1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright (c) 2018 Intel Corporation */ 3 4 #include <linux/module.h> 5 #include <linux/types.h> 6 #include <linux/if_vlan.h> 7 #include <linux/tcp.h> 8 #include <linux/udp.h> 9 #include <linux/ip.h> 10 #include <linux/pm_runtime.h> 11 #include <net/pkt_sched.h> 12 #include <linux/bpf_trace.h> 13 #include <net/xdp_sock_drv.h> 14 #include <linux/pci.h> 15 16 #include <net/ipv6.h> 17 18 #include "igc.h" 19 #include "igc_hw.h" 20 #include "igc_tsn.h" 21 #include "igc_xdp.h" 22 23 #define DRV_SUMMARY "Intel(R) 2.5G Ethernet Linux Driver" 24 25 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK) 26 27 #define IGC_XDP_PASS 0 28 #define IGC_XDP_CONSUMED BIT(0) 29 #define IGC_XDP_TX BIT(1) 30 #define IGC_XDP_REDIRECT BIT(2) 31 32 static int debug = -1; 33 34 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>"); 35 MODULE_DESCRIPTION(DRV_SUMMARY); 36 MODULE_LICENSE("GPL v2"); 37 module_param(debug, int, 0); 38 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)"); 39 40 char igc_driver_name[] = "igc"; 41 static const char igc_driver_string[] = DRV_SUMMARY; 42 static const char igc_copyright[] = 43 "Copyright(c) 2018 Intel Corporation."; 44 45 static const struct igc_info *igc_info_tbl[] = { 46 [board_base] = &igc_base_info, 47 }; 48 49 static const struct pci_device_id igc_pci_tbl[] = { 50 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_LM), board_base }, 51 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_V), board_base }, 52 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_I), board_base }, 53 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I220_V), board_base }, 54 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_K), board_base }, 55 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_K2), board_base }, 56 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_K), board_base }, 57 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_LMVP), board_base }, 58 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_LMVP), board_base }, 59 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_IT), board_base }, 60 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_LM), board_base }, 61 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_V), board_base }, 62 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_IT), board_base }, 63 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I221_V), board_base }, 64 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_BLANK_NVM), board_base }, 65 { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_BLANK_NVM), board_base }, 66 /* required last entry */ 67 {0, } 68 }; 69 70 MODULE_DEVICE_TABLE(pci, igc_pci_tbl); 71 72 enum latency_range { 73 lowest_latency = 0, 74 low_latency = 1, 75 bulk_latency = 2, 76 latency_invalid = 255 77 }; 78 79 void igc_reset(struct igc_adapter *adapter) 80 { 81 struct net_device *dev = adapter->netdev; 82 struct igc_hw *hw = &adapter->hw; 83 struct igc_fc_info *fc = &hw->fc; 84 u32 pba, hwm; 85 86 /* Repartition PBA for greater than 9k MTU if required */ 87 pba = IGC_PBA_34K; 88 89 /* flow control settings 90 * The high water mark must be low enough to fit one full frame 91 * after transmitting the pause frame. As such we must have enough 92 * space to allow for us to complete our current transmit and then 93 * receive the frame that is in progress from the link partner. 