1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright (c) 2018, Intel Corporation. */ 3 4 #include "ice.h" 5 #include "ice_base.h" 6 #include "ice_flow.h" 7 #include "ice_lib.h" 8 #include "ice_fltr.h" 9 #include "ice_dcb_lib.h" 10 #include "ice_vsi_vlan_ops.h" 11 12 /** 13 * ice_vsi_type_str - maps VSI type enum to string equivalents 14 * @vsi_type: VSI type enum 15 */ 16 const char *ice_vsi_type_str(enum ice_vsi_type vsi_type) 17 { 18 switch (vsi_type) { 19 case ICE_VSI_PF: 20 return "ICE_VSI_PF"; 21 case ICE_VSI_VF: 22 return "ICE_VSI_VF"; 23 case ICE_VSI_CTRL: 24 return "ICE_VSI_CTRL"; 25 case ICE_VSI_CHNL: 26 return "ICE_VSI_CHNL"; 27 case ICE_VSI_LB: 28 return "ICE_VSI_LB"; 29 default: 30 return "unknown"; 31 } 32 } 33 34 /** 35 * ice_vsi_ctrl_all_rx_rings - Start or stop a VSI's Rx rings 36 * @vsi: the VSI being configured 37 * @ena: start or stop the Rx rings 38 * 39 * First enable/disable all of the Rx rings, flush any remaining writes, and 40 * then verify that they have all been enabled/disabled successfully. This will 41 * let all of the register writes complete when enabling/disabling the Rx rings 42 * before waiting for the change in hardware to complete. 43 */ 44 static int ice_vsi_ctrl_all_rx_rings(struct ice_vsi *vsi, bool ena) 45 { 46 int ret = 0; 47 u16 i; 48 49 ice_for_each_rxq(vsi, i) 50 ice_vsi_ctrl_one_rx_ring(vsi, ena, i, false); 51 52 ice_flush(&vsi->back->hw); 53 54 ice_for_each_rxq(vsi, i) { 55 ret = ice_vsi_wait_one_rx_ring(vsi, ena, i); 56 if (ret) 57 break; 58 } 59 60 return ret; 61 } 62 63 /** 64 * ice_vsi_alloc_arrays - Allocate queue and vector pointer arrays for the VSI 65 * @vsi: VSI pointer 66 * 67 * On error: returns error code (negative) 68 * On success: returns 0 69 */ 70 static int ice_vsi_alloc_arrays(struct ice_vsi *vsi) 71 { 72 struct ice_pf *pf = vsi->back; 73 struct device *dev; 74 75 dev = ice_pf_to_dev(pf); 76 if (vsi->type == ICE_VSI_CHNL) 77 return 0; 78 79 /* allocate memory for both Tx and Rx ring pointers */ 80 vsi->tx_rings = devm_kcalloc(dev, vsi->alloc_txq, 81 sizeof(*vsi->tx_rings), GFP_KERNEL); 82 if (!vsi->tx_rings) 83 return -ENOMEM; 84 85 vsi->rx_rings = devm_kcalloc(dev, vsi->alloc_rxq, 86 sizeof(*vsi->rx_rings), GFP_KERNEL); 87 if (!vsi->rx_rings) 88 goto err_rings; 89 90 /* txq_map needs to have enough space to track both Tx (stack) rings 91 * and XDP rings; at this point vsi->num_xdp_txq might not be set, 92 * so use num_possible_cpus() as we want to always provide XDP ring 93 * per CPU, regardless of queue count settings from user that might 94 * have come from ethtool's set_channels() callback; 95 */ 96 vsi->txq_map = devm_kcalloc(dev, (vsi->alloc_txq + num_possible_cpus()), 97 sizeof(*vsi->txq_map), GFP_KERNEL); 98 99 if (!vsi->txq_map) 100 goto err_txq_map; 101 102 vsi->rxq_map = devm_kcalloc(dev, vsi->alloc_rxq, 103 sizeof(*vsi->rxq_map), GFP_KERNEL); 104 if (!vsi->rxq_map) 105 goto err_rxq_map; 106 107 /* There is no need to allocate q_vectors for a loopback VSI. */ 108 if (vsi->type == ICE_VSI_LB) 109 return 0; 110 111 /* allocate memory for q_vector pointers */ 112 vsi->q_vectors = devm_kcalloc(dev, vsi->num_q_vectors, 113 sizeof(*vsi->q_vectors), GFP_KERNEL); 114 if (!vsi->q_vectors) 115 goto err_vectors; 116 117 vsi->af_xdp_zc_qps = bitmap_zalloc(max_t(int, vsi->alloc_txq, vsi->alloc_rxq), GFP_KERNEL); 118 if (!vsi->af_xdp_zc_qps) 119 goto err_zc_qps; 120 121 return 0; 122 123 err_zc_qps: 124 devm_kfree(dev, vsi->q_vectors); 125 err_vectors: 126 devm_kfree(dev, vsi->rxq_map); 127 err_rxq_map: 128 devm_kfree(dev, vsi->txq_map); 129 err_txq_map: 130 devm_kfree(dev, vsi->rx_rings); 131 err_rings: 132 devm_kfree(dev, vsi->tx_rings); 133 return -ENOMEM; 134 } 135 136 /** 137 * ice_vsi_set_num_desc - Set number of descriptors for queues on this VSI 138 * @vsi: the VSI being configured 139 */ 140 static void ice_vsi_set_num_desc(struct ice_vsi *vsi) 141 { 142 switch (vsi->type) { 143 case ICE_VSI_PF: 144 case ICE_VSI_CTRL: 145 case ICE_VSI_LB: 146 /* a user could change the values of num_[tr]x_desc using 147 * ethtool -G so we should keep those values instead of 148 * overwriting them with the defaults. 149 */ 150 if (!vsi->num_rx_desc) 151 vsi->num_rx_desc = ICE_DFLT_NUM_RX_DESC; 152 if (!vsi->num_tx_desc) 153 vsi->num_tx_desc = ICE_DFLT_NUM_TX_DESC; 154 break; 155 default: 156 dev_dbg(ice_pf_to_dev(vsi->back), "Not setting number of Tx/Rx descriptors for VSI type %d\n", 157 vsi->type); 158 break; 159 } 160 } 161 162 /** 163 * ice_vsi_set_num_qs - Set number of queues, descriptors and vectors for a VSI 164 * @vsi: the VSI being configured 165 * 166 * Return 0 on success and a negative value on error 167 */ 168 static void ice_vsi_set_num_qs(struct ice_vsi *vsi) 169 { 170 enum ice_vsi_type vsi_type = vsi->type; 171 struct ice_pf *pf = vsi->back; 172 struct ice_vf *vf = vsi->vf; 173 174 if (WARN_ON(vsi_type == ICE_VSI_VF && !vf)) 175 return; 176 177 switch (vsi_type) { 178 case ICE_VSI_PF: 179 if (vsi->req_txq) { 180 vsi->alloc_txq = vsi->req_txq; 181 vsi->num_txq = vsi->req_txq; 182 } else { 183 vsi->alloc_txq = min3(pf->num_lan_msix, 184 ice_get_avail_txq_count(pf), 185 (u16)num_online_cpus()); 186 } 187 188 pf->num_lan_tx = vsi->alloc_txq; 189 190 /* only 1 Rx queue unless RSS is enabled */ 191 if (!test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 192 vsi->alloc_rxq = 1; 193 } else { 194 if (vsi->req_rxq) { 195 vsi->alloc_rxq = vsi->req_rxq; 196 vsi->num_rxq = vsi->req_rxq; 197 } else { 198 vsi->alloc_rxq = min3(pf->num_lan_msix, 199 ice_get_avail_rxq_count(pf), 200 (u16)num_online_cpus()); 201 } 202 } 203 204 pf->num_lan_rx = vsi->alloc_rxq; 205 206 vsi->num_q_vectors = min_t(int, pf->num_lan_msix, 207 max_t(int, vsi->alloc_rxq, 208 vsi->alloc_txq)); 209 break; 210 case ICE_VSI_VF: 211 if (vf->num_req_qs) 212 vf->num_vf_qs = vf->num_req_qs; 213 vsi->alloc_txq = vf->num_vf_qs; 214 vsi->alloc_rxq = vf->num_vf_qs; 215 /* pf->vfs.num_msix_per includes (VF miscellaneous vector + 216 * data queue interrupts). Since vsi->num_q_vectors is number 217 * of queues vectors, subtract 1 (ICE_NONQ_VECS_VF) from the 218 * original vector count 219 */ 220 vsi->num_q_vectors = vf->num_msix - ICE_NONQ_VECS_VF; 221 break; 222 case ICE_VSI_CTRL: 223 vsi->alloc_txq = 1; 224 vsi->alloc_rxq = 1; 225 vsi->num_q_vectors = 1; 226 break; 227 case ICE_VSI_CHNL: 228 vsi->alloc_txq = 0; 229 vsi->alloc_rxq = 0; 230 break; 231 case ICE_VSI_LB: 232 vsi->alloc_txq = 1; 233 vsi->alloc_rxq = 1; 234 break; 235 default: 236 dev_warn(ice_pf_to_dev(pf), "Unknown VSI type %d\n", vsi_type); 237 break; 238 } 239 240 ice_vsi_set_num_desc(vsi); 241 } 242 243 /** 244 * ice_get_free_slot - get the next non-NULL location index in array 245 * @array: array to search 246 * @size: size of the array 247 * @curr: last known occupied index to be used as a search hint 248 * 249 * void * is being used to keep the functionality generic. This lets us use this 250 * function on any array of pointers. 251 */ 252 static int ice_get_free_slot(void *array, int size, int curr) 253 { 254 int **tmp_array = (int **)array; 255 int next; 256 257 if (curr < (size - 1) && !tmp_array[curr + 1]) { 258 next = curr + 1; 259 } else { 260 int i = 0; 261 262 while ((i < size) && (tmp_array[i])) 263 i++; 264 if (i == size) 265 next = ICE_NO_VSI; 266 else 267 next = i; 268 } 269 return next; 270 } 271 272 /** 273 * ice_vsi_delete_from_hw - delete a VSI from the switch 274 * @vsi: pointer to VSI being removed 275 */ 276 static void ice_vsi_delete_from_hw(struct ice_vsi *vsi) 277 { 278 struct ice_pf *pf = vsi->back; 279 struct ice_vsi_ctx *ctxt; 280 int status; 281 282 ice_fltr_remove_all(vsi); 283 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); 284 if (!ctxt) 285 return; 286 287 if (vsi->type == ICE_VSI_VF) 288 ctxt->vf_num = vsi->vf->vf_id; 289 ctxt->vsi_num = vsi->vsi_num; 290 291 memcpy(&ctxt->info, &vsi->info, sizeof(ctxt->info)); 292 293 status = ice_free_vsi(&pf->hw, vsi->idx, ctxt, false, NULL); 294 if (status) 295 dev_err(ice_pf_to_dev(pf), "Failed to delete VSI %i in FW - error: %d\n", 296 vsi->vsi_num, status); 297 298 kfree(ctxt); 299 } 300 301 /** 302 * ice_vsi_free_arrays - De-allocate queue and vector pointer arrays for the VSI 303 * @vsi: pointer to VSI being cleared 304 */ 305 static void ice_vsi_free_arrays(struct ice_vsi *vsi) 306 { 307 struct ice_pf *pf = vsi->back; 308 struct device *dev; 309 310 dev = ice_pf_to_dev(pf); 311 312 bitmap_free(vsi->af_xdp_zc_qps); 313 vsi->af_xdp_zc_qps = NULL; 314 /* free the ring and vector containers */ 315 devm_kfree(dev, vsi->q_vectors); 316 vsi->q_vectors = NULL; 317 devm_kfree(dev, vsi->tx_rings); 318 vsi->tx_rings = NULL; 319 devm_kfree(dev, vsi->rx_rings); 320 vsi->rx_rings = NULL; 321 devm_kfree(dev, vsi->txq_map); 322 vsi->txq_map = NULL; 323 devm_kfree(dev, vsi->rxq_map); 324 vsi->rxq_map = NULL; 325 } 326 327 /** 328 * ice_vsi_free_stats - Free the ring statistics structures 329 * @vsi: VSI pointer 330 */ 331 static void ice_vsi_free_stats(struct ice_vsi *vsi) 332 { 333 struct ice_vsi_stats *vsi_stat; 334 struct ice_pf *pf = vsi->back; 335 int i; 336 337 if (vsi->type == ICE_VSI_CHNL) 338 return; 339 if (!pf->vsi_stats) 340 return; 341 342 vsi_stat = pf->vsi_stats[vsi->idx]; 343 if (!vsi_stat) 344 return; 345 346 ice_for_each_alloc_txq(vsi, i) { 347 if (vsi_stat->tx_ring_stats[i]) { 348 kfree_rcu(vsi_stat->tx_ring_stats[i], rcu); 349 WRITE_ONCE(vsi_stat->tx_ring_stats[i], NULL); 350 } 351 } 352 353 ice_for_each_alloc_rxq(vsi, i) { 354 if (vsi_stat->rx_ring_stats[i]) { 355 kfree_rcu(vsi_stat->rx_ring_stats[i], rcu); 356 WRITE_ONCE(vsi_stat->rx_ring_stats[i], NULL); 357 } 358 } 359 360 kfree(vsi_stat->tx_ring_stats); 361 kfree(vsi_stat->rx_ring_stats); 362 kfree(vsi_stat); 363 pf->vsi_stats[vsi->idx] = NULL; 364 } 365 366 /** 367 * ice_vsi_alloc_ring_stats - Allocates Tx and Rx ring stats for the VSI 368 * @vsi: VSI which is having stats allocated 369 */ 370 static int ice_vsi_alloc_ring_stats(struct ice_vsi *vsi) 371 { 372 struct ice_ring_stats **tx_ring_stats; 373 struct ice_ring_stats **rx_ring_stats; 374 struct ice_vsi_stats *vsi_stats; 375 struct ice_pf *pf = vsi->back; 376 u16 i; 377 378 vsi_stats = pf->vsi_stats[vsi->idx]; 379 tx_ring_stats = vsi_stats->tx_ring_stats; 380 rx_ring_stats = vsi_stats->rx_ring_stats; 381 382 /* Allocate Tx ring stats */ 383 ice_for_each_alloc_txq(vsi, i) { 384 struct ice_ring_stats *ring_stats; 385 struct ice_tx_ring *ring; 386 387 ring = vsi->tx_rings[i]; 388 ring_stats = tx_ring_stats[i]; 389 390 if (!ring_stats) { 391 ring_stats = kzalloc(sizeof(*ring_stats), GFP_KERNEL); 392 if (!ring_stats) 393 goto err_out; 394 395 WRITE_ONCE(tx_ring_stats[i], ring_stats); 396 } 397 398 ring->ring_stats = ring_stats; 399 } 400 401 /* Allocate Rx ring stats */ 402 ice_for_each_alloc_rxq(vsi, i) { 403 struct ice_ring_stats *ring_stats; 404 struct ice_rx_ring *ring; 405 406 ring = vsi->rx_rings[i]; 407 ring_stats = rx_ring_stats[i]; 408 409 if (!ring_stats) { 410 ring_stats = kzalloc(sizeof(*ring_stats), GFP_KERNEL); 411 if (!ring_stats) 412 goto err_out; 413 414 WRITE_ONCE(rx_ring_stats[i], ring_stats); 415 } 416 417 ring->ring_stats = ring_stats; 418 } 419 420 return 0; 421 422 err_out: 423 ice_vsi_free_stats(vsi); 424 return -ENOMEM; 425 } 426 427 /** 428 * ice_vsi_free - clean up and deallocate the provided VSI 429 * @vsi: pointer to VSI being cleared 430 * 431 * This deallocates the VSI's queue resources, removes it from the PF's 432 * VSI array if necessary, and deallocates the VSI 433 */ 434 static void ice_vsi_free(struct ice_vsi *vsi) 435 { 436 struct ice_pf *pf = NULL; 437 struct device *dev; 438 439 if (!vsi || !vsi->back) 440 return; 441 442 pf = vsi->back; 443 dev = ice_pf_to_dev(pf); 444 445 if (!pf->vsi[vsi->idx] || pf->vsi[vsi->idx] != vsi) { 446 dev_dbg(dev, "vsi does not exist at pf->vsi[%d]\n", vsi->idx); 447 return; 448 } 449 450 mutex_lock(&pf->sw_mutex); 451 /* updates the PF for this cleared VSI */ 452 453 pf->vsi[vsi->idx] = NULL; 454 pf->next_vsi = vsi->idx; 455 456 ice_vsi_free_stats(vsi); 457 ice_vsi_free_arrays(vsi); 458 mutex_unlock(&pf->sw_mutex); 459 devm_kfree(dev, vsi); 460 } 461 462 void ice_vsi_delete(struct ice_vsi *vsi) 463 { 464 ice_vsi_delete_from_hw(vsi); 465 ice_vsi_free(vsi); 466 } 467 468 /** 469 * ice_msix_clean_ctrl_vsi - MSIX mode interrupt handler for ctrl VSI 470 * @irq: interrupt number 471 * @data: pointer to a q_vector 472 */ 473 static irqreturn_t ice_msix_clean_ctrl_vsi(int __always_unused irq, void *data) 474 { 475 struct ice_q_vector *q_vector = (struct ice_q_vector *)data; 476 477 if (!q_vector->tx.tx_ring) 478 return IRQ_HANDLED; 479 480 #define FDIR_RX_DESC_CLEAN_BUDGET 64 481 ice_clean_rx_irq(q_vector->rx.rx_ring, FDIR_RX_DESC_CLEAN_BUDGET); 482 ice_clean_ctrl_tx_irq(q_vector->tx.tx_ring); 483 484 return IRQ_HANDLED; 485 } 486 487 /** 488 * ice_msix_clean_rings - MSIX mode Interrupt Handler 489 * @irq: interrupt number 490 * @data: pointer to a q_vector 491 */ 492 static irqreturn_t ice_msix_clean_rings(int __always_unused irq, void *data) 493 { 494 struct ice_q_vector *q_vector = (struct ice_q_vector *)data; 495 496 if (!q_vector->tx.tx_ring && !q_vector->rx.rx_ring) 497 return IRQ_HANDLED; 498 499 q_vector->total_events++; 500 501 napi_schedule(&q_vector->napi); 502 503 return IRQ_HANDLED; 504 } 505 506 /** 507 * ice_vsi_alloc_stat_arrays - Allocate statistics arrays 508 * @vsi: VSI pointer 509 */ 510 static int ice_vsi_alloc_stat_arrays(struct ice_vsi *vsi) 511 { 512 struct ice_vsi_stats *vsi_stat; 513 struct ice_pf *pf = vsi->back; 514 515 if (vsi->type == ICE_VSI_CHNL) 516 return 0; 517 if (!pf->vsi_stats) 518 return -ENOENT; 519 520 if (pf->vsi_stats[vsi->idx]) 521 /* realloc will happen in rebuild path */ 522 return 0; 523 524 vsi_stat = kzalloc(sizeof(*vsi_stat), GFP_KERNEL); 525 if (!vsi_stat) 526 return -ENOMEM; 527 528 vsi_stat->tx_ring_stats = 529 kcalloc(vsi->alloc_txq, sizeof(*vsi_stat->tx_ring_stats), 530 GFP_KERNEL); 531 if (!vsi_stat->tx_ring_stats) 532 goto err_alloc_tx; 533 534 vsi_stat->rx_ring_stats = 535 kcalloc(vsi->alloc_rxq, sizeof(*vsi_stat->rx_ring_stats), 536 GFP_KERNEL); 537 if (!vsi_stat->rx_ring_stats) 538 goto err_alloc_rx; 539 540 pf->vsi_stats[vsi->idx] = vsi_stat; 541 542 return 0; 543 544 err_alloc_rx: 545 kfree(vsi_stat->rx_ring_stats); 546 err_alloc_tx: 547 kfree(vsi_stat->tx_ring_stats); 548 kfree(vsi_stat); 549 pf->vsi_stats[vsi->idx] = NULL; 550 return -ENOMEM; 551 } 552 553 /** 554 * ice_vsi_alloc_def - set default values for already allocated VSI 555 * @vsi: ptr to VSI 556 * @ch: ptr to channel 557 */ 558 static int 559 ice_vsi_alloc_def(struct ice_vsi *vsi, struct ice_channel *ch) 560 { 561 if (vsi->type != ICE_VSI_CHNL) { 562 ice_vsi_set_num_qs(vsi); 563 if (ice_vsi_alloc_arrays(vsi)) 564 return -ENOMEM; 565 } 566 567 switch (vsi->type) { 568 case ICE_VSI_PF: 569 /* Setup default MSIX irq handler for VSI */ 570 vsi->irq_handler = ice_msix_clean_rings; 571 break; 572 case ICE_VSI_CTRL: 573 /* Setup ctrl VSI MSIX irq handler */ 574 vsi->irq_handler = ice_msix_clean_ctrl_vsi; 575 break; 576 case ICE_VSI_CHNL: 577 if (!ch) 578 return -EINVAL; 579 580 vsi->num_rxq = ch->num_rxq; 581 vsi->num_txq = ch->num_txq; 582 vsi->next_base_q = ch->base_q; 583 break; 584 case ICE_VSI_VF: 585 case ICE_VSI_LB: 586 break; 587 default: 588 ice_vsi_free_arrays(vsi); 589 return -EINVAL; 590 } 591 592 return 0; 593 } 594 595 /** 596 * ice_vsi_alloc - Allocates the next available struct VSI in the PF 597 * @pf: board private structure 598 * 599 * Reserves a VSI index from the PF and allocates an empty VSI structure 600 * without a type. The VSI structure must later be initialized by calling 601 * ice_vsi_cfg(). 