3569 /** 3570 * ice_set_dflt_vsi - set the default forwarding VSI 3571 * @vsi: VSI getting set as the default forwarding VSI on the switch 3572 * 3573 * If the VSI passed in is already the default VSI and it's enabled just return 3574 * success. 3575 * 3576 * Otherwise try to set the VSI passed in as the switch's default VSI and 3577 * return the result. 3578 */ 3579 int ice_set_dflt_vsi(struct ice_vsi *vsi) 3580 { 3581 struct device *dev; 3582 int status; 3583 3584 if (!vsi) 3585 return -EINVAL; 3586 3587 dev = ice_pf_to_dev(vsi->back); 3588 3589 if (ice_lag_is_switchdev_running(vsi->back)) { 3590 dev_dbg(dev, "VSI %d passed is a part of LAG containing interfaces in switchdev mode, nothing to do\n", 3591 vsi->vsi_num); 3592 return 0; 3593 } 3594 3595 /* the VSI passed in is already the default VSI */ 3596 if (ice_is_vsi_dflt_vsi(vsi)) { 3597 dev_dbg(dev, "VSI %d passed in is already the default forwarding VSI, nothing to do\n", 3598 vsi->vsi_num); 3599 return 0; 3600 } 3601 3602 status = ice_cfg_dflt_vsi(vsi->port_info, vsi->idx, true, ICE_FLTR_RX); 3603 if (status) { 3604 dev_err(dev, "Failed to set VSI %d as the default forwarding VSI, error %d\n", 3605 vsi->vsi_num, status); 3606 return status; 3607 } 3608 3609 return 0; 3610 } 3611 3612 /** 3613 * ice_clear_dflt_vsi - clear the default forwarding VSI 3614 * @vsi: VSI to remove from filter list 3615 * 3616 * If the switch has no default VSI or it's not enabled then return error. 3617 * 3618 * Otherwise try to clear the default VSI and return the result. 3619 */ 3620 int ice_clear_dflt_vsi(struct ice_vsi *vsi) 3621 { 3622 struct device *dev; 3623 int status; 3624 3625 if (!vsi) 3626 return -EINVAL; 3627 3628 dev = ice_pf_to_dev(vsi->back); 3629 3630 /* there is no default VSI configured */ 3631 if (!ice_is_dflt_vsi_in_use(vsi->port_info)) 3632 return -ENODEV; 3633 3634 status = ice_cfg_dflt_vsi(vsi->port_info, vsi->idx, false, 3635 ICE_FLTR_RX); 3636 if (status) { 3637 dev_err(dev, "Failed to clear the default forwarding VSI %d, error %d\n", 3638 vsi->vsi_num, status); 3639 return -EIO; 3640 } 3641 3642 return 0; 3643 } 3644 3645 /** 3646 * ice_get_link_speed_mbps - get link speed in Mbps 3647 * @vsi: the VSI whose link speed is being queried 3648 * 3649 * Return current VSI link speed and 0 if the speed is unknown. 3650 */ 3651 int ice_get_link_speed_mbps(struct ice_vsi *vsi) 3652 { 3653 unsigned int link_speed; 3654 3655 link_speed = vsi->port_info->phy.link_info.link_speed; 3656 3657 return (int)ice_get_link_speed(fls(link_speed) - 1); 3658 } 3659 3660 /** 3661 * ice_get_link_speed_kbps - get link speed in Kbps 3662 * @vsi: the VSI whose link speed is being queried 3663 * 3664 * Return current VSI link speed and 0 if the speed is unknown. 