94 * Set it to: 95 * - the full Rx FIFO size minus one full Tx plus one full Rx frame 96 */ 97 hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE); 98 99 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */ 100 fc->low_water = fc->high_water - 16; 101 fc->pause_time = 0xFFFF; 102 fc->send_xon = 1; 103 fc->current_mode = fc->requested_mode; 104 105 hw->mac.ops.reset_hw(hw); 106 107 if (hw->mac.ops.init_hw(hw)) 108 netdev_err(dev, "Error on hardware initialization\n"); 109 110 /* Re-establish EEE setting */ 111 igc_set_eee_i225(hw, true, true, true); 112 113 if (!netif_running(adapter->netdev)) 114 igc_power_down_phy_copper_base(&adapter->hw); 115 116 /* Enable HW to recognize an 802.1Q VLAN Ethernet packet */ 117 wr32(IGC_VET, ETH_P_8021Q); 118 119 /* Re-enable PTP, where applicable. */ 120 igc_ptp_reset(adapter); 121 122 /* Re-enable TSN offloading, where applicable. */ 123 igc_tsn_reset(adapter); 124 125 igc_get_phy_info(hw); 126 } 127 128 /** 129 * igc_power_up_link - Power up the phy link 130 * @adapter: address of board private structure 131 */ 132 static void igc_power_up_link(struct igc_adapter *adapter) 133 { 134 igc_reset_phy(&adapter->hw); 135 136 igc_power_up_phy_copper(&adapter->hw); 137 138 igc_setup_link(&adapter->hw); 139 } 140 141 /** 142 * igc_release_hw_control - release control of the h/w to f/w 143 * @adapter: address of board private structure 144 * 145 * igc_release_hw_control resets CTRL_EXT:DRV_LOAD bit. 146 * For ASF and Pass Through versions of f/w this means that the 147 * driver is no longer loaded. 148 */ 149 static void igc_release_hw_control(struct igc_adapter *adapter) 150 { 151 struct igc_hw *hw = &adapter->hw; 152 u32 ctrl_ext; 153 154 if (!pci_device_is_present(adapter->pdev)) 155 return; 156 157 /* Let firmware take over control of h/w */ 158 ctrl_ext = rd32(IGC_CTRL_EXT); 159 wr32(IGC_CTRL_EXT, 160 ctrl_ext & ~IGC_CTRL_EXT_DRV_LOAD); 161 } 162 163 /** 164 * igc_get_hw_control - get control of the h/w from f/w 165 * @adapter: address of board private structure 166 * 167 * igc_get_hw_control sets CTRL_EXT:DRV_LOAD bit. 168 * For ASF and Pass Through versions of f/w this means that 169 * the driver is loaded. 170 */ 171 static void igc_get_hw_control(struct igc_adapter *adapter) 172 { 173 struct igc_hw *hw = &adapter->hw; 174 u32 ctrl_ext; 175 176 /* Let firmware know the driver has taken over */ 177 ctrl_ext = rd32(IGC_CTRL_EXT); 178 wr32(IGC_CTRL_EXT, 179 ctrl_ext | IGC_CTRL_EXT_DRV_LOAD); 180 } 181 182 static void igc_unmap_tx_buffer(struct device *dev, struct igc_tx_buffer *buf) 183 { 184 dma_unmap_single(dev, dma_unmap_addr(buf, dma), 185 dma_unmap_len(buf, len), DMA_TO_DEVICE); 186 187 dma_unmap_len_set(buf, len, 0); 188 } 189 190 /** 191 * igc_clean_tx_ring - Free Tx Buffers 192 * @tx_ring: ring to be cleaned 193 */ 194 static void igc_clean_tx_ring(struct igc_ring *tx_ring) 195 { 196 u16 i = tx_ring->next_to_clean; 197 struct igc_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i]; 