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 VFs. 216 * Each ring is associated to the corresponding VF_PR netdev. 217 */ 218 vsi->alloc_txq = ice_get_num_vfs(pf); 219 vsi->alloc_rxq = vsi->alloc_txq; 220 vsi->num_q_vectors = 1; 221 break; 222 case ICE_VSI_VF: 223 if (vf->num_req_qs) 224 vf->num_vf_qs = vf->num_req_qs; 225 vsi->alloc_txq = vf->num_vf_qs; 226 vsi->alloc_rxq = vf->num_vf_qs; 227 /* pf->vfs.num_msix_per includes (VF miscellaneous vector + 228 * data queue interrupts). Since vsi->num_q_vectors is number 229 * of queues vectors, subtract 1 (ICE_NONQ_VECS_VF) from the 230 * original vector count 231 */ 232 vsi->num_q_vectors = pf->vfs.num_msix_per - ICE_NONQ_VECS_VF; 233 break; 234 case ICE_VSI_CTRL: 235 vsi->alloc_txq = 1; 236 vsi->alloc_rxq = 1; 237 vsi->num_q_vectors = 1; 238 break; 239 case ICE_VSI_CHNL: 240 vsi->alloc_txq = 0; 241 vsi->alloc_rxq = 0; 242 break; 243 case ICE_VSI_LB: 244 vsi->alloc_txq = 1; 245 vsi->alloc_rxq = 1; 246 break; 247 default: 248 dev_warn(ice_pf_to_dev(pf), "Unknown VSI type %d\n", vsi_type); 249 break; 250 } 251 252 ice_vsi_set_num_desc(vsi); 253 } 254 255 /** 256 * ice_get_free_slot - get the next non-NULL location index in array 257 * @array: array to search 258 * @size: size of the array 259 * @curr: last known occupied index to be used as a search hint 260 * 261 * void * is being used to keep the functionality generic. This lets us use this 262 * function on any array of pointers. 263 */ 264 static int ice_get_free_slot(void *array, int size, int curr) 265 { 266 int **tmp_array = (int **)array; 267 int next; 268 269 if (curr < (size - 1) && !tmp_array[curr + 1]) { 270 next = curr + 1; 271 } else { 272 int i = 0; 273 274 while ((i < size) && (tmp_array[i])) 275 i++; 276 if (i == size) 277 next = ICE_NO_VSI; 278 else 279 next = i; 280 } 281 return next; 282 } 283 284 /** 285 * ice_vsi_delete_from_hw - delete a VSI from the switch 286 * @vsi: pointer to VSI being removed 287 */ 288 static void ice_vsi_delete_from_hw(struct ice_vsi *vsi) 289 { 290 struct ice_pf *pf = vsi->back; 291 struct ice_vsi_ctx *ctxt; 292 int status; 293 294 ice_fltr_remove_all(vsi); 295 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); 296 if (!ctxt) 297 return; 298 299 if (vsi->type == ICE_VSI_VF) 300 ctxt->vf_num = vsi->vf->vf_id; 301 ctxt->vsi_num = vsi->vsi_num; 302 303 memcpy(&ctxt->info, &vsi->info, sizeof(ctxt->info)); 304 305 status = ice_free_vsi(&pf->hw, vsi->idx, ctxt, false, NULL); 306 if (status) 307 dev_err(ice_pf_to_dev(pf), "Failed to delete VSI %i in FW - error: %d\n", 308 vsi->vsi_num, status); 309 310 kfree(ctxt); 311 } 312 313 /** 314 * ice_vsi_free_arrays - De-allocate queue and vector pointer arrays for the VSI 315 * @vsi: pointer to VSI being cleared 316 */ 317 static void ice_vsi_free_arrays(struct ice_vsi *vsi) 318 { 319 struct ice_pf *pf = vsi->back; 320 struct device *dev; 321 322 dev = ice_pf_to_dev(pf); 323 324 if (vsi->af_xdp_zc_qps) { 325 bitmap_free(vsi->af_xdp_zc_qps); 326 vsi->af_xdp_zc_qps = NULL; 327 } 328 /* free the ring and vector containers */ 329 if (vsi->q_vectors) { 330 devm_kfree(dev, vsi->q_vectors); 331 vsi->q_vectors = NULL; 332 } 333 if (vsi->tx_rings) { 334 devm_kfree(dev, vsi->tx_rings); 335 vsi->tx_rings = NULL; 336 } 337 if (vsi->rx_rings) { 338 devm_kfree(dev, vsi->rx_rings); 339 vsi->rx_rings = NULL; 340 } 341 if (vsi->txq_map) { 342 devm_kfree(dev, vsi->txq_map); 343 vsi->txq_map = NULL; 344 } 345 if (vsi->rxq_map) { 346 devm_kfree(dev, vsi->rxq_map); 347 vsi->rxq_map = NULL; 348 } 349 } 350 351 /** 352 * ice_vsi_free_stats - Free the ring statistics structures 353 * @vsi: VSI pointer 354 */ 355 static void ice_vsi_free_stats(struct ice_vsi *vsi) 356 { 357 struct ice_vsi_stats *vsi_stat; 358 struct ice_pf *pf = vsi->back; 359 int i; 360 361 if (vsi->type == ICE_VSI_CHNL) 362 return; 363 if (!pf->vsi_stats) 364 return; 365 366 vsi_stat = pf->vsi_stats[vsi->idx]; 367 if (!vsi_stat) 368 return; 369 370 ice_for_each_alloc_txq(vsi, i) { 371 if (vsi_stat->tx_ring_stats[i]) { 372 kfree_rcu(vsi_stat->tx_ring_stats[i], rcu); 373 WRITE_ONCE(vsi_stat->tx_ring_stats[i], NULL); 374 } 375 } 376 377 ice_for_each_alloc_rxq(vsi, i) { 378 if (vsi_stat->rx_ring_stats[i]) { 379 kfree_rcu(vsi_stat->rx_ring_stats[i], rcu); 380 WRITE_ONCE(vsi_stat->rx_ring_stats[i], NULL); 381 } 382 } 383 384 kfree(vsi_stat->tx_ring_stats); 385 kfree(vsi_stat->rx_ring_stats); 386 kfree(vsi_stat); 387 pf->vsi_stats[vsi->idx] = NULL; 388 } 389 390 /** 391 * ice_vsi_alloc_ring_stats - Allocates Tx and Rx ring stats for the VSI 392 * @vsi: VSI which is having stats allocated 393 */ 394 static int ice_vsi_alloc_ring_stats(struct ice_vsi *vsi) 395 { 396 struct ice_ring_stats **tx_ring_stats; 397 struct ice_ring_stats **rx_ring_stats; 398 struct ice_vsi_stats *vsi_stats; 399 struct ice_pf *pf = vsi->back; 400 u16 i; 401 402 vsi_stats = pf->vsi_stats[vsi->idx]; 403 tx_ring_stats = vsi_stats->tx_ring_stats; 404 rx_ring_stats = vsi_stats->rx_ring_stats; 405 406 /* Allocate Tx ring stats */ 407 ice_for_each_alloc_txq(vsi, i) { 408 struct ice_ring_stats *ring_stats; 409 struct ice_tx_ring *ring; 410 411 ring = vsi->tx_rings[i]; 412 ring_stats = tx_ring_stats[i]; 413 414 if (!ring_stats) { 415 ring_stats = kzalloc(sizeof(*ring_stats), GFP_KERNEL); 416 if (!ring_stats) 417 goto err_out; 418 419 WRITE_ONCE(tx_ring_stats[i], ring_stats); 420 } 421 422 ring->ring_stats = ring_stats; 423 } 424 425 /* Allocate Rx ring stats */ 426 ice_for_each_alloc_rxq(vsi, i) { 427 struct ice_ring_stats *ring_stats; 428 struct ice_rx_ring *ring; 429 430 ring = vsi->rx_rings[i]; 431 ring_stats = rx_ring_stats[i]; 432 433 if (!ring_stats) { 434 ring_stats = kzalloc(sizeof(*ring_stats), GFP_KERNEL); 435 if (!ring_stats) 436 goto err_out; 437 438 WRITE_ONCE(rx_ring_stats[i], ring_stats); 439 } 440 441 ring->ring_stats = ring_stats; 442 } 443 444 return 0; 445 446 err_out: 447 ice_vsi_free_stats(vsi); 448 return -ENOMEM; 449 } 450 451 /** 452 * ice_vsi_free - clean up and deallocate the provided VSI 453 * @vsi: pointer to VSI being cleared 454 * 455 * This deallocates the VSI's queue resources, removes it from the PF's 456 * VSI array if necessary, and deallocates the VSI 457 */ 458 static void ice_vsi_free(struct ice_vsi *vsi) 459 { 460 struct ice_pf *pf = NULL; 461 struct device *dev; 462 463 if (!vsi || !vsi->back) 464 return; 465 466 pf = vsi->back; 467 dev = ice_pf_to_dev(pf); 468 469 if (!pf->vsi[vsi->idx] || pf->vsi[vsi->idx] != vsi) { 470 dev_dbg(dev, "vsi does not exist at pf->vsi[%d]\n", vsi->idx); 471 return; 472 } 473 474 mutex_lock(&pf->sw_mutex); 475 /* updates the PF for this cleared VSI */ 476 477 pf->vsi[vsi->idx] = NULL; 478 pf->next_vsi = vsi->idx; 479 480 ice_vsi_free_stats(vsi); 481 ice_vsi_free_arrays(vsi); 482 mutex_unlock(&pf->sw_mutex); 483 devm_kfree(dev, vsi); 484 } 485 486 void ice_vsi_delete(struct ice_vsi *vsi) 487 { 488 ice_vsi_delete_from_hw(vsi); 489 ice_vsi_free(vsi); 490 } 491 492 /** 493 * ice_msix_clean_ctrl_vsi - MSIX mode interrupt handler for ctrl VSI 494 * @irq: interrupt number 495 * @data: pointer to a q_vector 496 */ 497 static irqreturn_t ice_msix_clean_ctrl_vsi(int __always_unused irq, void *data) 498 { 499 struct ice_q_vector *q_vector = (struct ice_q_vector *)data; 500 501 if (!q_vector->tx.tx_ring) 502 return IRQ_HANDLED; 503 504 #define FDIR_RX_DESC_CLEAN_BUDGET 64 505 ice_clean_rx_irq(q_vector->rx.rx_ring, FDIR_RX_DESC_CLEAN_BUDGET); 506 ice_clean_ctrl_tx_irq(q_vector->tx.tx_ring); 507 508 return IRQ_HANDLED; 509 } 510 511 /** 512 * ice_msix_clean_rings - MSIX mode Interrupt Handler 513 * @irq: interrupt number 514 * @data: pointer to a q_vector 515 */ 516 static irqreturn_t ice_msix_clean_rings(int __always_unused irq, void *data) 517 { 518 struct ice_q_vector *q_vector = (struct ice_q_vector *)data; 519 520 if (!q_vector->tx.tx_ring && !q_vector->rx.rx_ring) 521 return IRQ_HANDLED; 522 523 q_vector->total_events++; 524 525 napi_schedule(&q_vector->napi); 526 527 return IRQ_HANDLED; 528 } 529 530 static irqreturn_t ice_eswitch_msix_clean_rings(int __always_unused irq, void *data) 531 { 532 struct ice_q_vector *q_vector = (struct ice_q_vector *)data; 533 struct ice_pf *pf = q_vector->vsi->back; 534 struct ice_vf *vf; 535 unsigned int bkt; 536 537 if (!q_vector->tx.tx_ring && !q_vector->rx.rx_ring) 538 return IRQ_HANDLED; 539 540 rcu_read_lock(); 541 ice_for_each_vf_rcu(pf, bkt, vf) 542 napi_schedule(&vf->repr->q_vector->napi); 543 rcu_read_unlock(); 544 545 return IRQ_HANDLED; 546 } 547 548 /** 549 * ice_vsi_alloc_stat_arrays - Allocate statistics arrays 550 * @vsi: VSI pointer 551 */ 552 static int ice_vsi_alloc_stat_arrays(struct ice_vsi *vsi) 553 { 554 struct ice_vsi_stats *vsi_stat; 555 struct ice_pf *pf = vsi->back; 556 557 if (vsi->type == ICE_VSI_CHNL) 558 return 0; 559 if (!pf->vsi_stats) 560 return -ENOENT; 561 562 if (pf->vsi_stats[vsi->idx]) 563 /* realloc will happen in rebuild path */ 564 return 0; 565 566 vsi_stat = kzalloc(sizeof(*vsi_stat), GFP_KERNEL); 567 if (!vsi_stat) 568 return -ENOMEM; 569 570 vsi_stat->tx_ring_stats = 571 kcalloc(vsi->alloc_txq, sizeof(*vsi_stat->tx_ring_stats), 572 GFP_KERNEL); 573 if (!vsi_stat->tx_ring_stats) 574 goto err_alloc_tx; 575 576 vsi_stat->rx_ring_stats = 577 kcalloc(vsi->alloc_rxq, sizeof(*vsi_stat->rx_ring_stats), 578 GFP_KERNEL); 579 if (!vsi_stat->rx_ring_stats) 580 goto err_alloc_rx; 581 582 pf->vsi_stats[vsi->idx] = vsi_stat; 583 584 return 0; 585 586 err_alloc_rx: 587 kfree(vsi_stat->rx_ring_stats); 588 err_alloc_tx: 589 kfree(vsi_stat->tx_ring_stats); 590 kfree(vsi_stat); 591 pf->vsi_stats[vsi->idx] = NULL; 592 return -ENOMEM; 593 } 594 595 /** 596 * ice_vsi_alloc_def - set default values for already allocated VSI 597 * @vsi: ptr to VSI 598 * @ch: ptr to channel 599 */ 600 static int 601 ice_vsi_alloc_def(struct ice_vsi *vsi, struct ice_channel *ch) 602 { 603 if (vsi->type != ICE_VSI_CHNL) { 604 ice_vsi_set_num_qs(vsi); 605 if (ice_vsi_alloc_arrays(vsi)) 606 return -ENOMEM; 607 } 608 609 switch (vsi->type) { 610 case ICE_VSI_SWITCHDEV_CTRL: 611 /* Setup eswitch MSIX irq handler for VSI */ 612 vsi->irq_handler = ice_eswitch_msix_clean_rings; 613 break; 614 case ICE_VSI_PF: 615 /* Setup default MSIX irq handler for VSI */ 616 vsi->irq_handler = ice_msix_clean_rings; 617 break; 618 case ICE_VSI_CTRL: 619 /* Setup ctrl VSI MSIX irq handler */ 620 vsi->irq_handler = ice_msix_clean_ctrl_vsi; 621 break; 622 case ICE_VSI_CHNL: 623 if (!ch) 624 return -EINVAL; 625 626 vsi->num_rxq = ch->num_rxq; 627 vsi->num_txq = ch->num_txq; 628 vsi->next_base_q = ch->base_q; 629 break; 630 case ICE_VSI_VF: 631 case ICE_VSI_LB: 632 break; 633 default: 634 ice_vsi_free_arrays(vsi); 635 return -EINVAL; 636 } 637 638 return 0; 639 } 640 641 /** 642 * ice_vsi_alloc - Allocates the next available struct VSI in the PF 643 * @pf: board private structure 644 * 645 * Reserves a VSI index from the PF and allocates an empty VSI structure 646 * without a type. The VSI structure must later be initialized by calling 647 * ice_vsi_cfg(). 648 * 649 * returns a pointer to a VSI on success, NULL on failure. 650 */ 651 static struct ice_vsi *ice_vsi_alloc(struct ice_pf *pf) 652 { 653 struct device *dev = ice_pf_to_dev(pf); 654 struct ice_vsi *vsi = NULL; 655 656 /* Need to protect the allocation of the VSIs at the PF level */ 657 mutex_lock(&pf->sw_mutex); 658 659 /* If we have already allocated our maximum number of VSIs, 660 * pf->next_vsi will be ICE_NO_VSI. If not, pf->next_vsi index 661 * is available to be populated 662 */ 663 if (pf->next_vsi == ICE_NO_VSI) { 664 dev_dbg(dev, "out of VSI slots!\n"); 665 goto unlock_pf; 666 } 667 668 vsi = devm_kzalloc(dev, sizeof(*vsi), GFP_KERNEL); 669 if (!vsi) 670 goto unlock_pf; 671 672 vsi->back = pf; 673 set_bit(ICE_VSI_DOWN, vsi->state); 674 675 /* fill slot and make note of the index */ 676 vsi->idx = pf->next_vsi; 677 pf->vsi[pf->next_vsi] = vsi; 678 679 /* prepare pf->next_vsi for next use */ 680 pf->next_vsi = ice_get_free_slot(pf->vsi, pf->num_alloc_vsi, 681 pf->next_vsi); 682 683 unlock_pf: 684 mutex_unlock(&pf->sw_mutex); 685 return vsi; 686 } 687 688 /** 689 * ice_alloc_fd_res - Allocate FD resource for a VSI 690 * @vsi: pointer to the ice_vsi 691 * 692 * This allocates the FD resources 693 * 694 * Returns 0 on success, -EPERM on no-op or -EIO on failure 695 */ 696 static int ice_alloc_fd_res(struct ice_vsi *vsi) 697 { 698 struct ice_pf *pf = vsi->back; 699 u32 g_val, b_val; 700 701 /* Flow Director filters are only allocated/assigned to the PF VSI or 702 * CHNL VSI which passes the traffic. The CTRL VSI is only used to 703 * add/delete filters so resources are not allocated to it 704 */ 705 if (!test_bit(ICE_FLAG_FD_ENA, pf->flags)) 706 return -EPERM; 707 708 if (!(vsi->type == ICE_VSI_PF || vsi->type == ICE_VSI_VF || 709 vsi->type == ICE_VSI_CHNL)) 710 return -EPERM; 711 712 /* FD filters from guaranteed pool per VSI */ 713 g_val = pf->hw.func_caps.fd_fltr_guar; 714 if (!g_val) 715 return -EPERM; 716 717 /* FD filters from best effort pool */ 718 b_val = pf->hw.func_caps.fd_fltr_best_effort; 719 if (!b_val) 720 return -EPERM; 721 722 /* PF main VSI gets only 64 FD resources from guaranteed pool 723 * when ADQ is configured. 