3665 */ 3666 int ice_get_link_speed_kbps(struct ice_vsi *vsi) 3667 { 3668 int speed_mbps; 3669 3670 speed_mbps = ice_get_link_speed_mbps(vsi); 3671 3672 return speed_mbps * 1000; 3673 } 3674 3675 /** 3676 * ice_set_min_bw_limit - setup minimum BW limit for Tx based on min_tx_rate 3677 * @vsi: VSI to be configured 3678 * @min_tx_rate: min Tx rate in Kbps to be configured as BW limit 3679 * 3680 * If the min_tx_rate is specified as 0 that means to clear the minimum BW limit 3681 * profile, otherwise a non-zero value will force a minimum BW limit for the VSI 3682 * on TC 0. 3683 */ 3684 int ice_set_min_bw_limit(struct ice_vsi *vsi, u64 min_tx_rate) 3685 { 3686 struct ice_pf *pf = vsi->back; 3687 struct device *dev; 3688 int status; 3689 int speed; 3690 3691 dev = ice_pf_to_dev(pf); 3692 if (!vsi->port_info) { 3693 dev_dbg(dev, "VSI %d, type %u specified doesn't have valid port_info\n", 3694 vsi->idx, vsi->type); 3695 return -EINVAL; 3696 } 3697 3698 speed = ice_get_link_speed_kbps(vsi); 3699 if (min_tx_rate > (u64)speed) { 3700 dev_err(dev, "invalid min Tx rate %llu Kbps specified for %s %d is greater than current link speed %u Kbps\n", 3701 min_tx_rate, ice_vsi_type_str(vsi->type), vsi->idx, 3702 speed); 3703 return -EINVAL; 3704 } 3705 3706 /* Configure min BW for VSI limit */ 3707 if (min_tx_rate) { 3708 status = ice_cfg_vsi_bw_lmt_per_tc(vsi->port_info, vsi->idx, 0, 3709 ICE_MIN_BW, min_tx_rate); 3710 if (status) { 3711 dev_err(dev, "failed to set min Tx rate(%llu Kbps) for %s %d\n", 3712 min_tx_rate, ice_vsi_type_str(vsi->type), 3713 vsi->idx); 3714 return status; 3715 } 3716 3717 dev_dbg(dev, "set min Tx rate(%llu Kbps) for %s\n", 3718 min_tx_rate, ice_vsi_type_str(vsi->type)); 3719 } else { 3720 status = ice_cfg_vsi_bw_dflt_lmt_per_tc(vsi->port_info, 3721 vsi->idx, 0, 3722 ICE_MIN_BW); 3723 if (status) { 3724 dev_err(dev, "failed to clear min Tx rate configuration for %s %d\n", 3725 ice_vsi_type_str(vsi->type), vsi->idx); 3726 return status; 3727 } 3728 3729 dev_dbg(dev, "cleared min Tx rate configuration for %s %d\n", 3730 ice_vsi_type_str(vsi->type), vsi->idx); 3731 } 3732 3733 return 0; 3734 } 3735 3736 /** 3737 * ice_set_max_bw_limit - setup maximum BW limit for Tx based on max_tx_rate 3738 * @vsi: VSI to be configured 3739 * @max_tx_rate: max Tx rate in Kbps to be configured as BW limit 3740 * 3741 * If the max_tx_rate is specified as 0 that means to clear the maximum BW limit 3742 * profile, otherwise a non-zero value will force a maximum BW limit for the VSI 3743 * on TC 0. 