198 u32 xsk_frames = 0; 199 200 while (i != tx_ring->next_to_use) { 201 union igc_adv_tx_desc *eop_desc, *tx_desc; 202 203 switch (tx_buffer->type) { 204 case IGC_TX_BUFFER_TYPE_XSK: 205 xsk_frames++; 206 break; 207 case IGC_TX_BUFFER_TYPE_XDP: 208 xdp_return_frame(tx_buffer->xdpf); 209 igc_unmap_tx_buffer(tx_ring->dev, tx_buffer); 210 break; 211 case IGC_TX_BUFFER_TYPE_SKB: 212 dev_kfree_skb_any(tx_buffer->skb); 213 igc_unmap_tx_buffer(tx_ring->dev, tx_buffer); 214 break; 215 default: 216 netdev_warn_once(tx_ring->netdev, "Unknown Tx buffer type\n"); 217 break; 218 } 219 220 /* check for eop_desc to determine the end of the packet */ 221 eop_desc = tx_buffer->next_to_watch; 222 tx_desc = IGC_TX_DESC(tx_ring, i); 223 224 /* unmap remaining buffers */ 225 while (tx_desc != eop_desc) { 226 tx_buffer++; 227 tx_desc++; 228 i++; 229 if (unlikely(i == tx_ring->count)) { 230 i = 0; 231 tx_buffer = tx_ring->tx_buffer_info; 232 tx_desc = IGC_TX_DESC(tx_ring, 0); 233 } 234 235 /* unmap any remaining paged data */ 236 if (dma_unmap_len(tx_buffer, len)) 237 igc_unmap_tx_buffer(tx_ring->dev, tx_buffer); 238 } 239 240 tx_buffer->next_to_watch = NULL; 241 242 /* move us one more past the eop_desc for start of next pkt */ 243 tx_buffer++; 244 i++; 245 if (unlikely(i == tx_ring->count)) { 246 i = 0; 247 tx_buffer = tx_ring->tx_buffer_info; 248 } 249 } 250 251 if (tx_ring->xsk_pool && xsk_frames) 252 xsk_tx_completed(tx_ring->xsk_pool, xsk_frames); 253 254 /* reset BQL for queue */ 255 netdev_tx_reset_queue(txring_txq(tx_ring)); 256 257 /* Zero out the buffer ring */ 258 memset(tx_ring->tx_buffer_info, 0, 259 sizeof(*tx_ring->tx_buffer_info) * tx_ring->count); 260 261 /* Zero out the descriptor ring */ 262 memset(tx_ring->desc, 0, tx_ring->size); 263 264 /* reset next_to_use and next_to_clean */ 265 tx_ring->next_to_use = 0; 266 tx_ring->next_to_clean = 0; 267 } 268 269 /** 270 * igc_free_tx_resources - Free Tx Resources per Queue 271 * @tx_ring: Tx descriptor ring for a specific queue 272 * 273 * Free all transmit software resources 274 */ 275 void igc_free_tx_resources(struct igc_ring *tx_ring) 276 { 277 igc_disable_tx_ring(tx_ring); 278 279 vfree(tx_ring->tx_buffer_info); 280 tx_ring->tx_buffer_info = NULL; 281 282 /* if not set, then don't free */ 283 if (!tx_ring->desc) 284 return; 285 286 dma_free_coherent(tx_ring->dev, tx_ring->size, 287 tx_ring->desc, tx_ring->dma); 288 289 tx_ring->desc = NULL; 290 } 291 292 /** 293 * igc_free_all_tx_resources - Free Tx Resources for All Queues 294 * @adapter: board private structure 295 * 296 * Free all transmit software resources 297 */ 298 static void igc_free_all_tx_resources(struct igc_adapter *adapter) 299 { 300 int i; 301 302 for (i = 0; i < adapter->num_tx_queues; i++) 303 igc_free_tx_resources(adapter->tx_ring[i]); 304 } 305 306 /** 307 * igc_clean_all_tx_rings - Free Tx Buffers for all queues 308 * @adapter: board private structure 309 */ 310 static void igc_clean_all_tx_rings(struct igc_adapter *adapter) 311 { 312 int i; 313 314 for (i = 0; i < adapter->num_tx_queues; i++) 315 if (adapter->tx_ring[i]) 316 igc_clean_tx_ring(adapter->tx_ring[i]); 317 } 318 