724 */ 725 #define ICE_PF_VSI_GFLTR 64 726 727 /* determine FD filter resources per VSI from shared(best effort) and 728 * dedicated pool 729 */ 730 if (vsi->type == ICE_VSI_PF) { 731 vsi->num_gfltr = g_val; 732 /* if MQPRIO is configured, main VSI doesn't get all FD 733 * resources from guaranteed pool. PF VSI gets 64 FD resources 734 */ 735 if (test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) { 736 if (g_val < ICE_PF_VSI_GFLTR) 737 return -EPERM; 738 /* allow bare minimum entries for PF VSI */ 739 vsi->num_gfltr = ICE_PF_VSI_GFLTR; 740 } 741 742 /* each VSI gets same "best_effort" quota */ 743 vsi->num_bfltr = b_val; 744 } else if (vsi->type == ICE_VSI_VF) { 745 vsi->num_gfltr = 0; 746 747 /* each VSI gets same "best_effort" quota */ 748 vsi->num_bfltr = b_val; 749 } else { 750 struct ice_vsi *main_vsi; 751 int numtc; 752 753 main_vsi = ice_get_main_vsi(pf); 754 if (!main_vsi) 755 return -EPERM; 756 757 if (!main_vsi->all_numtc) 758 return -EINVAL; 759 760 /* figure out ADQ numtc */ 761 numtc = main_vsi->all_numtc - ICE_CHNL_START_TC; 762 763 /* only one TC but still asking resources for channels, 764 * invalid config 765 */ 766 if (numtc < ICE_CHNL_START_TC) 767 return -EPERM; 768 769 g_val -= ICE_PF_VSI_GFLTR; 770 /* channel VSIs gets equal share from guaranteed pool */ 771 vsi->num_gfltr = g_val / numtc; 772 773 /* each VSI gets same "best_effort" quota */ 774 vsi->num_bfltr = b_val; 775 } 776 777 return 0; 778 } 779 780 /** 781 * ice_vsi_get_qs - Assign queues from PF to VSI 782 * @vsi: the VSI to assign queues to 783 * 784 * Returns 0 on success and a negative value on error 785 */ 786 static int ice_vsi_get_qs(struct ice_vsi *vsi) 787 { 788 struct ice_pf *pf = vsi->back; 789 struct ice_qs_cfg tx_qs_cfg = { 790 .qs_mutex = &pf->avail_q_mutex, 791 .pf_map = pf->avail_txqs, 792 .pf_map_size = pf->max_pf_txqs, 793 .q_count = vsi->alloc_txq, 794 .scatter_count = ICE_MAX_SCATTER_TXQS, 795 .vsi_map = vsi->txq_map, 796 .vsi_map_offset = 0, 797 .mapping_mode = ICE_VSI_MAP_CONTIG 798 }; 799 struct ice_qs_cfg rx_qs_cfg = { 800 .qs_mutex = &pf->avail_q_mutex, 801 .pf_map = pf->avail_rxqs, 802 .pf_map_size = pf->max_pf_rxqs, 803 .q_count = vsi->alloc_rxq, 804 .scatter_count = ICE_MAX_SCATTER_RXQS, 805 .vsi_map = vsi->rxq_map, 806 .vsi_map_offset = 0, 807 .mapping_mode = ICE_VSI_MAP_CONTIG 808 }; 809 int ret; 810 811 if (vsi->type == ICE_VSI_CHNL) 812 return 0; 813 814 ret = __ice_vsi_get_qs(&tx_qs_cfg); 815 if (ret) 816 return ret; 817 vsi->tx_mapping_mode = tx_qs_cfg.mapping_mode; 818 819 ret = __ice_vsi_get_qs(&rx_qs_cfg); 820 if (ret) 821 return ret; 822 vsi->rx_mapping_mode = rx_qs_cfg.mapping_mode; 823 824 return 0; 825 } 826 827 /** 828 * ice_vsi_put_qs - Release queues from VSI to PF 829 * @vsi: the VSI that is going to release queues 830 */ 831 static void ice_vsi_put_qs(struct ice_vsi *vsi) 832 { 833 struct ice_pf *pf = vsi->back; 834 int i; 835 836 mutex_lock(&pf->avail_q_mutex); 837 838 ice_for_each_alloc_txq(vsi, i) { 839 clear_bit(vsi->txq_map[i], pf->avail_txqs); 840 vsi->txq_map[i] = ICE_INVAL_Q_INDEX; 841 } 842 843 ice_for_each_alloc_rxq(vsi, i) { 844 clear_bit(vsi->rxq_map[i], pf->avail_rxqs); 845 vsi->rxq_map[i] = ICE_INVAL_Q_INDEX; 846 } 847 848 mutex_unlock(&pf->avail_q_mutex); 849 } 850 851 /** 852 * ice_is_safe_mode 853 * @pf: pointer to the PF struct 854 * 855 * returns true if driver is in safe mode, false otherwise 856 */ 857 bool ice_is_safe_mode(struct ice_pf *pf) 858 { 859 return !test_bit(ICE_FLAG_ADV_FEATURES, pf->flags); 860 } 861 862 /** 863 * ice_is_rdma_ena 864 * @pf: pointer to the PF struct 865 * 866 * returns true if RDMA is currently supported, false otherwise 867 */ 868 bool ice_is_rdma_ena(struct ice_pf *pf) 869 { 870 return test_bit(ICE_FLAG_RDMA_ENA, pf->flags); 871 } 872 873 /** 874 * ice_vsi_clean_rss_flow_fld - Delete RSS configuration 875 * @vsi: the VSI being cleaned up 876 * 877 * This function deletes RSS input set for all flows that were configured 878 * for this VSI 879 */ 880 static void ice_vsi_clean_rss_flow_fld(struct ice_vsi *vsi) 881 { 882 struct ice_pf *pf = vsi->back; 883 int status; 884 885 if (ice_is_safe_mode(pf)) 886 return; 887 888 status = ice_rem_vsi_rss_cfg(&pf->hw, vsi->idx); 889 if (status) 890 dev_dbg(ice_pf_to_dev(pf), "ice_rem_vsi_rss_cfg failed for vsi = %d, error = %d\n", 891 vsi->vsi_num, status); 892 } 893 894 /** 895 * ice_rss_clean - Delete RSS related VSI structures and configuration 896 * @vsi: the VSI being removed 897 */ 898 static void ice_rss_clean(struct ice_vsi *vsi) 899 { 900 struct ice_pf *pf = vsi->back; 901 struct device *dev; 902 903 dev = ice_pf_to_dev(pf); 904 905 if (vsi->rss_hkey_user) 906 devm_kfree(dev, vsi->rss_hkey_user); 907 if (vsi->rss_lut_user) 908 devm_kfree(dev, vsi->rss_lut_user); 909 910 ice_vsi_clean_rss_flow_fld(vsi); 911 /* remove RSS replay list */ 912 if (!ice_is_safe_mode(pf)) 913 ice_rem_vsi_rss_list(&pf->hw, vsi->idx); 914 } 915 916 /** 917 * ice_vsi_set_rss_params - Setup RSS capabilities per VSI type 918 * @vsi: the VSI being configured 919 */ 920 static void ice_vsi_set_rss_params(struct ice_vsi *vsi) 921 { 922 struct ice_hw_common_caps *cap; 923 struct ice_pf *pf = vsi->back; 924 925 if (!test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 926 vsi->rss_size = 1; 927 return; 928 } 929 930 cap = &pf->hw.func_caps.common_cap; 931 switch (vsi->type) { 932 case ICE_VSI_CHNL: 933 case ICE_VSI_PF: 934 /* PF VSI will inherit RSS instance of PF */ 935 vsi->rss_table_size = (u16)cap->rss_table_size; 936 if (vsi->type == ICE_VSI_CHNL) 937 vsi->rss_size = min_t(u16, vsi->num_rxq, 938 BIT(cap->rss_table_entry_width)); 939 else 940 vsi->rss_size = min_t(u16, num_online_cpus(), 941 BIT(cap->rss_table_entry_width)); 942 vsi->rss_lut_type = ICE_AQC_GSET_RSS_LUT_TABLE_TYPE_PF; 943 break; 944 case ICE_VSI_SWITCHDEV_CTRL: 945 vsi->rss_table_size = ICE_VSIQF_HLUT_ARRAY_SIZE; 946 vsi->rss_size = min_t(u16, num_online_cpus(), 947 BIT(cap->rss_table_entry_width)); 948 vsi->rss_lut_type = ICE_AQC_GSET_RSS_LUT_TABLE_TYPE_VSI; 949 break; 950 case ICE_VSI_VF: 951 /* VF VSI will get a small RSS table. 952 * For VSI_LUT, LUT size should be set to 64 bytes. 953 */ 954 vsi->rss_table_size = ICE_VSIQF_HLUT_ARRAY_SIZE; 955 vsi->rss_size = ICE_MAX_RSS_QS_PER_VF; 956 vsi->rss_lut_type = ICE_AQC_GSET_RSS_LUT_TABLE_TYPE_VSI; 957 break; 958 case ICE_VSI_LB: 959 break; 960 default: 961 dev_dbg(ice_pf_to_dev(pf), "Unsupported VSI type %s\n", 962 ice_vsi_type_str(vsi->type)); 963 break; 964 } 965 } 966 967 /** 968 * ice_set_dflt_vsi_ctx - Set default VSI context before adding a VSI 969 * @hw: HW structure used to determine the VLAN mode of the device 970 * @ctxt: the VSI context being set 971 * 972 * This initializes a default VSI context for all sections except the Queues. 973 */ 974 static void ice_set_dflt_vsi_ctx(struct ice_hw *hw, struct ice_vsi_ctx *ctxt) 975 { 976 u32 table = 0; 977 978 memset(&ctxt->info, 0, sizeof(ctxt->info)); 979 /* VSI's should be allocated from shared pool */ 980 ctxt->alloc_from_pool = true; 981 /* Src pruning enabled by default */ 982 ctxt->info.sw_flags = ICE_AQ_VSI_SW_FLAG_SRC_PRUNE; 983 /* Traffic from VSI can be sent to LAN */ 984 ctxt->info.sw_flags2 = ICE_AQ_VSI_SW_FLAG_LAN_ENA; 985 /* allow all untagged/tagged packets by default on Tx */ 986 ctxt->info.inner_vlan_flags = ((ICE_AQ_VSI_INNER_VLAN_TX_MODE_ALL & 987 ICE_AQ_VSI_INNER_VLAN_TX_MODE_M) >> 988 ICE_AQ_VSI_INNER_VLAN_TX_MODE_S); 989 /* SVM - by default bits 3 and 4 in inner_vlan_flags are 0's which 990 * results in legacy behavior (show VLAN, DEI, and UP) in descriptor. 991 * 992 * DVM - leave inner VLAN in packet by default 993 */ 994 if (ice_is_dvm_ena(hw)) { 995 ctxt->info.inner_vlan_flags |= 996 ICE_AQ_VSI_INNER_VLAN_EMODE_NOTHING; 997 ctxt->info.outer_vlan_flags = 998 (ICE_AQ_VSI_OUTER_VLAN_TX_MODE_ALL << 999 ICE_AQ_VSI_OUTER_VLAN_TX_MODE_S) & 1000 ICE_AQ_VSI_OUTER_VLAN_TX_MODE_M; 1001 ctxt->info.outer_vlan_flags |= 1002 (ICE_AQ_VSI_OUTER_TAG_VLAN_8100 << 1003 ICE_AQ_VSI_OUTER_TAG_TYPE_S) & 1004 ICE_AQ_VSI_OUTER_TAG_TYPE_M; 1005 ctxt->info.outer_vlan_flags |= 1006 FIELD_PREP(ICE_AQ_VSI_OUTER_VLAN_EMODE_M, 1007 ICE_AQ_VSI_OUTER_VLAN_EMODE_NOTHING); 1008 } 1009 /* Have 1:1 UP mapping for both ingress/egress tables */ 1010 table |= ICE_UP_TABLE_TRANSLATE(0, 0); 1011 table |= ICE_UP_TABLE_TRANSLATE(1, 1); 1012 table |= ICE_UP_TABLE_TRANSLATE(2, 2); 1013 table |= ICE_UP_TABLE_TRANSLATE(3, 3); 1014 table |= ICE_UP_TABLE_TRANSLATE(4, 4); 1015 table |= ICE_UP_TABLE_TRANSLATE(5, 5); 1016 table |= ICE_UP_TABLE_TRANSLATE(6, 6); 1017 table |= ICE_UP_TABLE_TRANSLATE(7, 7); 1018 ctxt->info.ingress_table = cpu_to_le32(table); 1019 ctxt->info.egress_table = cpu_to_le32(table); 1020 /* Have 1:1 UP mapping for outer to inner UP table */ 1021 ctxt->info.outer_up_table = cpu_to_le32(table); 1022 /* No Outer tag support outer_tag_flags remains to zero */ 1023 } 1024 1025 /** 1026 * ice_vsi_setup_q_map - Setup a VSI queue map 1027 * @vsi: the VSI being configured 1028 * @ctxt: VSI context structure 1029 */ 1030 static int ice_vsi_setup_q_map(struct ice_vsi *vsi, struct ice_vsi_ctx *ctxt) 1031 { 1032 u16 offset = 0, qmap = 0, tx_count = 0, rx_count = 0, pow = 0; 1033 u16 num_txq_per_tc, num_rxq_per_tc; 1034 u16 qcount_tx = vsi->alloc_txq; 1035 u16 qcount_rx = vsi->alloc_rxq; 1036 u8 netdev_tc = 0; 1037 int i; 1038 1039 if (!vsi->tc_cfg.numtc) { 1040 /* at least TC0 should be enabled by default */ 1041 vsi->tc_cfg.numtc = 1; 1042 vsi->tc_cfg.ena_tc = 1; 1043 } 1044 1045 num_rxq_per_tc = min_t(u16, qcount_rx / vsi->tc_cfg.numtc, ICE_MAX_RXQS_PER_TC); 1046 if (!num_rxq_per_tc) 1047 num_rxq_per_tc = 1; 1048 num_txq_per_tc = qcount_tx / vsi->tc_cfg.numtc; 1049 if (!num_txq_per_tc) 1050 num_txq_per_tc = 1; 1051 1052 /* find the (rounded up) power-of-2 of qcount */ 1053 pow = (u16)order_base_2(num_rxq_per_tc); 1054 1055 /* TC mapping is a function of the number of Rx queues assigned to the 1056 * VSI for each traffic class and the offset of these queues. 1057 * The first 10 bits are for queue offset for TC0, next 4 bits for no:of 1058 * queues allocated to TC0. No:of queues is a power-of-2. 1059 * 1060 * If TC is not enabled, the queue offset is set to 0, and allocate one 1061 * queue, this way, traffic for the given TC will be sent to the default 1062 * queue. 1063 * 1064 * Setup number and offset of Rx queues for all TCs for the VSI 1065 */ 1066 ice_for_each_traffic_class(i) { 1067 if (!(vsi->tc_cfg.ena_tc & BIT(i))) { 1068 /* TC is not enabled */ 1069 vsi->tc_cfg.tc_info[i].qoffset = 0; 1070 vsi->tc_cfg.tc_info[i].qcount_rx = 1; 1071 vsi->tc_cfg.tc_info[i].qcount_tx = 1; 1072 vsi->tc_cfg.tc_info[i].netdev_tc = 0; 1073 ctxt->info.tc_mapping[i] = 0; 1074 continue; 1075 } 1076 1077 /* TC is enabled */ 1078 vsi->tc_cfg.tc_info[i].qoffset = offset; 1079 vsi->tc_cfg.tc_info[i].qcount_rx = num_rxq_per_tc; 1080 vsi->tc_cfg.tc_info[i].qcount_tx = num_txq_per_tc; 1081 vsi->tc_cfg.tc_info[i].netdev_tc = netdev_tc++; 1082 1083 qmap = ((offset << ICE_AQ_VSI_TC_Q_OFFSET_S) & 1084 ICE_AQ_VSI_TC_Q_OFFSET_M) | 1085 ((pow << ICE_AQ_VSI_TC_Q_NUM_S) & 1086 ICE_AQ_VSI_TC_Q_NUM_M); 1087 offset += num_rxq_per_tc; 1088 tx_count += num_txq_per_tc; 1089 ctxt->info.tc_mapping[i] = cpu_to_le16(qmap); 1090 } 1091 1092 /* if offset is non-zero, means it is calculated correctly based on 1093 * enabled TCs for a given VSI otherwise qcount_rx will always 1094 * be correct and non-zero because it is based off - VSI's 1095 * allocated Rx queues which is at least 1 (hence qcount_tx will be 1096 * at least 1) 1097 */ 1098 if (offset) 1099 rx_count = offset; 1100 else 1101 rx_count = num_rxq_per_tc; 1102 1103 if (rx_count > vsi->alloc_rxq) { 1104 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Rx queues (%u), than were allocated (%u)!\n", 1105 rx_count, vsi->alloc_rxq); 1106 return -EINVAL; 1107 } 1108 1109 if (tx_count > vsi->alloc_txq) { 1110 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Tx queues (%u), than were allocated (%u)!\n", 1111 tx_count, vsi->alloc_txq); 1112 return -EINVAL; 1113 } 1114 1115 vsi->num_txq = tx_count; 1116 vsi->num_rxq = rx_count; 1117 1118 if (vsi->type == ICE_VSI_VF && vsi->num_txq != vsi->num_rxq) { 1119 dev_dbg(ice_pf_to_dev(vsi->back), "VF VSI should have same number of Tx and Rx queues. Hence making them equal\n"); 1120 /* since there is a chance that num_rxq could have been changed 1121 * in the above for loop, make num_txq equal to num_rxq. 1122 */ 1123 vsi->num_txq = vsi->num_rxq; 1124 } 1125 1126 /* Rx queue mapping */ 1127 ctxt->info.mapping_flags |= cpu_to_le16(ICE_AQ_VSI_Q_MAP_CONTIG); 1128 /* q_mapping buffer holds the info for the first queue allocated for 1129 * this VSI in the PF space and also the number of queues associated 1130 * with this VSI. 1131 */ 1132 ctxt->info.q_mapping[0] = cpu_to_le16(vsi->rxq_map[0]); 1133 ctxt->info.q_mapping[1] = cpu_to_le16(vsi->num_rxq); 1134 1135 return 0; 1136 } 1137 1138 /** 1139 * ice_set_fd_vsi_ctx - Set FD VSI context before adding a VSI 1140 * @ctxt: the VSI context being set 1141 * @vsi: the VSI being configured 1142 */ 1143 static void ice_set_fd_vsi_ctx(struct ice_vsi_ctx *ctxt, struct ice_vsi *vsi) 1144 { 1145 u8 dflt_q_group, dflt_q_prio; 1146 u16 dflt_q, report_q, val; 1147 1148 if (vsi->type != ICE_VSI_PF && vsi->type != ICE_VSI_CTRL && 1149 vsi->type != ICE_VSI_VF && vsi->type != ICE_VSI_CHNL) 1150 return; 1151 1152 val = ICE_AQ_VSI_PROP_FLOW_DIR_VALID; 1153 ctxt->info.valid_sections |= cpu_to_le16(val); 1154 dflt_q = 0; 1155 dflt_q_group = 0; 1156 report_q = 0; 1157 dflt_q_prio = 0; 1158 1159 /* enable flow director filtering/programming */ 1160 val = ICE_AQ_VSI_FD_ENABLE | ICE_AQ_VSI_FD_PROG_ENABLE; 1161 ctxt->info.fd_options = cpu_to_le16(val); 1162 /* max of allocated flow director filters */ 1163 ctxt->info.max_fd_fltr_dedicated = 1164 cpu_to_le16(vsi->num_gfltr); 1165 /* max of shared flow director filters any VSI may program */ 1166 ctxt->info.max_fd_fltr_shared = 1167 cpu_to_le16(vsi->num_bfltr); 1168 /* default queue index within the VSI of the default FD */ 1169 val = ((dflt_q << ICE_AQ_VSI_FD_DEF_Q_S) & 1170 ICE_AQ_VSI_FD_DEF_Q_M); 1171 /* target queue or queue group to the FD filter */ 1172 val |= ((dflt_q_group << ICE_AQ_VSI_FD_DEF_GRP_S) & 1173 ICE_AQ_VSI_FD_DEF_GRP_M); 1174 ctxt->info.fd_def_q = cpu_to_le16(val); 1175 /* queue index on which FD filter completion is reported */ 1176 val = ((report_q << ICE_AQ_VSI_FD_REPORT_Q_S) & 1177 ICE_AQ_VSI_FD_REPORT_Q_M); 1178 /* priority of the default qindex action */ 1179 val |= ((dflt_q_prio << ICE_AQ_VSI_FD_DEF_PRIORITY_S) & 1180 ICE_AQ_VSI_FD_DEF_PRIORITY_M); 1181 ctxt->info.fd_report_opt = cpu_to_le16(val); 1182 } 1183 1184 /** 1185 * ice_set_rss_vsi_ctx - Set RSS VSI context before adding a VSI 1186 * @ctxt: the VSI context being set 1187 * @vsi: the VSI being configured 1188 */ 1189 static void ice_set_rss_vsi_ctx(struct ice_vsi_ctx *ctxt, struct ice_vsi *vsi) 1190 { 1191 u8 lut_type, hash_type; 1192 struct device *dev; 1193 struct ice_pf *pf; 1194 1195 pf = vsi->back; 1196 dev = ice_pf_to_dev(pf); 1197 1198 switch (vsi->type) { 1199 case ICE_VSI_CHNL: 1200 case ICE_VSI_PF: 1201 /* PF VSI will inherit RSS instance of PF */ 1202 lut_type = ICE_AQ_VSI_Q_OPT_RSS_LUT_PF; 1203 hash_type = ICE_AQ_VSI_Q_OPT_RSS_TPLZ; 1204 break; 1205 case ICE_VSI_VF: 1206 /* VF VSI will gets a small RSS table which is a VSI LUT type */ 1207 lut_type = ICE_AQ_VSI_Q_OPT_RSS_LUT_VSI; 1208 hash_type = ICE_AQ_VSI_Q_OPT_RSS_TPLZ; 1209 break; 1210 default: 1211 dev_dbg(dev, "Unsupported VSI type %s\n", 1212 ice_vsi_type_str(vsi->type)); 1213 return; 1214 } 1215 1216 ctxt->info.q_opt_rss = ((lut_type << ICE_AQ_VSI_Q_OPT_RSS_LUT_S) & 1217 ICE_AQ_VSI_Q_OPT_RSS_LUT_M) | 1218 ((hash_type << ICE_AQ_VSI_Q_OPT_RSS_HASH_S) & 1219 ICE_AQ_VSI_Q_OPT_RSS_HASH_M); 1220 } 1221 1222 static void 1223 ice_chnl_vsi_setup_q_map(struct ice_vsi *vsi, struct ice_vsi_ctx *ctxt) 1224 { 1225 struct ice_pf *pf = vsi->back; 1226 u16 qcount, qmap; 1227 u8 offset = 0; 1228 int pow; 1229 1230 qcount = min_t(int, vsi->num_rxq, pf->num_lan_msix); 1231 1232 pow = order_base_2(qcount); 1233 qmap = ((offset << ICE_AQ_VSI_TC_Q_OFFSET_S) & 1234 ICE_AQ_VSI_TC_Q_OFFSET_M) | 1235 ((pow << ICE_AQ_VSI_TC_Q_NUM_S) & 1236 ICE_AQ_VSI_TC_Q_NUM_M); 1237 1238 ctxt->info.tc_mapping[0] = cpu_to_le16(qmap); 1239 ctxt->info.mapping_flags |= cpu_to_le16(ICE_AQ_VSI_Q_MAP_CONTIG); 1240 ctxt->info.q_mapping[0] = cpu_to_le16(vsi->next_base_q); 1241 ctxt->info.q_mapping[1] = cpu_to_le16(qcount); 1242 } 1243 1244 /** 1245 * ice_vsi_init - Create and initialize a VSI 1246 * @vsi: the VSI being configured 1247 * @vsi_flags: VSI configuration flags 1248 * 1249 * Set ICE_FLAG_VSI_INIT to initialize a new VSI context, clear it to 1250 * reconfigure an existing context. 