602 * 603 * returns a pointer to a VSI on success, NULL on failure. 604 */ 605 static struct ice_vsi *ice_vsi_alloc(struct ice_pf *pf) 606 { 607 struct device *dev = ice_pf_to_dev(pf); 608 struct ice_vsi *vsi = NULL; 609 610 /* Need to protect the allocation of the VSIs at the PF level */ 611 mutex_lock(&pf->sw_mutex); 612 613 /* If we have already allocated our maximum number of VSIs, 614 * pf->next_vsi will be ICE_NO_VSI. If not, pf->next_vsi index 615 * is available to be populated 616 */ 617 if (pf->next_vsi == ICE_NO_VSI) { 618 dev_dbg(dev, "out of VSI slots!\n"); 619 goto unlock_pf; 620 } 621 622 vsi = devm_kzalloc(dev, sizeof(*vsi), GFP_KERNEL); 623 if (!vsi) 624 goto unlock_pf; 625 626 vsi->back = pf; 627 set_bit(ICE_VSI_DOWN, vsi->state); 628 629 /* fill slot and make note of the index */ 630 vsi->idx = pf->next_vsi; 631 pf->vsi[pf->next_vsi] = vsi; 632 633 /* prepare pf->next_vsi for next use */ 634 pf->next_vsi = ice_get_free_slot(pf->vsi, pf->num_alloc_vsi, 635 pf->next_vsi); 636 637 unlock_pf: 638 mutex_unlock(&pf->sw_mutex); 639 return vsi; 640 } 641 642 /** 643 * ice_alloc_fd_res - Allocate FD resource for a VSI 644 * @vsi: pointer to the ice_vsi 645 * 646 * This allocates the FD resources 647 * 648 * Returns 0 on success, -EPERM on no-op or -EIO on failure 649 */ 650 static int ice_alloc_fd_res(struct ice_vsi *vsi) 651 { 652 struct ice_pf *pf = vsi->back; 653 u32 g_val, b_val; 654 655 /* Flow Director filters are only allocated/assigned to the PF VSI or 656 * CHNL VSI which passes the traffic. The CTRL VSI is only used to 657 * add/delete filters so resources are not allocated to it 658 */ 659 if (!test_bit(ICE_FLAG_FD_ENA, pf->flags)) 660 return -EPERM; 661 662 if (!(vsi->type == ICE_VSI_PF || vsi->type == ICE_VSI_VF || 663 vsi->type == ICE_VSI_CHNL)) 664 return -EPERM; 665 666 /* FD filters from guaranteed pool per VSI */ 667 g_val = pf->hw.func_caps.fd_fltr_guar; 668 if (!g_val) 669 return -EPERM; 670 671 /* FD filters from best effort pool */ 672 b_val = pf->hw.func_caps.fd_fltr_best_effort; 673 if (!b_val) 674 return -EPERM; 675 676 /* PF main VSI gets only 64 FD resources from guaranteed pool 677 * when ADQ is configured. 678 */ 679 #define ICE_PF_VSI_GFLTR 64 680 681 /* determine FD filter resources per VSI from shared(best effort) and 682 * dedicated pool 683 */ 684 if (vsi->type == ICE_VSI_PF) { 685 vsi->num_gfltr = g_val; 686 /* if MQPRIO is configured, main VSI doesn't get all FD 687 * resources from guaranteed pool. PF VSI gets 64 FD resources 688 */ 689 if (test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) { 690 if (g_val < ICE_PF_VSI_GFLTR) 691 return -EPERM; 692 /* allow bare minimum entries for PF VSI */ 693 vsi->num_gfltr = ICE_PF_VSI_GFLTR; 694 } 695 696 /* each VSI gets same "best_effort" quota */ 697 vsi->num_bfltr = b_val; 698 } else if (vsi->type == ICE_VSI_VF) { 699 vsi->num_gfltr = 0; 700 701 /* each VSI gets same "best_effort" quota */ 702 vsi->num_bfltr = b_val; 703 } else { 704 struct ice_vsi *main_vsi; 705 int numtc; 706 707 main_vsi = ice_get_main_vsi(pf); 708 if (!main_vsi) 709 return -EPERM; 710 711 if (!main_vsi->all_numtc) 712 return -EINVAL; 713 714 /* figure out ADQ numtc */ 715 numtc = main_vsi->all_numtc - ICE_CHNL_START_TC; 716 717 /* only one TC but still asking resources for channels, 718 * invalid config 719 */ 720 if (numtc < ICE_CHNL_START_TC) 721 return -EPERM; 722 723 g_val -= ICE_PF_VSI_GFLTR; 724 /* channel VSIs gets equal share from guaranteed pool */ 725 vsi->num_gfltr = g_val / numtc; 726 727 /* each VSI gets same "best_effort" quota */ 728 vsi->num_bfltr = b_val; 729 } 730 731 return 0; 732 } 733 734 /** 735 * ice_vsi_get_qs - Assign queues from PF to VSI 736 * @vsi: the VSI to assign queues to 737 * 738 * Returns 0 on success and a negative value on error 739 */ 740 static int ice_vsi_get_qs(struct ice_vsi *vsi) 741 { 742 struct ice_pf *pf = vsi->back; 743 struct ice_qs_cfg tx_qs_cfg = { 744 .qs_mutex = &pf->avail_q_mutex, 745 .pf_map = pf->avail_txqs, 746 .pf_map_size = pf->max_pf_txqs, 747 .q_count = vsi->alloc_txq, 748 .scatter_count = ICE_MAX_SCATTER_TXQS, 749 .vsi_map = vsi->txq_map, 750 .vsi_map_offset = 0, 751 .mapping_mode = ICE_VSI_MAP_CONTIG 752 }; 753 struct ice_qs_cfg rx_qs_cfg = { 754 .qs_mutex = &pf->avail_q_mutex, 755 .pf_map = pf->avail_rxqs, 756 .pf_map_size = pf->max_pf_rxqs, 757 .q_count = vsi->alloc_rxq, 758 .scatter_count = ICE_MAX_SCATTER_RXQS, 759 .vsi_map = vsi->rxq_map, 760 .vsi_map_offset = 0, 761 .mapping_mode = ICE_VSI_MAP_CONTIG 762 }; 763 int ret; 764 765 if (vsi->type == ICE_VSI_CHNL) 766 return 0; 767 768 ret = __ice_vsi_get_qs(&tx_qs_cfg); 769 if (ret) 770 return ret; 771 vsi->tx_mapping_mode = tx_qs_cfg.mapping_mode; 772 773 ret = __ice_vsi_get_qs(&rx_qs_cfg); 774 if (ret) 775 return ret; 776 vsi->rx_mapping_mode = rx_qs_cfg.mapping_mode; 777 778 return 0; 779 } 780 781 /** 782 * ice_vsi_put_qs - Release queues from VSI to PF 783 * @vsi: the VSI that is going to release queues 784 */ 785 static void ice_vsi_put_qs(struct ice_vsi *vsi) 786 { 787 struct ice_pf *pf = vsi->back; 788 int i; 789 790 mutex_lock(&pf->avail_q_mutex); 791 792 ice_for_each_alloc_txq(vsi, i) { 793 clear_bit(vsi->txq_map[i], pf->avail_txqs); 794 vsi->txq_map[i] = ICE_INVAL_Q_INDEX; 795 } 796 797 ice_for_each_alloc_rxq(vsi, i) { 798 clear_bit(vsi->rxq_map[i], pf->avail_rxqs); 799 vsi->rxq_map[i] = ICE_INVAL_Q_INDEX; 800 } 801 802 mutex_unlock(&pf->avail_q_mutex); 803 } 804 805 /** 806 * ice_is_safe_mode 807 * @pf: pointer to the PF struct 808 * 809 * returns true if driver is in safe mode, false otherwise 810 */ 811 bool ice_is_safe_mode(struct ice_pf *pf) 812 { 813 return !test_bit(ICE_FLAG_ADV_FEATURES, pf->flags); 814 } 815 816 /** 817 * ice_is_rdma_ena 818 * @pf: pointer to the PF struct 819 * 820 * returns true if RDMA is currently supported, false otherwise 821 */ 822 bool ice_is_rdma_ena(struct ice_pf *pf) 823 { 824 return test_bit(ICE_FLAG_RDMA_ENA, pf->flags); 825 } 826 827 /** 828 * ice_vsi_clean_rss_flow_fld - Delete RSS configuration 829 * @vsi: the VSI being cleaned up 830 * 831 * This function deletes RSS input set for all flows that were configured 832 * for this VSI 833 */ 834 static void ice_vsi_clean_rss_flow_fld(struct ice_vsi *vsi) 835 { 836 struct ice_pf *pf = vsi->back; 837 int status; 838 839 if (ice_is_safe_mode(pf)) 840 return; 841 842 status = ice_rem_vsi_rss_cfg(&pf->hw, vsi->idx); 843 if (status) 844 dev_dbg(ice_pf_to_dev(pf), "ice_rem_vsi_rss_cfg failed for vsi = %d, error = %d\n", 845 vsi->vsi_num, status); 846 } 847 848 /** 849 * ice_rss_clean - Delete RSS related VSI structures and configuration 850 * @vsi: the VSI being removed 851 */ 852 static void ice_rss_clean(struct ice_vsi *vsi) 853 { 854 struct ice_pf *pf = vsi->back; 855 struct device *dev; 856 857 dev = ice_pf_to_dev(pf); 858 859 devm_kfree(dev, vsi->rss_hkey_user); 860 devm_kfree(dev, vsi->rss_lut_user); 861 862 ice_vsi_clean_rss_flow_fld(vsi); 863 /* remove RSS replay list */ 864 if (!ice_is_safe_mode(pf)) 865 ice_rem_vsi_rss_list(&pf->hw, vsi->idx); 866 } 867 868 /** 869 * ice_vsi_set_rss_params - Setup RSS capabilities per VSI type 870 * @vsi: the VSI being configured 871 */ 872 static void ice_vsi_set_rss_params(struct ice_vsi *vsi) 873 { 874 struct ice_hw_common_caps *cap; 875 struct ice_pf *pf = vsi->back; 876 u16 max_rss_size; 877 878 if (!test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 879 vsi->rss_size = 1; 880 return; 881 } 882 883 cap = &pf->hw.func_caps.common_cap; 884 max_rss_size = BIT(cap->rss_table_entry_width); 885 switch (vsi->type) { 886 case ICE_VSI_CHNL: 887 case ICE_VSI_PF: 888 /* PF VSI will inherit RSS instance of PF */ 889 vsi->rss_table_size = (u16)cap->rss_table_size; 890 if (vsi->type == ICE_VSI_CHNL) 891 vsi->rss_size = min_t(u16, vsi->num_rxq, max_rss_size); 892 else 893 vsi->rss_size = min_t(u16, num_online_cpus(), 894 max_rss_size); 895 vsi->rss_lut_type = ICE_LUT_PF; 896 break; 897 case ICE_VSI_VF: 898 /* VF VSI will get a small RSS table. 899 * For VSI_LUT, LUT size should be set to 64 bytes. 900 */ 901 vsi->rss_table_size = ICE_LUT_VSI_SIZE; 902 vsi->rss_size = ICE_MAX_RSS_QS_PER_VF; 903 vsi->rss_lut_type = ICE_LUT_VSI; 904 break; 905 case ICE_VSI_LB: 906 break; 907 default: 908 dev_dbg(ice_pf_to_dev(pf), "Unsupported VSI type %s\n", 909 ice_vsi_type_str(vsi->type)); 910 break; 911 } 912 } 913 914 /** 915 * ice_set_dflt_vsi_ctx - Set default VSI context before adding a VSI 916 * @hw: HW structure used to determine the VLAN mode of the device 917 * @ctxt: the VSI context being set 918 * 919 * This initializes a default VSI context for all sections except the Queues. 920 */ 921 static void ice_set_dflt_vsi_ctx(struct ice_hw *hw, struct ice_vsi_ctx *ctxt) 922 { 923 u32 table = 0; 924 925 memset(&ctxt->info, 0, sizeof(ctxt->info)); 926 /* VSI's should be allocated from shared pool */ 927 ctxt->alloc_from_pool = true; 928 /* Src pruning enabled by default */ 929 ctxt->info.sw_flags = ICE_AQ_VSI_SW_FLAG_SRC_PRUNE; 930 /* Traffic from VSI can be sent to LAN */ 931 ctxt->info.sw_flags2 = ICE_AQ_VSI_SW_FLAG_LAN_ENA; 932 /* allow all untagged/tagged packets by default on Tx */ 933 ctxt->info.inner_vlan_flags = FIELD_PREP(ICE_AQ_VSI_INNER_VLAN_TX_MODE_M, 934 ICE_AQ_VSI_INNER_VLAN_TX_MODE_ALL); 935 /* SVM - by default bits 3 and 4 in inner_vlan_flags are 0's which 936 * results in legacy behavior (show VLAN, DEI, and UP) in descriptor. 937 * 938 * DVM - leave inner VLAN in packet by default 939 */ 940 if (ice_is_dvm_ena(hw)) { 941 ctxt->info.inner_vlan_flags |= 942 FIELD_PREP(ICE_AQ_VSI_INNER_VLAN_EMODE_M, 943 ICE_AQ_VSI_INNER_VLAN_EMODE_NOTHING); 944 ctxt->info.outer_vlan_flags = 945 FIELD_PREP(ICE_AQ_VSI_OUTER_VLAN_TX_MODE_M, 946 ICE_AQ_VSI_OUTER_VLAN_TX_MODE_ALL); 947 ctxt->info.outer_vlan_flags |= 948 FIELD_PREP(ICE_AQ_VSI_OUTER_TAG_TYPE_M, 949 ICE_AQ_VSI_OUTER_TAG_VLAN_8100); 950 ctxt->info.outer_vlan_flags |= 951 FIELD_PREP(ICE_AQ_VSI_OUTER_VLAN_EMODE_M, 952 ICE_AQ_VSI_OUTER_VLAN_EMODE_NOTHING); 953 } 954 /* Have 1:1 UP mapping for both ingress/egress tables */ 955 table |= ICE_UP_TABLE_TRANSLATE(0, 0); 956 table |= ICE_UP_TABLE_TRANSLATE(1, 1); 957 table |= ICE_UP_TABLE_TRANSLATE(2, 2); 958 table |= ICE_UP_TABLE_TRANSLATE(3, 3); 959 table |= ICE_UP_TABLE_TRANSLATE(4, 4); 960 table |= ICE_UP_TABLE_TRANSLATE(5, 5); 961 table |= ICE_UP_TABLE_TRANSLATE(6, 6); 962 table |= ICE_UP_TABLE_TRANSLATE(7, 7); 963 ctxt->info.ingress_table = cpu_to_le32(table); 964 ctxt->info.egress_table = cpu_to_le32(table); 965 /* Have 1:1 UP mapping for outer to inner UP table */ 966 ctxt->info.outer_up_table = cpu_to_le32(table); 967 /* No Outer tag support outer_tag_flags remains to zero */ 968 } 969 970 /** 971 * ice_vsi_setup_q_map - Setup a VSI queue map 972 * @vsi: the VSI being configured 973 * @ctxt: VSI context structure 974 */ 975 static int ice_vsi_setup_q_map(struct ice_vsi *vsi, struct ice_vsi_ctx *ctxt) 976 { 977 u16 offset = 0, qmap = 0, tx_count = 0, rx_count = 0, pow = 0; 978 u16 num_txq_per_tc, num_rxq_per_tc; 979 u16 qcount_tx = vsi->alloc_txq; 980 u16 qcount_rx = vsi->alloc_rxq; 981 u8 netdev_tc = 0; 982 int i; 983 984 if (!vsi->tc_cfg.numtc) { 985 /* at least TC0 should be enabled by default */ 986 vsi->tc_cfg.numtc = 1; 987 vsi->tc_cfg.ena_tc = 1; 988 } 989 990 num_rxq_per_tc = min_t(u16, qcount_rx / vsi->tc_cfg.numtc, ICE_MAX_RXQS_PER_TC); 991 if (!num_rxq_per_tc) 992 num_rxq_per_tc = 1; 993 num_txq_per_tc = qcount_tx / vsi->tc_cfg.numtc; 994 if (!num_txq_per_tc) 995 num_txq_per_tc = 1; 996 997 /* find the (rounded up) power-of-2 of qcount */ 998 pow = (u16)order_base_2(num_rxq_per_tc); 999 1000 /* TC mapping is a function of the number of Rx queues assigned to the 1001 * VSI for each traffic class and the offset of these queues. 1002 * The first 10 bits are for queue offset for TC0, next 4 bits for no:of 1003 * queues allocated to TC0. No:of queues is a power-of-2. 1004 * 1005 * If TC is not enabled, the queue offset is set to 0, and allocate one 1006 * queue, this way, traffic for the given TC will be sent to the default 1007 * queue. 1008 * 1009 * Setup number and offset of Rx queues for all TCs for the VSI 1010 */ 1011 ice_for_each_traffic_class(i) { 1012 if (!(vsi->tc_cfg.ena_tc & BIT(i))) { 1013 /* TC is not enabled */ 1014 vsi->tc_cfg.tc_info[i].qoffset = 0; 1015 vsi->tc_cfg.tc_info[i].qcount_rx = 1; 1016 vsi->tc_cfg.tc_info[i].qcount_tx = 1; 1017 vsi->tc_cfg.tc_info[i].netdev_tc = 0; 1018 ctxt->info.tc_mapping[i] = 0; 1019 continue; 1020 } 1021 1022 /* TC is enabled */ 1023 vsi->tc_cfg.tc_info[i].qoffset = offset; 1024 vsi->tc_cfg.tc_info[i].qcount_rx = num_rxq_per_tc; 1025 vsi->tc_cfg.tc_info[i].qcount_tx = num_txq_per_tc; 1026 vsi->tc_cfg.tc_info[i].netdev_tc = netdev_tc++; 1027 1028 qmap = FIELD_PREP(ICE_AQ_VSI_TC_Q_OFFSET_M, offset); 1029 qmap |= FIELD_PREP(ICE_AQ_VSI_TC_Q_NUM_M, pow); 1030 offset += num_rxq_per_tc; 1031 tx_count += num_txq_per_tc; 1032 ctxt->info.tc_mapping[i] = cpu_to_le16(qmap); 1033 } 1034 1035 /* if offset is non-zero, means it is calculated correctly based on 1036 * enabled TCs for a given VSI otherwise qcount_rx will always 1037 * be correct and non-zero because it is based off - VSI's 1038 * allocated Rx queues which is at least 1 (hence qcount_tx will be 1039 * at least 1) 1040 */ 1041 if (offset) 1042 rx_count = offset; 1043 else 1044 rx_count = num_rxq_per_tc; 1045 1046 if (rx_count > vsi->alloc_rxq) { 1047 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Rx queues (%u), than were allocated (%u)!\n", 1048 rx_count, vsi->alloc_rxq); 1049 return -EINVAL; 1050 } 1051 1052 if (tx_count > vsi->alloc_txq) { 1053 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Tx queues (%u), than were allocated (%u)!