319 static void igc_disable_tx_ring_hw(struct igc_ring *ring) 320 { 321 struct igc_hw *hw = &ring->q_vector->adapter->hw; 322 u8 idx = ring->reg_idx; 323 u32 txdctl; 324 325 txdctl = rd32(IGC_TXDCTL(idx)); 326 txdctl &= ~IGC_TXDCTL_QUEUE_ENABLE; 327 txdctl |= IGC_TXDCTL_SWFLUSH; 328 wr32(IGC_TXDCTL(idx), txdctl); 329 } 330 331 /** 332 * igc_disable_all_tx_rings_hw - Disable all transmit queue operation 333 * @adapter: board private structure 334 */ 335 static void igc_disable_all_tx_rings_hw(struct igc_adapter *adapter) 336 { 337 int i; 338 339 for (i = 0; i < adapter->num_tx_queues; i++) { 340 struct igc_ring *tx_ring = adapter->tx_ring[i]; 341 342 igc_disable_tx_ring_hw(tx_ring); 343 } 344 } 345 346 /** 347 * igc_setup_tx_resources - allocate Tx resources (Descriptors) 348 * @tx_ring: tx descriptor ring (for a specific queue) to setup 349 * 350 * Return 0 on success, negative on failure 351 */ 352 int igc_setup_tx_resources(struct igc_ring *tx_ring) 353 { 354 struct net_device *ndev = tx_ring->netdev; 355 struct device *dev = tx_ring->dev; 356 int size = 0; 357 358 size = sizeof(struct igc_tx_buffer) * tx_ring->count; 359 tx_ring->tx_buffer_info = vzalloc(size); 360 if (!tx_ring->tx_buffer_info) 361 goto err; 362 363 /* round up to nearest 4K */ 364 tx_ring->size = tx_ring->count * sizeof(union igc_adv_tx_desc); 365 tx_ring->size = ALIGN(tx_ring->size, 4096); 366 367 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size, 368 &tx_ring->dma, GFP_KERNEL); 369 370 if (!tx_ring->desc) 371 goto err; 372 373 tx_ring->next_to_use = 0; 374 tx_ring->next_to_clean = 0; 375 376 return 0; 377 378 err: 379 vfree(tx_ring->tx_buffer_info); 380 netdev_err(ndev, "Unable to allocate memory for Tx descriptor ring\n"); 381 return -ENOMEM; 382 } 383 384 /** 385 * igc_setup_all_tx_resources - wrapper to allocate Tx resources for all queues 386 * @adapter: board private structure 387 * 388 * Return 0 on success, negative on failure 389 */ 390 static int igc_setup_all_tx_resources(struct igc_adapter *adapter) 391 { 392 struct net_device *dev = adapter->netdev; 393 int i, err = 0; 394 395 for (i = 0; i < adapter->num_tx_queues; i++) { 396 err = igc_setup_tx_resources(adapter->tx_ring[i]); 397 if (err) { 398 netdev_err(dev, "Error on Tx queue %u setup\n", i); 399 for (i--; i >= 0; i--) 400 igc_free_tx_resources(adapter->tx_ring[i]); 401 break; 402 } 403 } 404 405 return err; 406 } 407 408 static void igc_clean_rx_ring_page_shared(struct igc_ring *rx_ring) 409 { 410 u16 i = rx_ring->next_to_clean; 411 412 dev_kfree_skb(rx_ring->skb); 413 rx_ring->skb = NULL; 414 415 /* Free all the Rx ring sk_buffs */ 416 while (i != rx_ring->next_to_alloc) { 417 struct igc_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i]; 418 419 /* Invalidate cache lines that may have been written to by 420 * device so that we avoid corrupting memory. 