1251 * 1252 * This initializes a VSI context depending on the VSI type to be added and 1253 * passes it down to the add_vsi aq command to create a new VSI. 1254 */ 1255 static int ice_vsi_init(struct ice_vsi *vsi, u32 vsi_flags) 1256 { 1257 struct ice_pf *pf = vsi->back; 1258 struct ice_hw *hw = &pf->hw; 1259 struct ice_vsi_ctx *ctxt; 1260 struct device *dev; 1261 int ret = 0; 1262 1263 dev = ice_pf_to_dev(pf); 1264 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); 1265 if (!ctxt) 1266 return -ENOMEM; 1267 1268 switch (vsi->type) { 1269 case ICE_VSI_CTRL: 1270 case ICE_VSI_LB: 1271 case ICE_VSI_PF: 1272 ctxt->flags = ICE_AQ_VSI_TYPE_PF; 1273 break; 1274 case ICE_VSI_SWITCHDEV_CTRL: 1275 case ICE_VSI_CHNL: 1276 ctxt->flags = ICE_AQ_VSI_TYPE_VMDQ2; 1277 break; 1278 case ICE_VSI_VF: 1279 ctxt->flags = ICE_AQ_VSI_TYPE_VF; 1280 /* VF number here is the absolute VF number (0-255) */ 1281 ctxt->vf_num = vsi->vf->vf_id + hw->func_caps.vf_base_id; 1282 break; 1283 default: 1284 ret = -ENODEV; 1285 goto out; 1286 } 1287 1288 /* Handle VLAN pruning for channel VSI if main VSI has VLAN 1289 * prune enabled 1290 */ 1291 if (vsi->type == ICE_VSI_CHNL) { 1292 struct ice_vsi *main_vsi; 1293 1294 main_vsi = ice_get_main_vsi(pf); 1295 if (main_vsi && ice_vsi_is_vlan_pruning_ena(main_vsi)) 1296 ctxt->info.sw_flags2 |= 1297 ICE_AQ_VSI_SW_FLAG_RX_VLAN_PRUNE_ENA; 1298 else 1299 ctxt->info.sw_flags2 &= 1300 ~ICE_AQ_VSI_SW_FLAG_RX_VLAN_PRUNE_ENA; 1301 } 1302 1303 ice_set_dflt_vsi_ctx(hw, ctxt); 1304 if (test_bit(ICE_FLAG_FD_ENA, pf->flags)) 1305 ice_set_fd_vsi_ctx(ctxt, vsi); 1306 /* if the switch is in VEB mode, allow VSI loopback */ 1307 if (vsi->vsw->bridge_mode == BRIDGE_MODE_VEB) 1308 ctxt->info.sw_flags |= ICE_AQ_VSI_SW_FLAG_ALLOW_LB; 1309 1310 /* Set LUT type and HASH type if RSS is enabled */ 1311 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags) && 1312 vsi->type != ICE_VSI_CTRL) { 1313 ice_set_rss_vsi_ctx(ctxt, vsi); 1314 /* if updating VSI context, make sure to set valid_section: 1315 * to indicate which section of VSI context being updated 1316 */ 1317 if (!(vsi_flags & ICE_VSI_FLAG_INIT)) 1318 ctxt->info.valid_sections |= 1319 cpu_to_le16(ICE_AQ_VSI_PROP_Q_OPT_VALID); 1320 } 1321 1322 ctxt->info.sw_id = vsi->port_info->sw_id; 1323 if (vsi->type == ICE_VSI_CHNL) { 1324 ice_chnl_vsi_setup_q_map(vsi, ctxt); 1325 } else { 1326 ret = ice_vsi_setup_q_map(vsi, ctxt); 1327 if (ret) 1328 goto out; 1329 1330 if (!(vsi_flags & ICE_VSI_FLAG_INIT)) 1331 /* means VSI being updated */ 1332 /* must to indicate which section of VSI context are 1333 * being modified 1334 */ 1335 ctxt->info.valid_sections |= 1336 cpu_to_le16(ICE_AQ_VSI_PROP_RXQ_MAP_VALID); 1337 } 1338 1339 /* Allow control frames out of main VSI */ 1340 if (vsi->type == ICE_VSI_PF) { 1341 ctxt->info.sec_flags |= ICE_AQ_VSI_SEC_FLAG_ALLOW_DEST_OVRD; 1342 ctxt->info.valid_sections |= 1343 cpu_to_le16(ICE_AQ_VSI_PROP_SECURITY_VALID); 1344 } 1345 1346 if (vsi_flags & ICE_VSI_FLAG_INIT) { 1347 ret = ice_add_vsi(hw, vsi->idx, ctxt, NULL); 1348 if (ret) { 1349 dev_err(dev, "Add VSI failed, err %d\n", ret); 1350 ret = -EIO; 1351 goto out; 1352 } 1353 } else { 1354 ret = ice_update_vsi(hw, vsi->idx, ctxt, NULL); 1355 if (ret) { 1356 dev_err(dev, "Update VSI failed, err %d\n", ret); 1357 ret = -EIO; 1358 goto out; 1359 } 1360 } 1361 1362 /* keep context for update VSI operations */ 1363 vsi->info = ctxt->info; 1364 1365 /* record VSI number returned */ 1366 vsi->vsi_num = ctxt->vsi_num; 1367 1368 out: 1369 kfree(ctxt); 1370 return ret; 1371 } 1372 1373 /** 1374 * ice_vsi_clear_rings - Deallocates the Tx and Rx rings for VSI 1375 * @vsi: the VSI having rings deallocated 1376 */ 1377 static void ice_vsi_clear_rings(struct ice_vsi *vsi) 1378 { 1379 int i; 1380 1381 /* Avoid stale references by clearing map from vector to ring */ 1382 if (vsi->q_vectors) { 1383 ice_for_each_q_vector(vsi, i) { 1384 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 1385 1386 if (q_vector) { 1387 q_vector->tx.tx_ring = NULL; 1388 q_vector->rx.rx_ring = NULL; 1389 } 1390 } 1391 } 1392 1393 if (vsi->tx_rings) { 1394 ice_for_each_alloc_txq(vsi, i) { 1395 if (vsi->tx_rings[i]) { 1396 kfree_rcu(vsi->tx_rings[i], rcu); 1397 WRITE_ONCE(vsi->tx_rings[i], NULL); 1398 } 1399 } 1400 } 1401 if (vsi->rx_rings) { 1402 ice_for_each_alloc_rxq(vsi, i) { 1403 if (vsi->rx_rings[i]) { 1404 kfree_rcu(vsi->rx_rings[i], rcu); 1405 WRITE_ONCE(vsi->rx_rings[i], NULL); 1406 } 1407 } 1408 } 1409 } 1410 1411 /** 1412 * ice_vsi_alloc_rings - Allocates Tx and Rx rings for the VSI 1413 * @vsi: VSI which is having rings allocated 1414 */ 1415 static int ice_vsi_alloc_rings(struct ice_vsi *vsi) 1416 { 1417 bool dvm_ena = ice_is_dvm_ena(&vsi->back->hw); 1418 struct ice_pf *pf = vsi->back; 1419 struct device *dev; 1420 u16 i; 1421 1422 dev = ice_pf_to_dev(pf); 1423 /* Allocate Tx rings */ 1424 ice_for_each_alloc_txq(vsi, i) { 1425 struct ice_tx_ring *ring; 1426 1427 /* allocate with kzalloc(), free with kfree_rcu() */ 1428 ring = kzalloc(sizeof(*ring), GFP_KERNEL); 1429 1430 if (!ring) 1431 goto err_out; 1432 1433 ring->q_index = i; 1434 ring->reg_idx = vsi->txq_map[i]; 1435 ring->vsi = vsi; 1436 ring->tx_tstamps = &pf->ptp.port.tx; 1437 ring->dev = dev; 1438 ring->count = vsi->num_tx_desc; 1439 ring->txq_teid = ICE_INVAL_TEID; 1440 if (dvm_ena) 1441 ring->flags |= ICE_TX_FLAGS_RING_VLAN_L2TAG2; 1442 else 1443 ring->flags |= ICE_TX_FLAGS_RING_VLAN_L2TAG1; 1444 WRITE_ONCE(vsi->tx_rings[i], ring); 1445 } 1446 1447 /* Allocate Rx rings */ 1448 ice_for_each_alloc_rxq(vsi, i) { 1449 struct ice_rx_ring *ring; 1450 1451 /* allocate with kzalloc(), free with kfree_rcu() */ 1452 ring = kzalloc(sizeof(*ring), GFP_KERNEL); 1453 if (!ring) 1454 goto err_out; 1455 1456 ring->q_index = i; 1457 ring->reg_idx = vsi->rxq_map[i]; 1458 ring->vsi = vsi; 1459 ring->netdev = vsi->netdev; 1460 ring->dev = dev; 1461 ring->count = vsi->num_rx_desc; 1462 ring->cached_phctime = pf->ptp.cached_phc_time; 1463 WRITE_ONCE(vsi->rx_rings[i], ring); 1464 } 1465 1466 return 0; 1467 1468 err_out: 1469 ice_vsi_clear_rings(vsi); 1470 return -ENOMEM; 1471 } 1472 1473 /** 1474 * ice_vsi_manage_rss_lut - disable/enable RSS 1475 * @vsi: the VSI being changed 1476 * @ena: boolean value indicating if this is an enable or disable request 1477 * 1478 * In the event of disable request for RSS, this function will zero out RSS 1479 * LUT, while in the event of enable request for RSS, it will reconfigure RSS 1480 * LUT. 1481 */ 1482 void ice_vsi_manage_rss_lut(struct ice_vsi *vsi, bool ena) 1483 { 1484 u8 *lut; 1485 1486 lut = kzalloc(vsi->rss_table_size, GFP_KERNEL); 1487 if (!lut) 1488 return; 1489 1490 if (ena) { 1491 if (vsi->rss_lut_user) 1492 memcpy(lut, vsi->rss_lut_user, vsi->rss_table_size); 1493 else 1494 ice_fill_rss_lut(lut, vsi->rss_table_size, 1495 vsi->rss_size); 1496 } 1497 1498 ice_set_rss_lut(vsi, lut, vsi->rss_table_size); 1499 kfree(lut); 1500 } 1501 1502 /** 1503 * ice_vsi_cfg_crc_strip - Configure CRC stripping for a VSI 1504 * @vsi: VSI to be configured 1505 * @disable: set to true to have FCS / CRC in the frame data 1506 */ 1507 void ice_vsi_cfg_crc_strip(struct ice_vsi *vsi, bool disable) 1508 { 1509 int i; 1510 1511 ice_for_each_rxq(vsi, i) 1512 if (disable) 1513 vsi->rx_rings[i]->flags |= ICE_RX_FLAGS_CRC_STRIP_DIS; 1514 else 1515 vsi->rx_rings[i]->flags &= ~ICE_RX_FLAGS_CRC_STRIP_DIS; 1516 } 1517 1518 /** 1519 * ice_vsi_cfg_rss_lut_key - Configure RSS params for a VSI 1520 * @vsi: VSI to be configured 1521 */ 1522 int ice_vsi_cfg_rss_lut_key(struct ice_vsi *vsi) 1523 { 1524 struct ice_pf *pf = vsi->back; 1525 struct device *dev; 1526 u8 *lut, *key; 1527 int err; 1528 1529 dev = ice_pf_to_dev(pf); 1530 if (vsi->type == ICE_VSI_PF && vsi->ch_rss_size && 1531 (test_bit(ICE_FLAG_TC_MQPRIO, pf->flags))) { 1532 vsi->rss_size = min_t(u16, vsi->rss_size, vsi->ch_rss_size); 1533 } else { 1534 vsi->rss_size = min_t(u16, vsi->rss_size, vsi->num_rxq); 1535 1536 /* If orig_rss_size is valid and it is less than determined 1537 * main VSI's rss_size, update main VSI's rss_size to be 1538 * orig_rss_size so that when tc-qdisc is deleted, main VSI 1539 * RSS table gets programmed to be correct (whatever it was 1540 * to begin with (prior to setup-tc for ADQ config) 1541 */ 1542 if (vsi->orig_rss_size && vsi->rss_size < vsi->orig_rss_size && 1543 vsi->orig_rss_size <= vsi->num_rxq) { 1544 vsi->rss_size = vsi->orig_rss_size; 1545 /* now orig_rss_size is used, reset it to zero */ 1546 vsi->orig_rss_size = 0; 1547 } 1548 } 1549 1550 lut = kzalloc(vsi->rss_table_size, GFP_KERNEL); 1551 if (!lut) 1552 return -ENOMEM; 1553 1554 if (vsi->rss_lut_user) 1555 memcpy(lut, vsi->rss_lut_user, vsi->rss_table_size); 1556 else 1557 ice_fill_rss_lut(lut, vsi->rss_table_size, vsi->rss_size); 1558 1559 err = ice_set_rss_lut(vsi, lut, vsi->rss_table_size); 1560 if (err) { 1561 dev_err(dev, "set_rss_lut failed, error %d\n", err); 1562 goto ice_vsi_cfg_rss_exit; 1563 } 1564 1565 key = kzalloc(ICE_GET_SET_RSS_KEY_EXTEND_KEY_SIZE, GFP_KERNEL); 1566 if (!key) { 1567 err = -ENOMEM; 1568 goto ice_vsi_cfg_rss_exit; 1569 } 1570 1571 if (vsi->rss_hkey_user) 1572 memcpy(key, vsi->rss_hkey_user, ICE_GET_SET_RSS_KEY_EXTEND_KEY_SIZE); 1573 else 1574 netdev_rss_key_fill((void *)key, ICE_GET_SET_RSS_KEY_EXTEND_KEY_SIZE); 1575 1576 err = ice_set_rss_key(vsi, key); 1577 if (err) 1578 dev_err(dev, "set_rss_key failed, error %d\n", err); 1579 1580 kfree(key); 1581 ice_vsi_cfg_rss_exit: 1582 kfree(lut); 1583 return err; 1584 } 1585 1586 /** 1587 * ice_vsi_set_vf_rss_flow_fld - Sets VF VSI RSS input set for different flows 1588 * @vsi: VSI to be configured 1589 * 1590 * This function will only be called during the VF VSI setup. Upon successful 1591 * completion of package download, this function will configure default RSS 1592 * input sets for VF VSI. 1593 */ 1594 static void ice_vsi_set_vf_rss_flow_fld(struct ice_vsi *vsi) 1595 { 1596 struct ice_pf *pf = vsi->back; 1597 struct device *dev; 1598 int status; 1599 1600 dev = ice_pf_to_dev(pf); 1601 if (ice_is_safe_mode(pf)) { 1602 dev_dbg(dev, "Advanced RSS disabled. Package download failed, vsi num = %d\n", 1603 vsi->vsi_num); 1604 return; 1605 } 1606 1607 status = ice_add_avf_rss_cfg(&pf->hw, vsi->idx, ICE_DEFAULT_RSS_HENA); 1608 if (status) 1609 dev_dbg(dev, "ice_add_avf_rss_cfg failed for vsi = %d, error = %d\n", 1610 vsi->vsi_num, status); 1611 } 1612 1613 /** 1614 * ice_vsi_set_rss_flow_fld - Sets RSS input set for different flows 1615 * @vsi: VSI to be configured 1616 * 1617 * This function will only be called after successful download package call 1618 * during initialization of PF. Since the downloaded package will erase the 1619 * RSS section, this function will configure RSS input sets for different 1620 * flow types. The last profile added has the highest priority, therefore 2 1621 * tuple profiles (i.e. IPv4 src/dst) are added before 4 tuple profiles 1622 * (i.e. IPv4 src/dst TCP src/dst port). 1623 */ 1624 static void ice_vsi_set_rss_flow_fld(struct ice_vsi *vsi) 1625 { 1626 u16 vsi_handle = vsi->idx, vsi_num = vsi->vsi_num; 1627 struct ice_pf *pf = vsi->back; 1628 struct ice_hw *hw = &pf->hw; 1629 struct device *dev; 1630 int status; 1631 1632 dev = ice_pf_to_dev(pf); 1633 if (ice_is_safe_mode(pf)) { 1634 dev_dbg(dev, "Advanced RSS disabled. Package download failed, vsi num = %d\n", 1635 vsi_num); 1636 return; 1637 } 1638 /* configure RSS for IPv4 with input set IP src/dst */ 1639 status = ice_add_rss_cfg(hw, vsi_handle, ICE_FLOW_HASH_IPV4, 1640 ICE_FLOW_SEG_HDR_IPV4); 1641 if (status) 1642 dev_dbg(dev, "ice_add_rss_cfg failed for ipv4 flow, vsi = %d, error = %d\n", 1643 vsi_num, status); 1644 1645 /* configure RSS for IPv6 with input set IPv6 src/dst */ 1646 status = ice_add_rss_cfg(hw, vsi_handle, ICE_FLOW_HASH_IPV6, 1647 ICE_FLOW_SEG_HDR_IPV6); 1648 if (status) 1649 dev_dbg(dev, "ice_add_rss_cfg failed for ipv6 flow, vsi = %d, error = %d\n", 1650 vsi_num, status); 1651 1652 /* configure RSS for tcp4 with input set IP src/dst, TCP src/dst */ 1653 status = ice_add_rss_cfg(hw, vsi_handle, ICE_HASH_TCP_IPV4, 1654 ICE_FLOW_SEG_HDR_TCP | ICE_FLOW_SEG_HDR_IPV4); 1655 if (status) 1656 dev_dbg(dev, "ice_add_rss_cfg failed for tcp4 flow, vsi = %d, error = %d\n", 1657 vsi_num, status); 1658 1659 /* configure RSS for udp4 with input set IP src/dst, UDP src/dst */ 1660 status = ice_add_rss_cfg(hw, vsi_handle, ICE_HASH_UDP_IPV4, 1661 ICE_FLOW_SEG_HDR_UDP | ICE_FLOW_SEG_HDR_IPV4); 1662 if (status) 1663 dev_dbg(dev, "ice_add_rss_cfg failed for udp4 flow, vsi = %d, error = %d\n", 1664 vsi_num, status); 1665 1666 /* configure RSS for sctp4 with input set IP src/dst */ 1667 status = ice_add_rss_cfg(hw, vsi_handle, ICE_FLOW_HASH_IPV4, 1668 ICE_FLOW_SEG_HDR_SCTP | ICE_FLOW_SEG_HDR_IPV4); 1669 if (status) 1670 dev_dbg(dev, "ice_add_rss_cfg failed for sctp4 flow, vsi = %d, error = %d\n", 1671 vsi_num, status); 1672 1673 /* configure RSS for tcp6 with input set IPv6 src/dst, TCP src/dst */ 1674 status = ice_add_rss_cfg(hw, vsi_handle, ICE_HASH_TCP_IPV6, 1675 ICE_FLOW_SEG_HDR_TCP | ICE_FLOW_SEG_HDR_IPV6); 1676 if (status) 1677 dev_dbg(dev, "ice_add_rss_cfg failed for tcp6 flow, vsi = %d, error = %d\n", 1678 vsi_num, status); 1679 1680 /* configure RSS for udp6 with input set IPv6 src/dst, UDP src/dst */ 1681 status = ice_add_rss_cfg(hw, vsi_handle, ICE_HASH_UDP_IPV6, 1682 ICE_FLOW_SEG_HDR_UDP | ICE_FLOW_SEG_HDR_IPV6); 1683 if (status) 1684 dev_dbg(dev, "ice_add_rss_cfg failed for udp6 flow, vsi = %d, error = %d\n", 1685 vsi_num, status); 1686 1687 /* configure RSS for sctp6 with input set IPv6 src/dst */ 1688 status = ice_add_rss_cfg(hw, vsi_handle, ICE_FLOW_HASH_IPV6, 1689 ICE_FLOW_SEG_HDR_SCTP | ICE_FLOW_SEG_HDR_IPV6); 1690 if (status) 1691 dev_dbg(dev, "ice_add_rss_cfg failed for sctp6 flow, vsi = %d, error = %d\n", 1692 vsi_num, status); 1693 1694 status = ice_add_rss_cfg(hw, vsi_handle, ICE_FLOW_HASH_ESP_SPI, 1695 ICE_FLOW_SEG_HDR_ESP); 1696 if (status) 1697 dev_dbg(dev, "ice_add_rss_cfg failed for esp/spi flow, vsi = %d, error = %d\n", 1698 vsi_num, status); 1699 } 1700 1701 /** 1702 * ice_pf_state_is_nominal - checks the PF for nominal state 1703 * @pf: pointer to PF to check 1704 * 1705 * Check the PF's state for a collection of bits that would indicate 1706 * the PF is in a state that would inhibit normal operation for 1707 * driver functionality. 