\n", 1054 tx_count, vsi->alloc_txq); 1055 return -EINVAL; 1056 } 1057 1058 vsi->num_txq = tx_count; 1059 vsi->num_rxq = rx_count; 1060 1061 if (vsi->type == ICE_VSI_VF && vsi->num_txq != vsi->num_rxq) { 1062 dev_dbg(ice_pf_to_dev(vsi->back), "VF VSI should have same number of Tx and Rx queues. Hence making them equal\n"); 1063 /* since there is a chance that num_rxq could have been changed 1064 * in the above for loop, make num_txq equal to num_rxq. 1065 */ 1066 vsi->num_txq = vsi->num_rxq; 1067 } 1068 1069 /* Rx queue mapping */ 1070 ctxt->info.mapping_flags |= cpu_to_le16(ICE_AQ_VSI_Q_MAP_CONTIG); 1071 /* q_mapping buffer holds the info for the first queue allocated for 1072 * this VSI in the PF space and also the number of queues associated 1073 * with this VSI. 1074 */ 1075 ctxt->info.q_mapping[0] = cpu_to_le16(vsi->rxq_map[0]); 1076 ctxt->info.q_mapping[1] = cpu_to_le16(vsi->num_rxq); 1077 1078 return 0; 1079 } 1080 1081 /** 1082 * ice_set_fd_vsi_ctx - Set FD VSI context before adding a VSI 1083 * @ctxt: the VSI context being set 1084 * @vsi: the VSI being configured 1085 */ 1086 static void ice_set_fd_vsi_ctx(struct ice_vsi_ctx *ctxt, struct ice_vsi *vsi) 1087 { 1088 u8 dflt_q_group, dflt_q_prio; 1089 u16 dflt_q, report_q, val; 1090 1091 if (vsi->type != ICE_VSI_PF && vsi->type != ICE_VSI_CTRL && 1092 vsi->type != ICE_VSI_VF && vsi->type != ICE_VSI_CHNL) 1093 return; 1094 1095 val = ICE_AQ_VSI_PROP_FLOW_DIR_VALID; 1096 ctxt->info.valid_sections |= cpu_to_le16(val); 1097 dflt_q = 0; 1098 dflt_q_group = 0; 1099 report_q = 0; 1100 dflt_q_prio = 0; 1101 1102 /* enable flow director filtering/programming */ 1103 val = ICE_AQ_VSI_FD_ENABLE | ICE_AQ_VSI_FD_PROG_ENABLE; 1104 ctxt->info.fd_options = cpu_to_le16(val); 1105 /* max of allocated flow director filters */ 1106 ctxt->info.max_fd_fltr_dedicated = 1107 cpu_to_le16(vsi->num_gfltr); 1108 /* max of shared flow director filters any VSI may program */ 1109 ctxt->info.max_fd_fltr_shared = 1110 cpu_to_le16(vsi->num_bfltr); 1111 /* default queue index within the VSI of the default FD */ 1112 val = FIELD_PREP(ICE_AQ_VSI_FD_DEF_Q_M, dflt_q); 1113 /* target queue or queue group to the FD filter */ 1114 val |= FIELD_PREP(ICE_AQ_VSI_FD_DEF_GRP_M, dflt_q_group); 1115 ctxt->info.fd_def_q = cpu_to_le16(val); 1116 /* queue index on which FD filter completion is reported */ 1117 val = FIELD_PREP(ICE_AQ_VSI_FD_REPORT_Q_M, report_q); 1118 /* priority of the default qindex action */ 1119 val |= FIELD_PREP(ICE_AQ_VSI_FD_DEF_PRIORITY_M, dflt_q_prio); 1120 ctxt->info.fd_report_opt = cpu_to_le16(val); 1121 } 1122 1123 /** 1124 * ice_set_rss_vsi_ctx - Set RSS VSI context before adding a VSI 1125 * @ctxt: the VSI context being set 1126 * @vsi: the VSI being configured 1127 */ 1128 static void ice_set_rss_vsi_ctx(struct ice_vsi_ctx *ctxt, struct ice_vsi *vsi) 1129 { 1130 u8 lut_type, hash_type; 1131 struct device *dev; 1132 struct ice_pf *pf; 1133 1134 pf = vsi->back; 1135 dev = ice_pf_to_dev(pf); 1136 1137 switch (vsi->type) { 1138 case ICE_VSI_CHNL: 1139 case ICE_VSI_PF: 1140 /* PF VSI will inherit RSS instance of PF */ 1141 lut_type = ICE_AQ_VSI_Q_OPT_RSS_LUT_PF; 1142 break; 1143 case ICE_VSI_VF: 1144 /* VF VSI will gets a small RSS table which is a VSI LUT type */ 1145 lut_type = ICE_AQ_VSI_Q_OPT_RSS_LUT_VSI; 1146 break; 1147 default: 1148 dev_dbg(dev, "Unsupported VSI type %s\n", 1149 ice_vsi_type_str(vsi->type)); 1150 return; 1151 } 1152 1153 hash_type = ICE_AQ_VSI_Q_OPT_RSS_HASH_TPLZ; 1154 vsi->rss_hfunc = hash_type; 1155 1156 ctxt->info.q_opt_rss = 1157 FIELD_PREP(ICE_AQ_VSI_Q_OPT_RSS_LUT_M, lut_type) | 1158 FIELD_PREP(ICE_AQ_VSI_Q_OPT_RSS_HASH_M, hash_type); 1159 } 1160 1161 static void 1162 ice_chnl_vsi_setup_q_map(struct ice_vsi *vsi, struct ice_vsi_ctx *ctxt) 1163 { 1164 struct ice_pf *pf = vsi->back; 1165 u16 qcount, qmap; 1166 u8 offset = 0; 1167 int pow; 1168 1169 qcount = min_t(int, vsi->num_rxq, pf->num_lan_msix); 1170 1171 pow = order_base_2(qcount); 1172 qmap = FIELD_PREP(ICE_AQ_VSI_TC_Q_OFFSET_M, offset); 1173 qmap |= FIELD_PREP(ICE_AQ_VSI_TC_Q_NUM_M, pow); 1174 1175 ctxt->info.tc_mapping[0] = cpu_to_le16(qmap); 1176 ctxt->info.mapping_flags |= cpu_to_le16(ICE_AQ_VSI_Q_MAP_CONTIG); 1177 ctxt->info.q_mapping[0] = cpu_to_le16(vsi->next_base_q); 1178 ctxt->info.q_mapping[1] = cpu_to_le16(qcount); 1179 } 1180 1181 /** 1182 * ice_vsi_is_vlan_pruning_ena - check if VLAN pruning is enabled or not 1183 * @vsi: VSI to check whether or not VLAN pruning is enabled. 1184 * 1185 * returns true if Rx VLAN pruning is enabled and false otherwise. 1186 */ 1187 static bool ice_vsi_is_vlan_pruning_ena(struct ice_vsi *vsi) 1188 { 1189 return vsi->info.sw_flags2 & ICE_AQ_VSI_SW_FLAG_RX_VLAN_PRUNE_ENA; 1190 } 1191 1192 /** 1193 * ice_vsi_init - Create and initialize a VSI 1194 * @vsi: the VSI being configured 1195 * @vsi_flags: VSI configuration flags 1196 * 1197 * Set ICE_FLAG_VSI_INIT to initialize a new VSI context, clear it to 1198 * reconfigure an existing context. 1199 * 1200 * This initializes a VSI context depending on the VSI type to be added and 1201 * passes it down to the add_vsi aq command to create a new VSI. 1202 */ 1203 static int ice_vsi_init(struct ice_vsi *vsi, u32 vsi_flags) 1204 { 1205 struct ice_pf *pf = vsi->back; 1206 struct ice_hw *hw = &pf->hw; 1207 struct ice_vsi_ctx *ctxt; 1208 struct device *dev; 1209 int ret = 0; 1210 1211 dev = ice_pf_to_dev(pf); 1212 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); 1213 if (!ctxt) 1214 return -ENOMEM; 1215 1216 switch (vsi->type) { 1217 case ICE_VSI_CTRL: 1218 case ICE_VSI_LB: 1219 case ICE_VSI_PF: 1220 ctxt->flags = ICE_AQ_VSI_TYPE_PF; 1221 break; 1222 case ICE_VSI_CHNL: 1223 ctxt->flags = ICE_AQ_VSI_TYPE_VMDQ2; 1224 break; 1225 case ICE_VSI_VF: 1226 ctxt->flags = ICE_AQ_VSI_TYPE_VF; 1227 /* VF number here is the absolute VF number (0-255) */ 1228 ctxt->vf_num = vsi->vf->vf_id + hw->func_caps.vf_base_id; 1229 break; 1230 default: 1231 ret = -ENODEV; 1232 goto out; 1233 } 1234 1235 /* Handle VLAN pruning for channel VSI if main VSI has VLAN 1236 * prune enabled 1237 */ 1238 if (vsi->type == ICE_VSI_CHNL) { 1239 struct ice_vsi *main_vsi; 1240 1241 main_vsi = ice_get_main_vsi(pf); 1242 if (main_vsi && ice_vsi_is_vlan_pruning_ena(main_vsi)) 1243 ctxt->info.sw_flags2 |= 1244 ICE_AQ_VSI_SW_FLAG_RX_VLAN_PRUNE_ENA; 1245 else 1246 ctxt->info.sw_flags2 &= 1247 ~ICE_AQ_VSI_SW_FLAG_RX_VLAN_PRUNE_ENA; 1248 } 1249 1250 ice_set_dflt_vsi_ctx(hw, ctxt); 1251 if (test_bit(ICE_FLAG_FD_ENA, pf->flags)) 1252 ice_set_fd_vsi_ctx(ctxt, vsi); 1253 /* if the switch is in VEB mode, allow VSI loopback */ 1254 if (vsi->vsw->bridge_mode == BRIDGE_MODE_VEB) 1255 ctxt->info.sw_flags |= ICE_AQ_VSI_SW_FLAG_ALLOW_LB; 1256 1257 /* Set LUT type and HASH type if RSS is enabled */ 1258 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags) && 1259 vsi->type != ICE_VSI_CTRL) { 1260 ice_set_rss_vsi_ctx(ctxt, vsi); 1261 /* if updating VSI context, make sure to set valid_section: 1262 * to indicate which section of VSI context being updated 1263 */ 1264 if (!(vsi_flags & ICE_VSI_FLAG_INIT)) 1265 ctxt->info.valid_sections |= 1266 cpu_to_le16(ICE_AQ_VSI_PROP_Q_OPT_VALID); 1267 } 1268 1269 ctxt->info.sw_id = vsi->port_info->sw_id; 1270 if (vsi->type == ICE_VSI_CHNL) { 1271 ice_chnl_vsi_setup_q_map(vsi, ctxt); 1272 } else { 1273 ret = ice_vsi_setup_q_map(vsi, ctxt); 1274 if (ret) 1275 goto out; 1276 1277 if (!(vsi_flags & ICE_VSI_FLAG_INIT)) 1278 /* means VSI being updated */ 1279 /* must to indicate which section of VSI context are 1280 * being modified 1281 */ 1282 ctxt->info.valid_sections |= 1283 cpu_to_le16(ICE_AQ_VSI_PROP_RXQ_MAP_VALID); 1284 } 1285 1286 /* Allow control frames out of main VSI */ 1287 if (vsi->type == ICE_VSI_PF) { 1288 ctxt->info.sec_flags |= ICE_AQ_VSI_SEC_FLAG_ALLOW_DEST_OVRD; 1289 ctxt->info.valid_sections |= 1290 cpu_to_le16(ICE_AQ_VSI_PROP_SECURITY_VALID); 1291 } 1292 1293 if (vsi_flags & ICE_VSI_FLAG_INIT) { 1294 ret = ice_add_vsi(hw, vsi->idx, ctxt, NULL); 1295 if (ret) { 1296 dev_err(dev, "Add VSI failed, err %d\n", ret); 1297 ret = -EIO; 1298 goto out; 1299 } 1300 } else { 1301 ret = ice_update_vsi(hw, vsi->idx, ctxt, NULL); 1302 if (ret) { 1303 dev_err(dev, "Update VSI failed, err %d\n", ret); 1304 ret = -EIO; 1305 goto out; 1306 } 1307 } 1308 1309 /* keep context for update VSI operations */ 1310 vsi->info = ctxt->info; 1311 1312 /* record VSI number returned */ 1313 vsi->vsi_num = ctxt->vsi_num; 1314 1315 out: 1316 kfree(ctxt); 1317 return ret; 1318 } 1319 1320 /** 1321 * ice_vsi_clear_rings - Deallocates the Tx and Rx rings for VSI 1322 * @vsi: the VSI having rings deallocated 1323 */ 1324 static void ice_vsi_clear_rings(struct ice_vsi *vsi) 1325 { 1326 int i; 1327 1328 /* Avoid stale references by clearing map from vector to ring */ 1329 if (vsi->q_vectors) { 1330 ice_for_each_q_vector(vsi, i) { 1331 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 1332 1333 if (q_vector) { 1334 q_vector->tx.tx_ring = NULL; 1335 q_vector->rx.rx_ring = NULL; 1336 } 1337 } 1338 } 1339 1340 if (vsi->tx_rings) { 1341 ice_for_each_alloc_txq(vsi, i) { 1342 if (vsi->tx_rings[i]) { 1343 kfree_rcu(vsi->tx_rings[i], rcu); 1344 WRITE_ONCE(vsi->tx_rings[i], NULL); 1345 } 1346 } 1347 } 1348 if (vsi->rx_rings) { 1349 ice_for_each_alloc_rxq(vsi, i) { 1350 if (vsi->rx_rings[i]) { 1351 kfree_rcu(vsi->rx_rings[i], rcu); 1352 WRITE_ONCE(vsi->rx_rings[i], NULL); 1353 } 1354 } 1355 } 1356 } 1357 1358 /** 1359 * ice_vsi_alloc_rings - Allocates Tx and Rx rings for the VSI 1360 * @vsi: VSI which is having rings allocated 1361 */ 1362 static int ice_vsi_alloc_rings(struct ice_vsi *vsi) 1363 { 1364 bool dvm_ena = ice_is_dvm_ena(&vsi->back->hw); 1365 struct ice_pf *pf = vsi->back; 1366 struct device *dev; 1367 u16 i; 1368 1369 dev = ice_pf_to_dev(pf); 1370 /* Allocate Tx rings */ 1371 ice_for_each_alloc_txq(vsi, i) { 1372 struct ice_tx_ring *ring; 1373 1374 /* allocate with kzalloc(), free with kfree_rcu() */ 1375 ring = kzalloc(sizeof(*ring), GFP_KERNEL); 1376 1377 if (!ring) 1378 goto err_out; 1379 1380 ring->q_index = i; 1381 ring->reg_idx = vsi->txq_map[i]; 1382 ring->vsi = vsi; 1383 ring->tx_tstamps = &pf->ptp.port.tx; 1384 ring->dev = dev; 1385 ring->count = vsi->num_tx_desc; 1386 ring->txq_teid = ICE_INVAL_TEID; 1387 if (dvm_ena) 1388 ring->flags |= ICE_TX_FLAGS_RING_VLAN_L2TAG2; 1389 else 1390 ring->flags |= ICE_TX_FLAGS_RING_VLAN_L2TAG1; 1391 WRITE_ONCE(vsi->tx_rings[i], ring); 1392 } 1393 1394 /* Allocate Rx rings */ 1395 ice_for_each_alloc_rxq(vsi, i) { 1396 struct ice_rx_ring *ring; 1397 1398 /* allocate with kzalloc(), free with kfree_rcu() */ 1399 ring = kzalloc(sizeof(*ring), GFP_KERNEL); 1400 if (!ring) 1401 goto err_out; 1402 1403 ring->q_index = i; 1404 ring->reg_idx = vsi->rxq_map[i]; 1405 ring->vsi = vsi; 1406 ring->netdev = vsi->netdev; 1407 ring->dev = dev; 1408 ring->count = vsi->num_rx_desc; 1409 ring->cached_phctime = pf->ptp.cached_phc_time; 1410 WRITE_ONCE(vsi->rx_rings[i], ring); 1411 } 1412 1413 return 0; 1414 1415 err_out: 1416 ice_vsi_clear_rings(vsi); 1417 return -ENOMEM; 1418 } 1419 1420 /** 1421 * ice_vsi_manage_rss_lut - disable/enable RSS 1422 * @vsi: the VSI being changed 1423 * @ena: boolean value indicating if this is an enable or disable request 1424 * 1425 * In the event of disable request for RSS, this function will zero out RSS 1426 * LUT, while in the event of enable request for RSS, it will reconfigure RSS 1427 * LUT. 1428 */ 1429 void ice_vsi_manage_rss_lut(struct ice_vsi *vsi, bool ena) 1430 { 1431 u8 *lut; 1432 1433 lut = kzalloc(vsi->rss_table_size, GFP_KERNEL); 1434 if (!lut) 1435 return; 1436 1437 if (ena) { 1438 if (vsi->rss_lut_user) 1439 memcpy(lut, vsi->rss_lut_user, vsi->rss_table_size); 1440 else 1441 ice_fill_rss_lut(lut, vsi->rss_table_size, 1442 vsi->rss_size); 1443 } 1444 1445 ice_set_rss_lut(vsi, lut, vsi->rss_table_size); 1446 kfree(lut); 1447 } 1448 1449 /** 1450 * ice_vsi_cfg_crc_strip - Configure CRC stripping for a VSI 1451 * @vsi: VSI to be configured 1452 * @disable: set to true to have FCS / CRC in the frame data 1453 */ 1454 void ice_vsi_cfg_crc_strip(struct ice_vsi *vsi, bool disable) 1455 { 1456 int i; 1457 1458 ice_for_each_rxq(vsi, i) 1459 if (disable) 1460 vsi->rx_rings[i]->flags |= ICE_RX_FLAGS_CRC_STRIP_DIS; 1461 else 1462 vsi->rx_rings[i]->flags &= ~ICE_RX_FLAGS_CRC_STRIP_DIS; 1463 } 1464 1465 /** 1466 * ice_vsi_cfg_rss_lut_key - Configure RSS params for a VSI 1467 * @vsi: VSI to be configured 1468 */ 1469 int ice_vsi_cfg_rss_lut_key(struct ice_vsi *vsi) 1470 { 1471 struct ice_pf *pf = vsi->back; 1472 struct device *dev; 1473 u8 *lut, *key; 1474 int err; 1475 1476 dev = ice_pf_to_dev(pf); 1477 if (vsi->type == ICE_VSI_PF && vsi->ch_rss_size && 1478 (test_bit(ICE_FLAG_TC_MQPRIO, pf->flags))) { 1479 vsi->rss_size = min_t(u16, vsi->rss_size, vsi->ch_rss_size); 1480 } else { 1481 vsi->rss_size = min_t(u16, vsi->rss_size, vsi->num_rxq); 1482 1483 /* If orig_rss_size is valid and it is less than determined 1484 * main VSI's rss_size, update main VSI's rss_size to be 1485 * orig_rss_size so that when tc-qdisc is deleted, main VSI 1486 * RSS table gets programmed to be correct (whatever it was 1487 * to begin with (prior to setup-tc for ADQ config) 1488 */ 1489 if (vsi->orig_rss_size && vsi->rss_size < vsi->orig_rss_size && 1490 vsi->orig_rss_size <= vsi->num_rxq) { 1491 vsi->rss_size = vsi->orig_rss_size; 1492 /* now orig_rss_size is used, reset it to zero */ 1493 vsi->orig_rss_size = 0; 1494 } 1495 } 1496 1497 lut = kzalloc(vsi->rss_table_size, GFP_KERNEL); 1498 if (!lut) 1499 return -ENOMEM; 1500 1501 if (vsi->rss_lut_user) 1502 memcpy(lut, vsi->rss_lut_user, vsi->rss_table_size); 1503 else 1504 ice_fill_rss_lut(lut, vsi->rss_table_size, vsi->rss_size); 1505 1506 err = ice_set_rss_lut(vsi, lut, vsi->rss_table_size); 1507 if (err) { 1508 dev_err(dev, "set_rss_lut failed, error %d\n", err); 1509 goto ice_vsi_cfg_rss_exit; 1510 } 1511 1512 key = kzalloc(ICE_GET_SET_RSS_KEY_EXTEND_KEY_SIZE, GFP_KERNEL); 1513 if (!key) { 1514 err = -ENOMEM; 1515 goto ice_vsi_cfg_rss_exit; 1516 } 1517 1518 if (vsi->rss_hkey_user) 1519 memcpy(key, vsi->rss_hkey_user, ICE_GET_SET_RSS_KEY_EXTEND_KEY_SIZE); 1520 else 1521 netdev_rss_key_fill((void *)key, ICE_GET_SET_RSS_KEY_EXTEND_KEY_SIZE); 1522 1523 err = ice_set_rss_key(vsi, key); 1524 if (err) 1525 dev_err(dev, "set_rss_key failed, error %d\n", err); 1526 1527 kfree(key); 1528 ice_vsi_cfg_rss_exit: 1529 kfree(lut); 1530 return err; 1531 } 1532 1533 /** 1534 * ice_vsi_set_vf_rss_flow_fld - Sets VF VSI RSS input set for different flows 1535 * @vsi: VSI to be configured 1536 * 1537 * This function will only be called during the VF VSI setup. Upon successful 1538 * completion of package download, this function will configure default RSS 1539 * input sets for VF VSI. 1540 */ 1541 static void ice_vsi_set_vf_rss_flow_fld(struct ice_vsi *vsi) 1542 { 1543 struct ice_pf *pf = vsi->back; 1544 struct device *dev; 1545 int status; 1546 1547 dev = ice_pf_to_dev(pf); 1548 if (ice_is_safe_mode(pf)) { 1549 dev_dbg(dev, "Advanced RSS disabled. Package download failed, vsi num = %d\n", 1550 vsi->vsi_num); 1551 return; 1552 } 1553 1554 status = ice_add_avf_rss_cfg(&pf->hw, vsi, ICE_DEFAULT_RSS_HENA); 1555 if (status) 1556 dev_dbg(dev, "ice_add_avf_rss_cfg failed for vsi = %d, error = %d\n", 1557 vsi->vsi_num, status); 1558 } 1559 1560 static const struct ice_rss_hash_cfg default_rss_cfgs[] = { 1561 /* configure RSS for IPv4 with input set IP src/dst */ 1562 {ICE_FLOW_SEG_HDR_IPV4, ICE_FLOW_HASH_IPV4, ICE_RSS_ANY_HEADERS, false}, 