421 */ 422 dma_sync_single_range_for_cpu(rx_ring->dev, 423 buffer_info->dma, 424 buffer_info->page_offset, 425 igc_rx_bufsz(rx_ring), 426 DMA_FROM_DEVICE); 427 428 /* free resources associated with mapping */ 429 dma_unmap_page_attrs(rx_ring->dev, 430 buffer_info->dma, 431 igc_rx_pg_size(rx_ring), 432 DMA_FROM_DEVICE, 433 IGC_RX_DMA_ATTR); 434 __page_frag_cache_drain(buffer_info->page, 435 buffer_info->pagecnt_bias); 436 437 i++; 438 if (i == rx_ring->count) 439 i = 0; 440 } 441 } 442 443 static void igc_clean_rx_ring_xsk_pool(struct igc_ring *ring) 444 { 445 struct igc_rx_buffer *bi; 446 u16 i; 447 448 for (i = 0; i < ring->count; i++) { 449 bi = &ring->rx_buffer_info[i]; 450 if (!bi->xdp) 451 continue; 452 453 xsk_buff_free(bi->xdp); 454 bi->xdp = NULL; 455 } 456 } 457 458 /** 459 * igc_clean_rx_ring - Free Rx Buffers per Queue 460 * @ring: ring to free buffers from 461 */ 462 static void igc_clean_rx_ring(struct igc_ring *ring) 463 { 464 if (ring->xsk_pool) 465 igc_clean_rx_ring_xsk_pool(ring); 466 else 467 igc_clean_rx_ring_page_shared(ring); 468 469 clear_ring_uses_large_buffer(ring); 470 471 ring->next_to_alloc = 0; 472 ring->next_to_clean = 0; 473 ring->next_to_use = 0; 474 } 475 476 /** 477 * igc_clean_all_rx_rings - Free Rx Buffers for all queues 478 * @adapter: board private structure 479 */ 480 static void igc_clean_all_rx_rings(struct igc_adapter *adapter) 481 { 482 int i; 483 484 for (i = 0; i < adapter->num_rx_queues; i++) 485 if (adapter->rx_ring[i]) 486 igc_clean_rx_ring(adapter->rx_ring[i]); 487 } 488 489 /** 490 * igc_free_rx_resources - Free Rx Resources 491 * @rx_ring: ring to clean the resources from 492 * 493 * Free all receive software resources 494 */ 495 void igc_free_rx_resources(struct igc_ring *rx_ring) 496 { 497 igc_clean_rx_ring(rx_ring); 498 499 xdp_rxq_info_unreg(&rx_ring->xdp_rxq); 500 501 vfree(rx_ring->rx_buffer_info); 502 rx_ring->rx_buffer_info = NULL; 503 504 /* if not set, then don't free */ 505 if (!rx_ring->desc) 506 return; 507 508 dma_free_coherent(rx_ring->dev, rx_ring->size, 509 rx_ring->desc, rx_ring->dma); 510 511 rx_ring->desc = NULL; 512 } 513 514 /** 515 * igc_free_all_rx_resources - Free Rx Resources for All Queues 516 * @adapter: board private structure 517 * 518 * Free all receive software resources 519 */ 520 static void igc_free_all_rx_resources(struct igc_adapter *adapter) 521 { 522 int i; 523 524 for (i = 0; i < adapter->num_rx_queues; i++) 525 igc_free_rx_resources(adapter->rx_ring[i]); 526 } 527 528 /** 529 * igc_setup_rx_resources - allocate Rx resources (Descriptors) 530 * @rx_ring: rx descriptor ring (for a specific queue) to setup 531 * 532 * Returns 0 on success, negative on failure 533 */ 534 int igc_setup_rx_resources(struct igc_ring *rx_ring) 535 { 536 struct net_device *ndev = rx_ring->netdev; 537 struct device *dev = rx_ring->dev; 538 u8 index = rx_ring->queue_index; 539 int size, desc_len, res; 540 541 /* XDP RX-queue info */ 542 if (xdp_rxq_info_is_reg(&rx_ring->xdp_rxq)) 543 xdp_rxq_info_unreg(&rx_ring->xdp_rxq); 544 res = xdp_rxq_info_reg(&rx_ring->xdp_rxq, ndev, index, 545 rx_ring->q_vector->napi.napi_id); 546 if (res < 0) { 547 netdev_err(ndev, "Failed to register xdp_rxq index %u\n", 548 index); 549 return res; 550 } 551 552 size = sizeof(struct igc_rx_buffer) * rx_ring->count; 553 rx_ring->rx_buffer_info = vzalloc(size); 554 if (!rx_ring->rx_buffer_info) 555 goto err; 556 557 desc_len = sizeof(union igc_adv_rx_desc); 558 559 /* Round up to nearest 4K */ 