1708 * 1709 * Returns true if PF is in a nominal state, false otherwise 1710 */ 1711 bool ice_pf_state_is_nominal(struct ice_pf *pf) 1712 { 1713 DECLARE_BITMAP(check_bits, ICE_STATE_NBITS) = { 0 }; 1714 1715 if (!pf) 1716 return false; 1717 1718 bitmap_set(check_bits, 0, ICE_STATE_NOMINAL_CHECK_BITS); 1719 if (bitmap_intersects(pf->state, check_bits, ICE_STATE_NBITS)) 1720 return false; 1721 1722 return true; 1723 } 1724 1725 /** 1726 * ice_update_eth_stats - Update VSI-specific ethernet statistics counters 1727 * @vsi: the VSI to be updated 1728 */ 1729 void ice_update_eth_stats(struct ice_vsi *vsi) 1730 { 1731 struct ice_eth_stats *prev_es, *cur_es; 1732 struct ice_hw *hw = &vsi->back->hw; 1733 struct ice_pf *pf = vsi->back; 1734 u16 vsi_num = vsi->vsi_num; /* HW absolute index of a VSI */ 1735 1736 prev_es = &vsi->eth_stats_prev; 1737 cur_es = &vsi->eth_stats; 1738 1739 if (ice_is_reset_in_progress(pf->state)) 1740 vsi->stat_offsets_loaded = false; 1741 1742 ice_stat_update40(hw, GLV_GORCL(vsi_num), vsi->stat_offsets_loaded, 1743 &prev_es->rx_bytes, &cur_es->rx_bytes); 1744 1745 ice_stat_update40(hw, GLV_UPRCL(vsi_num), vsi->stat_offsets_loaded, 1746 &prev_es->rx_unicast, &cur_es->rx_unicast); 1747 1748 ice_stat_update40(hw, GLV_MPRCL(vsi_num), vsi->stat_offsets_loaded, 1749 &prev_es->rx_multicast, &cur_es->rx_multicast); 1750 1751 ice_stat_update40(hw, GLV_BPRCL(vsi_num), vsi->stat_offsets_loaded, 1752 &prev_es->rx_broadcast, &cur_es->rx_broadcast); 1753 1754 ice_stat_update32(hw, GLV_RDPC(vsi_num), vsi->stat_offsets_loaded, 1755 &prev_es->rx_discards, &cur_es->rx_discards); 1756 1757 ice_stat_update40(hw, GLV_GOTCL(vsi_num), vsi->stat_offsets_loaded, 1758 &prev_es->tx_bytes, &cur_es->tx_bytes); 1759 1760 ice_stat_update40(hw, GLV_UPTCL(vsi_num), vsi->stat_offsets_loaded, 1761 &prev_es->tx_unicast, &cur_es->tx_unicast); 1762 1763 ice_stat_update40(hw, GLV_MPTCL(vsi_num), vsi->stat_offsets_loaded, 1764 &prev_es->tx_multicast, &cur_es->tx_multicast); 1765 1766 ice_stat_update40(hw, GLV_BPTCL(vsi_num), vsi->stat_offsets_loaded, 1767 &prev_es->tx_broadcast, &cur_es->tx_broadcast); 1768 1769 ice_stat_update32(hw, GLV_TEPC(vsi_num), vsi->stat_offsets_loaded, 1770 &prev_es->tx_errors, &cur_es->tx_errors); 1771 1772 vsi->stat_offsets_loaded = true; 1773 } 1774 1775 /** 1776 * ice_vsi_cfg_frame_size - setup max frame size and Rx buffer length 1777 * @vsi: VSI 1778 */ 1779 void ice_vsi_cfg_frame_size(struct ice_vsi *vsi) 1780 { 1781 if (!vsi->netdev || test_bit(ICE_FLAG_LEGACY_RX, vsi->back->flags)) { 1782 vsi->max_frame = ICE_MAX_FRAME_LEGACY_RX; 1783 vsi->rx_buf_len = ICE_RXBUF_1664; 1784 #if (PAGE_SIZE < 8192) 1785 } else if (!ICE_2K_TOO_SMALL_WITH_PADDING && 1786 (vsi->netdev->mtu <= ETH_DATA_LEN)) { 1787 vsi->max_frame = ICE_RXBUF_1536 - NET_IP_ALIGN; 1788 vsi->rx_buf_len = ICE_RXBUF_1536 - NET_IP_ALIGN; 1789 #endif 1790 } else { 1791 vsi->max_frame = ICE_AQ_SET_MAC_FRAME_SIZE_MAX; 1792 vsi->rx_buf_len = ICE_RXBUF_3072; 1793 } 1794 } 1795 1796 /** 1797 * ice_write_qrxflxp_cntxt - write/configure QRXFLXP_CNTXT register 1798 * @hw: HW pointer 1799 * @pf_q: index of the Rx queue in the PF's queue space 1800 * @rxdid: flexible descriptor RXDID 1801 * @prio: priority for the RXDID for this queue 1802 * @ena_ts: true to enable timestamp and false to disable timestamp 1803 */ 1804 void 1805 ice_write_qrxflxp_cntxt(struct ice_hw *hw, u16 pf_q, u32 rxdid, u32 prio, 1806 bool ena_ts) 1807 { 1808 int regval = rd32(hw, QRXFLXP_CNTXT(pf_q)); 1809 1810 /* clear any previous values */ 1811 regval &= ~(QRXFLXP_CNTXT_RXDID_IDX_M | 1812 QRXFLXP_CNTXT_RXDID_PRIO_M | 1813 QRXFLXP_CNTXT_TS_M); 1814 1815 regval |= (rxdid << QRXFLXP_CNTXT_RXDID_IDX_S) & 1816 QRXFLXP_CNTXT_RXDID_IDX_M; 1817 1818 regval |= (prio << QRXFLXP_CNTXT_RXDID_PRIO_S) & 1819 QRXFLXP_CNTXT_RXDID_PRIO_M; 1820 1821 if (ena_ts) 1822 /* Enable TimeSync on this queue */ 1823 regval |= QRXFLXP_CNTXT_TS_M; 1824 1825 wr32(hw, QRXFLXP_CNTXT(pf_q), regval); 1826 } 1827 1828 int ice_vsi_cfg_single_rxq(struct ice_vsi *vsi, u16 q_idx) 1829 { 1830 if (q_idx >= vsi->num_rxq) 1831 return -EINVAL; 1832 1833 return ice_vsi_cfg_rxq(vsi->rx_rings[q_idx]); 1834 } 1835 1836 int ice_vsi_cfg_single_txq(struct ice_vsi *vsi, struct ice_tx_ring **tx_rings, u16 q_idx) 1837 { 1838 struct ice_aqc_add_tx_qgrp *qg_buf; 1839 int err; 1840 1841 if (q_idx >= vsi->alloc_txq || !tx_rings || !tx_rings[q_idx]) 1842 return -EINVAL; 1843 1844 qg_buf = kzalloc(struct_size(qg_buf, txqs, 1), GFP_KERNEL); 1845 if (!qg_buf) 1846 return -ENOMEM; 1847 1848 qg_buf->num_txqs = 1; 1849 1850 err = ice_vsi_cfg_txq(vsi, tx_rings[q_idx], qg_buf); 1851 kfree(qg_buf); 1852 return err; 1853 } 1854 1855 /** 1856 * ice_vsi_cfg_rxqs - Configure the VSI for Rx 1857 * @vsi: the VSI being configured 1858 * 1859 * Return 0 on success and a negative value on error 1860 * Configure the Rx VSI for operation. 1861 */ 1862 int ice_vsi_cfg_rxqs(struct ice_vsi *vsi) 1863 { 1864 u16 i; 1865 1866 if (vsi->type == ICE_VSI_VF) 1867 goto setup_rings; 1868 1869 ice_vsi_cfg_frame_size(vsi); 1870 setup_rings: 1871 /* set up individual rings */ 1872 ice_for_each_rxq(vsi, i) { 1873 int err = ice_vsi_cfg_rxq(vsi->rx_rings[i]); 1874 1875 if (err) 1876 return err; 1877 } 1878 1879 return 0; 1880 } 1881 1882 /** 1883 * ice_vsi_cfg_txqs - Configure the VSI for Tx 1884 * @vsi: the VSI being configured 1885 * @rings: Tx ring array to be configured 1886 * @count: number of Tx ring array elements 1887 * 1888 * Return 0 on success and a negative value on error 1889 * Configure the Tx VSI for operation. 1890 */ 1891 static int 1892 ice_vsi_cfg_txqs(struct ice_vsi *vsi, struct ice_tx_ring **rings, u16 count) 1893 { 1894 struct ice_aqc_add_tx_qgrp *qg_buf; 1895 u16 q_idx = 0; 1896 int err = 0; 1897 1898 qg_buf = kzalloc(struct_size(qg_buf, txqs, 1), GFP_KERNEL); 1899 if (!qg_buf) 1900 return -ENOMEM; 1901 1902 qg_buf->num_txqs = 1; 1903 1904 for (q_idx = 0; q_idx < count; q_idx++) { 1905 err = ice_vsi_cfg_txq(vsi, rings[q_idx], qg_buf); 1906 if (err) 1907 goto err_cfg_txqs; 1908 } 1909 1910 err_cfg_txqs: 1911 kfree(qg_buf); 1912 return err; 1913 } 1914 1915 /** 1916 * ice_vsi_cfg_lan_txqs - Configure the VSI for Tx 1917 * @vsi: the VSI being configured 1918 * 1919 * Return 0 on success and a negative value on error 1920 * Configure the Tx VSI for operation. 1921 */ 1922 int ice_vsi_cfg_lan_txqs(struct ice_vsi *vsi) 1923 { 1924 return ice_vsi_cfg_txqs(vsi, vsi->tx_rings, vsi->num_txq); 1925 } 1926 1927 /** 1928 * ice_vsi_cfg_xdp_txqs - Configure Tx queues dedicated for XDP in given VSI 1929 * @vsi: the VSI being configured 1930 * 1931 * Return 0 on success and a negative value on error 1932 * Configure the Tx queues dedicated for XDP in given VSI for operation. 1933 */ 1934 int ice_vsi_cfg_xdp_txqs(struct ice_vsi *vsi) 1935 { 1936 int ret; 1937 int i; 1938 1939 ret = ice_vsi_cfg_txqs(vsi, vsi->xdp_rings, vsi->num_xdp_txq); 1940 if (ret) 1941 return ret; 1942 1943 ice_for_each_rxq(vsi, i) 1944 ice_tx_xsk_pool(vsi, i); 1945 1946 return 0; 1947 } 1948 1949 /** 1950 * ice_intrl_usec_to_reg - convert interrupt rate limit to register value 1951 * @intrl: interrupt rate limit in usecs 1952 * @gran: interrupt rate limit granularity in usecs 1953 * 1954 * This function converts a decimal interrupt rate limit in usecs to the format 1955 * expected by firmware. 1956 */ 1957 static u32 ice_intrl_usec_to_reg(u8 intrl, u8 gran) 1958 { 1959 u32 val = intrl / gran; 1960 1961 if (val) 1962 return val | GLINT_RATE_INTRL_ENA_M; 1963 return 0; 1964 } 1965 1966 /** 1967 * ice_write_intrl - write throttle rate limit to interrupt specific register 1968 * @q_vector: pointer to interrupt specific structure 1969 * @intrl: throttle rate limit in microseconds to write 1970 */ 1971 void ice_write_intrl(struct ice_q_vector *q_vector, u8 intrl) 1972 { 1973 struct ice_hw *hw = &q_vector->vsi->back->hw; 1974 1975 wr32(hw, GLINT_RATE(q_vector->reg_idx), 1976 ice_intrl_usec_to_reg(intrl, ICE_INTRL_GRAN_ABOVE_25)); 1977 } 1978 1979 static struct ice_q_vector *ice_pull_qvec_from_rc(struct ice_ring_container *rc) 1980 { 1981 switch (rc->type) { 1982 case ICE_RX_CONTAINER: 1983 if (rc->rx_ring) 1984 return rc->rx_ring->q_vector; 1985 break; 1986 case ICE_TX_CONTAINER: 1987 if (rc->tx_ring) 1988 return rc->tx_ring->q_vector; 1989 break; 1990 default: 1991 break; 1992 } 1993 1994 return NULL; 1995 } 1996 1997 /** 1998 * __ice_write_itr - write throttle rate to register 1999 * @q_vector: pointer to interrupt data structure 2000 * @rc: pointer to ring container 2001 * @itr: throttle rate in microseconds to write 2002 */ 2003 static void __ice_write_itr(struct ice_q_vector *q_vector, 2004 struct ice_ring_container *rc, u16 itr) 2005 { 2006 struct ice_hw *hw = &q_vector->vsi->back->hw; 2007 2008 wr32(hw, GLINT_ITR(rc->itr_idx, q_vector->reg_idx), 2009 ITR_REG_ALIGN(itr) >> ICE_ITR_GRAN_S); 2010 } 2011 2012 /** 2013 * ice_write_itr - write throttle rate to queue specific register 2014 * @rc: pointer to ring container 2015 * @itr: throttle rate in microseconds to write 2016 */ 2017 void ice_write_itr(struct ice_ring_container *rc, u16 itr) 2018 { 2019 struct ice_q_vector *q_vector; 2020 2021 q_vector = ice_pull_qvec_from_rc(rc); 2022 if (!q_vector) 2023 return; 2024 2025 __ice_write_itr(q_vector, rc, itr); 2026 } 2027 2028 /** 2029 * ice_set_q_vector_intrl - set up interrupt rate limiting 2030 * @q_vector: the vector to be configured 2031 * 2032 * Interrupt rate limiting is local to the vector, not per-queue so we must 2033 * detect if either ring container has dynamic moderation enabled to decide 2034 * what to set the interrupt rate limit to via INTRL settings. In the case that 2035 * dynamic moderation is disabled on both, write the value with the cached 2036 * setting to make sure INTRL register matches the user visible value. 2037 */ 2038 void ice_set_q_vector_intrl(struct ice_q_vector *q_vector) 2039 { 2040 if (ITR_IS_DYNAMIC(&q_vector->tx) || ITR_IS_DYNAMIC(&q_vector->rx)) { 2041 /* in the case of dynamic enabled, cap each vector to no more 2042 * than (4 us) 250,000 ints/sec, which allows low latency 2043 * but still less than 500,000 interrupts per second, which 2044 * reduces CPU a bit in the case of the lowest latency 2045 * setting. The 4 here is a value in microseconds. 2046 */ 2047 ice_write_intrl(q_vector, 4); 2048 } else { 2049 ice_write_intrl(q_vector, q_vector->intrl); 2050 } 2051 } 2052 2053 /** 2054 * ice_vsi_cfg_msix - MSIX mode Interrupt Config in the HW 2055 * @vsi: the VSI being configured 2056 * 2057 * This configures MSIX mode interrupts for the PF VSI, and should not be used 2058 * for the VF VSI. 2059 */ 2060 void ice_vsi_cfg_msix(struct ice_vsi *vsi) 2061 { 2062 struct ice_pf *pf = vsi->back; 2063 struct ice_hw *hw = &pf->hw; 2064 u16 txq = 0, rxq = 0; 2065 int i, q; 2066 2067 ice_for_each_q_vector(vsi, i) { 2068 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 2069 u16 reg_idx = q_vector->reg_idx; 2070 2071 ice_cfg_itr(hw, q_vector); 2072 2073 /* Both Transmit Queue Interrupt Cause Control register 2074 * and Receive Queue Interrupt Cause control register 2075 * expects MSIX_INDX field to be the vector index 2076 * within the function space and not the absolute 2077 * vector index across PF or across device. 2078 * For SR-IOV VF VSIs queue vector index always starts 2079 * with 1 since first vector index(0) is used for OICR 2080 * in VF space. Since VMDq and other PF VSIs are within 2081 * the PF function space, use the vector index that is 2082 * tracked for this PF. 2083 */ 2084 for (q = 0; q < q_vector->num_ring_tx; q++) { 2085 ice_cfg_txq_interrupt(vsi, txq, reg_idx, 2086 q_vector->tx.itr_idx); 2087 txq++; 2088 } 2089 2090 for (q = 0; q < q_vector->num_ring_rx; q++) { 2091 ice_cfg_rxq_interrupt(vsi, rxq, reg_idx, 2092 q_vector->rx.itr_idx); 2093 rxq++; 2094 } 2095 } 2096 } 2097 2098 /** 2099 * ice_vsi_start_all_rx_rings - start/enable all of a VSI's Rx rings 2100 * @vsi: the VSI whose rings are to be enabled 2101 * 2102 * Returns 0 on success and a negative value on error 2103 */ 2104 int ice_vsi_start_all_rx_rings(struct ice_vsi *vsi) 2105 { 2106 return ice_vsi_ctrl_all_rx_rings(vsi, true); 2107 } 2108 2109 /** 2110 * ice_vsi_stop_all_rx_rings - stop/disable all of a VSI's Rx rings 2111 * @vsi: the VSI whose rings are to be disabled 2112 * 2113 * Returns 0 on success and a negative value on error 2114 */ 2115 int ice_vsi_stop_all_rx_rings(struct ice_vsi *vsi) 2116 { 2117 return ice_vsi_ctrl_all_rx_rings(vsi, false); 2118 } 2119 2120 /** 2121 * ice_vsi_stop_tx_rings - Disable Tx rings 2122 * @vsi: the VSI being configured 2123 * @rst_src: reset source 2124 * @rel_vmvf_num: Relative ID of VF/VM 2125 * @rings: Tx ring array to be stopped 2126 * @count: number of Tx ring array elements 2127 */ 2128 static int 2129 ice_vsi_stop_tx_rings(struct ice_vsi *vsi, enum ice_disq_rst_src rst_src, 2130 u16 rel_vmvf_num, struct ice_tx_ring **rings, u16 count) 2131 { 2132 u16 q_idx; 2133 2134 if (vsi->num_txq > ICE_LAN_TXQ_MAX_QDIS) 2135 return -EINVAL; 2136 2137 for (q_idx = 0; q_idx < count; q_idx++) { 2138 struct ice_txq_meta txq_meta = { }; 2139 int status; 2140 2141 if (!rings || !rings[q_idx]) 2142 return -EINVAL; 2143 2144 ice_fill_txq_meta(vsi, rings[q_idx], &txq_meta); 2145 status = ice_vsi_stop_tx_ring(vsi, rst_src, rel_vmvf_num, 2146 rings[q_idx], &txq_meta); 2147 2148 if (status) 2149 return status; 2150 } 2151 2152 return 0; 2153 } 2154 2155 /** 2156 * ice_vsi_stop_lan_tx_rings - Disable LAN Tx rings 2157 * @vsi: the VSI being configured 2158 * @rst_src: reset source 2159 * @rel_vmvf_num: Relative ID of VF/VM 2160 */ 2161 int 2162 ice_vsi_stop_lan_tx_rings(struct ice_vsi *vsi, enum ice_disq_rst_src rst_src, 2163 u16 rel_vmvf_num) 2164 { 2165 return ice_vsi_stop_tx_rings(vsi, rst_src, rel_vmvf_num, vsi->tx_rings, vsi->num_txq); 2166 } 2167 2168 /** 2169 * ice_vsi_stop_xdp_tx_rings - Disable XDP Tx rings 2170 * @vsi: the VSI being configured 2171 */ 2172 int ice_vsi_stop_xdp_tx_rings(struct ice_vsi *vsi) 2173 { 2174 return ice_vsi_stop_tx_rings(vsi, ICE_NO_RESET, 0, vsi->xdp_rings, vsi->num_xdp_txq); 2175 } 2176 2177 /** 2178 * ice_vsi_is_rx_queue_active 2179 * @vsi: the VSI being configured 2180 * 2181 * Return true if at least one queue is active. 