1563 /* configure RSS for IPv6 with input set IPv6 src/dst */ 1564 {ICE_FLOW_SEG_HDR_IPV6, ICE_FLOW_HASH_IPV6, ICE_RSS_ANY_HEADERS, false}, 1565 /* configure RSS for tcp4 with input set IP src/dst, TCP src/dst */ 1566 {ICE_FLOW_SEG_HDR_TCP | ICE_FLOW_SEG_HDR_IPV4, 1567 ICE_HASH_TCP_IPV4, ICE_RSS_ANY_HEADERS, false}, 1568 /* configure RSS for udp4 with input set IP src/dst, UDP src/dst */ 1569 {ICE_FLOW_SEG_HDR_UDP | ICE_FLOW_SEG_HDR_IPV4, 1570 ICE_HASH_UDP_IPV4, ICE_RSS_ANY_HEADERS, false}, 1571 /* configure RSS for sctp4 with input set IP src/dst - only support 1572 * RSS on SCTPv4 on outer headers (non-tunneled) 1573 */ 1574 {ICE_FLOW_SEG_HDR_SCTP | ICE_FLOW_SEG_HDR_IPV4, 1575 ICE_HASH_SCTP_IPV4, ICE_RSS_OUTER_HEADERS, false}, 1576 /* configure RSS for gtpc4 with input set IPv4 src/dst */ 1577 {ICE_FLOW_SEG_HDR_GTPC | ICE_FLOW_SEG_HDR_IPV4, 1578 ICE_FLOW_HASH_IPV4, ICE_RSS_OUTER_HEADERS, false}, 1579 /* configure RSS for gtpc4t with input set IPv4 src/dst */ 1580 {ICE_FLOW_SEG_HDR_GTPC_TEID | ICE_FLOW_SEG_HDR_IPV4, 1581 ICE_FLOW_HASH_GTP_C_IPV4_TEID, ICE_RSS_OUTER_HEADERS, false}, 1582 /* configure RSS for gtpu4 with input set IPv4 src/dst */ 1583 {ICE_FLOW_SEG_HDR_GTPU_IP | ICE_FLOW_SEG_HDR_IPV4, 1584 ICE_FLOW_HASH_GTP_U_IPV4_TEID, ICE_RSS_OUTER_HEADERS, false}, 1585 /* configure RSS for gtpu4e with input set IPv4 src/dst */ 1586 {ICE_FLOW_SEG_HDR_GTPU_EH | ICE_FLOW_SEG_HDR_IPV4, 1587 ICE_FLOW_HASH_GTP_U_IPV4_EH, ICE_RSS_OUTER_HEADERS, false}, 1588 /* configure RSS for gtpu4u with input set IPv4 src/dst */ 1589 { ICE_FLOW_SEG_HDR_GTPU_UP | ICE_FLOW_SEG_HDR_IPV4, 1590 ICE_FLOW_HASH_GTP_U_IPV4_UP, ICE_RSS_OUTER_HEADERS, false}, 1591 /* configure RSS for gtpu4d with input set IPv4 src/dst */ 1592 {ICE_FLOW_SEG_HDR_GTPU_DWN | ICE_FLOW_SEG_HDR_IPV4, 1593 ICE_FLOW_HASH_GTP_U_IPV4_DWN, ICE_RSS_OUTER_HEADERS, false}, 1594 1595 /* configure RSS for tcp6 with input set IPv6 src/dst, TCP src/dst */ 1596 {ICE_FLOW_SEG_HDR_TCP | ICE_FLOW_SEG_HDR_IPV6, 1597 ICE_HASH_TCP_IPV6, ICE_RSS_ANY_HEADERS, false}, 1598 /* configure RSS for udp6 with input set IPv6 src/dst, UDP src/dst */ 1599 {ICE_FLOW_SEG_HDR_UDP | ICE_FLOW_SEG_HDR_IPV6, 1600 ICE_HASH_UDP_IPV6, ICE_RSS_ANY_HEADERS, false}, 1601 /* configure RSS for sctp6 with input set IPv6 src/dst - only support 1602 * RSS on SCTPv6 on outer headers (non-tunneled) 1603 */ 1604 {ICE_FLOW_SEG_HDR_SCTP | ICE_FLOW_SEG_HDR_IPV6, 1605 ICE_HASH_SCTP_IPV6, ICE_RSS_OUTER_HEADERS, false}, 1606 /* configure RSS for IPSEC ESP SPI with input set MAC_IPV4_SPI */ 1607 {ICE_FLOW_SEG_HDR_ESP, 1608 ICE_FLOW_HASH_ESP_SPI, ICE_RSS_OUTER_HEADERS, false}, 1609 /* configure RSS for gtpc6 with input set IPv6 src/dst */ 1610 {ICE_FLOW_SEG_HDR_GTPC | ICE_FLOW_SEG_HDR_IPV6, 1611 ICE_FLOW_HASH_IPV6, ICE_RSS_OUTER_HEADERS, false}, 1612 /* configure RSS for gtpc6t with input set IPv6 src/dst */ 1613 {ICE_FLOW_SEG_HDR_GTPC_TEID | ICE_FLOW_SEG_HDR_IPV6, 1614 ICE_FLOW_HASH_GTP_C_IPV6_TEID, ICE_RSS_OUTER_HEADERS, false}, 1615 /* configure RSS for gtpu6 with input set IPv6 src/dst */ 1616 {ICE_FLOW_SEG_HDR_GTPU_IP | ICE_FLOW_SEG_HDR_IPV6, 1617 ICE_FLOW_HASH_GTP_U_IPV6_TEID, ICE_RSS_OUTER_HEADERS, false}, 1618 /* configure RSS for gtpu6e with input set IPv6 src/dst */ 1619 {ICE_FLOW_SEG_HDR_GTPU_EH | ICE_FLOW_SEG_HDR_IPV6, 1620 ICE_FLOW_HASH_GTP_U_IPV6_EH, ICE_RSS_OUTER_HEADERS, false}, 1621 /* configure RSS for gtpu6u with input set IPv6 src/dst */ 1622 { ICE_FLOW_SEG_HDR_GTPU_UP | ICE_FLOW_SEG_HDR_IPV6, 1623 ICE_FLOW_HASH_GTP_U_IPV6_UP, ICE_RSS_OUTER_HEADERS, false}, 1624 /* configure RSS for gtpu6d with input set IPv6 src/dst */ 1625 {ICE_FLOW_SEG_HDR_GTPU_DWN | ICE_FLOW_SEG_HDR_IPV6, 1626 ICE_FLOW_HASH_GTP_U_IPV6_DWN, ICE_RSS_OUTER_HEADERS, false}, 1627 }; 1628 1629 /** 1630 * ice_vsi_set_rss_flow_fld - Sets RSS input set for different flows 1631 * @vsi: VSI to be configured 1632 * 1633 * This function will only be called after successful download package call 1634 * during initialization of PF. Since the downloaded package will erase the 1635 * RSS section, this function will configure RSS input sets for different 1636 * flow types. The last profile added has the highest priority, therefore 2 1637 * tuple profiles (i.e. IPv4 src/dst) are added before 4 tuple profiles 1638 * (i.e. IPv4 src/dst TCP src/dst port). 1639 */ 1640 static void ice_vsi_set_rss_flow_fld(struct ice_vsi *vsi) 1641 { 1642 u16 vsi_num = vsi->vsi_num; 1643 struct ice_pf *pf = vsi->back; 1644 struct ice_hw *hw = &pf->hw; 1645 struct device *dev; 1646 int status; 1647 u32 i; 1648 1649 dev = ice_pf_to_dev(pf); 1650 if (ice_is_safe_mode(pf)) { 1651 dev_dbg(dev, "Advanced RSS disabled. Package download failed, vsi num = %d\n", 1652 vsi_num); 1653 return; 1654 } 1655 for (i = 0; i < ARRAY_SIZE(default_rss_cfgs); i++) { 1656 const struct ice_rss_hash_cfg *cfg = &default_rss_cfgs[i]; 1657 1658 status = ice_add_rss_cfg(hw, vsi, cfg); 1659 if (status) 1660 dev_dbg(dev, "ice_add_rss_cfg failed, addl_hdrs = %x, hash_flds = %llx, hdr_type = %d, symm = %d\n", 1661 cfg->addl_hdrs, cfg->hash_flds, 1662 cfg->hdr_type, cfg->symm); 1663 } 1664 } 1665 1666 /** 1667 * ice_pf_state_is_nominal - checks the PF for nominal state 1668 * @pf: pointer to PF to check 1669 * 1670 * Check the PF's state for a collection of bits that would indicate 1671 * the PF is in a state that would inhibit normal operation for 1672 * driver functionality. 1673 * 1674 * Returns true if PF is in a nominal state, false otherwise 1675 */ 1676 bool ice_pf_state_is_nominal(struct ice_pf *pf) 1677 { 1678 DECLARE_BITMAP(check_bits, ICE_STATE_NBITS) = { 0 }; 1679 1680 if (!pf) 1681 return false; 1682 1683 bitmap_set(check_bits, 0, ICE_STATE_NOMINAL_CHECK_BITS); 1684 if (bitmap_intersects(pf->state, check_bits, ICE_STATE_NBITS)) 1685 return false; 1686 1687 return true; 1688 } 1689 1690 /** 1691 * ice_update_eth_stats - Update VSI-specific ethernet statistics counters 1692 * @vsi: the VSI to be updated 1693 */ 1694 void ice_update_eth_stats(struct ice_vsi *vsi) 1695 { 1696 struct ice_eth_stats *prev_es, *cur_es; 1697 struct ice_hw *hw = &vsi->back->hw; 1698 struct ice_pf *pf = vsi->back; 1699 u16 vsi_num = vsi->vsi_num; /* HW absolute index of a VSI */ 1700 1701 prev_es = &vsi->eth_stats_prev; 1702 cur_es = &vsi->eth_stats; 1703 1704 if (ice_is_reset_in_progress(pf->state)) 1705 vsi->stat_offsets_loaded = false; 1706 1707 ice_stat_update40(hw, GLV_GORCL(vsi_num), vsi->stat_offsets_loaded, 1708 &prev_es->rx_bytes, &cur_es->rx_bytes); 1709 1710 ice_stat_update40(hw, GLV_UPRCL(vsi_num), vsi->stat_offsets_loaded, 1711 &prev_es->rx_unicast, &cur_es->rx_unicast); 1712 1713 ice_stat_update40(hw, GLV_MPRCL(vsi_num), vsi->stat_offsets_loaded, 1714 &prev_es->rx_multicast, &cur_es->rx_multicast); 1715 1716 ice_stat_update40(hw, GLV_BPRCL(vsi_num), vsi->stat_offsets_loaded, 1717 &prev_es->rx_broadcast, &cur_es->rx_broadcast); 1718 1719 ice_stat_update32(hw, GLV_RDPC(vsi_num), vsi->stat_offsets_loaded, 1720 &prev_es->rx_discards, &cur_es->rx_discards); 1721 1722 ice_stat_update40(hw, GLV_GOTCL(vsi_num), vsi->stat_offsets_loaded, 1723 &prev_es->tx_bytes, &cur_es->tx_bytes); 1724 1725 ice_stat_update40(hw, GLV_UPTCL(vsi_num), vsi->stat_offsets_loaded, 1726 &prev_es->tx_unicast, &cur_es->tx_unicast); 1727 1728 ice_stat_update40(hw, GLV_MPTCL(vsi_num), vsi->stat_offsets_loaded, 1729 &prev_es->tx_multicast, &cur_es->tx_multicast); 1730 1731 ice_stat_update40(hw, GLV_BPTCL(vsi_num), vsi->stat_offsets_loaded, 1732 &prev_es->tx_broadcast, &cur_es->tx_broadcast); 1733 1734 ice_stat_update32(hw, GLV_TEPC(vsi_num), vsi->stat_offsets_loaded, 1735 &prev_es->tx_errors, &cur_es->tx_errors); 1736 1737 vsi->stat_offsets_loaded = true; 1738 } 1739 1740 /** 1741 * ice_write_qrxflxp_cntxt - write/configure QRXFLXP_CNTXT register 1742 * @hw: HW pointer 1743 * @pf_q: index of the Rx queue in the PF's queue space 1744 * @rxdid: flexible descriptor RXDID 1745 * @prio: priority for the RXDID for this queue 1746 * @ena_ts: true to enable timestamp and false to disable timestamp 1747 */ 1748 void 1749 ice_write_qrxflxp_cntxt(struct ice_hw *hw, u16 pf_q, u32 rxdid, u32 prio, 1750 bool ena_ts) 1751 { 1752 int regval = rd32(hw, QRXFLXP_CNTXT(pf_q)); 1753 1754 /* clear any previous values */ 1755 regval &= ~(QRXFLXP_CNTXT_RXDID_IDX_M | 1756 QRXFLXP_CNTXT_RXDID_PRIO_M | 1757 QRXFLXP_CNTXT_TS_M); 1758 1759 regval |= FIELD_PREP(QRXFLXP_CNTXT_RXDID_IDX_M, rxdid); 1760 regval |= FIELD_PREP(QRXFLXP_CNTXT_RXDID_PRIO_M, prio); 1761 1762 if (ena_ts) 1763 /* Enable TimeSync on this queue */ 1764 regval |= QRXFLXP_CNTXT_TS_M; 1765 1766 wr32(hw, QRXFLXP_CNTXT(pf_q), regval); 1767 } 1768 1769 /** 1770 * ice_intrl_usec_to_reg - convert interrupt rate limit to register value 1771 * @intrl: interrupt rate limit in usecs 1772 * @gran: interrupt rate limit granularity in usecs 1773 * 1774 * This function converts a decimal interrupt rate limit in usecs to the format 1775 * expected by firmware. 1776 */ 1777 static u32 ice_intrl_usec_to_reg(u8 intrl, u8 gran) 1778 { 1779 u32 val = intrl / gran; 1780 1781 if (val) 1782 return val | GLINT_RATE_INTRL_ENA_M; 1783 return 0; 1784 } 1785 1786 /** 1787 * ice_write_intrl - write throttle rate limit to interrupt specific register 1788 * @q_vector: pointer to interrupt specific structure 1789 * @intrl: throttle rate limit in microseconds to write 1790 */ 1791 void ice_write_intrl(struct ice_q_vector *q_vector, u8 intrl) 1792 { 1793 struct ice_hw *hw = &q_vector->vsi->back->hw; 1794 1795 wr32(hw, GLINT_RATE(q_vector->reg_idx), 1796 ice_intrl_usec_to_reg(intrl, ICE_INTRL_GRAN_ABOVE_25)); 1797 } 1798 1799 static struct ice_q_vector *ice_pull_qvec_from_rc(struct ice_ring_container *rc) 1800 { 1801 switch (rc->type) { 1802 case ICE_RX_CONTAINER: 1803 if (rc->rx_ring) 1804 return rc->rx_ring->q_vector; 1805 break; 1806 case ICE_TX_CONTAINER: 1807 if (rc->tx_ring) 1808 return rc->tx_ring->q_vector; 1809 break; 1810 default: 1811 break; 1812 } 1813 1814 return NULL; 1815 } 1816 1817 /** 1818 * __ice_write_itr - write throttle rate to register 1819 * @q_vector: pointer to interrupt data structure 1820 * @rc: pointer to ring container 1821 * @itr: throttle rate in microseconds to write 1822 */ 1823 static void __ice_write_itr(struct ice_q_vector *q_vector, 1824 struct ice_ring_container *rc, u16 itr) 1825 { 1826 struct ice_hw *hw = &q_vector->vsi->back->hw; 1827 1828 wr32(hw, GLINT_ITR(rc->itr_idx, q_vector->reg_idx), 1829 ITR_REG_ALIGN(itr) >> ICE_ITR_GRAN_S); 1830 } 1831 1832 /** 1833 * ice_write_itr - write throttle rate to queue specific register 1834 * @rc: pointer to ring container 1835 * @itr: throttle rate in microseconds to write 1836 */ 1837 void ice_write_itr(struct ice_ring_container *rc, u16 itr) 1838 { 1839 struct ice_q_vector *q_vector; 1840 1841 q_vector = ice_pull_qvec_from_rc(rc); 1842 if (!q_vector) 1843 return; 1844 1845 __ice_write_itr(q_vector, rc, itr); 1846 } 1847 1848 /** 1849 * ice_set_q_vector_intrl - set up interrupt rate limiting 1850 * @q_vector: the vector to be configured 1851 * 1852 * Interrupt rate limiting is local to the vector, not per-queue so we must 1853 * detect if either ring container has dynamic moderation enabled to decide 1854 * what to set the interrupt rate limit to via INTRL settings. In the case that 1855 * dynamic moderation is disabled on both, write the value with the cached 1856 * setting to make sure INTRL register matches the user visible value. 1857 */ 1858 void ice_set_q_vector_intrl(struct ice_q_vector *q_vector) 1859 { 1860 if (ITR_IS_DYNAMIC(&q_vector->tx) || ITR_IS_DYNAMIC(&q_vector->rx)) { 1861 /* in the case of dynamic enabled, cap each vector to no more 1862 * than (4 us) 250,000 ints/sec, which allows low latency 1863 * but still less than 500,000 interrupts per second, which 1864 * reduces CPU a bit in the case of the lowest latency 1865 * setting. The 4 here is a value in microseconds. 1866 */ 1867 ice_write_intrl(q_vector, 4); 1868 } else { 1869 ice_write_intrl(q_vector, q_vector->intrl); 1870 } 1871 } 1872 1873 /** 1874 * ice_vsi_cfg_msix - MSIX mode Interrupt Config in the HW 1875 * @vsi: the VSI being configured 1876 * 1877 * This configures MSIX mode interrupts for the PF VSI, and should not be used 1878 * for the VF VSI. 1879 */ 1880 void ice_vsi_cfg_msix(struct ice_vsi *vsi) 1881 { 1882 struct ice_pf *pf = vsi->back; 1883 struct ice_hw *hw = &pf->hw; 1884 u16 txq = 0, rxq = 0; 1885 int i, q; 1886 1887 ice_for_each_q_vector(vsi, i) { 1888 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 1889 u16 reg_idx = q_vector->reg_idx; 1890 1891 ice_cfg_itr(hw, q_vector); 1892 1893 /* Both Transmit Queue Interrupt Cause Control register 1894 * and Receive Queue Interrupt Cause control register 1895 * expects MSIX_INDX field to be the vector index 1896 * within the function space and not the absolute 1897 * vector index across PF or across device. 1898 * For SR-IOV VF VSIs queue vector index always starts 1899 * with 1 since first vector index(0) is used for OICR 1900 * in VF space. Since VMDq and other PF VSIs are within 1901 * the PF function space, use the vector index that is 1902 * tracked for this PF. 1903 */ 1904 for (q = 0; q < q_vector->num_ring_tx; q++) { 1905 ice_cfg_txq_interrupt(vsi, txq, reg_idx, 1906 q_vector->tx.itr_idx); 1907 txq++; 1908 } 1909 1910 for (q = 0; q < q_vector->num_ring_rx; q++) { 1911 ice_cfg_rxq_interrupt(vsi, rxq, reg_idx, 1912 q_vector->rx.itr_idx); 1913 rxq++; 1914 } 1915 } 1916 } 1917 1918 /** 1919 * ice_vsi_start_all_rx_rings - start/enable all of a VSI's Rx rings 1920 * @vsi: the VSI whose rings are to be enabled 1921 * 1922 * Returns 0 on success and a negative value on error 1923 */ 1924 int ice_vsi_start_all_rx_rings(struct ice_vsi *vsi) 1925 { 1926 return ice_vsi_ctrl_all_rx_rings(vsi, true); 1927 } 1928 1929 /** 1930 * ice_vsi_stop_all_rx_rings - stop/disable all of a VSI's Rx rings 1931 * @vsi: the VSI whose rings are to be disabled 1932 * 1933 * Returns 0 on success and a negative value on error 1934 */ 1935 int ice_vsi_stop_all_rx_rings(struct ice_vsi *vsi) 1936 { 1937 return ice_vsi_ctrl_all_rx_rings(vsi, false); 1938 } 1939 1940 /** 1941 * ice_vsi_stop_tx_rings - Disable Tx rings 1942 * @vsi: the VSI being configured 1943 * @rst_src: reset source 1944 * @rel_vmvf_num: Relative ID of VF/VM 1945 * @rings: Tx ring array to be stopped 1946 * @count: number of Tx ring array elements 1947 */ 1948 static int 1949 ice_vsi_stop_tx_rings(struct ice_vsi *vsi, enum ice_disq_rst_src rst_src, 1950 u16 rel_vmvf_num, struct ice_tx_ring **rings, u16 count) 1951 { 1952 u16 q_idx; 1953 1954 if (vsi->num_txq > ICE_LAN_TXQ_MAX_QDIS) 1955 return -EINVAL; 1956 1957 for (q_idx = 0; q_idx < count; q_idx++) { 1958 struct ice_txq_meta txq_meta = { }; 1959 int status; 1960 1961 if (!rings || !rings[q_idx]) 1962 return -EINVAL; 1963 1964 ice_fill_txq_meta(vsi, rings[q_idx], &txq_meta); 1965 status = ice_vsi_stop_tx_ring(vsi, rst_src, rel_vmvf_num, 1966 rings[q_idx], &txq_meta); 1967 1968 if (status) 1969 return status; 1970 } 1971 1972 return 0; 1973 } 1974 1975 /** 1976 * ice_vsi_stop_lan_tx_rings - Disable LAN Tx rings 1977 * @vsi: the VSI being configured 1978 * @rst_src: reset source 1979 * @rel_vmvf_num: Relative ID of VF/VM 1980 */ 1981 int 1982 ice_vsi_stop_lan_tx_rings(struct ice_vsi *vsi, enum ice_disq_rst_src rst_src, 1983 u16 rel_vmvf_num) 1984 { 1985 return ice_vsi_stop_tx_rings(vsi, rst_src, rel_vmvf_num, vsi->tx_rings, vsi->num_txq); 1986 } 1987 1988 /** 1989 * ice_vsi_stop_xdp_tx_rings - Disable XDP Tx rings 1990 * @vsi: the VSI being configured 1991 */ 1992 int ice_vsi_stop_xdp_tx_rings(struct ice_vsi *vsi) 1993 { 1994 return ice_vsi_stop_tx_rings(vsi, ICE_NO_RESET, 0, vsi->xdp_rings, vsi->num_xdp_txq); 1995 } 1996 1997 /** 1998 * ice_vsi_is_rx_queue_active 1999 * @vsi: the VSI being configured 2000 * 2001 * Return true if at least one queue is active. 