560 rx_ring->size = rx_ring->count * desc_len; 561 rx_ring->size = ALIGN(rx_ring->size, 4096); 562 563 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size, 564 &rx_ring->dma, GFP_KERNEL); 565 566 if (!rx_ring->desc) 567 goto err; 568 569 rx_ring->next_to_alloc = 0; 570 rx_ring->next_to_clean = 0; 571 rx_ring->next_to_use = 0; 572 573 return 0; 574 575 err: 576 xdp_rxq_info_unreg(&rx_ring->xdp_rxq); 577 vfree(rx_ring->rx_buffer_info); 578 rx_ring->rx_buffer_info = NULL; 579 netdev_err(ndev, "Unable to allocate memory for Rx descriptor ring\n"); 580 return -ENOMEM; 581 } 582 583 /** 584 * igc_setup_all_rx_resources - wrapper to allocate Rx resources 585 * (Descriptors) for all queues 586 * @adapter: board private structure 587 * 588 * Return 0 on success, negative on failure 589 */ 590 static int igc_setup_all_rx_resources(struct igc_adapter *adapter) 591 { 592 struct net_device *dev = adapter->netdev; 593 int i, err = 0; 594 595 for (i = 0; i < adapter->num_rx_queues; i++) { 596 err = igc_setup_rx_resources(adapter->rx_ring[i]); 597 if (err) { 598 netdev_err(dev, "Error on Rx queue %u setup\n", i); 599 for (i--; i >= 0; i--) 600 igc_free_rx_resources(adapter->rx_ring[i]); 601 break; 602 } 603 } 604 605 return err; 606 } 607 608 static struct xsk_buff_pool *igc_get_xsk_pool(struct igc_adapter *adapter, 609 struct igc_ring *ring) 610 { 611 if (!igc_xdp_is_enabled(adapter) || 612 !test_bit(IGC_RING_FLAG_AF_XDP_ZC, &ring->flags)) 613 return NULL; 614 615 return xsk_get_pool_from_qid(ring->netdev, ring->queue_index); 616 } 617 618 /** 619 * igc_configure_rx_ring - Configure a receive ring after Reset 620 * @adapter: board private structure 621 * @ring: receive ring to be configured 622 * 623 * Configure the Rx unit of the MAC after a reset. 624 */ 625 static void igc_configure_rx_ring(struct igc_adapter *adapter, 626 struct igc_ring *ring) 627 { 628 struct igc_hw *hw = &adapter->hw; 629 union igc_adv_rx_desc *rx_desc; 630 int reg_idx = ring->reg_idx; 631 u32 srrctl = 0, rxdctl = 0; 632 u64 rdba = ring->dma; 633 u32 buf_size; 634 635 xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq); 636 ring->xsk_pool = igc_get_xsk_pool(adapter, ring); 637 if (ring->xsk_pool) { 638 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq, 639 MEM_TYPE_XSK_BUFF_POOL, 640 NULL)); 641 xsk_pool_set_rxq_info(ring->xsk_pool, &ring->xdp_rxq); 642 } else { 643 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq, 644 MEM_TYPE_PAGE_SHARED, 645 NULL)); 646 } 647 648 if (igc_xdp_is_enabled(adapter)) 649 set_ring_uses_large_buffer(ring); 650 651 /* disable the queue */ 652 wr32(IGC_RXDCTL(reg_idx), 0); 653 654 /* Set DMA base address registers */ 655 wr32(IGC_RDBAL(reg_idx), 656 rdba & 0x00000000ffffffffULL); 657 wr32(IGC_RDBAH(reg_idx), rdba >> 32); 658 wr32(IGC_RDLEN(reg_idx), 659 ring->count * sizeof(union igc_adv_rx_desc)); 660 661 /* initialize head and tail */ 662 ring->tail = adapter->io_addr + IGC_RDT(reg_idx); 663 wr32(IGC_RDH(reg_idx), 0); 664 writel(0, ring->tail); 665 666 /* reset next-to- use/clean to place SW in sync with hardware */ 667 ring->next_to_clean = 0; 668 ring->next_to_use = 0; 669 670 if (ring->xsk_pool) 671 