2182 */ 2183 bool ice_vsi_is_rx_queue_active(struct ice_vsi *vsi) 2184 { 2185 struct ice_pf *pf = vsi->back; 2186 struct ice_hw *hw = &pf->hw; 2187 int i; 2188 2189 ice_for_each_rxq(vsi, i) { 2190 u32 rx_reg; 2191 int pf_q; 2192 2193 pf_q = vsi->rxq_map[i]; 2194 rx_reg = rd32(hw, QRX_CTRL(pf_q)); 2195 if (rx_reg & QRX_CTRL_QENA_STAT_M) 2196 return true; 2197 } 2198 2199 return false; 2200 } 2201 2202 /** 2203 * ice_vsi_is_vlan_pruning_ena - check if VLAN pruning is enabled or not 2204 * @vsi: VSI to check whether or not VLAN pruning is enabled. 2205 * 2206 * returns true if Rx VLAN pruning is enabled and false otherwise. 2207 */ 2208 bool ice_vsi_is_vlan_pruning_ena(struct ice_vsi *vsi) 2209 { 2210 if (!vsi) 2211 return false; 2212 2213 return (vsi->info.sw_flags2 & ICE_AQ_VSI_SW_FLAG_RX_VLAN_PRUNE_ENA); 2214 } 2215 2216 static void ice_vsi_set_tc_cfg(struct ice_vsi *vsi) 2217 { 2218 if (!test_bit(ICE_FLAG_DCB_ENA, vsi->back->flags)) { 2219 vsi->tc_cfg.ena_tc = ICE_DFLT_TRAFFIC_CLASS; 2220 vsi->tc_cfg.numtc = 1; 2221 return; 2222 } 2223 2224 /* set VSI TC information based on DCB config */ 2225 ice_vsi_set_dcb_tc_cfg(vsi); 2226 } 2227 2228 /** 2229 * ice_cfg_sw_lldp - Config switch rules for LLDP packet handling 2230 * @vsi: the VSI being configured 2231 * @tx: bool to determine Tx or Rx rule 2232 * @create: bool to determine create or remove Rule 2233 */ 2234 void ice_cfg_sw_lldp(struct ice_vsi *vsi, bool tx, bool create) 2235 { 2236 int (*eth_fltr)(struct ice_vsi *v, u16 type, u16 flag, 2237 enum ice_sw_fwd_act_type act); 2238 struct ice_pf *pf = vsi->back; 2239 struct device *dev; 2240 int status; 2241 2242 dev = ice_pf_to_dev(pf); 2243 eth_fltr = create ? ice_fltr_add_eth : ice_fltr_remove_eth; 2244 2245 if (tx) { 2246 status = eth_fltr(vsi, ETH_P_LLDP, ICE_FLTR_TX, 2247 ICE_DROP_PACKET); 2248 } else { 2249 if (ice_fw_supports_lldp_fltr_ctrl(&pf->hw)) { 2250 status = ice_lldp_fltr_add_remove(&pf->hw, vsi->vsi_num, 2251 create); 2252 } else { 2253 status = eth_fltr(vsi, ETH_P_LLDP, ICE_FLTR_RX, 2254 ICE_FWD_TO_VSI); 2255 } 2256 } 2257 2258 if (status) 2259 dev_dbg(dev, "Fail %s %s LLDP rule on VSI %i error: %d\n", 2260 create ? "adding" : "removing", tx ? "TX" : "RX", 2261 vsi->vsi_num, status); 2262 } 2263 2264 /** 2265 * ice_set_agg_vsi - sets up scheduler aggregator node and move VSI into it 2266 * @vsi: pointer to the VSI 2267 * 2268 * This function will allocate new scheduler aggregator now if needed and will 2269 * move specified VSI into it. 2270 */ 2271 static void ice_set_agg_vsi(struct ice_vsi *vsi) 2272 { 2273 struct device *dev = ice_pf_to_dev(vsi->back); 2274 struct ice_agg_node *agg_node_iter = NULL; 2275 u32 agg_id = ICE_INVALID_AGG_NODE_ID; 2276 struct ice_agg_node *agg_node = NULL; 2277 int node_offset, max_agg_nodes = 0; 2278 struct ice_port_info *port_info; 2279 struct ice_pf *pf = vsi->back; 2280 u32 agg_node_id_start = 0; 2281 int status; 2282 2283 /* create (as needed) scheduler aggregator node and move VSI into 2284 * corresponding aggregator node 2285 * - PF aggregator node to contains VSIs of type _PF and _CTRL 2286 * - VF aggregator nodes will contain VF VSI 2287 */ 2288 port_info = pf->hw.port_info; 2289 if (!port_info) 2290 return; 2291 2292 switch (vsi->type) { 2293 case ICE_VSI_CTRL: 2294 case ICE_VSI_CHNL: 2295 case ICE_VSI_LB: 2296 case ICE_VSI_PF: 2297 case ICE_VSI_SWITCHDEV_CTRL: 2298 max_agg_nodes = ICE_MAX_PF_AGG_NODES; 2299 agg_node_id_start = ICE_PF_AGG_NODE_ID_START; 2300 agg_node_iter = &pf->pf_agg_node[0]; 2301 break; 2302 case ICE_VSI_VF: 2303 /* user can create 'n' VFs on a given PF, but since max children 2304 * per aggregator node can be only 64. Following code handles 2305 * aggregator(s) for VF VSIs, either selects a agg_node which 2306 * was already created provided num_vsis < 64, otherwise 2307 * select next available node, which will be created 2308 */ 2309 max_agg_nodes = ICE_MAX_VF_AGG_NODES; 2310 agg_node_id_start = ICE_VF_AGG_NODE_ID_START; 2311 agg_node_iter = &pf->vf_agg_node[0]; 2312 break; 2313 default: 2314 /* other VSI type, handle later if needed */ 2315 dev_dbg(dev, "unexpected VSI type %s\n", 2316 ice_vsi_type_str(vsi->type)); 2317 return; 2318 } 2319 2320 /* find the appropriate aggregator node */ 2321 for (node_offset = 0; node_offset < max_agg_nodes; node_offset++) { 2322 /* see if we can find space in previously created 2323 * node if num_vsis < 64, otherwise skip 2324 */ 2325 if (agg_node_iter->num_vsis && 2326 agg_node_iter->num_vsis == ICE_MAX_VSIS_IN_AGG_NODE) { 2327 agg_node_iter++; 2328 continue; 2329 } 2330 2331 if (agg_node_iter->valid && 2332 agg_node_iter->agg_id != ICE_INVALID_AGG_NODE_ID) { 2333 agg_id = agg_node_iter->agg_id; 2334 agg_node = agg_node_iter; 2335 break; 2336 } 2337 2338 /* find unclaimed agg_id */ 2339 if (agg_node_iter->agg_id == ICE_INVALID_AGG_NODE_ID) { 2340 agg_id = node_offset + agg_node_id_start; 2341 agg_node = agg_node_iter; 2342 break; 2343 } 2344 /* move to next agg_node */ 2345 agg_node_iter++; 2346 } 2347 2348 if (!agg_node) 2349 return; 2350 2351 /* if selected aggregator node was not created, create it */ 2352 if (!agg_node->valid) { 2353 status = ice_cfg_agg(port_info, agg_id, ICE_AGG_TYPE_AGG, 2354 (u8)vsi->tc_cfg.ena_tc); 2355 if (status) { 2356 dev_err(dev, "unable to create aggregator node with agg_id %u\n", 2357 agg_id); 2358 return; 2359 } 2360 /* aggregator node is created, store the needed info */ 2361 agg_node->valid = true; 2362 agg_node->agg_id = agg_id; 2363 } 2364 2365 /* move VSI to corresponding aggregator node */ 2366 status = ice_move_vsi_to_agg(port_info, agg_id, vsi->idx, 2367 (u8)vsi->tc_cfg.ena_tc); 2368 if (status) { 2369 dev_err(dev, "unable to move VSI idx %u into aggregator %u node", 2370 vsi->idx, agg_id); 2371 return; 2372 } 2373 2374 /* keep active children count for aggregator node */ 2375 agg_node->num_vsis++; 2376 2377 /* cache the 'agg_id' in VSI, so that after reset - VSI will be moved 2378 * to aggregator node 2379 */ 2380 vsi->agg_node = agg_node; 2381 dev_dbg(dev, "successfully moved VSI idx %u tc_bitmap 0x%x) into aggregator node %d which has num_vsis %u\n", 2382 vsi->idx, vsi->tc_cfg.ena_tc, vsi->agg_node->agg_id, 2383 vsi->agg_node->num_vsis); 2384 } 2385 2386 static int ice_vsi_cfg_tc_lan(struct ice_pf *pf, struct ice_vsi *vsi) 2387 { 2388 u16 max_txqs[ICE_MAX_TRAFFIC_CLASS] = { 0 }; 2389 struct device *dev = ice_pf_to_dev(pf); 2390 int ret, i; 2391 2392 /* configure VSI nodes based on number of queues and TC's */ 2393 ice_for_each_traffic_class(i) { 2394 if (!(vsi->tc_cfg.ena_tc & BIT(i))) 2395 continue; 2396 2397 if (vsi->type == ICE_VSI_CHNL) { 2398 if (!vsi->alloc_txq && vsi->num_txq) 2399 max_txqs[i] = vsi->num_txq; 2400 else 2401 max_txqs[i] = pf->num_lan_tx; 2402 } else { 2403 max_txqs[i] = vsi->alloc_txq; 2404 } 2405 } 2406 2407 dev_dbg(dev, "vsi->tc_cfg.ena_tc = %d\n", vsi->tc_cfg.ena_tc); 2408 ret = ice_cfg_vsi_lan(vsi->port_info, vsi->idx, vsi->tc_cfg.ena_tc, 2409 max_txqs); 2410 if (ret) { 2411 dev_err(dev, "VSI %d failed lan queue config, error %d\n", 2412 vsi->vsi_num, ret); 2413 return ret; 2414 } 2415 2416 return 0; 2417 } 2418 2419 /** 2420 * ice_vsi_cfg_def - configure default VSI based on the type 2421 * @vsi: pointer to VSI 2422 * @params: the parameters to configure this VSI with 2423 */ 2424 static int 2425 ice_vsi_cfg_def(struct ice_vsi *vsi, struct ice_vsi_cfg_params *params) 2426 { 2427 struct device *dev = ice_pf_to_dev(vsi->back); 2428 struct ice_pf *pf = vsi->back; 2429 int ret; 2430 2431 vsi->vsw = pf->first_sw; 2432 2433 ret = ice_vsi_alloc_def(vsi, params->ch); 2434 if (ret) 2435 return ret; 2436 2437 /* allocate memory for Tx/Rx ring stat pointers */ 2438 ret = ice_vsi_alloc_stat_arrays(vsi); 2439 if (ret) 2440 goto unroll_vsi_alloc; 2441 2442 ice_alloc_fd_res(vsi); 2443 2444 ret = ice_vsi_get_qs(vsi); 2445 if (ret) { 2446 dev_err(dev, "Failed to allocate queues. vsi->idx = %d\n", 2447 vsi->idx); 2448 goto unroll_vsi_alloc_stat; 2449 } 2450 2451 /* set RSS capabilities */ 2452 ice_vsi_set_rss_params(vsi); 2453 2454 /* set TC configuration */ 2455 ice_vsi_set_tc_cfg(vsi); 2456 2457 /* create the VSI */ 2458 ret = ice_vsi_init(vsi, params->flags); 2459 if (ret) 2460 goto unroll_get_qs; 2461 2462 ice_vsi_init_vlan_ops(vsi); 2463 2464 switch (vsi->type) { 2465 case ICE_VSI_CTRL: 2466 case ICE_VSI_SWITCHDEV_CTRL: 2467 case ICE_VSI_PF: 2468 ret = ice_vsi_alloc_q_vectors(vsi); 2469 if (ret) 2470 goto unroll_vsi_init; 2471 2472 ret = ice_vsi_alloc_rings(vsi); 2473 if (ret) 2474 goto unroll_vector_base; 2475 2476 ret = ice_vsi_alloc_ring_stats(vsi); 2477 if (ret) 2478 goto unroll_vector_base; 2479 2480 ice_vsi_map_rings_to_vectors(vsi); 2481 if (ice_is_xdp_ena_vsi(vsi)) { 2482 ret = ice_vsi_determine_xdp_res(vsi); 2483 if (ret) 2484 goto unroll_vector_base; 2485 ret = ice_prepare_xdp_rings(vsi, vsi->xdp_prog); 2486 if (ret) 2487 goto unroll_vector_base; 2488 } 2489 2490 /* ICE_VSI_CTRL does not need RSS so skip RSS processing */ 2491 if (vsi->type != ICE_VSI_CTRL) 2492 /* Do not exit if configuring RSS had an issue, at 2493 * least receive traffic on first queue. Hence no 2494 * need to capture return value 2495 */ 2496 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 2497 ice_vsi_cfg_rss_lut_key(vsi); 2498 ice_vsi_set_rss_flow_fld(vsi); 2499 } 2500 ice_init_arfs(vsi); 2501 break; 2502 case ICE_VSI_CHNL: 2503 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 2504 ice_vsi_cfg_rss_lut_key(vsi); 2505 ice_vsi_set_rss_flow_fld(vsi); 2506 } 2507 break; 2508 case ICE_VSI_VF: 2509 /* VF driver will take care of creating netdev for this type and 2510 * map queues to vectors through Virtchnl, PF driver only 2511 * creates a VSI and corresponding structures for bookkeeping 2512 * purpose 2513 */ 2514 ret = ice_vsi_alloc_q_vectors(vsi); 2515 if (ret) 2516 goto unroll_vsi_init; 2517 2518 ret = ice_vsi_alloc_rings(vsi); 2519 if (ret) 2520 goto unroll_alloc_q_vector; 2521 2522 ret = ice_vsi_alloc_ring_stats(vsi); 2523 if (ret) 2524 goto unroll_vector_base; 2525 /* Do not exit if configuring RSS had an issue, at least 2526 * receive traffic on first queue. Hence no need to capture 2527 * return value 2528 */ 2529 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 2530 ice_vsi_cfg_rss_lut_key(vsi); 2531 ice_vsi_set_vf_rss_flow_fld(vsi); 2532 } 2533 break; 2534 case ICE_VSI_LB: 2535 ret = ice_vsi_alloc_rings(vsi); 2536 if (ret) 2537 goto unroll_vsi_init; 2538 2539 ret = ice_vsi_alloc_ring_stats(vsi); 2540 if (ret) 2541 goto unroll_vector_base; 2542 2543 break; 2544 default: 2545 /* clean up the resources and exit */ 2546 ret = -EINVAL; 2547 goto unroll_vsi_init; 2548 } 2549 2550 return 0; 2551 2552 unroll_vector_base: 2553 /* reclaim SW interrupts back to the common pool */ 2554 unroll_alloc_q_vector: 2555 ice_vsi_free_q_vectors(vsi); 2556 unroll_vsi_init: 2557 ice_vsi_delete_from_hw(vsi); 2558 unroll_get_qs: 2559 ice_vsi_put_qs(vsi); 2560 unroll_vsi_alloc_stat: 2561 ice_vsi_free_stats(vsi); 2562 unroll_vsi_alloc: 2563 ice_vsi_free_arrays(vsi); 2564 return ret; 2565 } 2566 2567 /** 2568 * ice_vsi_cfg - configure a previously allocated VSI 2569 * @vsi: pointer to VSI 2570 * @params: parameters used to configure this VSI 2571 */ 2572 int ice_vsi_cfg(struct ice_vsi *vsi, struct ice_vsi_cfg_params *params) 2573 { 2574 struct ice_pf *pf = vsi->back; 2575 int ret; 2576 2577 if (WARN_ON(params->type == ICE_VSI_VF && !params->vf)) 2578 return -EINVAL; 2579 2580 vsi->type = params->type; 2581 vsi->port_info = params->pi; 2582 2583 /* For VSIs which don't have a connected VF, this will be NULL */ 2584 vsi->vf = params->vf; 2585 2586 ret = ice_vsi_cfg_def(vsi, params); 2587 if (ret) 2588 return ret; 2589 2590 ret = ice_vsi_cfg_tc_lan(vsi->back, vsi); 2591 if (ret) 2592 ice_vsi_decfg(vsi); 2593 2594 if (vsi->type == ICE_VSI_CTRL) { 2595 if (vsi->vf) { 2596 WARN_ON(vsi->vf->ctrl_vsi_idx != ICE_NO_VSI); 2597 vsi->vf->ctrl_vsi_idx = vsi->idx; 2598 } else { 2599 WARN_ON(pf->ctrl_vsi_idx != ICE_NO_VSI); 2600 pf->ctrl_vsi_idx = vsi->idx; 2601 } 2602 } 2603 2604 return ret; 2605 } 2606 2607 /** 2608 * ice_vsi_decfg - remove all VSI configuration 2609 * @vsi: pointer to VSI 2610 */ 2611 void ice_vsi_decfg(struct ice_vsi *vsi) 2612 { 2613 struct ice_pf *pf = vsi->back; 2614 int err; 2615 2616 /* The Rx rule will only exist to remove if the LLDP FW 2617 * engine is currently stopped 2618 */ 2619 if (!ice_is_safe_mode(pf) && vsi->type == ICE_VSI_PF && 2620 !test_bit(ICE_FLAG_FW_LLDP_AGENT, pf->flags)) 2621 ice_cfg_sw_lldp(vsi, false, false); 2622 2623 ice_rm_vsi_lan_cfg(vsi->port_info, vsi->idx); 2624 err = ice_rm_vsi_rdma_cfg(vsi->port_info, vsi->idx); 2625 if (err) 2626 dev_err(ice_pf_to_dev(pf), "Failed to remove RDMA scheduler config for VSI %u, err %d\n", 2627 vsi->vsi_num, err); 2628 2629 if (ice_is_xdp_ena_vsi(vsi)) 2630 /* return value check can be skipped here, it always returns 2631 * 0 if reset is in progress 2632 */ 2633 ice_destroy_xdp_rings(vsi); 2634 2635 ice_vsi_clear_rings(vsi); 2636 ice_vsi_free_q_vectors(vsi); 2637 ice_vsi_put_qs(vsi); 2638 ice_vsi_free_arrays(vsi); 2639 2640 /* SR-IOV determines needed MSIX resources all at once instead of per 2641 * VSI since when VFs are spawned we know how many VFs there are and how 2642 * many interrupts each VF needs. SR-IOV MSIX resources are also 2643 * cleared in the same manner. 2644 */ 2645 2646 if (vsi->type == ICE_VSI_VF && 2647 vsi->agg_node && vsi->agg_node->valid) 2648 vsi->agg_node->num_vsis--; 2649 if (vsi->agg_node) { 2650 vsi->agg_node->valid = false; 2651 vsi->agg_node->agg_id = 0; 2652 } 2653 } 2654 2655 /** 2656 * ice_vsi_setup - Set up a VSI by a given type 2657 * @pf: board private structure 2658 * @params: parameters to use when creating the VSI 2659 * 2660 * This allocates the sw VSI structure and its queue resources. 