2002 */ 2003 bool ice_vsi_is_rx_queue_active(struct ice_vsi *vsi) 2004 { 2005 struct ice_pf *pf = vsi->back; 2006 struct ice_hw *hw = &pf->hw; 2007 int i; 2008 2009 ice_for_each_rxq(vsi, i) { 2010 u32 rx_reg; 2011 int pf_q; 2012 2013 pf_q = vsi->rxq_map[i]; 2014 rx_reg = rd32(hw, QRX_CTRL(pf_q)); 2015 if (rx_reg & QRX_CTRL_QENA_STAT_M) 2016 return true; 2017 } 2018 2019 return false; 2020 } 2021 2022 static void ice_vsi_set_tc_cfg(struct ice_vsi *vsi) 2023 { 2024 if (!test_bit(ICE_FLAG_DCB_ENA, vsi->back->flags)) { 2025 vsi->tc_cfg.ena_tc = ICE_DFLT_TRAFFIC_CLASS; 2026 vsi->tc_cfg.numtc = 1; 2027 return; 2028 } 2029 2030 /* set VSI TC information based on DCB config */ 2031 ice_vsi_set_dcb_tc_cfg(vsi); 2032 } 2033 2034 /** 2035 * ice_cfg_sw_lldp - Config switch rules for LLDP packet handling 2036 * @vsi: the VSI being configured 2037 * @tx: bool to determine Tx or Rx rule 2038 * @create: bool to determine create or remove Rule 2039 */ 2040 void ice_cfg_sw_lldp(struct ice_vsi *vsi, bool tx, bool create) 2041 { 2042 int (*eth_fltr)(struct ice_vsi *v, u16 type, u16 flag, 2043 enum ice_sw_fwd_act_type act); 2044 struct ice_pf *pf = vsi->back; 2045 struct device *dev; 2046 int status; 2047 2048 dev = ice_pf_to_dev(pf); 2049 eth_fltr = create ? ice_fltr_add_eth : ice_fltr_remove_eth; 2050 2051 if (tx) { 2052 status = eth_fltr(vsi, ETH_P_LLDP, ICE_FLTR_TX, 2053 ICE_DROP_PACKET); 2054 } else { 2055 if (ice_fw_supports_lldp_fltr_ctrl(&pf->hw)) { 2056 status = ice_lldp_fltr_add_remove(&pf->hw, vsi->vsi_num, 2057 create); 2058 } else { 2059 status = eth_fltr(vsi, ETH_P_LLDP, ICE_FLTR_RX, 2060 ICE_FWD_TO_VSI); 2061 } 2062 } 2063 2064 if (status) 2065 dev_dbg(dev, "Fail %s %s LLDP rule on VSI %i error: %d\n", 2066 create ? "adding" : "removing", tx ? "TX" : "RX", 2067 vsi->vsi_num, status); 2068 } 2069 2070 /** 2071 * ice_set_agg_vsi - sets up scheduler aggregator node and move VSI into it 2072 * @vsi: pointer to the VSI 2073 * 2074 * This function will allocate new scheduler aggregator now if needed and will 2075 * move specified VSI into it. 2076 */ 2077 static void ice_set_agg_vsi(struct ice_vsi *vsi) 2078 { 2079 struct device *dev = ice_pf_to_dev(vsi->back); 2080 struct ice_agg_node *agg_node_iter = NULL; 2081 u32 agg_id = ICE_INVALID_AGG_NODE_ID; 2082 struct ice_agg_node *agg_node = NULL; 2083 int node_offset, max_agg_nodes = 0; 2084 struct ice_port_info *port_info; 2085 struct ice_pf *pf = vsi->back; 2086 u32 agg_node_id_start = 0; 2087 int status; 2088 2089 /* create (as needed) scheduler aggregator node and move VSI into 2090 * corresponding aggregator node 2091 * - PF aggregator node to contains VSIs of type _PF and _CTRL 2092 * - VF aggregator nodes will contain VF VSI 2093 */ 2094 port_info = pf->hw.port_info; 2095 if (!port_info) 2096 return; 2097 2098 switch (vsi->type) { 2099 case ICE_VSI_CTRL: 2100 case ICE_VSI_CHNL: 2101 case ICE_VSI_LB: 2102 case ICE_VSI_PF: 2103 max_agg_nodes = ICE_MAX_PF_AGG_NODES; 2104 agg_node_id_start = ICE_PF_AGG_NODE_ID_START; 2105 agg_node_iter = &pf->pf_agg_node[0]; 2106 break; 2107 case ICE_VSI_VF: 2108 /* user can create 'n' VFs on a given PF, but since max children 2109 * per aggregator node can be only 64. Following code handles 2110 * aggregator(s) for VF VSIs, either selects a agg_node which 2111 * was already created provided num_vsis < 64, otherwise 2112 * select next available node, which will be created 2113 */ 2114 max_agg_nodes = ICE_MAX_VF_AGG_NODES; 2115 agg_node_id_start = ICE_VF_AGG_NODE_ID_START; 2116 agg_node_iter = &pf->vf_agg_node[0]; 2117 break; 2118 default: 2119 /* other VSI type, handle later if needed */ 2120 dev_dbg(dev, "unexpected VSI type %s\n", 2121 ice_vsi_type_str(vsi->type)); 2122 return; 2123 } 2124 2125 /* find the appropriate aggregator node */ 2126 for (node_offset = 0; node_offset < max_agg_nodes; node_offset++) { 2127 /* see if we can find space in previously created 2128 * node if num_vsis < 64, otherwise skip 2129 */ 2130 if (agg_node_iter->num_vsis && 2131 agg_node_iter->num_vsis == ICE_MAX_VSIS_IN_AGG_NODE) { 2132 agg_node_iter++; 2133 continue; 2134 } 2135 2136 if (agg_node_iter->valid && 2137 agg_node_iter->agg_id != ICE_INVALID_AGG_NODE_ID) { 2138 agg_id = agg_node_iter->agg_id; 2139 agg_node = agg_node_iter; 2140 break; 2141 } 2142 2143 /* find unclaimed agg_id */ 2144 if (agg_node_iter->agg_id == ICE_INVALID_AGG_NODE_ID) { 2145 agg_id = node_offset + agg_node_id_start; 2146 agg_node = agg_node_iter; 2147 break; 2148 } 2149 /* move to next agg_node */ 2150 agg_node_iter++; 2151 } 2152 2153 if (!agg_node) 2154 return; 2155 2156 /* if selected aggregator node was not created, create it */ 2157 if (!agg_node->valid) { 2158 status = ice_cfg_agg(port_info, agg_id, ICE_AGG_TYPE_AGG, 2159 (u8)vsi->tc_cfg.ena_tc); 2160 if (status) { 2161 dev_err(dev, "unable to create aggregator node with agg_id %u\n", 2162 agg_id); 2163 return; 2164 } 2165 /* aggregator node is created, store the needed info */ 2166 agg_node->valid = true; 2167 agg_node->agg_id = agg_id; 2168 } 2169 2170 /* move VSI to corresponding aggregator node */ 2171 status = ice_move_vsi_to_agg(port_info, agg_id, vsi->idx, 2172 (u8)vsi->tc_cfg.ena_tc); 2173 if (status) { 2174 dev_err(dev, "unable to move VSI idx %u into aggregator %u node", 2175 vsi->idx, agg_id); 2176 return; 2177 } 2178 2179 /* keep active children count for aggregator node */ 2180 agg_node->num_vsis++; 2181 2182 /* cache the 'agg_id' in VSI, so that after reset - VSI will be moved 2183 * to aggregator node 2184 */ 2185 vsi->agg_node = agg_node; 2186 dev_dbg(dev, "successfully moved VSI idx %u tc_bitmap 0x%x) into aggregator node %d which has num_vsis %u\n", 2187 vsi->idx, vsi->tc_cfg.ena_tc, vsi->agg_node->agg_id, 2188 vsi->agg_node->num_vsis); 2189 } 2190 2191 static int ice_vsi_cfg_tc_lan(struct ice_pf *pf, struct ice_vsi *vsi) 2192 { 2193 u16 max_txqs[ICE_MAX_TRAFFIC_CLASS] = { 0 }; 2194 struct device *dev = ice_pf_to_dev(pf); 2195 int ret, i; 2196 2197 /* configure VSI nodes based on number of queues and TC's */ 2198 ice_for_each_traffic_class(i) { 2199 if (!(vsi->tc_cfg.ena_tc & BIT(i))) 2200 continue; 2201 2202 if (vsi->type == ICE_VSI_CHNL) { 2203 if (!vsi->alloc_txq && vsi->num_txq) 2204 max_txqs[i] = vsi->num_txq; 2205 else 2206 max_txqs[i] = pf->num_lan_tx; 2207 } else { 2208 max_txqs[i] = vsi->alloc_txq; 2209 } 2210 2211 if (vsi->type == ICE_VSI_PF) 2212 max_txqs[i] += vsi->num_xdp_txq; 2213 } 2214 2215 dev_dbg(dev, "vsi->tc_cfg.ena_tc = %d\n", vsi->tc_cfg.ena_tc); 2216 ret = ice_cfg_vsi_lan(vsi->port_info, vsi->idx, vsi->tc_cfg.ena_tc, 2217 max_txqs); 2218 if (ret) { 2219 dev_err(dev, "VSI %d failed lan queue config, error %d\n", 2220 vsi->vsi_num, ret); 2221 return ret; 2222 } 2223 2224 return 0; 2225 } 2226 2227 /** 2228 * ice_vsi_cfg_def - configure default VSI based on the type 2229 * @vsi: pointer to VSI 2230 */ 2231 static int ice_vsi_cfg_def(struct ice_vsi *vsi) 2232 { 2233 struct device *dev = ice_pf_to_dev(vsi->back); 2234 struct ice_pf *pf = vsi->back; 2235 int ret; 2236 2237 vsi->vsw = pf->first_sw; 2238 2239 ret = ice_vsi_alloc_def(vsi, vsi->ch); 2240 if (ret) 2241 return ret; 2242 2243 /* allocate memory for Tx/Rx ring stat pointers */ 2244 ret = ice_vsi_alloc_stat_arrays(vsi); 2245 if (ret) 2246 goto unroll_vsi_alloc; 2247 2248 ice_alloc_fd_res(vsi); 2249 2250 ret = ice_vsi_get_qs(vsi); 2251 if (ret) { 2252 dev_err(dev, "Failed to allocate queues. vsi->idx = %d\n", 2253 vsi->idx); 2254 goto unroll_vsi_alloc_stat; 2255 } 2256 2257 /* set RSS capabilities */ 2258 ice_vsi_set_rss_params(vsi); 2259 2260 /* set TC configuration */ 2261 ice_vsi_set_tc_cfg(vsi); 2262 2263 /* create the VSI */ 2264 ret = ice_vsi_init(vsi, vsi->flags); 2265 if (ret) 2266 goto unroll_get_qs; 2267 2268 ice_vsi_init_vlan_ops(vsi); 2269 2270 switch (vsi->type) { 2271 case ICE_VSI_CTRL: 2272 case ICE_VSI_PF: 2273 ret = ice_vsi_alloc_q_vectors(vsi); 2274 if (ret) 2275 goto unroll_vsi_init; 2276 2277 ret = ice_vsi_alloc_rings(vsi); 2278 if (ret) 2279 goto unroll_vector_base; 2280 2281 ret = ice_vsi_alloc_ring_stats(vsi); 2282 if (ret) 2283 goto unroll_vector_base; 2284 2285 ice_vsi_map_rings_to_vectors(vsi); 2286 2287 /* Associate q_vector rings to napi */ 2288 ice_vsi_set_napi_queues(vsi); 2289 2290 vsi->stat_offsets_loaded = false; 2291 2292 if (ice_is_xdp_ena_vsi(vsi)) { 2293 ret = ice_vsi_determine_xdp_res(vsi); 2294 if (ret) 2295 goto unroll_vector_base; 2296 ret = ice_prepare_xdp_rings(vsi, vsi->xdp_prog); 2297 if (ret) 2298 goto unroll_vector_base; 2299 } 2300 2301 /* ICE_VSI_CTRL does not need RSS so skip RSS processing */ 2302 if (vsi->type != ICE_VSI_CTRL) 2303 /* Do not exit if configuring RSS had an issue, at 2304 * least receive traffic on first queue. Hence no 2305 * need to capture return value 2306 */ 2307 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 2308 ice_vsi_cfg_rss_lut_key(vsi); 2309 ice_vsi_set_rss_flow_fld(vsi); 2310 } 2311 ice_init_arfs(vsi); 2312 break; 2313 case ICE_VSI_CHNL: 2314 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 2315 ice_vsi_cfg_rss_lut_key(vsi); 2316 ice_vsi_set_rss_flow_fld(vsi); 2317 } 2318 break; 2319 case ICE_VSI_VF: 2320 /* VF driver will take care of creating netdev for this type and 2321 * map queues to vectors through Virtchnl, PF driver only 2322 * creates a VSI and corresponding structures for bookkeeping 2323 * purpose 2324 */ 2325 ret = ice_vsi_alloc_q_vectors(vsi); 2326 if (ret) 2327 goto unroll_vsi_init; 2328 2329 ret = ice_vsi_alloc_rings(vsi); 2330 if (ret) 2331 goto unroll_alloc_q_vector; 2332 2333 ret = ice_vsi_alloc_ring_stats(vsi); 2334 if (ret) 2335 goto unroll_vector_base; 2336 2337 vsi->stat_offsets_loaded = false; 2338 2339 /* Do not exit if configuring RSS had an issue, at least 2340 * receive traffic on first queue. Hence no need to capture 2341 * return value 2342 */ 2343 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) { 2344 ice_vsi_cfg_rss_lut_key(vsi); 2345 ice_vsi_set_vf_rss_flow_fld(vsi); 2346 } 2347 break; 2348 case ICE_VSI_LB: 2349 ret = ice_vsi_alloc_rings(vsi); 2350 if (ret) 2351 goto unroll_vsi_init; 2352 2353 ret = ice_vsi_alloc_ring_stats(vsi); 2354 if (ret) 2355 goto unroll_vector_base; 2356 2357 break; 2358 default: 2359 /* clean up the resources and exit */ 2360 ret = -EINVAL; 2361 goto unroll_vsi_init; 2362 } 2363 2364 return 0; 2365 2366 unroll_vector_base: 2367 /* reclaim SW interrupts back to the common pool */ 2368 unroll_alloc_q_vector: 2369 ice_vsi_free_q_vectors(vsi); 2370 unroll_vsi_init: 2371 ice_vsi_delete_from_hw(vsi); 2372 unroll_get_qs: 2373 ice_vsi_put_qs(vsi); 2374 unroll_vsi_alloc_stat: 2375 ice_vsi_free_stats(vsi); 2376 unroll_vsi_alloc: 2377 ice_vsi_free_arrays(vsi); 2378 return ret; 2379 } 2380 2381 /** 2382 * ice_vsi_cfg - configure a previously allocated VSI 2383 * @vsi: pointer to VSI 2384 */ 2385 int ice_vsi_cfg(struct ice_vsi *vsi) 2386 { 2387 struct ice_pf *pf = vsi->back; 2388 int ret; 2389 2390 if (WARN_ON(vsi->type == ICE_VSI_VF && !vsi->vf)) 2391 return -EINVAL; 2392 2393 ret = ice_vsi_cfg_def(vsi); 2394 if (ret) 2395 return ret; 2396 2397 ret = ice_vsi_cfg_tc_lan(vsi->back, vsi); 2398 if (ret) 2399 ice_vsi_decfg(vsi); 2400 2401 if (vsi->type == ICE_VSI_CTRL) { 2402 if (vsi->vf) { 2403 WARN_ON(vsi->vf->ctrl_vsi_idx != ICE_NO_VSI); 2404 vsi->vf->ctrl_vsi_idx = vsi->idx; 2405 } else { 2406 WARN_ON(pf->ctrl_vsi_idx != ICE_NO_VSI); 2407 pf->ctrl_vsi_idx = vsi->idx; 2408 } 2409 } 2410 2411 return ret; 2412 } 2413 2414 /** 2415 * ice_vsi_decfg - remove all VSI configuration 2416 * @vsi: pointer to VSI 2417 */ 2418 void ice_vsi_decfg(struct ice_vsi *vsi) 2419 { 2420 struct ice_pf *pf = vsi->back; 2421 int err; 2422 2423 /* The Rx rule will only exist to remove if the LLDP FW 2424 * engine is currently stopped 2425 */ 2426 if (!ice_is_safe_mode(pf) && vsi->type == ICE_VSI_PF && 2427 !test_bit(ICE_FLAG_FW_LLDP_AGENT, pf->flags)) 2428 ice_cfg_sw_lldp(vsi, false, false); 2429 2430 ice_rm_vsi_lan_cfg(vsi->port_info, vsi->idx); 2431 err = ice_rm_vsi_rdma_cfg(vsi->port_info, vsi->idx); 2432 if (err) 2433 dev_err(ice_pf_to_dev(pf), "Failed to remove RDMA scheduler config for VSI %u, err %d\n", 2434 vsi->vsi_num, err); 2435 2436 if (ice_is_xdp_ena_vsi(vsi)) 2437 /* return value check can be skipped here, it always returns 2438 * 0 if reset is in progress 2439 */ 2440 ice_destroy_xdp_rings(vsi); 2441 2442 ice_vsi_clear_rings(vsi); 2443 ice_vsi_free_q_vectors(vsi); 2444 ice_vsi_put_qs(vsi); 2445 ice_vsi_free_arrays(vsi); 2446 2447 /* SR-IOV determines needed MSIX resources all at once instead of per 2448 * VSI since when VFs are spawned we know how many VFs there are and how 2449 * many interrupts each VF needs. SR-IOV MSIX resources are also 2450 * cleared in the same manner. 2451 */ 2452 2453 if (vsi->type == ICE_VSI_VF && 2454 vsi->agg_node && vsi->agg_node->valid) 2455 vsi->agg_node->num_vsis--; 2456 } 2457 2458 /** 2459 * ice_vsi_setup - Set up a VSI by a given type 2460 * @pf: board private structure 2461 * @params: parameters to use when creating the VSI 2462 * 2463 * This allocates the sw VSI structure and its queue resources. 2464 * 2465 * Returns pointer to the successfully allocated and configured VSI sw struct on 2466 * success, NULL on failure. 2467 */ 2468 struct ice_vsi * 2469 ice_vsi_setup(struct ice_pf *pf, struct ice_vsi_cfg_params *params) 2470 { 2471 struct device *dev = ice_pf_to_dev(pf); 2472 struct ice_vsi *vsi; 2473 int ret; 2474 2475 /* ice_vsi_setup can only initialize a new VSI, and we must have 2476 * a port_info structure for it. 2477 */ 2478 if (WARN_ON(!(params->flags & ICE_VSI_FLAG_INIT)) || 2479 WARN_ON(!params->port_info)) 2480 return NULL; 2481 2482 vsi = ice_vsi_alloc(pf); 2483 if (!vsi) { 2484 dev_err(dev, "could not allocate VSI\n"); 2485 return NULL; 2486 } 2487 2488 vsi->params = *params; 2489 ret = ice_vsi_cfg(vsi); 2490 if (ret) 2491 goto err_vsi_cfg; 2492 2493 /* Add switch rule to drop all Tx Flow Control Frames, of look up 2494 * type ETHERTYPE from VSIs, and restrict malicious VF from sending 2495 * out PAUSE or PFC frames. If enabled, FW can still send FC frames. 