3744 */ 3745 int ice_set_max_bw_limit(struct ice_vsi *vsi, u64 max_tx_rate) 3746 { 3747 struct ice_pf *pf = vsi->back; 3748 struct device *dev; 3749 int status; 3750 int speed; 3751 3752 dev = ice_pf_to_dev(pf); 3753 if (!vsi->port_info) { 3754 dev_dbg(dev, "VSI %d, type %u specified doesn't have valid port_info\n", 3755 vsi->idx, vsi->type); 3756 return -EINVAL; 3757 } 3758 3759 speed = ice_get_link_speed_kbps(vsi); 3760 if (max_tx_rate > (u64)speed) { 3761 dev_err(dev, "invalid max Tx rate %llu Kbps specified for %s %d is greater than current link speed %u Kbps\n", 3762 max_tx_rate, ice_vsi_type_str(vsi->type), vsi->idx, 3763 speed); 3764 return -EINVAL; 3765 } 3766 3767 /* Configure max BW for VSI limit */ 3768 if (max_tx_rate) { 3769 status = ice_cfg_vsi_bw_lmt_per_tc(vsi->port_info, vsi->idx, 0, 3770 ICE_MAX_BW, max_tx_rate); 3771 if (status) { 3772 dev_err(dev, "failed setting max Tx rate(%llu Kbps) for %s %d\n", 3773 max_tx_rate, ice_vsi_type_str(vsi->type), 3774 vsi->idx); 3775 return status; 3776 } 3777 3778 dev_dbg(dev, "set max Tx rate(%llu Kbps) for %s %d\n", 3779 max_tx_rate, ice_vsi_type_str(vsi->type), vsi->idx); 3780 } else { 3781 status = ice_cfg_vsi_bw_dflt_lmt_per_tc(vsi->port_info, 3782 vsi->idx, 0, 3783 ICE_MAX_BW); 3784 if (status) { 3785 dev_err(dev, "failed clearing max Tx rate configuration for %s %d\n", 3786 ice_vsi_type_str(vsi->type), vsi->idx); 3787 return status; 3788 } 3789 3790 dev_dbg(dev, "cleared max Tx rate configuration for %s %d\n", 3791 ice_vsi_type_str(vsi->type), vsi->idx); 3792 } 3793 3794 return 0; 3795 } 3796 3797 /** 3798 * ice_set_link - turn on/off physical link 3799 * @vsi: VSI to modify physical link on 3800 * @ena: turn on/off physical link 3801 */ 3802 int ice_set_link(struct ice_vsi *vsi, bool ena) 3803 { 3804 struct device *dev = ice_pf_to_dev(vsi->back); 3805 struct ice_port_info *pi = vsi->port_info; 3806 struct ice_hw *hw = pi->hw; 3807 int status; 3808 3809 if (vsi->type != ICE_VSI_PF) 3810 return -EINVAL; 3811 3812 status = ice_aq_set_link_restart_an(pi, ena, NULL); 3813 3814 /* if link is owned by manageability, FW will return ICE_AQ_RC_EMODE. 3815 * this is not a fatal error, so print a warning message and return 3816 * a success code. Return an error if FW returns an error code other 3817 * than ICE_AQ_RC_EMODE 3818 */ 3819 if (status == -EIO) { 3820 if (hw->adminq.sq_last_status == ICE_AQ_RC_EMODE) 3821 dev_dbg(dev, "can't set link to %s, err %d aq_err %s. not fatal, continuing\n", 3822 (ena ? "ON" : "OFF"), status, 3823 ice_aq_str(hw->adminq.sq_last_status)); 3824 } else if (status) { 3825 dev_err(dev, "can't set link to %s, err %d aq_err %s\n", 3826 (ena ? "ON" : "OFF"), status, 3827 ice_aq_str(hw->adminq.sq_last_status)); 3828 return status; 3829 } 3830 3831 return 0; 3832 } 3833 3834 /** 3835 * ice_vsi_add_vlan_zero - add VLAN 0 filter(s) for this VSI 3836 * @vsi: VSI used to add VLAN filters 3837 * 3838 * In Single VLAN Mode (SVM), single VLAN filters via ICE_SW_LKUP_VLAN are based 3839 * on the inner VLAN