2661 * 2662 * Returns pointer to the successfully allocated and configured VSI sw struct on 2663 * success, NULL on failure. 2664 */ 2665 struct ice_vsi * 2666 ice_vsi_setup(struct ice_pf *pf, struct ice_vsi_cfg_params *params) 2667 { 2668 struct device *dev = ice_pf_to_dev(pf); 2669 struct ice_vsi *vsi; 2670 int ret; 2671 2672 /* ice_vsi_setup can only initialize a new VSI, and we must have 2673 * a port_info structure for it. 2674 */ 2675 if (WARN_ON(!(params->flags & ICE_VSI_FLAG_INIT)) || 2676 WARN_ON(!params->pi)) 2677 return NULL; 2678 2679 vsi = ice_vsi_alloc(pf); 2680 if (!vsi) { 2681 dev_err(dev, "could not allocate VSI\n"); 2682 return NULL; 2683 } 2684 2685 ret = ice_vsi_cfg(vsi, params); 2686 if (ret) 2687 goto err_vsi_cfg; 2688 2689 /* Add switch rule to drop all Tx Flow Control Frames, of look up 2690 * type ETHERTYPE from VSIs, and restrict malicious VF from sending 2691 * out PAUSE or PFC frames. If enabled, FW can still send FC frames. 2692 * The rule is added once for PF VSI in order to create appropriate 2693 * recipe, since VSI/VSI list is ignored with drop action... 2694 * Also add rules to handle LLDP Tx packets. Tx LLDP packets need to 2695 * be dropped so that VFs cannot send LLDP packets to reconfig DCB 2696 * settings in the HW. 2697 */ 2698 if (!ice_is_safe_mode(pf) && vsi->type == ICE_VSI_PF) { 2699 ice_fltr_add_eth(vsi, ETH_P_PAUSE, ICE_FLTR_TX, 2700 ICE_DROP_PACKET); 2701 ice_cfg_sw_lldp(vsi, true, true); 2702 } 2703 2704 if (!vsi->agg_node) 2705 ice_set_agg_vsi(vsi); 2706 2707 return vsi; 2708 2709 err_vsi_cfg: 2710 ice_vsi_free(vsi); 2711 2712 return NULL; 2713 } 2714 2715 /** 2716 * ice_vsi_release_msix - Clear the queue to Interrupt mapping in HW 2717 * @vsi: the VSI being cleaned up 2718 */ 2719 static void ice_vsi_release_msix(struct ice_vsi *vsi) 2720 { 2721 struct ice_pf *pf = vsi->back; 2722 struct ice_hw *hw = &pf->hw; 2723 u32 txq = 0; 2724 u32 rxq = 0; 2725 int i, q; 2726 2727 ice_for_each_q_vector(vsi, i) { 2728 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 2729 2730 ice_write_intrl(q_vector, 0); 2731 for (q = 0; q < q_vector->num_ring_tx; q++) { 2732 ice_write_itr(&q_vector->tx, 0); 2733 wr32(hw, QINT_TQCTL(vsi->txq_map[txq]), 0); 2734 if (ice_is_xdp_ena_vsi(vsi)) { 2735 u32 xdp_txq = txq + vsi->num_xdp_txq; 2736 2737 wr32(hw, QINT_TQCTL(vsi->txq_map[xdp_txq]), 0); 2738 } 2739 txq++; 2740 } 2741 2742 for (q = 0; q < q_vector->num_ring_rx; q++) { 2743 ice_write_itr(&q_vector->rx, 0); 2744 wr32(hw, QINT_RQCTL(vsi->rxq_map[rxq]), 0); 2745 rxq++; 2746 } 2747 } 2748 2749 ice_flush(hw); 2750 } 2751 2752 /** 2753 * ice_vsi_free_irq - Free the IRQ association with the OS 2754 * @vsi: the VSI being configured 2755 */ 2756 void ice_vsi_free_irq(struct ice_vsi *vsi) 2757 { 2758 struct ice_pf *pf = vsi->back; 2759 int i; 2760 2761 if (!vsi->q_vectors || !vsi->irqs_ready) 2762 return; 2763 2764 ice_vsi_release_msix(vsi); 2765 if (vsi->type == ICE_VSI_VF) 2766 return; 2767 2768 vsi->irqs_ready = false; 2769 ice_free_cpu_rx_rmap(vsi); 2770 2771 ice_for_each_q_vector(vsi, i) { 2772 int irq_num; 2773 2774 irq_num = vsi->q_vectors[i]->irq.virq; 2775 2776 /* free only the irqs that were actually requested */ 2777 if (!vsi->q_vectors[i] || 2778 !(vsi->q_vectors[i]->num_ring_tx || 2779 vsi->q_vectors[i]->num_ring_rx)) 2780 continue; 2781 2782 /* clear the affinity notifier in the IRQ descriptor */ 2783 if (!IS_ENABLED(CONFIG_RFS_ACCEL)) 2784 irq_set_affinity_notifier(irq_num, NULL); 2785 2786 /* clear the affinity_mask in the IRQ descriptor */ 2787 irq_set_affinity_hint(irq_num, NULL); 2788 synchronize_irq(irq_num); 2789 devm_free_irq(ice_pf_to_dev(pf), irq_num, vsi->q_vectors[i]); 2790 } 2791 } 2792 2793 /** 2794 * ice_vsi_free_tx_rings - Free Tx resources for VSI queues 2795 * @vsi: the VSI having resources freed 2796 */ 2797 void ice_vsi_free_tx_rings(struct ice_vsi *vsi) 2798 { 2799 int i; 2800 2801 if (!vsi->tx_rings) 2802 return; 2803 2804 ice_for_each_txq(vsi, i) 2805 if (vsi->tx_rings[i] && vsi->tx_rings[i]->desc) 2806 ice_free_tx_ring(vsi->tx_rings[i]); 2807 } 2808 2809 /** 2810 * ice_vsi_free_rx_rings - Free Rx resources for VSI queues 2811 * @vsi: the VSI having resources freed 2812 */ 2813 void ice_vsi_free_rx_rings(struct ice_vsi *vsi) 2814 { 2815 int i; 2816 2817 if (!vsi->rx_rings) 2818 return; 2819 2820 ice_for_each_rxq(vsi, i) 2821 if (vsi->rx_rings[i] && vsi->rx_rings[i]->desc) 2822 ice_free_rx_ring(vsi->rx_rings[i]); 2823 } 2824 2825 /** 2826 * ice_vsi_close - Shut down a VSI 2827 * @vsi: the VSI being shut down 2828 */ 2829 void ice_vsi_close(struct ice_vsi *vsi) 2830 { 2831 if (!test_and_set_bit(ICE_VSI_DOWN, vsi->state)) 2832 ice_down(vsi); 2833 2834 ice_vsi_free_irq(vsi); 2835 ice_vsi_free_tx_rings(vsi); 2836 ice_vsi_free_rx_rings(vsi); 2837 } 2838 2839 /** 2840 * ice_ena_vsi - resume a VSI 2841 * @vsi: the VSI being resume 2842 * @locked: is the rtnl_lock already held 2843 */ 2844 int ice_ena_vsi(struct ice_vsi *vsi, bool locked) 2845 { 2846 int err = 0; 2847 2848 if (!test_bit(ICE_VSI_NEEDS_RESTART, vsi->state)) 2849 return 0; 2850 2851 clear_bit(ICE_VSI_NEEDS_RESTART, vsi->state); 2852 2853 if (vsi->netdev && vsi->type == ICE_VSI_PF) { 2854 if (netif_running(vsi->netdev)) { 2855 if (!locked) 2856 rtnl_lock(); 2857 2858 err = ice_open_internal(vsi->netdev); 2859 2860 if (!locked) 2861 rtnl_unlock(); 2862 } 2863 } else if (vsi->type == ICE_VSI_CTRL) { 2864 err = ice_vsi_open_ctrl(vsi); 2865 } 2866 2867 return err; 2868 } 2869 2870 /** 2871 * ice_dis_vsi - pause a VSI 2872 * @vsi: the VSI being paused 2873 * @locked: is the rtnl_lock already held 2874 */ 2875 void ice_dis_vsi(struct ice_vsi *vsi, bool locked) 2876 { 2877 if (test_bit(ICE_VSI_DOWN, vsi->state)) 2878 return; 2879 2880 set_bit(ICE_VSI_NEEDS_RESTART, vsi->state); 2881 2882 if (vsi->type == ICE_VSI_PF && vsi->netdev) { 2883 if (netif_running(vsi->netdev)) { 2884 if (!locked) 2885 rtnl_lock(); 2886 2887 ice_vsi_close(vsi); 2888 2889 if (!locked) 2890 rtnl_unlock(); 2891 } else { 2892 ice_vsi_close(vsi); 2893 } 2894 } else if (vsi->type == ICE_VSI_CTRL || 2895 vsi->type == ICE_VSI_SWITCHDEV_CTRL) { 2896 ice_vsi_close(vsi); 2897 } 2898 } 2899 2900 /** 2901 * ice_vsi_dis_irq - Mask off queue interrupt generation on the VSI 2902 * @vsi: the VSI being un-configured 2903 */ 2904 void ice_vsi_dis_irq(struct ice_vsi *vsi) 2905 { 2906 struct ice_pf *pf = vsi->back; 2907 struct ice_hw *hw = &pf->hw; 2908 u32 val; 2909 int i; 2910 2911 /* disable interrupt causation from each queue */ 2912 if (vsi->tx_rings) { 2913 ice_for_each_txq(vsi, i) { 2914 if (vsi->tx_rings[i]) { 2915 u16 reg; 2916 2917 reg = vsi->tx_rings[i]->reg_idx; 2918 val = rd32(hw, QINT_TQCTL(reg)); 2919 val &= ~QINT_TQCTL_CAUSE_ENA_M; 2920 wr32(hw, QINT_TQCTL(reg), val); 2921 } 2922 } 2923 } 2924 2925 if (vsi->rx_rings) { 2926 ice_for_each_rxq(vsi, i) { 2927 if (vsi->rx_rings[i]) { 2928 u16 reg; 2929 2930 reg = vsi->rx_rings[i]->reg_idx; 2931 val = rd32(hw, QINT_RQCTL(reg)); 2932 val &= ~QINT_RQCTL_CAUSE_ENA_M; 2933 wr32(hw, QINT_RQCTL(reg), val); 2934 } 2935 } 2936 } 2937 2938 /* disable each interrupt */ 2939 ice_for_each_q_vector(vsi, i) { 2940 if (!vsi->q_vectors[i]) 2941 continue; 2942 wr32(hw, GLINT_DYN_CTL(vsi->q_vectors[i]->reg_idx), 0); 2943 } 2944 2945 ice_flush(hw); 2946 2947 /* don't call synchronize_irq() for VF's from the host */ 2948 if (vsi->type == ICE_VSI_VF) 2949 return; 2950 2951 ice_for_each_q_vector(vsi, i) 2952 synchronize_irq(vsi->q_vectors[i]->irq.virq); 2953 } 2954 2955 /** 2956 * ice_napi_del - Remove NAPI handler for the VSI 2957 * @vsi: VSI for which NAPI handler is to be removed 2958 */ 2959 void ice_napi_del(struct ice_vsi *vsi) 2960 { 2961 int v_idx; 2962 2963 if (!vsi->netdev) 2964 return; 2965 2966 ice_for_each_q_vector(vsi, v_idx) 2967 netif_napi_del(&vsi->q_vectors[v_idx]->napi); 2968 } 2969 2970 /** 2971 * ice_vsi_release - Delete a VSI and free its resources 2972 * @vsi: the VSI being removed 2973 * 2974 * Returns 0 on success or < 0 on error 2975 */ 2976 int ice_vsi_release(struct ice_vsi *vsi) 2977 { 2978 struct ice_pf *pf; 2979 2980 if (!vsi->back) 2981 return -ENODEV; 2982 pf = vsi->back; 2983 2984 /* do not unregister while driver is in the reset recovery pending 2985 * state. Since reset/rebuild happens through PF service task workqueue, 2986 * it's not a good idea to unregister netdev that is associated to the 2987 * PF that is running the work queue items currently. This is done to 2988 * avoid check_flush_dependency() warning on this wq 2989 */ 2990 if (vsi->netdev && !ice_is_reset_in_progress(pf->state) && 2991 (test_bit(ICE_VSI_NETDEV_REGISTERED, vsi->state))) { 2992 unregister_netdev(vsi->netdev); 2993 clear_bit(ICE_VSI_NETDEV_REGISTERED, vsi->state); 2994 } 2995 2996 if (vsi->type == ICE_VSI_PF) 2997 ice_devlink_destroy_pf_port(pf); 2998 2999 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) 3000 ice_rss_clean(vsi); 3001 3002 ice_vsi_close(vsi); 3003 ice_vsi_decfg(vsi); 3004 3005 if (vsi->netdev) { 3006 if (test_bit(ICE_VSI_NETDEV_REGISTERED, vsi->state)) { 3007 unregister_netdev(vsi->netdev); 3008 clear_bit(ICE_VSI_NETDEV_REGISTERED, vsi->state); 3009 } 3010 if (test_bit(ICE_VSI_NETDEV_ALLOCD, vsi->state)) { 3011 free_netdev(vsi->netdev); 3012 vsi->netdev = NULL; 3013 clear_bit(ICE_VSI_NETDEV_ALLOCD, vsi->state); 3014 } 3015 } 3016 3017 /* retain SW VSI data structure since it is needed to unregister and 3018 * free VSI netdev when PF is not in reset recovery pending state,\ 3019 * for ex: during rmmod. 3020 */ 3021 if (!ice_is_reset_in_progress(pf->state)) 3022 ice_vsi_delete(vsi); 3023 3024 return 0; 3025 } 3026 3027 /** 3028 * ice_vsi_rebuild_get_coalesce - get coalesce from all q_vectors 3029 * @vsi: VSI connected with q_vectors 3030 * @coalesce: array of struct with stored coalesce 3031 * 3032 * Returns array size. 3033 */ 3034 static int 3035 ice_vsi_rebuild_get_coalesce(struct ice_vsi *vsi, 3036 struct ice_coalesce_stored *coalesce) 3037 { 3038 int i; 3039 3040 ice_for_each_q_vector(vsi, i) { 3041 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 3042 3043 coalesce[i].itr_tx = q_vector->tx.itr_settings; 3044 coalesce[i].itr_rx = q_vector->rx.itr_settings; 3045 coalesce[i].intrl = q_vector->intrl; 3046 3047 if (i < vsi->num_txq) 3048 coalesce[i].tx_valid = true; 3049 if (i < vsi->num_rxq) 3050 coalesce[i].rx_valid = true; 3051 } 3052 3053 return vsi->num_q_vectors; 3054 } 3055 3056 /** 3057 * ice_vsi_rebuild_set_coalesce - set coalesce from earlier saved arrays 3058 * @vsi: VSI connected with q_vectors 3059 * @coalesce: pointer to array of struct with stored coalesce 3060 * @size: size of coalesce array 3061 * 3062 * Before this function, ice_vsi_rebuild_get_coalesce should be called to save 3063 * ITR params in arrays. If size is 0 or coalesce wasn't stored set coalesce 3064 * to default value. 3065 */ 3066 static void 3067 ice_vsi_rebuild_set_coalesce(struct ice_vsi *vsi, 3068 struct ice_coalesce_stored *coalesce, int size) 3069 { 3070 struct ice_ring_container *rc; 3071 int i; 3072 3073 if ((size && !coalesce) || !vsi) 3074 return; 3075 3076 /* There are a couple of cases that have to be handled here: 3077 * 1. The case where the number of queue vectors stays the same, but 3078 * the number of Tx or Rx rings changes (the first for loop) 3079 * 2. The case where the number of queue vectors increased (the 3080 * second for loop) 3081 */ 3082 for (i = 0; i < size && i < vsi->num_q_vectors; i++) { 3083 /* There are 2 cases to handle here and they are the same for 3084 * both Tx and Rx: 3085 * if the entry was valid previously (coalesce[i].[tr]x_valid 3086 * and the loop variable is less than the number of rings 3087 * allocated, then write the previous values 3088 * 3089 * if the entry was not valid previously, but the number of 3090 * rings is less than are allocated (this means the number of 3091 * rings increased from previously), then write out the 3092 * values in the first element 3093 * 3094 * Also, always write the ITR, even if in ITR_IS_DYNAMIC 3095 * as there is no harm because the dynamic algorithm 3096 * will just overwrite. 3097 */ 3098 if (i < vsi->alloc_rxq && coalesce[i].rx_valid) { 3099 rc = &vsi->q_vectors[i]->rx; 3100 rc->itr_settings = coalesce[i].itr_rx; 3101 ice_write_itr(rc, rc->itr_setting); 3102 } else if (i < vsi->alloc_rxq) { 3103 rc = &vsi->q_vectors[i]->rx; 3104 rc->itr_settings = coalesce[0].itr_rx; 3105 ice_write_itr(rc, rc->itr_setting); 3106 } 3107 3108 if (i < vsi->alloc_txq && coalesce[i].tx_valid) { 3109 rc = &vsi->q_vectors[i]->tx; 3110 rc->itr_settings = coalesce[i].itr_tx; 3111 ice_write_itr(rc, rc->itr_setting); 3112 } else if (i < vsi->alloc_txq) { 3113 rc = &vsi->q_vectors[i]->tx; 3114 rc->itr_settings = coalesce[0].itr_tx; 3115 ice_write_itr(rc, rc->itr_setting); 3116 } 3117 3118 vsi->q_vectors[i]->intrl = coalesce[i].intrl; 3119 ice_set_q_vector_intrl(vsi->q_vectors[i]); 3120 } 3121 3122 /* the number of queue vectors increased so write whatever is in 3123 * the first element 3124 */ 3125 for (; i < vsi->num_q_vectors; i++) { 3126 /* transmit */ 3127 rc = &vsi->q_vectors[i]->tx; 3128 rc->itr_settings = coalesce[0].itr_tx; 3129 ice_write_itr(rc, rc->itr_setting); 3130 3131 /* receive */ 3132 rc = &vsi->q_vectors[i]->rx; 3133 rc->itr_settings = coalesce[0].itr_rx; 3134 ice_write_itr(rc, rc->itr_setting); 3135 3136 vsi->q_vectors[i]->intrl = coalesce[0].intrl; 3137 ice_set_q_vector_intrl(vsi->q_vectors[i]); 3138 } 3139 } 3140 3141 /** 3142 * ice_vsi_realloc_stat_arrays - Frees unused stat structures 3143 * @vsi: VSI pointer 3144 * @prev_txq: Number of Tx rings before ring reallocation 3145 * @prev_rxq: Number of Rx rings before ring reallocation 3146 */ 3147 static void 3148 ice_vsi_realloc_stat_arrays(struct ice_vsi *vsi, int prev_txq, int prev_rxq) 3149 { 3150 struct ice_vsi_stats *vsi_stat; 3151 struct ice_pf *pf = vsi->back; 3152 int i; 3153 3154 if (!prev_txq || !prev_rxq) 3155 return; 3156 if (vsi->type == ICE_VSI_CHNL) 3157 return; 3158 3159 vsi_stat = pf->vsi_stats[vsi->idx]; 3160 3161 if (vsi->num_txq < prev_txq) { 3162 for (i = vsi->num_txq; i < prev_txq; i++) { 3163 if (vsi_stat->tx_ring_stats[i]) { 3164 kfree_rcu(vsi_stat->tx_ring_stats[i], rcu); 3165 WRITE_ONCE(vsi_stat->tx_ring_stats[i], NULL); 3166 } 3167 } 3168 } 3169 3170 if (vsi->num_rxq < prev_rxq) { 3171 for (i = vsi->num_rxq; i < prev_rxq; i++) { 3172 if (vsi_stat->rx_ring_stats[i]) { 3173 kfree_rcu(vsi_stat->rx_ring_stats[i], rcu); 3174 WRITE_ONCE(vsi_stat->rx_ring_stats[i], NULL); 3175 } 3176 } 3177 } 3178 } 3179 3180 /** 3181 * ice_vsi_rebuild - Rebuild VSI after reset 3182 * @vsi: VSI to be rebuild 3183 * @vsi_flags: flags used for VSI rebuild flow 3184 * 3185 * Set vsi_flags to ICE_VSI_FLAG_INIT to initialize a new VSI, or 3186 * ICE_VSI_FLAG_NO_INIT to rebuild an existing VSI in hardware. 