2496 * The rule is added once for PF VSI in order to create appropriate 2497 * recipe, since VSI/VSI list is ignored with drop action... 2498 * Also add rules to handle LLDP Tx packets. Tx LLDP packets need to 2499 * be dropped so that VFs cannot send LLDP packets to reconfig DCB 2500 * settings in the HW. 2501 */ 2502 if (!ice_is_safe_mode(pf) && vsi->type == ICE_VSI_PF) { 2503 ice_fltr_add_eth(vsi, ETH_P_PAUSE, ICE_FLTR_TX, 2504 ICE_DROP_PACKET); 2505 ice_cfg_sw_lldp(vsi, true, true); 2506 } 2507 2508 if (!vsi->agg_node) 2509 ice_set_agg_vsi(vsi); 2510 2511 return vsi; 2512 2513 err_vsi_cfg: 2514 ice_vsi_free(vsi); 2515 2516 return NULL; 2517 } 2518 2519 /** 2520 * ice_vsi_release_msix - Clear the queue to Interrupt mapping in HW 2521 * @vsi: the VSI being cleaned up 2522 */ 2523 static void ice_vsi_release_msix(struct ice_vsi *vsi) 2524 { 2525 struct ice_pf *pf = vsi->back; 2526 struct ice_hw *hw = &pf->hw; 2527 u32 txq = 0; 2528 u32 rxq = 0; 2529 int i, q; 2530 2531 ice_for_each_q_vector(vsi, i) { 2532 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 2533 2534 ice_write_intrl(q_vector, 0); 2535 for (q = 0; q < q_vector->num_ring_tx; q++) { 2536 ice_write_itr(&q_vector->tx, 0); 2537 wr32(hw, QINT_TQCTL(vsi->txq_map[txq]), 0); 2538 if (ice_is_xdp_ena_vsi(vsi)) { 2539 u32 xdp_txq = txq + vsi->num_xdp_txq; 2540 2541 wr32(hw, QINT_TQCTL(vsi->txq_map[xdp_txq]), 0); 2542 } 2543 txq++; 2544 } 2545 2546 for (q = 0; q < q_vector->num_ring_rx; q++) { 2547 ice_write_itr(&q_vector->rx, 0); 2548 wr32(hw, QINT_RQCTL(vsi->rxq_map[rxq]), 0); 2549 rxq++; 2550 } 2551 } 2552 2553 ice_flush(hw); 2554 } 2555 2556 /** 2557 * ice_vsi_free_irq - Free the IRQ association with the OS 2558 * @vsi: the VSI being configured 2559 */ 2560 void ice_vsi_free_irq(struct ice_vsi *vsi) 2561 { 2562 struct ice_pf *pf = vsi->back; 2563 int i; 2564 2565 if (!vsi->q_vectors || !vsi->irqs_ready) 2566 return; 2567 2568 ice_vsi_release_msix(vsi); 2569 if (vsi->type == ICE_VSI_VF) 2570 return; 2571 2572 vsi->irqs_ready = false; 2573 ice_free_cpu_rx_rmap(vsi); 2574 2575 ice_for_each_q_vector(vsi, i) { 2576 int irq_num; 2577 2578 irq_num = vsi->q_vectors[i]->irq.virq; 2579 2580 /* free only the irqs that were actually requested */ 2581 if (!vsi->q_vectors[i] || 2582 !(vsi->q_vectors[i]->num_ring_tx || 2583 vsi->q_vectors[i]->num_ring_rx)) 2584 continue; 2585 2586 /* clear the affinity notifier in the IRQ descriptor */ 2587 if (!IS_ENABLED(CONFIG_RFS_ACCEL)) 2588 irq_set_affinity_notifier(irq_num, NULL); 2589 2590 /* clear the affinity_mask in the IRQ descriptor */ 2591 irq_set_affinity_hint(irq_num, NULL); 2592 synchronize_irq(irq_num); 2593 devm_free_irq(ice_pf_to_dev(pf), irq_num, vsi->q_vectors[i]); 2594 } 2595 } 2596 2597 /** 2598 * ice_vsi_free_tx_rings - Free Tx resources for VSI queues 2599 * @vsi: the VSI having resources freed 2600 */ 2601 void ice_vsi_free_tx_rings(struct ice_vsi *vsi) 2602 { 2603 int i; 2604 2605 if (!vsi->tx_rings) 2606 return; 2607 2608 ice_for_each_txq(vsi, i) 2609 if (vsi->tx_rings[i] && vsi->tx_rings[i]->desc) 2610 ice_free_tx_ring(vsi->tx_rings[i]); 2611 } 2612 2613 /** 2614 * ice_vsi_free_rx_rings - Free Rx resources for VSI queues 2615 * @vsi: the VSI having resources freed 2616 */ 2617 void ice_vsi_free_rx_rings(struct ice_vsi *vsi) 2618 { 2619 int i; 2620 2621 if (!vsi->rx_rings) 2622 return; 2623 2624 ice_for_each_rxq(vsi, i) 2625 if (vsi->rx_rings[i] && vsi->rx_rings[i]->desc) 2626 ice_free_rx_ring(vsi->rx_rings[i]); 2627 } 2628 2629 /** 2630 * ice_vsi_close - Shut down a VSI 2631 * @vsi: the VSI being shut down 2632 */ 2633 void ice_vsi_close(struct ice_vsi *vsi) 2634 { 2635 if (!test_and_set_bit(ICE_VSI_DOWN, vsi->state)) 2636 ice_down(vsi); 2637 2638 ice_vsi_free_irq(vsi); 2639 ice_vsi_free_tx_rings(vsi); 2640 ice_vsi_free_rx_rings(vsi); 2641 } 2642 2643 /** 2644 * ice_ena_vsi - resume a VSI 2645 * @vsi: the VSI being resume 2646 * @locked: is the rtnl_lock already held 2647 */ 2648 int ice_ena_vsi(struct ice_vsi *vsi, bool locked) 2649 { 2650 int err = 0; 2651 2652 if (!test_bit(ICE_VSI_NEEDS_RESTART, vsi->state)) 2653 return 0; 2654 2655 clear_bit(ICE_VSI_NEEDS_RESTART, vsi->state); 2656 2657 if (vsi->netdev && vsi->type == ICE_VSI_PF) { 2658 if (netif_running(vsi->netdev)) { 2659 if (!locked) 2660 rtnl_lock(); 2661 2662 err = ice_open_internal(vsi->netdev); 2663 2664 if (!locked) 2665 rtnl_unlock(); 2666 } 2667 } else if (vsi->type == ICE_VSI_CTRL) { 2668 err = ice_vsi_open_ctrl(vsi); 2669 } 2670 2671 return err; 2672 } 2673 2674 /** 2675 * ice_dis_vsi - pause a VSI 2676 * @vsi: the VSI being paused 2677 * @locked: is the rtnl_lock already held 2678 */ 2679 void ice_dis_vsi(struct ice_vsi *vsi, bool locked) 2680 { 2681 if (test_bit(ICE_VSI_DOWN, vsi->state)) 2682 return; 2683 2684 set_bit(ICE_VSI_NEEDS_RESTART, vsi->state); 2685 2686 if (vsi->type == ICE_VSI_PF && vsi->netdev) { 2687 if (netif_running(vsi->netdev)) { 2688 if (!locked) 2689 rtnl_lock(); 2690 2691 ice_vsi_close(vsi); 2692 2693 if (!locked) 2694 rtnl_unlock(); 2695 } else { 2696 ice_vsi_close(vsi); 2697 } 2698 } else if (vsi->type == ICE_VSI_CTRL) { 2699 ice_vsi_close(vsi); 2700 } 2701 } 2702 2703 /** 2704 * __ice_queue_set_napi - Set the napi instance for the queue 2705 * @dev: device to which NAPI and queue belong 2706 * @queue_index: Index of queue 2707 * @type: queue type as RX or TX 2708 * @napi: NAPI context 2709 * @locked: is the rtnl_lock already held 2710 * 2711 * Set the napi instance for the queue. Caller indicates the lock status. 2712 */ 2713 static void 2714 __ice_queue_set_napi(struct net_device *dev, unsigned int queue_index, 2715 enum netdev_queue_type type, struct napi_struct *napi, 2716 bool locked) 2717 { 2718 if (!locked) 2719 rtnl_lock(); 2720 netif_queue_set_napi(dev, queue_index, type, napi); 2721 if (!locked) 2722 rtnl_unlock(); 2723 } 2724 2725 /** 2726 * ice_queue_set_napi - Set the napi instance for the queue 2727 * @vsi: VSI being configured 2728 * @queue_index: Index of queue 2729 * @type: queue type as RX or TX 2730 * @napi: NAPI context 2731 * 2732 * Set the napi instance for the queue. The rtnl lock state is derived from the 2733 * execution path. 2734 */ 2735 void 2736 ice_queue_set_napi(struct ice_vsi *vsi, unsigned int queue_index, 2737 enum netdev_queue_type type, struct napi_struct *napi) 2738 { 2739 struct ice_pf *pf = vsi->back; 2740 2741 if (!vsi->netdev) 2742 return; 2743 2744 if (current_work() == &pf->serv_task || 2745 test_bit(ICE_PREPARED_FOR_RESET, pf->state) || 2746 test_bit(ICE_DOWN, pf->state) || 2747 test_bit(ICE_SUSPENDED, pf->state)) 2748 __ice_queue_set_napi(vsi->netdev, queue_index, type, napi, 2749 false); 2750 else 2751 __ice_queue_set_napi(vsi->netdev, queue_index, type, napi, 2752 true); 2753 } 2754 2755 /** 2756 * __ice_q_vector_set_napi_queues - Map queue[s] associated with the napi 2757 * @q_vector: q_vector pointer 2758 * @locked: is the rtnl_lock already held 2759 * 2760 * Associate the q_vector napi with all the queue[s] on the vector. 2761 * Caller indicates the lock status. 2762 */ 2763 void __ice_q_vector_set_napi_queues(struct ice_q_vector *q_vector, bool locked) 2764 { 2765 struct ice_rx_ring *rx_ring; 2766 struct ice_tx_ring *tx_ring; 2767 2768 ice_for_each_rx_ring(rx_ring, q_vector->rx) 2769 __ice_queue_set_napi(q_vector->vsi->netdev, rx_ring->q_index, 2770 NETDEV_QUEUE_TYPE_RX, &q_vector->napi, 2771 locked); 2772 2773 ice_for_each_tx_ring(tx_ring, q_vector->tx) 2774 __ice_queue_set_napi(q_vector->vsi->netdev, tx_ring->q_index, 2775 NETDEV_QUEUE_TYPE_TX, &q_vector->napi, 2776 locked); 2777 /* Also set the interrupt number for the NAPI */ 2778 netif_napi_set_irq(&q_vector->napi, q_vector->irq.virq); 2779 } 2780 2781 /** 2782 * ice_q_vector_set_napi_queues - Map queue[s] associated with the napi 2783 * @q_vector: q_vector pointer 2784 * 2785 * Associate the q_vector napi with all the queue[s] on the vector 2786 */ 2787 void ice_q_vector_set_napi_queues(struct ice_q_vector *q_vector) 2788 { 2789 struct ice_rx_ring *rx_ring; 2790 struct ice_tx_ring *tx_ring; 2791 2792 ice_for_each_rx_ring(rx_ring, q_vector->rx) 2793 ice_queue_set_napi(q_vector->vsi, rx_ring->q_index, 2794 NETDEV_QUEUE_TYPE_RX, &q_vector->napi); 2795 2796 ice_for_each_tx_ring(tx_ring, q_vector->tx) 2797 ice_queue_set_napi(q_vector->vsi, tx_ring->q_index, 2798 NETDEV_QUEUE_TYPE_TX, &q_vector->napi); 2799 /* Also set the interrupt number for the NAPI */ 2800 netif_napi_set_irq(&q_vector->napi, q_vector->irq.virq); 2801 } 2802 2803 /** 2804 * ice_vsi_set_napi_queues 2805 * @vsi: VSI pointer 2806 * 2807 * Associate queue[s] with napi for all vectors 2808 */ 2809 void ice_vsi_set_napi_queues(struct ice_vsi *vsi) 2810 { 2811 int i; 2812 2813 if (!vsi->netdev) 2814 return; 2815 2816 ice_for_each_q_vector(vsi, i) 2817 ice_q_vector_set_napi_queues(vsi->q_vectors[i]); 2818 } 2819 2820 /** 2821 * ice_vsi_release - Delete a VSI and free its resources 2822 * @vsi: the VSI being removed 2823 * 2824 * Returns 0 on success or < 0 on error 2825 */ 2826 int ice_vsi_release(struct ice_vsi *vsi) 2827 { 2828 struct ice_pf *pf; 2829 2830 if (!vsi->back) 2831 return -ENODEV; 2832 pf = vsi->back; 2833 2834 if (test_bit(ICE_FLAG_RSS_ENA, pf->flags)) 2835 ice_rss_clean(vsi); 2836 2837 ice_vsi_close(vsi); 2838 ice_vsi_decfg(vsi); 2839 2840 /* retain SW VSI data structure since it is needed to unregister and 2841 * free VSI netdev when PF is not in reset recovery pending state,\ 2842 * for ex: during rmmod. 2843 */ 2844 if (!ice_is_reset_in_progress(pf->state)) 2845 ice_vsi_delete(vsi); 2846 2847 return 0; 2848 } 2849 2850 /** 2851 * ice_vsi_rebuild_get_coalesce - get coalesce from all q_vectors 2852 * @vsi: VSI connected with q_vectors 2853 * @coalesce: array of struct with stored coalesce 2854 * 2855 * Returns array size. 2856 */ 2857 static int 2858 ice_vsi_rebuild_get_coalesce(struct ice_vsi *vsi, 2859 struct ice_coalesce_stored *coalesce) 2860 { 2861 int i; 2862 2863 ice_for_each_q_vector(vsi, i) { 2864 struct ice_q_vector *q_vector = vsi->q_vectors[i]; 2865 2866 coalesce[i].itr_tx = q_vector->tx.itr_settings; 2867 coalesce[i].itr_rx = q_vector->rx.itr_settings; 2868 coalesce[i].intrl = q_vector->intrl; 2869 2870 if (i < vsi->num_txq) 2871 coalesce[i].tx_valid = true; 2872 if (i < vsi->num_rxq) 2873 coalesce[i].rx_valid = true; 2874 } 2875 2876 return vsi->num_q_vectors; 2877 } 2878 2879 /** 2880 * ice_vsi_rebuild_set_coalesce - set coalesce from earlier saved arrays 2881 * @vsi: VSI connected with q_vectors 2882 * @coalesce: pointer to array of struct with stored coalesce 2883 * @size: size of coalesce array 2884 * 2885 * Before this function, ice_vsi_rebuild_get_coalesce should be called to save 2886 * ITR params in arrays. If size is 0 or coalesce wasn't stored set coalesce 2887 * to default value. 2888 */ 2889 static void 2890 ice_vsi_rebuild_set_coalesce(struct ice_vsi *vsi, 2891 struct ice_coalesce_stored *coalesce, int size) 2892 { 2893 struct ice_ring_container *rc; 2894 int i; 2895 2896 if ((size && !coalesce) || !vsi) 2897 return; 2898 2899 /* There are a couple of cases that have to be handled here: 2900 * 1. The case where the number of queue vectors stays the same, but 2901 * the number of Tx or Rx rings changes (the first for loop) 2902 * 2. The case where the number of queue vectors increased (the 2903 * second for loop) 2904 */ 2905 for (i = 0; i < size && i < vsi->num_q_vectors; i++) { 2906 /* There are 2 cases to handle here and they are the same for 2907 * both Tx and Rx: 2908 * if the entry was valid previously (coalesce[i].[tr]x_valid 2909 * and the loop variable is less than the number of rings 2910 * allocated, then write the previous values 2911 * 2912 * if the entry was not valid previously, but the number of 2913 * rings is less than are allocated (this means the number of 2914 * rings increased from previously), then write out the 2915 * values in the first element 2916 * 2917 * Also, always write the ITR, even if in ITR_IS_DYNAMIC 2918 * as there is no harm because the dynamic algorithm 2919 * will just overwrite. 2920 */ 2921 if (i < vsi->alloc_rxq && coalesce[i].rx_valid) { 2922 rc = &vsi->q_vectors[i]->rx; 2923 rc->itr_settings = coalesce[i].itr_rx; 2924 ice_write_itr(rc, rc->itr_setting); 2925 } else if (i < vsi->alloc_rxq) { 2926 rc = &vsi->q_vectors[i]->rx; 2927 rc->itr_settings = coalesce[0].itr_rx; 2928 ice_write_itr(rc, rc->itr_setting); 2929 } 2930 2931 if (i < vsi->alloc_txq && coalesce[i].tx_valid) { 2932 rc = &vsi->q_vectors[i]->tx; 2933 rc->itr_settings = coalesce[i].itr_tx; 2934 ice_write_itr(rc, rc->itr_setting); 2935 } else if (i < vsi->alloc_txq) { 2936 rc = &vsi->q_vectors[i]->tx; 2937 rc->itr_settings = coalesce[0].itr_tx; 2938 ice_write_itr(rc, rc->itr_setting); 2939 } 2940 2941 vsi->q_vectors[i]->intrl = coalesce[i].intrl; 2942 ice_set_q_vector_intrl(vsi->q_vectors[i]); 2943 } 2944 2945 /* the number of queue vectors increased so write whatever is in 2946 * the first element 2947 */ 2948 for (; i < vsi->num_q_vectors; i++) { 2949 /* transmit */ 2950 rc = &vsi->q_vectors[i]->tx; 2951 rc->itr_settings = coalesce[0].itr_tx; 2952 ice_write_itr(rc, rc->itr_setting); 2953 2954 /* receive */ 2955 rc = &vsi->q_vectors[i]->rx; 2956 rc->itr_settings = coalesce[0].itr_rx; 2957 ice_write_itr(rc, rc->itr_setting); 2958 2959 vsi->q_vectors[i]->intrl = coalesce[0].intrl; 2960 ice_set_q_vector_intrl(vsi->q_vectors[i]); 2961 } 2962 } 2963 2964 /** 2965 * ice_vsi_realloc_stat_arrays - Frees unused stat structures or alloc new ones 2966 * @vsi: VSI pointer 2967 */ 2968 static int 2969 ice_vsi_realloc_stat_arrays(struct ice_vsi *vsi) 2970 { 2971 u16 req_txq = vsi->req_txq ? vsi->req_txq : vsi->alloc_txq; 2972 u16 req_rxq = vsi->req_rxq ? vsi->req_rxq : vsi->alloc_rxq; 2973 struct ice_ring_stats **tx_ring_stats; 2974 struct ice_ring_stats **rx_ring_stats; 2975 struct ice_vsi_stats *vsi_stat; 2976 struct ice_pf *pf = vsi->back; 2977 u16 prev_txq = vsi->alloc_txq; 2978 u16 prev_rxq = vsi->alloc_rxq; 2979 int i; 2980 2981 vsi_stat = pf->vsi_stats[vsi->idx]; 2982 2983 if (req_txq < prev_txq) { 2984 for (i = req_txq; i < prev_txq; i++) { 2985 if (vsi_stat->tx_ring_stats[i]) { 2986 kfree_rcu(vsi_stat->tx_ring_stats[i], rcu); 2987 WRITE_ONCE(vsi_stat->tx_ring_stats[i], NULL); 2988 } 2989 } 2990 } 2991 2992 tx_ring_stats = vsi_stat->tx_ring_stats; 2993 vsi_stat->tx_ring_stats = 2994 krealloc_array(vsi_stat->tx_ring_stats, req_txq, 2995 sizeof(*vsi_stat->tx_ring_stats), 2996 GFP_KERNEL | __GFP_ZERO); 2997 if (!vsi_stat->tx_ring_stats) { 2998 vsi_stat->tx_ring_stats = tx_ring_stats; 2999 return -ENOMEM; 3000 } 3001 3002 if (req_rxq < prev_rxq) { 3003 for (i = req_rxq; i < prev_rxq; i++) { 3004 if (vsi_stat->rx_ring_stats[i]) { 3005 kfree_rcu(vsi_stat->rx_ring_stats[i], rcu); 3006 WRITE_ONCE(vsi_stat->rx_ring_stats[i], NULL); 3007 } 3008 } 3009 } 3010 3011 rx_ring_stats = vsi_stat->rx_ring_stats; 3012 vsi_stat->rx_ring_stats = 3013 krealloc_array(vsi_stat->rx_ring_stats, req_rxq, 3014 sizeof(*vsi_stat->rx_ring_stats), 3015 GFP_KERNEL | __GFP_ZERO); 3016 if (!vsi_stat->rx_ring_stats) { 3017 vsi_stat->rx_ring_stats = rx_ring_stats; 3018 return -ENOMEM; 3019 } 3020 3021 return 0; 3022 } 3023 3024 /** 3025 * ice_vsi_rebuild - Rebuild VSI after reset 3026 * @vsi: VSI to be rebuild 3027 * @vsi_flags: flags used for VSI rebuild flow 3028 * 3029 * Set vsi_flags to ICE_VSI_FLAG_INIT to initialize a new VSI, or 3030 * ICE_VSI_FLAG_NO_INIT to rebuild an existing VSI in hardware. 