ID, so the VLAN TPID (i.e. 0x8100 or 0x888a8) doesn't 3840 * matter. In Double VLAN Mode (DVM), outer/single VLAN filters via 3841 * ICE_SW_LKUP_VLAN are based on the outer/single VLAN ID + VLAN TPID. 3842 * 3843 * For both modes add a VLAN 0 + no VLAN TPID filter to handle untagged traffic 3844 * when VLAN pruning is enabled. Also, this handles VLAN 0 priority tagged 3845 * traffic in SVM, since the VLAN TPID isn't part of filtering. 3846 * 3847 * If DVM is enabled then an explicit VLAN 0 + VLAN TPID filter needs to be 3848 * added to allow VLAN 0 priority tagged traffic in DVM, since the VLAN TPID is 3849 * part of filtering. 3850 */ 3851 int ice_vsi_add_vlan_zero(struct ice_vsi *vsi) 3852 { 3853 struct ice_vsi_vlan_ops *vlan_ops = ice_get_compat_vsi_vlan_ops(vsi); 3854 struct ice_vlan vlan; 3855 int err; 3856 3857 vlan = ICE_VLAN(0, 0, 0); 3858 err = vlan_ops->add_vlan(vsi, &vlan); 3859 if (err && err != -EEXIST) 3860 return err; 3861 3862 /* in SVM both VLAN 0 filters are identical */ 3863 if (!ice_is_dvm_ena(&vsi->back->hw)) 3864 return 0; 3865 3866 vlan = ICE_VLAN(ETH_P_8021Q, 0, 0); 3867 err = vlan_ops->add_vlan(vsi, &vlan); 3868 if (err && err != -EEXIST) 3869 return err; 3870 3871 return 0; 3872 } 3873 3874 /** 3875 * ice_vsi_del_vlan_zero - delete VLAN 0 filter(s) for this VSI 3876 * @vsi: VSI used to add VLAN filters 3877 * 3878 * Delete the VLAN 0 filters in the same manner that they were added in 3879 * ice_vsi_add_vlan_zero. 3880 */ 3881 int ice_vsi_del_vlan_zero(struct ice_vsi *vsi) 3882 { 3883 struct ice_vsi_vlan_ops *vlan_ops = ice_get_compat_vsi_vlan_ops(vsi); 3884 struct ice_vlan vlan; 3885 int err; 3886 3887 vlan = ICE_VLAN(0, 0, 0); 3888 err = vlan_ops->del_vlan(vsi, &vlan); 3889 if (err && err != -EEXIST) 3890 return err; 3891 3892 /* in SVM both VLAN 0 filters are identical */ 3893 if (!ice_is_dvm_ena(&vsi->back->hw)) 3894 return 0; 3895 3896 vlan = ICE_VLAN(ETH_P_8021Q, 0, 0); 3897 err = vlan_ops->del_vlan(vsi, &vlan); 3898 if (err && err != -EEXIST) 3899 return err; 3900 3901 /* when deleting the last VLAN filter, make sure to disable the VLAN 3902 * promisc mode so the filter isn't left by accident 3903 */ 3904 return ice_clear_vsi_promisc(&vsi->back->hw, vsi->idx, 3905 ICE_MCAST_VLAN_PROMISC_BITS, 0); 3906 } 3907 3908 /** 3909 * ice_vsi_num_zero_vlans - get number of VLAN 0 filters based on VLAN mode 3910 * @vsi: VSI used to get the VLAN mode 3911 * 3912 * If DVM is enabled then 2 VLAN 0 filters are added, else if SVM is enabled 3913 * then 1 VLAN 0 filter is added. See ice_vsi_add_vlan_zero for more details. 