3187 * 3188 * Returns 0 on success and negative value on failure 3189 */ 3190 int ice_vsi_rebuild(struct ice_vsi *vsi, u32 vsi_flags) 3191 { 3192 struct ice_vsi_cfg_params params = {}; 3193 struct ice_coalesce_stored *coalesce; 3194 int ret, prev_txq, prev_rxq; 3195 int prev_num_q_vectors = 0; 3196 struct ice_pf *pf; 3197 3198 if (!vsi) 3199 return -EINVAL; 3200 3201 params = ice_vsi_to_params(vsi); 3202 params.flags = vsi_flags; 3203 3204 pf = vsi->back; 3205 if (WARN_ON(vsi->type == ICE_VSI_VF && !vsi->vf)) 3206 return -EINVAL; 3207 3208 coalesce = kcalloc(vsi->num_q_vectors, 3209 sizeof(struct ice_coalesce_stored), GFP_KERNEL); 3210 if (!coalesce) 3211 return -ENOMEM; 3212 3213 prev_num_q_vectors = ice_vsi_rebuild_get_coalesce(vsi, coalesce); 3214 3215 prev_txq = vsi->num_txq; 3216 prev_rxq = vsi->num_rxq; 3217 3218 ice_vsi_decfg(vsi); 3219 ret = ice_vsi_cfg_def(vsi, ¶ms); 3220 if (ret) 3221 goto err_vsi_cfg; 3222 3223 ret = ice_vsi_cfg_tc_lan(pf, vsi); 3224 if (ret) { 3225 if (vsi_flags & ICE_VSI_FLAG_INIT) { 3226 ret = -EIO; 3227 goto err_vsi_cfg_tc_lan; 3228 } 3229 3230 kfree(coalesce); 3231 return ice_schedule_reset(pf, ICE_RESET_PFR); 3232 } 3233 3234 ice_vsi_realloc_stat_arrays(vsi, prev_txq, prev_rxq); 3235 3236 ice_vsi_rebuild_set_coalesce(vsi, coalesce, prev_num_q_vectors); 3237 kfree(coalesce); 3238 3239 return 0; 3240 3241 err_vsi_cfg_tc_lan: 3242 ice_vsi_decfg(vsi); 3243 err_vsi_cfg: 3244 kfree(coalesce); 3245 return ret; 3246 } 3247 3248 /** 3249 * ice_is_reset_in_progress - check for a reset in progress 3250 * @state: PF state field 3251 */ 3252 bool ice_is_reset_in_progress(unsigned long *state) 3253 { 3254 return test_bit(ICE_RESET_OICR_RECV, state) || 3255 test_bit(ICE_PFR_REQ, state) || 3256 test_bit(ICE_CORER_REQ, state) || 3257 test_bit(ICE_GLOBR_REQ, state); 3258 } 3259 3260 /** 3261 * ice_wait_for_reset - Wait for driver to finish reset and rebuild 3262 * @pf: pointer to the PF structure 3263 * @timeout: length of time to wait, in jiffies 3264 * 3265 * Wait (sleep) for a short time until the driver finishes cleaning up from 3266 * a device reset. The caller must be able to sleep. Use this to delay 3267 * operations that could fail while the driver is cleaning up after a device 3268 * reset. 3269 * 3270 * Returns 0 on success, -EBUSY if the reset is not finished within the 3271 * timeout, and -ERESTARTSYS if the thread was interrupted. 3272 */ 3273 int ice_wait_for_reset(struct ice_pf *pf, unsigned long timeout) 3274 { 3275 long ret; 3276 3277 ret = wait_event_interruptible_timeout(pf->reset_wait_queue, 3278 !ice_is_reset_in_progress(pf->state), 3279 timeout); 3280 if (ret < 0) 3281 return ret; 3282 else if (!ret) 3283 return -EBUSY; 3284 else 3285 return 0; 3286 } 3287 3288 /** 3289 * ice_vsi_update_q_map - update our copy of the VSI info with new queue map 3290 * @vsi: VSI being configured 3291 * @ctx: the context buffer returned from AQ VSI update command 3292 */ 3293 static void ice_vsi_update_q_map(struct ice_vsi *vsi, struct ice_vsi_ctx *ctx) 3294 { 3295 vsi->info.mapping_flags = ctx->info.mapping_flags; 3296 memcpy(&vsi->info.q_mapping, &ctx->info.q_mapping, 3297 sizeof(vsi->info.q_mapping)); 3298 memcpy(&vsi->info.tc_mapping, ctx->info.tc_mapping, 3299 sizeof(vsi->info.tc_mapping)); 3300 } 3301 3302 /** 3303 * ice_vsi_cfg_netdev_tc - Setup the netdev TC configuration 3304 * @vsi: the VSI being configured 3305 * @ena_tc: TC map to be enabled 3306 */ 3307 void ice_vsi_cfg_netdev_tc(struct ice_vsi *vsi, u8 ena_tc) 3308 { 3309 struct net_device *netdev = vsi->netdev; 3310 struct ice_pf *pf = vsi->back; 3311 int numtc = vsi->tc_cfg.numtc; 3312 struct ice_dcbx_cfg *dcbcfg; 3313 u8 netdev_tc; 3314 int i; 3315 3316 if (!netdev) 3317 return; 3318 3319 /* CHNL VSI doesn't have it's own netdev, hence, no netdev_tc */ 3320 if (vsi->type == ICE_VSI_CHNL) 3321 return; 3322 3323 if (!ena_tc) { 3324 netdev_reset_tc(netdev); 3325 return; 3326 } 3327 3328 if (vsi->type == ICE_VSI_PF && ice_is_adq_active(pf)) 3329 numtc = vsi->all_numtc; 3330 3331 if (netdev_set_num_tc(netdev, numtc)) 3332 return; 3333 3334 dcbcfg = &pf->hw.port_info->qos_cfg.local_dcbx_cfg; 3335 3336 ice_for_each_traffic_class(i) 3337 if (vsi->tc_cfg.ena_tc & BIT(i)) 3338 netdev_set_tc_queue(netdev, 3339 vsi->tc_cfg.tc_info[i].netdev_tc, 3340 vsi->tc_cfg.tc_info[i].qcount_tx, 3341 vsi->tc_cfg.tc_info[i].qoffset); 3342 /* setup TC queue map for CHNL TCs */ 3343 ice_for_each_chnl_tc(i) { 3344 if (!(vsi->all_enatc & BIT(i))) 3345 break; 3346 if (!vsi->mqprio_qopt.qopt.count[i]) 3347 break; 3348 netdev_set_tc_queue(netdev, i, 3349 vsi->mqprio_qopt.qopt.count[i], 3350 vsi->mqprio_qopt.qopt.offset[i]); 3351 } 3352 3353 if (test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3354 return; 3355 3356 for (i = 0; i < ICE_MAX_USER_PRIORITY; i++) { 3357 u8 ets_tc = dcbcfg->etscfg.prio_table[i]; 3358 3359 /* Get the mapped netdev TC# for the UP */ 3360 netdev_tc = vsi->tc_cfg.tc_info[ets_tc].netdev_tc; 3361 netdev_set_prio_tc_map(netdev, i, netdev_tc); 3362 } 3363 } 3364 3365 /** 3366 * ice_vsi_setup_q_map_mqprio - Prepares mqprio based tc_config 3367 * @vsi: the VSI being configured, 3368 * @ctxt: VSI context structure 3369 * @ena_tc: number of traffic classes to enable 3370 * 3371 * Prepares VSI tc_config to have queue configurations based on MQPRIO options. 3372 */ 3373 static int 3374 ice_vsi_setup_q_map_mqprio(struct ice_vsi *vsi, struct ice_vsi_ctx *ctxt, 3375 u8 ena_tc) 3376 { 3377 u16 pow, offset = 0, qcount_tx = 0, qcount_rx = 0, qmap; 3378 u16 tc0_offset = vsi->mqprio_qopt.qopt.offset[0]; 3379 int tc0_qcount = vsi->mqprio_qopt.qopt.count[0]; 3380 u16 new_txq, new_rxq; 3381 u8 netdev_tc = 0; 3382 int i; 3383 3384 vsi->tc_cfg.ena_tc = ena_tc ? ena_tc : 1; 3385 3386 pow = order_base_2(tc0_qcount); 3387 qmap = ((tc0_offset << ICE_AQ_VSI_TC_Q_OFFSET_S) & 3388 ICE_AQ_VSI_TC_Q_OFFSET_M) | 3389 ((pow << ICE_AQ_VSI_TC_Q_NUM_S) & ICE_AQ_VSI_TC_Q_NUM_M); 3390 3391 ice_for_each_traffic_class(i) { 3392 if (!(vsi->tc_cfg.ena_tc & BIT(i))) { 3393 /* TC is not enabled */ 3394 vsi->tc_cfg.tc_info[i].qoffset = 0; 3395 vsi->tc_cfg.tc_info[i].qcount_rx = 1; 3396 vsi->tc_cfg.tc_info[i].qcount_tx = 1; 3397 vsi->tc_cfg.tc_info[i].netdev_tc = 0; 3398 ctxt->info.tc_mapping[i] = 0; 3399 continue; 3400 } 3401 3402 offset = vsi->mqprio_qopt.qopt.offset[i]; 3403 qcount_rx = vsi->mqprio_qopt.qopt.count[i]; 3404 qcount_tx = vsi->mqprio_qopt.qopt.count[i]; 3405 vsi->tc_cfg.tc_info[i].qoffset = offset; 3406 vsi->tc_cfg.tc_info[i].qcount_rx = qcount_rx; 3407 vsi->tc_cfg.tc_info[i].qcount_tx = qcount_tx; 3408 vsi->tc_cfg.tc_info[i].netdev_tc = netdev_tc++; 3409 } 3410 3411 if (vsi->all_numtc && vsi->all_numtc != vsi->tc_cfg.numtc) { 3412 ice_for_each_chnl_tc(i) { 3413 if (!(vsi->all_enatc & BIT(i))) 3414 continue; 3415 offset = vsi->mqprio_qopt.qopt.offset[i]; 3416 qcount_rx = vsi->mqprio_qopt.qopt.count[i]; 3417 qcount_tx = vsi->mqprio_qopt.qopt.count[i]; 3418 } 3419 } 3420 3421 new_txq = offset + qcount_tx; 3422 if (new_txq > vsi->alloc_txq) { 3423 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Tx queues (%u), than were allocated (%u)!\n", 3424 new_txq, vsi->alloc_txq); 3425 return -EINVAL; 3426 } 3427 3428 new_rxq = offset + qcount_rx; 3429 if (new_rxq > vsi->alloc_rxq) { 3430 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Rx queues (%u), than were allocated (%u)!\n", 3431 new_rxq, vsi->alloc_rxq); 3432 return -EINVAL; 3433 } 3434 3435 /* Set actual Tx/Rx queue pairs */ 3436 vsi->num_txq = new_txq; 3437 vsi->num_rxq = new_rxq; 3438 3439 /* Setup queue TC[0].qmap for given VSI context */ 3440 ctxt->info.tc_mapping[0] = cpu_to_le16(qmap); 3441 ctxt->info.q_mapping[0] = cpu_to_le16(vsi->rxq_map[0]); 3442 ctxt->info.q_mapping[1] = cpu_to_le16(tc0_qcount); 3443 3444 /* Find queue count available for channel VSIs and starting offset 3445 * for channel VSIs 3446 */ 3447 if (tc0_qcount && tc0_qcount < vsi->num_rxq) { 3448 vsi->cnt_q_avail = vsi->num_rxq - tc0_qcount; 3449 vsi->next_base_q = tc0_qcount; 3450 } 3451 dev_dbg(ice_pf_to_dev(vsi->back), "vsi->num_txq = %d\n", vsi->num_txq); 3452 dev_dbg(ice_pf_to_dev(vsi->back), "vsi->num_rxq = %d\n", vsi->num_rxq); 3453 dev_dbg(ice_pf_to_dev(vsi->back), "all_numtc %u, all_enatc: 0x%04x, tc_cfg.numtc %u\n", 3454 vsi->all_numtc, vsi->all_enatc, vsi->tc_cfg.numtc); 3455 3456 return 0; 3457 } 3458 3459 /** 3460 * ice_vsi_cfg_tc - Configure VSI Tx Sched for given TC map 3461 * @vsi: VSI to be configured 3462 * @ena_tc: TC bitmap 3463 * 3464 * VSI queues expected to be quiesced before calling this function 3465 */ 3466 int ice_vsi_cfg_tc(struct ice_vsi *vsi, u8 ena_tc) 3467 { 3468 u16 max_txqs[ICE_MAX_TRAFFIC_CLASS] = { 0 }; 3469 struct ice_pf *pf = vsi->back; 3470 struct ice_tc_cfg old_tc_cfg; 3471 struct ice_vsi_ctx *ctx; 3472 struct device *dev; 3473 int i, ret = 0; 3474 u8 num_tc = 0; 3475 3476 dev = ice_pf_to_dev(pf); 3477 if (vsi->tc_cfg.ena_tc == ena_tc && 3478 vsi->mqprio_qopt.mode != TC_MQPRIO_MODE_CHANNEL) 3479 return 0; 3480 3481 ice_for_each_traffic_class(i) { 3482 /* build bitmap of enabled TCs */ 3483 if (ena_tc & BIT(i)) 3484 num_tc++; 3485 /* populate max_txqs per TC */ 3486 max_txqs[i] = vsi->alloc_txq; 3487 /* Update max_txqs if it is CHNL VSI, because alloc_t[r]xq are 3488 * zero for CHNL VSI, hence use num_txq instead as max_txqs 3489 */ 3490 if (vsi->type == ICE_VSI_CHNL && 3491 test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3492 max_txqs[i] = vsi->num_txq; 3493 } 3494 3495 memcpy(&old_tc_cfg, &vsi->tc_cfg, sizeof(old_tc_cfg)); 3496 vsi->tc_cfg.ena_tc = ena_tc; 3497 vsi->tc_cfg.numtc = num_tc; 3498 3499 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); 3500 if (!ctx) 3501 return -ENOMEM; 3502 3503 ctx->vf_num = 0; 3504 ctx->info = vsi->info; 3505 3506 if (vsi->type == ICE_VSI_PF && 3507 test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3508 ret = ice_vsi_setup_q_map_mqprio(vsi, ctx, ena_tc); 3509 else 3510 ret = ice_vsi_setup_q_map(vsi, ctx); 3511 3512 if (ret) { 3513 memcpy(&vsi->tc_cfg, &old_tc_cfg, sizeof(vsi->tc_cfg)); 3514 goto out; 3515 } 3516 3517 /* must to indicate which section of VSI context are being modified */ 3518 ctx->info.valid_sections = cpu_to_le16(ICE_AQ_VSI_PROP_RXQ_MAP_VALID); 3519 ret = ice_update_vsi(&pf->hw, vsi->idx, ctx, NULL); 3520 if (ret) { 3521 dev_info(dev, "Failed VSI Update\n"); 3522 goto out; 3523 } 3524 3525 if (vsi->type == ICE_VSI_PF && 3526 test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3527 ret = ice_cfg_vsi_lan(vsi->port_info, vsi->idx, 1, max_txqs); 3528 else 3529 ret = ice_cfg_vsi_lan(vsi->port_info, vsi->idx, 3530 vsi->tc_cfg.ena_tc, max_txqs); 3531 3532 if (ret) { 3533 dev_err(dev, "VSI %d failed TC config, error %d\n", 3534 vsi->vsi_num, ret); 3535 goto out; 3536 } 3537 ice_vsi_update_q_map(vsi, ctx); 3538 vsi->info.valid_sections = 0; 3539 3540 ice_vsi_cfg_netdev_tc(vsi, ena_tc); 3541 out: 3542 kfree(ctx); 3543 return ret; 3544 } 3545 3546 /** 3547 * ice_update_ring_stats - Update ring statistics 3548 * @stats: stats to be updated 3549 * @pkts: number of processed packets 3550 * @bytes: number of processed bytes 3551 * 3552 * This function assumes that caller has acquired a u64_stats_sync lock. 3553 */ 3554 static void ice_update_ring_stats(struct ice_q_stats *stats, u64 pkts, u64 bytes) 3555 { 3556 stats->bytes += bytes; 3557 stats->pkts += pkts; 3558 } 3559 3560 /** 3561 * ice_update_tx_ring_stats - Update Tx ring specific counters 3562 * @tx_ring: ring to update 3563 * @pkts: number of processed packets 3564 * @bytes: number of processed bytes 3565 */ 3566 void ice_update_tx_ring_stats(struct ice_tx_ring *tx_ring, u64 pkts, u64 bytes) 3567 { 3568 u64_stats_update_begin(&tx_ring->ring_stats->syncp); 3569 ice_update_ring_stats(&tx_ring->ring_stats->stats, pkts, bytes); 3570 u64_stats_update_end(&tx_ring->ring_stats->syncp); 3571 } 3572 3573 /** 3574 * ice_update_rx_ring_stats - Update Rx ring specific counters 3575 * @rx_ring: ring to update 3576 * @pkts: number of processed packets 3577 * @bytes: number of processed bytes 3578 */ 3579 void ice_update_rx_ring_stats(struct ice_rx_ring *rx_ring, u64 pkts, u64 bytes) 3580 { 3581 u64_stats_update_begin(&rx_ring->ring_stats->syncp); 3582 ice_update_ring_stats(&rx_ring->ring_stats->stats, pkts, bytes); 3583 u64_stats_update_end(&rx_ring->ring_stats->syncp); 3584 } 3585 3586 /** 3587 * ice_is_dflt_vsi_in_use - check if the default forwarding VSI is being used 3588 * @pi: port info of the switch with default VSI 3589 * 3590 * Return true if the there is a single VSI in default forwarding VSI list 3591 */ 3592 bool ice_is_dflt_vsi_in_use(struct ice_port_info *pi) 3593 { 3594 bool exists = false; 3595 3596 ice_check_if_dflt_vsi(pi, 0, &exists); 3597 return exists; 3598 } 3599 3600 /** 3601 * ice_is_vsi_dflt_vsi - check if the VSI passed in is the default VSI 3602 * @vsi: VSI to compare against default forwarding VSI 3603 * 3604 * If this VSI passed in is the default forwarding VSI then return true, else 3605 * return false 3606 */ 3607 bool ice_is_vsi_dflt_vsi(struct ice_vsi *vsi) 3608 { 3609 return ice_check_if_dflt_vsi(vsi->port_info, vsi->idx, NULL); 3610 } 3611 3612 /** 3613 * ice_set_dflt_vsi - set the default forwarding VSI 3614 * @vsi: VSI getting set as the default forwarding VSI on the switch 3615 * 3616 * If the VSI passed in is already the default VSI and it's enabled just return 3617 * success. 3618 * 3619 * Otherwise try to set the VSI passed in as the switch's default VSI and 3620 * return the result. 3621 */ 3622 int ice_set_dflt_vsi(struct ice_vsi *vsi) 3623 { 3624 struct device *dev; 3625 int status; 3626 3627 if (!vsi) 3628 return -EINVAL; 3629 3630 dev = ice_pf_to_dev(vsi->back); 3631 3632 /* the VSI passed in is already the default VSI */ 3633 if (ice_is_vsi_dflt_vsi(vsi)) { 3634 dev_dbg(dev, "VSI %d passed in is already the default forwarding VSI, nothing to do\n", 3635 vsi->vsi_num); 3636 return 0; 3637 } 3638 3639 status = ice_cfg_dflt_vsi(vsi->port_info, vsi->idx, true, ICE_FLTR_RX); 3640 if (status) { 3641 dev_err(dev, "Failed to set VSI %d as the default forwarding VSI, error %d\n", 3642 vsi->vsi_num, status); 3643 return status; 3644 } 3645 3646 return 0; 3647 } 3648 3649 /** 3650 * ice_clear_dflt_vsi - clear the default forwarding VSI 3651 * @vsi: VSI to remove from filter list 3652 * 3653 * If the switch has no default VSI or it's not enabled then return error. 