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