3031 * 3032 * Returns 0 on success and negative value on failure 3033 */ 3034 int ice_vsi_rebuild(struct ice_vsi *vsi, u32 vsi_flags) 3035 { 3036 struct ice_coalesce_stored *coalesce; 3037 int prev_num_q_vectors; 3038 struct ice_pf *pf; 3039 int ret; 3040 3041 if (!vsi) 3042 return -EINVAL; 3043 3044 vsi->flags = vsi_flags; 3045 pf = vsi->back; 3046 if (WARN_ON(vsi->type == ICE_VSI_VF && !vsi->vf)) 3047 return -EINVAL; 3048 3049 ret = ice_vsi_realloc_stat_arrays(vsi); 3050 if (ret) 3051 goto err_vsi_cfg; 3052 3053 ice_vsi_decfg(vsi); 3054 ret = ice_vsi_cfg_def(vsi); 3055 if (ret) 3056 goto err_vsi_cfg; 3057 3058 coalesce = kcalloc(vsi->num_q_vectors, 3059 sizeof(struct ice_coalesce_stored), GFP_KERNEL); 3060 if (!coalesce) 3061 return -ENOMEM; 3062 3063 prev_num_q_vectors = ice_vsi_rebuild_get_coalesce(vsi, coalesce); 3064 3065 ret = ice_vsi_cfg_tc_lan(pf, vsi); 3066 if (ret) { 3067 if (vsi_flags & ICE_VSI_FLAG_INIT) { 3068 ret = -EIO; 3069 goto err_vsi_cfg_tc_lan; 3070 } 3071 3072 kfree(coalesce); 3073 return ice_schedule_reset(pf, ICE_RESET_PFR); 3074 } 3075 3076 ice_vsi_rebuild_set_coalesce(vsi, coalesce, prev_num_q_vectors); 3077 kfree(coalesce); 3078 3079 return 0; 3080 3081 err_vsi_cfg_tc_lan: 3082 ice_vsi_decfg(vsi); 3083 kfree(coalesce); 3084 err_vsi_cfg: 3085 return ret; 3086 } 3087 3088 /** 3089 * ice_is_reset_in_progress - check for a reset in progress 3090 * @state: PF state field 3091 */ 3092 bool ice_is_reset_in_progress(unsigned long *state) 3093 { 3094 return test_bit(ICE_RESET_OICR_RECV, state) || 3095 test_bit(ICE_PFR_REQ, state) || 3096 test_bit(ICE_CORER_REQ, state) || 3097 test_bit(ICE_GLOBR_REQ, state); 3098 } 3099 3100 /** 3101 * ice_wait_for_reset - Wait for driver to finish reset and rebuild 3102 * @pf: pointer to the PF structure 3103 * @timeout: length of time to wait, in jiffies 3104 * 3105 * Wait (sleep) for a short time until the driver finishes cleaning up from 3106 * a device reset. The caller must be able to sleep. Use this to delay 3107 * operations that could fail while the driver is cleaning up after a device 3108 * reset. 3109 * 3110 * Returns 0 on success, -EBUSY if the reset is not finished within the 3111 * timeout, and -ERESTARTSYS if the thread was interrupted. 3112 */ 3113 int ice_wait_for_reset(struct ice_pf *pf, unsigned long timeout) 3114 { 3115 long ret; 3116 3117 ret = wait_event_interruptible_timeout(pf->reset_wait_queue, 3118 !ice_is_reset_in_progress(pf->state), 3119 timeout); 3120 if (ret < 0) 3121 return ret; 3122 else if (!ret) 3123 return -EBUSY; 3124 else 3125 return 0; 3126 } 3127 3128 /** 3129 * ice_vsi_update_q_map - update our copy of the VSI info with new queue map 3130 * @vsi: VSI being configured 3131 * @ctx: the context buffer returned from AQ VSI update command 3132 */ 3133 static void ice_vsi_update_q_map(struct ice_vsi *vsi, struct ice_vsi_ctx *ctx) 3134 { 3135 vsi->info.mapping_flags = ctx->info.mapping_flags; 3136 memcpy(&vsi->info.q_mapping, &ctx->info.q_mapping, 3137 sizeof(vsi->info.q_mapping)); 3138 memcpy(&vsi->info.tc_mapping, ctx->info.tc_mapping, 3139 sizeof(vsi->info.tc_mapping)); 3140 } 3141 3142 /** 3143 * ice_vsi_cfg_netdev_tc - Setup the netdev TC configuration 3144 * @vsi: the VSI being configured 3145 * @ena_tc: TC map to be enabled 3146 */ 3147 void ice_vsi_cfg_netdev_tc(struct ice_vsi *vsi, u8 ena_tc) 3148 { 3149 struct net_device *netdev = vsi->netdev; 3150 struct ice_pf *pf = vsi->back; 3151 int numtc = vsi->tc_cfg.numtc; 3152 struct ice_dcbx_cfg *dcbcfg; 3153 u8 netdev_tc; 3154 int i; 3155 3156 if (!netdev) 3157 return; 3158 3159 /* CHNL VSI doesn't have it's own netdev, hence, no netdev_tc */ 3160 if (vsi->type == ICE_VSI_CHNL) 3161 return; 3162 3163 if (!ena_tc) { 3164 netdev_reset_tc(netdev); 3165 return; 3166 } 3167 3168 if (vsi->type == ICE_VSI_PF && ice_is_adq_active(pf)) 3169 numtc = vsi->all_numtc; 3170 3171 if (netdev_set_num_tc(netdev, numtc)) 3172 return; 3173 3174 dcbcfg = &pf->hw.port_info->qos_cfg.local_dcbx_cfg; 3175 3176 ice_for_each_traffic_class(i) 3177 if (vsi->tc_cfg.ena_tc & BIT(i)) 3178 netdev_set_tc_queue(netdev, 3179 vsi->tc_cfg.tc_info[i].netdev_tc, 3180 vsi->tc_cfg.tc_info[i].qcount_tx, 3181 vsi->tc_cfg.tc_info[i].qoffset); 3182 /* setup TC queue map for CHNL TCs */ 3183 ice_for_each_chnl_tc(i) { 3184 if (!(vsi->all_enatc & BIT(i))) 3185 break; 3186 if (!vsi->mqprio_qopt.qopt.count[i]) 3187 break; 3188 netdev_set_tc_queue(netdev, i, 3189 vsi->mqprio_qopt.qopt.count[i], 3190 vsi->mqprio_qopt.qopt.offset[i]); 3191 } 3192 3193 if (test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3194 return; 3195 3196 for (i = 0; i < ICE_MAX_USER_PRIORITY; i++) { 3197 u8 ets_tc = dcbcfg->etscfg.prio_table[i]; 3198 3199 /* Get the mapped netdev TC# for the UP */ 3200 netdev_tc = vsi->tc_cfg.tc_info[ets_tc].netdev_tc; 3201 netdev_set_prio_tc_map(netdev, i, netdev_tc); 3202 } 3203 } 3204 3205 /** 3206 * ice_vsi_setup_q_map_mqprio - Prepares mqprio based tc_config 3207 * @vsi: the VSI being configured, 3208 * @ctxt: VSI context structure 3209 * @ena_tc: number of traffic classes to enable 3210 * 3211 * Prepares VSI tc_config to have queue configurations based on MQPRIO options. 3212 */ 3213 static int 3214 ice_vsi_setup_q_map_mqprio(struct ice_vsi *vsi, struct ice_vsi_ctx *ctxt, 3215 u8 ena_tc) 3216 { 3217 u16 pow, offset = 0, qcount_tx = 0, qcount_rx = 0, qmap; 3218 u16 tc0_offset = vsi->mqprio_qopt.qopt.offset[0]; 3219 int tc0_qcount = vsi->mqprio_qopt.qopt.count[0]; 3220 u16 new_txq, new_rxq; 3221 u8 netdev_tc = 0; 3222 int i; 3223 3224 vsi->tc_cfg.ena_tc = ena_tc ? ena_tc : 1; 3225 3226 pow = order_base_2(tc0_qcount); 3227 qmap = FIELD_PREP(ICE_AQ_VSI_TC_Q_OFFSET_M, tc0_offset); 3228 qmap |= FIELD_PREP(ICE_AQ_VSI_TC_Q_NUM_M, pow); 3229 3230 ice_for_each_traffic_class(i) { 3231 if (!(vsi->tc_cfg.ena_tc & BIT(i))) { 3232 /* TC is not enabled */ 3233 vsi->tc_cfg.tc_info[i].qoffset = 0; 3234 vsi->tc_cfg.tc_info[i].qcount_rx = 1; 3235 vsi->tc_cfg.tc_info[i].qcount_tx = 1; 3236 vsi->tc_cfg.tc_info[i].netdev_tc = 0; 3237 ctxt->info.tc_mapping[i] = 0; 3238 continue; 3239 } 3240 3241 offset = vsi->mqprio_qopt.qopt.offset[i]; 3242 qcount_rx = vsi->mqprio_qopt.qopt.count[i]; 3243 qcount_tx = vsi->mqprio_qopt.qopt.count[i]; 3244 vsi->tc_cfg.tc_info[i].qoffset = offset; 3245 vsi->tc_cfg.tc_info[i].qcount_rx = qcount_rx; 3246 vsi->tc_cfg.tc_info[i].qcount_tx = qcount_tx; 3247 vsi->tc_cfg.tc_info[i].netdev_tc = netdev_tc++; 3248 } 3249 3250 if (vsi->all_numtc && vsi->all_numtc != vsi->tc_cfg.numtc) { 3251 ice_for_each_chnl_tc(i) { 3252 if (!(vsi->all_enatc & BIT(i))) 3253 continue; 3254 offset = vsi->mqprio_qopt.qopt.offset[i]; 3255 qcount_rx = vsi->mqprio_qopt.qopt.count[i]; 3256 qcount_tx = vsi->mqprio_qopt.qopt.count[i]; 3257 } 3258 } 3259 3260 new_txq = offset + qcount_tx; 3261 if (new_txq > vsi->alloc_txq) { 3262 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Tx queues (%u), than were allocated (%u)!\n", 3263 new_txq, vsi->alloc_txq); 3264 return -EINVAL; 3265 } 3266 3267 new_rxq = offset + qcount_rx; 3268 if (new_rxq > vsi->alloc_rxq) { 3269 dev_err(ice_pf_to_dev(vsi->back), "Trying to use more Rx queues (%u), than were allocated (%u)!\n", 3270 new_rxq, vsi->alloc_rxq); 3271 return -EINVAL; 3272 } 3273 3274 /* Set actual Tx/Rx queue pairs */ 3275 vsi->num_txq = new_txq; 3276 vsi->num_rxq = new_rxq; 3277 3278 /* Setup queue TC[0].qmap for given VSI context */ 3279 ctxt->info.tc_mapping[0] = cpu_to_le16(qmap); 3280 ctxt->info.q_mapping[0] = cpu_to_le16(vsi->rxq_map[0]); 3281 ctxt->info.q_mapping[1] = cpu_to_le16(tc0_qcount); 3282 3283 /* Find queue count available for channel VSIs and starting offset 3284 * for channel VSIs 3285 */ 3286 if (tc0_qcount && tc0_qcount < vsi->num_rxq) { 3287 vsi->cnt_q_avail = vsi->num_rxq - tc0_qcount; 3288 vsi->next_base_q = tc0_qcount; 3289 } 3290 dev_dbg(ice_pf_to_dev(vsi->back), "vsi->num_txq = %d\n", vsi->num_txq); 3291 dev_dbg(ice_pf_to_dev(vsi->back), "vsi->num_rxq = %d\n", vsi->num_rxq); 3292 dev_dbg(ice_pf_to_dev(vsi->back), "all_numtc %u, all_enatc: 0x%04x, tc_cfg.numtc %u\n", 3293 vsi->all_numtc, vsi->all_enatc, vsi->tc_cfg.numtc); 3294 3295 return 0; 3296 } 3297 3298 /** 3299 * ice_vsi_cfg_tc - Configure VSI Tx Sched for given TC map 3300 * @vsi: VSI to be configured 3301 * @ena_tc: TC bitmap 3302 * 3303 * VSI queues expected to be quiesced before calling this function 3304 */ 3305 int ice_vsi_cfg_tc(struct ice_vsi *vsi, u8 ena_tc) 3306 { 3307 u16 max_txqs[ICE_MAX_TRAFFIC_CLASS] = { 0 }; 3308 struct ice_pf *pf = vsi->back; 3309 struct ice_tc_cfg old_tc_cfg; 3310 struct ice_vsi_ctx *ctx; 3311 struct device *dev; 3312 int i, ret = 0; 3313 u8 num_tc = 0; 3314 3315 dev = ice_pf_to_dev(pf); 3316 if (vsi->tc_cfg.ena_tc == ena_tc && 3317 vsi->mqprio_qopt.mode != TC_MQPRIO_MODE_CHANNEL) 3318 return 0; 3319 3320 ice_for_each_traffic_class(i) { 3321 /* build bitmap of enabled TCs */ 3322 if (ena_tc & BIT(i)) 3323 num_tc++; 3324 /* populate max_txqs per TC */ 3325 max_txqs[i] = vsi->alloc_txq; 3326 /* Update max_txqs if it is CHNL VSI, because alloc_t[r]xq are 3327 * zero for CHNL VSI, hence use num_txq instead as max_txqs 3328 */ 3329 if (vsi->type == ICE_VSI_CHNL && 3330 test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3331 max_txqs[i] = vsi->num_txq; 3332 } 3333 3334 memcpy(&old_tc_cfg, &vsi->tc_cfg, sizeof(old_tc_cfg)); 3335 vsi->tc_cfg.ena_tc = ena_tc; 3336 vsi->tc_cfg.numtc = num_tc; 3337 3338 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); 3339 if (!ctx) 3340 return -ENOMEM; 3341 3342 ctx->vf_num = 0; 3343 ctx->info = vsi->info; 3344 3345 if (vsi->type == ICE_VSI_PF && 3346 test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3347 ret = ice_vsi_setup_q_map_mqprio(vsi, ctx, ena_tc); 3348 else 3349 ret = ice_vsi_setup_q_map(vsi, ctx); 3350 3351 if (ret) { 3352 memcpy(&vsi->tc_cfg, &old_tc_cfg, sizeof(vsi->tc_cfg)); 3353 goto out; 3354 } 3355 3356 /* must to indicate which section of VSI context are being modified */ 3357 ctx->info.valid_sections = cpu_to_le16(ICE_AQ_VSI_PROP_RXQ_MAP_VALID); 3358 ret = ice_update_vsi(&pf->hw, vsi->idx, ctx, NULL); 3359 if (ret) { 3360 dev_info(dev, "Failed VSI Update\n"); 3361 goto out; 3362 } 3363 3364 if (vsi->type == ICE_VSI_PF && 3365 test_bit(ICE_FLAG_TC_MQPRIO, pf->flags)) 3366 ret = ice_cfg_vsi_lan(vsi->port_info, vsi->idx, 1, max_txqs); 3367 else 3368 ret = ice_cfg_vsi_lan(vsi->port_info, vsi->idx, 3369 vsi->tc_cfg.ena_tc, max_txqs); 3370 3371 if (ret) { 3372 dev_err(dev, "VSI %d failed TC config, error %d\n", 3373 vsi->vsi_num, ret); 3374 goto out; 3375 } 3376 ice_vsi_update_q_map(vsi, ctx); 3377 vsi->info.valid_sections = 0; 3378 3379 ice_vsi_cfg_netdev_tc(vsi, ena_tc); 3380 out: 3381 kfree(ctx); 3382 return ret; 3383 } 3384 3385 /** 3386 * ice_update_ring_stats - Update ring statistics 3387 * @stats: stats to be updated 3388 * @pkts: number of processed packets 3389 * @bytes: number of processed bytes 3390 * 3391 * This function assumes that caller has acquired a u64_stats_sync lock. 3392 */ 3393 static void ice_update_ring_stats(struct ice_q_stats *stats, u64 pkts, u64 bytes) 3394 { 3395 stats->bytes += bytes; 3396 stats->pkts += pkts; 3397 } 3398 3399 /** 3400 * ice_update_tx_ring_stats - Update Tx ring specific counters 3401 * @tx_ring: ring to update 3402 * @pkts: number of processed packets 3403 * @bytes: number of processed bytes 3404 */ 3405 void ice_update_tx_ring_stats(struct ice_tx_ring *tx_ring, u64 pkts, u64 bytes) 3406 { 3407 u64_stats_update_begin(&tx_ring->ring_stats->syncp); 3408 ice_update_ring_stats(&tx_ring->ring_stats->stats, pkts, bytes); 3409 u64_stats_update_end(&tx_ring->ring_stats->syncp); 3410 } 3411 3412 /** 3413 * ice_update_rx_ring_stats - Update Rx ring specific counters 3414 * @rx_ring: ring to update 3415 * @pkts: number of processed packets 3416 * @bytes: number of processed bytes 3417 */ 3418 void ice_update_rx_ring_stats(struct ice_rx_ring *rx_ring, u64 pkts, u64 bytes) 3419 { 3420 u64_stats_update_begin(&rx_ring->ring_stats->syncp); 3421 ice_update_ring_stats(&rx_ring->ring_stats->stats, pkts, bytes); 3422 u64_stats_update_end(&rx_ring->ring_stats->syncp); 3423 } 3424 3425 /** 3426 * ice_is_dflt_vsi_in_use - check if the default forwarding VSI is being used 3427 * @pi: port info of the switch with default VSI 3428 * 3429 * Return true if the there is a single VSI in default forwarding VSI list 3430 */ 3431 bool ice_is_dflt_vsi_in_use(struct ice_port_info *pi) 3432 { 3433 bool exists = false; 3434 3435 ice_check_if_dflt_vsi(pi, 0, &exists); 3436 return exists; 3437 } 3438 3439 /** 3440 * ice_is_vsi_dflt_vsi - check if the VSI passed in is the default VSI 3441 * @vsi: VSI to compare against default forwarding VSI 3442 * 3443 * If this VSI passed in is the default forwarding VSI then return true, else 3444 * return false 3445 */ 3446 bool ice_is_vsi_dflt_vsi(struct ice_vsi *vsi) 3447 { 3448 return ice_check_if_dflt_vsi(vsi->port_info, vsi->idx, NULL); 3449 } 3450 3451 /** 3452 * ice_set_dflt_vsi - set the default forwarding VSI 3453 * @vsi: VSI getting set as the default forwarding VSI on the switch 3454 * 3455 * If the VSI passed in is already the default VSI and it's enabled just return 3456 * success. 3457 * 3458 * Otherwise try to set the VSI passed in as the switch's default VSI and 3459 * return the result. 3460 */ 3461 int ice_set_dflt_vsi(struct ice_vsi *vsi) 3462 { 3463 struct device *dev; 3464 int status; 3465 3466 if (!vsi) 3467 return -EINVAL; 3468 3469 dev = ice_pf_to_dev(vsi->back); 3470 3471 if (ice_lag_is_switchdev_running(vsi->back)) { 3472 dev_dbg(dev, "VSI %d passed is a part of LAG containing interfaces in switchdev mode, nothing to do\n", 3473 vsi->vsi_num); 3474 return 0; 3475 } 3476 3477 /* the VSI passed in is already the default VSI */ 3478 if (ice_is_vsi_dflt_vsi(vsi)) { 3479 dev_dbg(dev, "VSI %d passed in is already the default forwarding VSI, nothing to do\n", 3480 vsi->vsi_num); 3481 return 0; 3482 } 3483 3484 status = ice_cfg_dflt_vsi(vsi->port_info, vsi->idx, true, ICE_FLTR_RX); 3485 if (status) { 3486 dev_err(dev, "Failed to set VSI %d as the default forwarding VSI, error %d\n", 3487 vsi->vsi_num, status); 3488 return status; 3489 } 3490 3491 return 0; 3492 } 3493 3494 /** 3495 * ice_clear_dflt_vsi - clear the default forwarding VSI 3496 * @vsi: VSI to remove from filter list 3497 * 3498 * If the switch has no default VSI or it's not enabled then return error. 3499 * 3500 * Otherwise try to clear the default VSI and return the result. 