3654 * 3655 * Otherwise try to clear the default VSI and return the result. 3656 */ 3657 int ice_clear_dflt_vsi(struct ice_vsi *vsi) 3658 { 3659 struct device *dev; 3660 int status; 3661 3662 if (!vsi) 3663 return -EINVAL; 3664 3665 dev = ice_pf_to_dev(vsi->back); 3666 3667 /* there is no default VSI configured */ 3668 if (!ice_is_dflt_vsi_in_use(vsi->port_info)) 3669 return -ENODEV; 3670 3671 status = ice_cfg_dflt_vsi(vsi->port_info, vsi->idx, false, 3672 ICE_FLTR_RX); 3673 if (status) { 3674 dev_err(dev, "Failed to clear the default forwarding VSI %d, error %d\n", 3675 vsi->vsi_num, status); 3676 return -EIO; 3677 } 3678 3679 return 0; 3680 } 3681 3682 /** 3683 * ice_get_link_speed_mbps - get link speed in Mbps 3684 * @vsi: the VSI whose link speed is being queried 3685 * 3686 * Return current VSI link speed and 0 if the speed is unknown. 3687 */ 3688 int ice_get_link_speed_mbps(struct ice_vsi *vsi) 3689 { 3690 unsigned int link_speed; 3691 3692 link_speed = vsi->port_info->phy.link_info.link_speed; 3693 3694 return (int)ice_get_link_speed(fls(link_speed) - 1); 3695 } 3696 3697 /** 3698 * ice_get_link_speed_kbps - get link speed in Kbps 3699 * @vsi: the VSI whose link speed is being queried 3700 * 3701 * Return current VSI link speed and 0 if the speed is unknown. 3702 */ 3703 int ice_get_link_speed_kbps(struct ice_vsi *vsi) 3704 { 3705 int speed_mbps; 3706 3707 speed_mbps = ice_get_link_speed_mbps(vsi); 3708 3709 return speed_mbps * 1000; 3710 } 3711 3712 /** 3713 * ice_set_min_bw_limit - setup minimum BW limit for Tx based on min_tx_rate 3714 * @vsi: VSI to be configured 3715 * @min_tx_rate: min Tx rate in Kbps to be configured as BW limit 3716 * 3717 * If the min_tx_rate is specified as 0 that means to clear the minimum BW limit 3718 * profile, otherwise a non-zero value will force a minimum BW limit for the VSI 3719 * on TC 0. 3720 */ 3721 int ice_set_min_bw_limit(struct ice_vsi *vsi, u64 min_tx_rate) 3722 { 3723 struct ice_pf *pf = vsi->back; 3724 struct device *dev; 3725 int status; 3726 int speed; 3727 3728 dev = ice_pf_to_dev(pf); 3729 if (!vsi->port_info) { 3730 dev_dbg(dev, "VSI %d, type %u specified doesn't have valid port_info\n", 3731 vsi->idx, vsi->type); 3732 return -EINVAL; 3733 } 3734 3735 speed = ice_get_link_speed_kbps(vsi); 3736 if (min_tx_rate > (u64)speed) { 3737 dev_err(dev, "invalid min Tx rate %llu Kbps specified for %s %d is greater than current link speed %u Kbps\n", 3738 min_tx_rate, ice_vsi_type_str(vsi->type), vsi->idx, 3739 speed); 3740 return -EINVAL; 3741 } 3742 3743 /* Configure min BW for VSI limit */ 3744 if (min_tx_rate) { 3745 status = ice_cfg_vsi_bw_lmt_per_tc(vsi->port_info, vsi->idx, 0, 3746 ICE_MIN_BW, min_tx_rate); 3747 if (status) { 3748 dev_err(dev, "failed to set min Tx rate(%llu Kbps) for %s %d\n", 3749 min_tx_rate, ice_vsi_type_str(vsi->type), 3750 vsi->idx); 3751 return status; 3752 } 3753 3754 dev_dbg(dev, "set min Tx rate(%llu Kbps) for %s\n", 3755 min_tx_rate, ice_vsi_type_str(vsi->type)); 3756 } else { 3757 status = ice_cfg_vsi_bw_dflt_lmt_per_tc(vsi->port_info, 3758 vsi->idx, 0, 3759 ICE_MIN_BW); 3760 if (status) { 3761 dev_err(dev, "failed to clear min Tx rate configuration for %s %d\n", 3762 ice_vsi_type_str(vsi->type), vsi->idx); 3763 return status; 3764 } 3765 3766 dev_dbg(dev, "cleared min Tx rate configuration for %s %d\n", 3767 ice_vsi_type_str(vsi->type), vsi->idx); 3768 } 3769 3770 return 0; 3771 } 3772 3773 /** 3774 * ice_set_max_bw_limit - setup maximum BW limit for Tx based on max_tx_rate 3775 * @vsi: VSI to be configured 3776 * @max_tx_rate: max Tx rate in Kbps to be configured as BW limit 3777 * 3778 * If the max_tx_rate is specified as 0 that means to clear the maximum BW limit 3779 * profile, otherwise a non-zero value will force a maximum BW limit for the VSI 3780 * on TC 0. 3781 */ 3782 int ice_set_max_bw_limit(struct ice_vsi *vsi, u64 max_tx_rate) 3783 { 3784 struct ice_pf *pf = vsi->back; 3785 struct device *dev; 3786 int status; 3787 int speed; 3788 3789 dev = ice_pf_to_dev(pf); 3790 if (!vsi->port_info) { 3791 dev_dbg(dev, "VSI %d, type %u specified doesn't have valid port_info\n", 3792 vsi->idx, vsi->type); 3793 return -EINVAL; 3794 } 3795 3796 speed = ice_get_link_speed_kbps(vsi); 3797 if (max_tx_rate > (u64)speed) { 3798 dev_err(dev, "invalid max Tx rate %llu Kbps specified for %s %d is greater than current link speed %u Kbps\n", 3799 max_tx_rate, ice_vsi_type_str(vsi->type), vsi->idx, 3800 speed); 3801 return -EINVAL; 3802 } 3803 3804 /* Configure max BW for VSI limit */ 3805 if (max_tx_rate) { 3806 status = ice_cfg_vsi_bw_lmt_per_tc(vsi->port_info, vsi->idx, 0, 3807 ICE_MAX_BW, max_tx_rate); 3808 if (status) { 3809 dev_err(dev, "failed setting max Tx rate(%llu Kbps) for %s %d\n", 3810 max_tx_rate, ice_vsi_type_str(vsi->type), 3811 vsi->idx); 3812 return status; 3813 } 3814 3815 dev_dbg(dev, "set max Tx rate(%llu Kbps) for %s %d\n", 3816 max_tx_rate, ice_vsi_type_str(vsi->type), vsi->idx); 3817 } else { 3818 status = ice_cfg_vsi_bw_dflt_lmt_per_tc(vsi->port_info, 3819 vsi->idx, 0, 3820 ICE_MAX_BW); 3821 if (status) { 3822 dev_err(dev, "failed clearing max Tx rate configuration for %s %d\n", 3823 ice_vsi_type_str(vsi->type), vsi->idx); 3824 return status; 3825 } 3826 3827 dev_dbg(dev, "cleared max Tx rate configuration for %s %d\n", 3828 ice_vsi_type_str(vsi->type), vsi->idx); 3829 } 3830 3831 return 0; 3832 } 3833 3834 /** 3835 * ice_set_link - turn on/off physical link 3836 * @vsi: VSI to modify physical link on 3837 * @ena: turn on/off physical link 3838 */ 3839 int ice_set_link(struct ice_vsi *vsi, bool ena) 3840 { 3841 struct device *dev = ice_pf_to_dev(vsi->back); 3842 struct ice_port_info *pi = vsi->port_info; 3843 struct ice_hw *hw = pi->hw; 3844 int status; 3845 3846 if (vsi->type != ICE_VSI_PF) 3847 return -EINVAL; 3848 3849 status = ice_aq_set_link_restart_an(pi, ena, NULL); 3850 3851 /* if link is owned by manageability, FW will return ICE_AQ_RC_EMODE. 3852 * this is not a fatal error, so print a warning message and return 3853 * a success code. Return an error if FW returns an error code other 3854 * than ICE_AQ_RC_EMODE 3855 */ 3856 if (status == -EIO) { 3857 if (hw->adminq.sq_last_status == ICE_AQ_RC_EMODE) 3858 dev_dbg(dev, "can't set link to %s, err %d aq_err %s. not fatal, continuing\n", 3859 (ena ? "ON" : "OFF"), status, 3860 ice_aq_str(hw->adminq.sq_last_status)); 3861 } else if (status) { 3862 dev_err(dev, "can't set link to %s, err %d aq_err %s\n", 3863 (ena ? "ON" : "OFF"), status, 3864 ice_aq_str(hw->adminq.sq_last_status)); 3865 return status; 3866 } 3867 3868 return 0; 3869 } 3870 3871 /** 3872 * ice_vsi_add_vlan_zero - add VLAN 0 filter(s) for this VSI 3873 * @vsi: VSI used to add VLAN filters 3874 * 3875 * In Single VLAN Mode (SVM), single VLAN filters via ICE_SW_LKUP_VLAN are based 3876 * on the inner VLAN ID, so the VLAN TPID (i.e. 0x8100 or 0x888a8) doesn't 3877 * matter. In Double VLAN Mode (DVM), outer/single VLAN filters via 3878 * ICE_SW_LKUP_VLAN are based on the outer/single VLAN ID + VLAN TPID. 3879 * 3880 * For both modes add a VLAN 0 + no VLAN TPID filter to handle untagged traffic 3881 * when VLAN pruning is enabled. Also, this handles VLAN 0 priority tagged 3882 * traffic in SVM, since the VLAN TPID isn't part of filtering. 3883 * 3884 * If DVM is enabled then an explicit VLAN 0 + VLAN TPID filter needs to be 3885 * added to allow VLAN 0 priority tagged traffic in DVM, since the VLAN TPID is 3886 * part of filtering. 3887 */ 3888 int ice_vsi_add_vlan_zero(struct ice_vsi *vsi) 3889 { 3890 struct ice_vsi_vlan_ops *vlan_ops = ice_get_compat_vsi_vlan_ops(vsi); 3891 struct ice_vlan vlan; 3892 int err; 3893 3894 vlan = ICE_VLAN(0, 0, 0); 3895 err = vlan_ops->add_vlan(vsi, &vlan); 3896 if (err && err != -EEXIST) 3897 return err; 3898 3899 /* in SVM both VLAN 0 filters are identical */ 3900 if (!ice_is_dvm_ena(&vsi->back->hw)) 3901 return 0; 3902 3903 vlan = ICE_VLAN(ETH_P_8021Q, 0, 0); 3904 err = vlan_ops->add_vlan(vsi, &vlan); 3905 if (err && err != -EEXIST) 3906 return err; 3907 3908 return 0; 3909 } 3910 3911 /** 3912 * ice_vsi_del_vlan_zero - delete VLAN 0 filter(s) for this VSI 3913 * @vsi: VSI used to add VLAN filters 3914 * 3915 * Delete the VLAN 0 filters in the same manner that they were added in 3916 * ice_vsi_add_vlan_zero. 3917 */ 3918 int ice_vsi_del_vlan_zero(struct ice_vsi *vsi) 3919 { 3920 struct ice_vsi_vlan_ops *vlan_ops = ice_get_compat_vsi_vlan_ops(vsi); 3921 struct ice_vlan vlan; 3922 int err; 3923 3924 vlan = ICE_VLAN(0, 0, 0); 3925 err = vlan_ops->del_vlan(vsi, &vlan); 3926 if (err && err != -EEXIST) 3927 return err; 3928 3929 /* in SVM both VLAN 0 filters are identical */ 3930 if (!ice_is_dvm_ena(&vsi->back->hw)) 3931 return 0; 3932 3933 vlan = ICE_VLAN(ETH_P_8021Q, 0, 0); 3934 err = vlan_ops->del_vlan(vsi, &vlan); 3935 if (err && err != -EEXIST) 3936 return err; 3937 3938 /* when deleting the last VLAN filter, make sure to disable the VLAN 3939 * promisc mode so the filter isn't left by accident 3940 */ 3941 return ice_clear_vsi_promisc(&vsi->back->hw, vsi->idx, 3942 ICE_MCAST_VLAN_PROMISC_BITS, 0); 3943 } 3944 3945 /** 3946 * ice_vsi_num_zero_vlans - get number of VLAN 0 filters based on VLAN mode 3947 * @vsi: VSI used to get the VLAN mode 3948 * 3949 * If DVM is enabled then 2 VLAN 0 filters are added, else if SVM is enabled 3950 * then 1 VLAN 0 filter is added. See ice_vsi_add_vlan_zero for more details. 3951 */ 3952 static u16 ice_vsi_num_zero_vlans(struct ice_vsi *vsi) 3953 { 3954 #define ICE_DVM_NUM_ZERO_VLAN_FLTRS 2 3955 #define ICE_SVM_NUM_ZERO_VLAN_FLTRS 1 3956 /* no VLAN 0 filter is created when a port VLAN is active */ 3957 if (vsi->type == ICE_VSI_VF) { 3958 if (WARN_ON(!vsi->vf)) 3959 return 0; 3960 3961 if (ice_vf_is_port_vlan_ena(vsi->vf)) 3962 return 0; 3963 } 3964 3965 if (ice_is_dvm_ena(&vsi->back->hw)) 3966 return ICE_DVM_NUM_ZERO_VLAN_FLTRS; 3967 else 3968 return ICE_SVM_NUM_ZERO_VLAN_FLTRS; 3969 } 3970 3971 /** 3972 * ice_vsi_has_non_zero_vlans - check if VSI has any non-zero VLANs 3973 * @vsi: VSI used to determine if any non-zero VLANs have been added 3974 */ 3975 bool ice_vsi_has_non_zero_vlans(struct ice_vsi *vsi) 3976 { 3977 return (vsi->num_vlan > ice_vsi_num_zero_vlans(vsi)); 3978 } 3979 3980 /** 3981 * ice_vsi_num_non_zero_vlans - get the number of non-zero VLANs for this VSI 3982 * @vsi: VSI used to get the number of non-zero VLANs added 3983 */ 3984 u16 ice_vsi_num_non_zero_vlans(struct ice_vsi *vsi) 3985 { 3986 return (vsi->num_vlan - ice_vsi_num_zero_vlans(vsi)); 3987 } 3988 3989 /** 3990 * ice_is_feature_supported 3991 * @pf: pointer to the struct ice_pf instance 3992 * @f: feature enum to be checked 3993 * 3994 * returns true if feature is supported, false otherwise 3995 */ 3996 bool ice_is_feature_supported(struct ice_pf *pf, enum ice_feature f) 3997 { 3998 if (f < 0 || f >= ICE_F_MAX) 3999 return false; 4000 4001 return test_bit(f, pf->features); 4002 } 4003 4004 /** 4005 * ice_set_feature_support 4006 * @pf: pointer to the struct ice_pf instance 4007 * @f: feature enum to set 4008 */ 4009 static void ice_set_feature_support(struct ice_pf *pf, enum ice_feature f) 4010 { 4011 if (f < 0 || f >= ICE_F_MAX) 4012 return; 4013 4014 set_bit(f, pf->features); 4015 } 4016 4017 /** 4018 * ice_clear_feature_support 4019 * @pf: pointer to the struct ice_pf instance 4020 * @f: feature enum to clear 4021 */ 4022 void ice_clear_feature_support(struct ice_pf *pf, enum ice_feature f) 4023 { 4024 if (f < 0 || f >= ICE_F_MAX) 4025 return; 4026 4027 clear_bit(f, pf->features); 4028 } 4029 4030 /** 4031 * ice_init_feature_support 4032 * @pf: pointer to the struct ice_pf instance 4033 * 4034 * called during init to setup supported feature 4035 */ 4036 void ice_init_feature_support(struct ice_pf *pf) 4037 { 4038 switch (pf->hw.device_id) { 4039 case ICE_DEV_ID_E810C_BACKPLANE: 4040 case ICE_DEV_ID_E810C_QSFP: 4041 case ICE_DEV_ID_E810C_SFP: 4042 ice_set_feature_support(pf, ICE_F_DSCP); 4043 ice_set_feature_support(pf, ICE_F_PTP_EXTTS); 4044 if (ice_is_e810t(&pf->hw)) { 4045 ice_set_feature_support(pf, ICE_F_SMA_CTRL); 4046 if (ice_gnss_is_gps_present(&pf->hw)) 4047 ice_set_feature_support(pf, ICE_F_GNSS); 4048 } 4049 break; 4050 default: 4051 break; 4052 } 4053 } 4054 4055 /** 4056 * ice_vsi_update_security - update security block in VSI 4057 * @vsi: pointer to VSI structure 4058 * @fill: function pointer to fill ctx 4059 */ 4060 int 4061 ice_vsi_update_security(struct ice_vsi *vsi, void (*fill)(struct ice_vsi_ctx *)) 4062 { 4063 struct ice_vsi_ctx ctx = { 0 }; 4064 4065 ctx.info = vsi->info; 4066 ctx.info.valid_sections = cpu_to_le16(ICE_AQ_VSI_PROP_SECURITY_VALID); 4067 fill(&ctx); 4068 4069 if (ice_update_vsi(&vsi->back->hw, vsi->idx, &ctx, NULL)) 4070 return -ENODEV; 4071 4072 vsi->info = ctx.info; 4073 return 0; 4074 } 4075 4076 /** 4077 * ice_vsi_ctx_set_antispoof - set antispoof function in VSI ctx 4078 * @ctx: pointer to VSI ctx structure 4079 */ 4080 void ice_vsi_ctx_set_antispoof(struct ice_vsi_ctx *ctx) 4081 { 4082 ctx->info.sec_flags |= ICE_AQ_VSI_SEC_FLAG_ENA_MAC_ANTI_SPOOF | 4083 (ICE_AQ_VSI_SEC_TX_VLAN_PRUNE_ENA << 4084 ICE_AQ_VSI_SEC_TX_PRUNE_ENA_S); 4085 } 4086 4087 /** 4088 * ice_vsi_ctx_clear_antispoof - clear antispoof function in VSI ctx 4089 * @ctx: pointer to VSI ctx structure 4090 */ 4091 void ice_vsi_ctx_clear_antispoof(struct ice_vsi_ctx *ctx) 4092 { 4093 ctx->info.sec_flags &= ~ICE_AQ_VSI_SEC_FLAG_ENA_MAC_ANTI_SPOOF & 4094 ~(ICE_AQ_VSI_SEC_TX_VLAN_PRUNE_ENA << 4095 ICE_AQ_VSI_SEC_TX_PRUNE_ENA_S); 4096 } 4097 4098 /** 4099 * ice_vsi_ctx_set_allow_override - allow destination override on VSI 4100 * @ctx: pointer to VSI ctx structure 4101 */ 4102 void ice_vsi_ctx_set_allow_override(struct ice_vsi_ctx *ctx) 4103 { 4104 ctx->info.sec_flags |= ICE_AQ_VSI_SEC_FLAG_ALLOW_DEST_OVRD; 4105 } 4106 4107 /** 4108 * ice_vsi_ctx_clear_allow_override - turn off destination override on VSI 4109 * @ctx: pointer to VSI ctx structure 4110 */ 4111 void ice_vsi_ctx_clear_allow_override(struct ice_vsi_ctx *ctx) 4112 { 4113 ctx->info.sec_flags &= ~ICE_AQ_VSI_SEC_FLAG_ALLOW_DEST_OVRD; 4114 } 4115