3914 */ 3915 static u16 ice_vsi_num_zero_vlans(struct ice_vsi *vsi) 3916 { 3917 #define ICE_DVM_NUM_ZERO_VLAN_FLTRS 2 3918 #define ICE_SVM_NUM_ZERO_VLAN_FLTRS 1 3919 /* no VLAN 0 filter is created when a port VLAN is active */ 3920 if (vsi->type == ICE_VSI_VF) { 3921 if (WARN_ON(!vsi->vf)) 3922 return 0; 3923 3924 if (ice_vf_is_port_vlan_ena(vsi->vf)) 3925 return 0; 3926 } 3927 3928 if (ice_is_dvm_ena(&vsi->back->hw)) 3929 return ICE_DVM_NUM_ZERO_VLAN_FLTRS; 3930 else 3931 return ICE_SVM_NUM_ZERO_VLAN_FLTRS; 3932 } 3933 3934 /** 3935 * ice_vsi_has_non_zero_vlans - check if VSI has any non-zero VLANs 3936 * @vsi: VSI used to determine if any non-zero VLANs have been added 3937 */ 3938 bool ice_vsi_has_non_zero_vlans(struct ice_vsi *vsi) 3939 { 3940 return (vsi->num_vlan > ice_vsi_num_zero_vlans(vsi)); 3941 } 3942 3943 /** 3944 * ice_vsi_num_non_zero_vlans - get the number of non-zero VLANs for this VSI 3945 * @vsi: VSI used to get the number of non-zero VLANs added 3946 */ 3947 u16 ice_vsi_num_non_zero_vlans(struct ice_vsi *vsi) 3948 { 3949 return (vsi->num_vlan - ice_vsi_num_zero_vlans(vsi)); 3950 } 3951 3952 /** 3953 * ice_is_feature_supported 3954 * @pf: pointer to the struct ice_pf instance 3955 * @f: feature enum to be checked 3956 * 3957 * returns true if feature is supported, false otherwise 3958 */ 3959 bool ice_is_feature_supported(struct ice_pf *pf, enum ice_feature f) 3960 { 3961 if (f < 0 || f >= ICE_F_MAX) 3962 return false; 3963 3964 return test_bit(f, pf->features); 3965 } 3966 3967 /** 3968 * ice_set_feature_support 3969 * @pf: pointer to the struct ice_pf instance 3970 * @f: feature enum to set 3971 */ 3972 void ice_set_feature_support(struct ice_pf *pf, enum ice_feature f) 3973 { 3974 if (f < 0 || f >= ICE_F_MAX) 3975 return; 3976 3977 set_bit(f, pf->features); 3978 } 3979 3980 /** 3981 * ice_clear_feature_support 3982 * @pf: pointer to the struct ice_pf instance 3983 * @f: feature enum to clear 3984 */ 3985 void ice_clear_feature_support(struct ice_pf *pf, enum ice_feature f) 3986 { 3987 if (f < 0 || f >= ICE_F_MAX) 3988 return; 3989 3990 clear_bit(f, pf->features); 3991 } 3992 3993 /** 3994 * ice_init_feature_support 3995 * @pf: pointer to the struct ice_pf instance 3996 * 3997 * called during init to setup supported feature 3998 */ 3999 void ice_init_feature_support(struct ice_pf *pf) 4000 { 4001 switch (pf->hw.device_id) { 4002 case ICE_DEV_ID_E810C_BACKPLANE: 4003 case ICE_DEV_ID_E810C_QSFP: 4004 case ICE_DEV_ID_E810C_SFP: 4005 case ICE_DEV_ID_E810_XXV_BACKPLANE: 4006 case ICE_DEV_ID_E810_XXV_QSFP: 4007 case ICE_DEV_ID_E810_XXV_SFP: 4008 ice_set_feature_support(pf, ICE_F_DSCP); 4009 if (ice_is_phy_rclk_in_netlist(&pf->hw)) 4010 ice_set_feature_support(pf, ICE_F_PHY_RCLK); 4011 /* If we don't own the timer - don't enable other caps */ 4012 if (!ice_pf_src_tmr_owned(pf)) 4013 break; 4014 if (ice_is_cgu_in_netlist(&pf->hw)) 4015 ice_set_feature_support(pf, ICE_F_CGU); 4016 if (ice_is_clock_mux_in_netlist(&pf->hw)) 4017 ice_set_feature_support(pf, ICE_F_SMA_CTRL); 4018 if (ice_gnss_is_gps_present(&pf->hw)) 4019 ice_set_feature_support(pf, ICE_F_GNSS); 4020 break; 4021 default: 4022 break; 4023 } 4024 } 4025 4026 /** 4027 * ice_vsi_update_security - update security block in VSI 4028 * @vsi: pointer to VSI structure 4029 * @fill: function pointer to fill ctx 4030 */ 4031 int 4032 ice_vsi_update_security(struct ice_vsi *vsi, void (*fill)(struct ice_vsi_ctx *)) 4033 { 4034 struct ice_vsi_ctx ctx = { 0 }; 4035 4036 ctx.info = vsi->info; 4037 ctx.info.valid_sections = cpu_to_le16(ICE_AQ_VSI_PROP_SECURITY_VALID); 4038 fill(&ctx); 4039 4040 if (ice_update_vsi(&vsi->back->hw, vsi->idx, &ctx, NULL)) 4041 return -ENODEV; 4042 4043 vsi->info = ctx.info; 4044 return 0; 4045 } 4046 4047 /** 4048 * ice_vsi_ctx_set_antispoof - set antispoof function in VSI ctx 4049 * @ctx: pointer to VSI ctx structure 4050 */ 4051 void ice_vsi_ctx_set_antispoof(struct ice_vsi_ctx *ctx) 4052 { 4053 ctx->info.sec_flags |= ICE_AQ_VSI_SEC_FLAG_ENA_MAC_ANTI_SPOOF | 4054 (ICE_AQ_VSI_SEC_TX_VLAN_PRUNE_ENA << 4055 ICE_AQ_VSI_SEC_TX_PRUNE_ENA_S); 4056 } 4057 4058 /** 4059 * ice_vsi_ctx_clear_antispoof - clear antispoof function in VSI ctx 4060 * @ctx: pointer to VSI ctx structure 4061 */ 4062 void ice_vsi_ctx_clear_antispoof(struct ice_vsi_ctx *ctx) 4063 { 4064 ctx->info.sec_flags &= ~ICE_AQ_VSI_SEC_FLAG_ENA_MAC_ANTI_SPOOF & 4065 ~(ICE_AQ_VSI_SEC_TX_VLAN_PRUNE_ENA << 4066 ICE_AQ_VSI_SEC_TX_PRUNE_ENA_S); 4067 } 4068 4069 /** 4070 * ice_vsi_ctx_set_allow_override - allow destination override on VSI 4071 * @ctx: pointer to VSI ctx structure 4072 */ 4073 void ice_vsi_ctx_set_allow_override(struct ice_vsi_ctx *ctx) 4074 { 4075 ctx->info.sec_flags |= ICE_AQ_VSI_SEC_FLAG_ALLOW_DEST_OVRD; 4076 } 4077 4078 /** 4079 * ice_vsi_ctx_clear_allow_override - turn off destination override on VSI 4080 * @ctx: pointer to VSI ctx structure 4081 */ 4082 void ice_vsi_ctx_clear_allow_override(struct ice_vsi_ctx *ctx) 4083 { 4084 ctx->info.sec_flags &= ~ICE_AQ_VSI_SEC_FLAG_ALLOW_DEST_OVRD; 4085 } 4086 4087 /** 4088 * ice_vsi_update_local_lb - update sw block in VSI with local loopback bit 4089 * @vsi: pointer to VSI structure 4090 * @set: set or unset the bit 4091 */ 4092 int 4093 ice_vsi_update_local_lb(struct ice_vsi *vsi, bool set) 4094 { 4095 struct ice_vsi_ctx ctx = { 4096 .info = vsi->info, 4097 }; 4098 4099 ctx.info.valid_sections = cpu_to_le16(ICE_AQ_VSI_PROP_SW_VALID); 4100 if (set) 4101 ctx.info.sw_flags |= ICE_AQ_VSI_SW_FLAG_LOCAL_LB; 4102 else 4103 ctx.info.sw_flags &= ~ICE_AQ_VSI_SW_FLAG_LOCAL_LB; 4104 4105 if (ice_update_vsi(&vsi->back->hw, vsi->idx, &ctx, NULL)) 4106 return -ENODEV; 4107 4108 vsi->info = ctx.info; 4109 return 0; 4110 } 4111