3501 */ 3502 int ice_clear_dflt_vsi(struct ice_vsi *vsi) 3503 { 3504 struct device *dev; 3505 int status; 3506 3507 if (!vsi) 3508 return -EINVAL; 3509 3510 dev = ice_pf_to_dev(vsi->back); 3511 3512 /* there is no default VSI configured */ 3513 if (!ice_is_dflt_vsi_in_use(vsi->port_info)) 3514 return -ENODEV; 3515 3516 status = ice_cfg_dflt_vsi(vsi->port_info, vsi->idx, false, 3517 ICE_FLTR_RX); 3518 if (status) { 3519 dev_err(dev, "Failed to clear the default forwarding VSI %d, error %d\n", 3520 vsi->vsi_num, status); 3521 return -EIO; 3522 } 3523 3524 return 0; 3525 } 3526 3527 /** 3528 * ice_get_link_speed_mbps - get link speed in Mbps 3529 * @vsi: the VSI whose link speed is being queried 3530 * 3531 * Return current VSI link speed and 0 if the speed is unknown. 3532 */ 3533 int ice_get_link_speed_mbps(struct ice_vsi *vsi) 3534 { 3535 unsigned int link_speed; 3536 3537 link_speed = vsi->port_info->phy.link_info.link_speed; 3538 3539 return (int)ice_get_link_speed(fls(link_speed) - 1); 3540 } 3541 3542 /** 3543 * ice_get_link_speed_kbps - get link speed in Kbps 3544 * @vsi: the VSI whose link speed is being queried 3545 * 3546 * Return current VSI link speed and 0 if the speed is unknown. 3547 */ 3548 int ice_get_link_speed_kbps(struct ice_vsi *vsi) 3549 { 3550 int speed_mbps; 3551 3552 speed_mbps = ice_get_link_speed_mbps(vsi); 3553 3554 return speed_mbps * 1000; 3555 } 3556 3557 /** 3558 * ice_set_min_bw_limit - setup minimum BW limit for Tx based on min_tx_rate 3559 * @vsi: VSI to be configured 3560 * @min_tx_rate: min Tx rate in Kbps to be configured as BW limit 3561 * 3562 * If the min_tx_rate is specified as 0 that means to clear the minimum BW limit 3563 * profile, otherwise a non-zero value will force a minimum BW limit for the VSI 3564 * on TC 0. 3565 */ 3566 int ice_set_min_bw_limit(struct ice_vsi *vsi, u64 min_tx_rate) 3567 { 3568 struct ice_pf *pf = vsi->back; 3569 struct device *dev; 3570 int status; 3571 int speed; 3572 3573 dev = ice_pf_to_dev(pf); 3574 if (!vsi->port_info) { 3575 dev_dbg(dev, "VSI %d, type %u specified doesn't have valid port_info\n", 3576 vsi->idx, vsi->type); 3577 return -EINVAL; 3578 } 3579 3580 speed = ice_get_link_speed_kbps(vsi); 3581 if (min_tx_rate > (u64)speed) { 3582 dev_err(dev, "invalid min Tx rate %llu Kbps specified for %s %d is greater than current link speed %u Kbps\n", 3583 min_tx_rate, ice_vsi_type_str(vsi->type), vsi->idx, 3584 speed); 3585 return -EINVAL; 3586 } 3587 3588 /* Configure min BW for VSI limit */ 3589 if (min_tx_rate) { 3590 status = ice_cfg_vsi_bw_lmt_per_tc(vsi->port_info, vsi->idx, 0, 3591 ICE_MIN_BW, min_tx_rate); 3592 if (status) { 3593 dev_err(dev, "failed to set min Tx rate(%llu Kbps) for %s %d\n", 3594 min_tx_rate, ice_vsi_type_str(vsi->type), 3595 vsi->idx); 3596 return status; 3597 } 3598 3599 dev_dbg(dev, "set min Tx rate(%llu Kbps) for %s\n", 3600 min_tx_rate, ice_vsi_type_str(vsi->type)); 3601 } else { 3602 status = ice_cfg_vsi_bw_dflt_lmt_per_tc(vsi->port_info, 3603 vsi->idx, 0, 3604 ICE_MIN_BW); 3605 if (status) { 3606 dev_err(dev, "failed to clear min Tx rate configuration for %s %d\n", 3607 ice_vsi_type_str(vsi->type), vsi->idx); 3608 return status; 3609 } 3610 3611 dev_dbg(dev, "cleared min Tx rate configuration for %s %d\n", 3612 ice_vsi_type_str(vsi->type), vsi->idx); 3613 } 3614 3615 return 0; 3616 } 3617 3618 /** 3619 * ice_set_max_bw_limit - setup maximum BW limit for Tx based on max_tx_rate 3620 * @vsi: VSI to be configured 3621 * @max_tx_rate: max Tx rate in Kbps to be configured as BW limit 3622 * 3623 * If the max_tx_rate is specified as 0 that means to clear the maximum BW limit 3624 * profile, otherwise a non-zero value will force a maximum BW limit for the VSI 3625 * on TC 0. 3626 */ 3627 int ice_set_max_bw_limit(struct ice_vsi *vsi, u64 max_tx_rate) 3628 { 3629 struct ice_pf *pf = vsi->back; 3630 struct device *dev; 3631 int status; 3632 int speed; 3633 3634 dev = ice_pf_to_dev(pf); 3635 if (!vsi->port_info) { 3636 dev_dbg(dev, "VSI %d, type %u specified doesn't have valid port_info\n", 3637 vsi->idx, vsi->type); 3638 return -EINVAL; 3639 } 3640 3641 speed = ice_get_link_speed_kbps(vsi); 3642 if (max_tx_rate > (u64)speed) { 3643 dev_err(dev, "invalid max Tx rate %llu Kbps specified for %s %d is greater than current link speed %u Kbps\n", 3644 max_tx_rate, ice_vsi_type_str(vsi->type), vsi->idx, 3645 speed); 3646 return -EINVAL; 3647 } 3648 3649 /* Configure max BW for VSI limit */ 3650 if (max_tx_rate) { 3651 status = ice_cfg_vsi_bw_lmt_per_tc(vsi->port_info, vsi->idx, 0, 3652 ICE_MAX_BW, max_tx_rate); 3653 if (status) { 3654 dev_err(dev, "failed setting max Tx rate(%llu Kbps) for %s %d\n", 3655 max_tx_rate, ice_vsi_type_str(vsi->type), 3656 vsi->idx); 3657 return status; 3658 } 3659 3660 dev_dbg(dev, "set max Tx rate(%llu Kbps) for %s %d\n", 3661 max_tx_rate, ice_vsi_type_str(vsi->type), vsi->idx); 3662 } else { 3663 status = ice_cfg_vsi_bw_dflt_lmt_per_tc(vsi->port_info, 3664 vsi->idx, 0, 3665 ICE_MAX_BW); 3666 if (status) { 3667 dev_err(dev, "failed clearing max Tx rate configuration for %s %d\n", 3668 ice_vsi_type_str(vsi->type), vsi->idx); 3669 return status; 3670 } 3671 3672 dev_dbg(dev, "cleared max Tx rate configuration for %s %d\n", 3673 ice_vsi_type_str(vsi->type), vsi->idx); 3674 } 3675 3676 return 0; 3677 } 3678 3679 /** 3680 * ice_set_link - turn on/off physical link 3681 * @vsi: VSI to modify physical link on 3682 * @ena: turn on/off physical link 3683 */ 3684 int ice_set_link(struct ice_vsi *vsi, bool ena) 3685 { 3686 struct device *dev = ice_pf_to_dev(vsi->back); 3687 struct ice_port_info *pi = vsi->port_info; 3688 struct ice_hw *hw = pi->hw; 3689 int status; 3690 3691 if (vsi->type != ICE_VSI_PF) 3692 return -EINVAL; 3693 3694 status = ice_aq_set_link_restart_an(pi, ena, NULL); 3695 3696 /* if link is owned by manageability, FW will return ICE_AQ_RC_EMODE. 3697 * this is not a fatal error, so print a warning message and return 3698 * a success code. Return an error if FW returns an error code other 3699 * than ICE_AQ_RC_EMODE 3700 */ 3701 if (status == -EIO) { 3702 if (hw->adminq.sq_last_status == ICE_AQ_RC_EMODE) 3703 dev_dbg(dev, "can't set link to %s, err %d aq_err %s. not fatal, continuing\n", 3704 (ena ? "ON" : "OFF"), status, 3705 ice_aq_str(hw->adminq.sq_last_status)); 3706 } else if (status) { 3707 dev_err(dev, "can't set link to %s, err %d aq_err %s\n", 3708 (ena ? "ON" : "OFF"), status, 3709 ice_aq_str(hw->adminq.sq_last_status)); 3710 return status; 3711 } 3712 3713 return 0; 3714 } 3715 3716 /** 3717 * ice_vsi_add_vlan_zero - add VLAN 0 filter(s) for this VSI 3718 * @vsi: VSI used to add VLAN filters 3719 * 3720 * In Single VLAN Mode (SVM), single VLAN filters via ICE_SW_LKUP_VLAN are based 3721 * on the inner VLAN ID, so the VLAN TPID (i.e. 0x8100 or 0x888a8) doesn't 3722 * matter. In Double VLAN Mode (DVM), outer/single VLAN filters via 3723 * ICE_SW_LKUP_VLAN are based on the outer/single VLAN ID + VLAN TPID. 3724 * 3725 * For both modes add a VLAN 0 + no VLAN TPID filter to handle untagged traffic 3726 * when VLAN pruning is enabled. Also, this handles VLAN 0 priority tagged 3727 * traffic in SVM, since the VLAN TPID isn't part of filtering. 3728 * 3729 * If DVM is enabled then an explicit VLAN 0 + VLAN TPID filter needs to be 3730 * added to allow VLAN 0 priority tagged traffic in DVM, since the VLAN TPID is 3731 * part of filtering. 3732 */ 3733 int ice_vsi_add_vlan_zero(struct ice_vsi *vsi) 3734 { 3735 struct ice_vsi_vlan_ops *vlan_ops = ice_get_compat_vsi_vlan_ops(vsi); 3736 struct ice_vlan vlan; 3737 int err; 3738 3739 vlan = ICE_VLAN(0, 0, 0); 3740 err = vlan_ops->add_vlan(vsi, &vlan); 3741 if (err && err != -EEXIST) 3742 return err; 3743 3744 /* in SVM both VLAN 0 filters are identical */ 3745 if (!ice_is_dvm_ena(&vsi->back->hw)) 3746 return 0; 3747 3748 vlan = ICE_VLAN(ETH_P_8021Q, 0, 0); 3749 err = vlan_ops->add_vlan(vsi, &vlan); 3750 if (err && err != -EEXIST) 3751 return err; 3752 3753 return 0; 3754 } 3755 3756 /** 3757 * ice_vsi_del_vlan_zero - delete VLAN 0 filter(s) for this VSI 3758 * @vsi: VSI used to add VLAN filters 3759 * 3760 * Delete the VLAN 0 filters in the same manner that they were added in 3761 * ice_vsi_add_vlan_zero. 3762 */ 3763 int ice_vsi_del_vlan_zero(struct ice_vsi *vsi) 3764 { 3765 struct ice_vsi_vlan_ops *vlan_ops = ice_get_compat_vsi_vlan_ops(vsi); 3766 struct ice_vlan vlan; 3767 int err; 3768 3769 vlan = ICE_VLAN(0, 0, 0); 3770 err = vlan_ops->del_vlan(vsi, &vlan); 3771 if (err && err != -EEXIST) 3772 return err; 3773 3774 /* in SVM both VLAN 0 filters are identical */ 3775 if (!ice_is_dvm_ena(&vsi->back->hw)) 3776 return 0; 3777 3778 vlan = ICE_VLAN(ETH_P_8021Q, 0, 0); 3779 err = vlan_ops->del_vlan(vsi, &vlan); 3780 if (err && err != -EEXIST) 3781 return err; 3782 3783 /* when deleting the last VLAN filter, make sure to disable the VLAN 3784 * promisc mode so the filter isn't left by accident 3785 */ 3786 return ice_clear_vsi_promisc(&vsi->back->hw, vsi->idx, 3787 ICE_MCAST_VLAN_PROMISC_BITS, 0); 3788 } 3789 3790 /** 3791 * ice_vsi_num_zero_vlans - get number of VLAN 0 filters based on VLAN mode 3792 * @vsi: VSI used to get the VLAN mode 3793 * 3794 * If DVM is enabled then 2 VLAN 0 filters are added, else if SVM is enabled 3795 * then 1 VLAN 0 filter is added. See ice_vsi_add_vlan_zero for more details. 3796 */ 3797 static u16 ice_vsi_num_zero_vlans(struct ice_vsi *vsi) 3798 { 3799 #define ICE_DVM_NUM_ZERO_VLAN_FLTRS 2 3800 #define ICE_SVM_NUM_ZERO_VLAN_FLTRS 1 3801 /* no VLAN 0 filter is created when a port VLAN is active */ 3802 if (vsi->type == ICE_VSI_VF) { 3803 if (WARN_ON(!vsi->vf)) 3804 return 0; 3805 3806 if (ice_vf_is_port_vlan_ena(vsi->vf)) 3807 return 0; 3808 } 3809 3810 if (ice_is_dvm_ena(&vsi->back->hw)) 3811 return ICE_DVM_NUM_ZERO_VLAN_FLTRS; 3812 else 3813 return ICE_SVM_NUM_ZERO_VLAN_FLTRS; 3814 } 3815 3816 /** 3817 * ice_vsi_has_non_zero_vlans - check if VSI has any non-zero VLANs 3818 * @vsi: VSI used to determine if any non-zero VLANs have been added 3819 */ 3820 bool ice_vsi_has_non_zero_vlans(struct ice_vsi *vsi) 3821 { 3822 return (vsi->num_vlan > ice_vsi_num_zero_vlans(vsi)); 3823 } 3824 3825 /** 3826 * ice_vsi_num_non_zero_vlans - get the number of non-zero VLANs for this VSI 3827 * @vsi: VSI used to get the number of non-zero VLANs added 3828 */ 3829 u16 ice_vsi_num_non_zero_vlans(struct ice_vsi *vsi) 3830 { 3831 return (vsi->num_vlan - ice_vsi_num_zero_vlans(vsi)); 3832 } 3833 3834 /** 3835 * ice_is_feature_supported 3836 * @pf: pointer to the struct ice_pf instance 3837 * @f: feature enum to be checked 3838 * 3839 * returns true if feature is supported, false otherwise 3840 */ 3841 bool ice_is_feature_supported(struct ice_pf *pf, enum ice_feature f) 3842 { 3843 if (f < 0 || f >= ICE_F_MAX) 3844 return false; 3845 3846 return test_bit(f, pf->features); 3847 } 3848 3849 /** 3850 * ice_set_feature_support 3851 * @pf: pointer to the struct ice_pf instance 3852 * @f: feature enum to set 3853 */ 3854 void ice_set_feature_support(struct ice_pf *pf, enum ice_feature f) 3855 { 3856 if (f < 0 || f >= ICE_F_MAX) 3857 return; 3858 3859 set_bit(f, pf->features); 3860 } 3861 3862 /** 3863 * ice_clear_feature_support 3864 * @pf: pointer to the struct ice_pf instance 3865 * @f: feature enum to clear 3866 */ 3867 void ice_clear_feature_support(struct ice_pf *pf, enum ice_feature f) 3868 { 3869 if (f < 0 || f >= ICE_F_MAX) 3870 return; 3871 3872 clear_bit(f, pf->features); 3873 } 3874 3875 /** 3876 * ice_init_feature_support 3877 * @pf: pointer to the struct ice_pf instance 3878 * 3879 * called during init to setup supported feature 3880 */ 3881 void ice_init_feature_support(struct ice_pf *pf) 3882 { 3883 switch (pf->hw.device_id) { 3884 case ICE_DEV_ID_E810C_BACKPLANE: 3885 case ICE_DEV_ID_E810C_QSFP: 3886 case ICE_DEV_ID_E810C_SFP: 3887 case ICE_DEV_ID_E810_XXV_BACKPLANE: 3888 case ICE_DEV_ID_E810_XXV_QSFP: 3889 case ICE_DEV_ID_E810_XXV_SFP: 3890 ice_set_feature_support(pf, ICE_F_DSCP); 3891 if (ice_is_phy_rclk_in_netlist(&pf->hw)) 3892 ice_set_feature_support(pf, ICE_F_PHY_RCLK); 3893 /* If we don't own the timer - don't enable other caps */ 3894 if (!ice_pf_src_tmr_owned(pf)) 3895 break; 3896 if (ice_is_cgu_in_netlist(&pf->hw)) 3897 ice_set_feature_support(pf, ICE_F_CGU); 3898 if (ice_is_clock_mux_in_netlist(&pf->hw)) 3899 ice_set_feature_support(pf, ICE_F_SMA_CTRL); 3900 if (ice_gnss_is_gps_present(&pf->hw)) 3901 ice_set_feature_support(pf, ICE_F_GNSS); 3902 break; 3903 default: 3904 break; 3905 } 3906 } 3907 3908 /** 3909 * ice_vsi_update_security - update security block in VSI 3910 * @vsi: pointer to VSI structure 3911 * @fill: function pointer to fill ctx 3912 */ 3913 int 3914 ice_vsi_update_security(struct ice_vsi *vsi, void (*fill)(struct ice_vsi_ctx *)) 3915 { 3916 struct ice_vsi_ctx ctx = { 0 }; 3917 3918 ctx.info = vsi->info; 3919 ctx.info.valid_sections = cpu_to_le16(ICE_AQ_VSI_PROP_SECURITY_VALID); 3920 fill(&ctx); 3921 3922 if (ice_update_vsi(&vsi->back->hw, vsi->idx, &ctx, NULL)) 3923 return -ENODEV; 3924 3925 vsi->info = ctx.info; 3926 return 0; 3927 } 3928 3929 /** 3930 * ice_vsi_ctx_set_antispoof - set antispoof function in VSI ctx 3931 * @ctx: pointer to VSI ctx structure 3932 */ 3933 void ice_vsi_ctx_set_antispoof(struct ice_vsi_ctx *ctx) 3934 { 3935 ctx->info.sec_flags |= ICE_AQ_VSI_SEC_FLAG_ENA_MAC_ANTI_SPOOF | 3936 (ICE_AQ_VSI_SEC_TX_VLAN_PRUNE_ENA << 3937 ICE_AQ_VSI_SEC_TX_PRUNE_ENA_S); 3938 } 3939 3940 /** 3941 * ice_vsi_ctx_clear_antispoof - clear antispoof function in VSI ctx 3942 * @ctx: pointer to VSI ctx structure 3943 */ 3944 void ice_vsi_ctx_clear_antispoof(struct ice_vsi_ctx *ctx) 3945 { 3946 ctx->info.sec_flags &= ~ICE_AQ_VSI_SEC_FLAG_ENA_MAC_ANTI_SPOOF & 3947 ~(ICE_AQ_VSI_SEC_TX_VLAN_PRUNE_ENA << 3948 ICE_AQ_VSI_SEC_TX_PRUNE_ENA_S); 3949 } 3950 3951 /** 3952 * ice_vsi_ctx_set_allow_override - allow destination override on VSI 3953 * @ctx: pointer to VSI ctx structure 3954 */ 3955 void ice_vsi_ctx_set_allow_override(struct ice_vsi_ctx *ctx) 3956 { 3957 ctx->info.sec_flags |= ICE_AQ_VSI_SEC_FLAG_ALLOW_DEST_OVRD; 3958 } 3959 3960 /** 3961 * ice_vsi_ctx_clear_allow_override - turn off destination override on VSI 3962 * @ctx: pointer to VSI ctx structure 3963 */ 3964 void ice_vsi_ctx_clear_allow_override(struct ice_vsi_ctx *ctx) 3965 { 3966 ctx->info.sec_flags &= ~ICE_AQ_VSI_SEC_FLAG_ALLOW_DEST_OVRD; 3967 } 3968 3969 /** 3970 * ice_vsi_update_local_lb - update sw block in VSI with local loopback bit 3971 * @vsi: pointer to VSI structure 3972 * @set: set or unset the bit 3973 */ 3974 int 3975 ice_vsi_update_local_lb(struct ice_vsi *vsi, bool set) 3976 { 3977 struct ice_vsi_ctx ctx = { 3978 .info = vsi->info, 3979 }; 3980 3981 ctx.info.valid_sections = cpu_to_le16(ICE_AQ_VSI_PROP_SW_VALID); 3982 if (set) 3983 ctx.info.sw_flags |= ICE_AQ_VSI_SW_FLAG_LOCAL_LB; 3984 else 3985 ctx.info.sw_flags &= ~ICE_AQ_VSI_SW_FLAG_LOCAL_LB; 3986 3987 if (ice_update_vsi(&vsi->back->hw, vsi->idx, &ctx, NULL)) 3988 return -ENODEV; 3989 3990 vsi->info = ctx.info; 3991 return 0; 3992 } 3993