buf_size = xsk_pool_get_rx_frame_size(ring->xsk_pool); 672 else if (ring_uses_large_buffer(ring)) 673 buf_size = IGC_RXBUFFER_3072; 674 else 675 buf_size = IGC_RXBUFFER_2048; 676 677 srrctl = rd32(IGC_SRRCTL(reg_idx)); 678 srrctl &= ~(IGC_SRRCTL_BSIZEPKT_MASK | IGC_SRRCTL_BSIZEHDR_MASK | 679 IGC_SRRCTL_DESCTYPE_MASK); 680 srrctl |= IGC_SRRCTL_BSIZEHDR(IGC_RX_HDR_LEN); 681 srrctl |= IGC_SRRCTL_BSIZEPKT(buf_size); 682 srrctl |= IGC_SRRCTL_DESCTYPE_ADV_ONEBUF; 683 684 wr32(IGC_SRRCTL(reg_idx), srrctl); 685 686 rxdctl |= IGC_RX_PTHRESH; 687 rxdctl |= IGC_RX_HTHRESH << 8; 688 rxdctl |= IGC_RX_WTHRESH << 16; 689 690 /* initialize rx_buffer_info */ 691 memset(ring->rx_buffer_info, 0, 692 sizeof(struct igc_rx_buffer) * ring->count); 693 694 /* initialize Rx descriptor 0 */ 695 rx_desc = IGC_RX_DESC(ring, 0); 696 rx_desc->wb.upper.length = 0; 697 698 /* enable receive descriptor fetching */ 699 rxdctl |= IGC_RXDCTL_QUEUE_ENABLE; 700 701 wr32(IGC_RXDCTL(reg_idx), rxdctl); 702 } 703 704 /** 705 * igc_configure_rx - Configure receive Unit after Reset 706 * @adapter: board private structure 707 * 708 * Configure the Rx unit of the MAC after a reset. 709 */ 710 static void igc_configure_rx(struct igc_adapter *adapter) 711 { 712 int i; 713 714 /* Setup the HW Rx Head and Tail Descriptor Pointers and 715 * the Base and Length of the Rx Descriptor Ring 716 */ 717 for (i = 0; i < adapter->num_rx_queues; i++) 718 igc_configure_rx_ring(adapter, adapter->rx_ring[i]); 719 } 720 721 /** 722 * igc_configure_tx_ring - Configure transmit ring after Reset 723 * @adapter: board private structure 724 * @ring: tx ring to configure 725 * 726 * Configure a transmit ring after a reset. 727 */ 728 static void igc_configure_tx_ring(struct igc_adapter *adapter, 729 struct igc_ring *ring) 730 { 731 struct igc_hw *hw = &adapter->hw; 732 int reg_idx = ring->reg_idx; 733 u64 tdba = ring->dma; 734 u32 txdctl = 0; 735 736 ring->xsk_pool = igc_get_xsk_pool(adapter, ring); 737 738 /* disable the queue */ 739 wr32(IGC_TXDCTL(reg_idx), 0); 740 wrfl(); 741 742 wr32(IGC_TDLEN(reg_idx), 743 ring->count * sizeof(union igc_adv_tx_desc)); 744 wr32(IGC_TDBAL(reg_idx), 745 tdba & 0x00000000ffffffffULL); 746 wr32(IGC_TDBAH(reg_idx), tdba >> 32); 747 748 ring->tail = adapter->io_addr + IGC_TDT(reg_idx); 749 wr32(IGC_TDH(reg_idx), 0); 750 writel(0, ring->tail); 751 752 txdctl |= IGC_TX_PTHRESH; 753 txdctl |= IGC_TX_HTHRESH << 8; 754 txdctl |= IGC_TX_WTHRESH << 16; 755 756 txdctl |= IGC_TXDCTL_QUEUE_ENABLE; 757 wr32(IGC_TXDCTL(reg_idx), txdctl); 758 } 759 760 /** 761 * igc_configure_tx - Configure transmit Unit after Reset 762 * @adapter: board private structure 763 * 764 * Configure the Tx unit of the MAC after a reset. 765 */ 766 static void igc_configure_tx(struct igc_adapter *adapter) 767 { 768 int i; 769 770 for (i = 0; i < adapter->num_tx_queues; i++) 771 igc_configure_tx_ring(adapter, adapter->tx_ring[i]); 772 } 773 774 /** 775 * igc_setup_mrqc - configure the multiple receive queue control registers 776 * @adapter: Board private structure 777 */ 778 static void igc_setup_mrqc(struct igc_adapter *adapter) 779 { 780 struct igc_hw *hw = &adapter->hw; 781 u32 j, num_rx_queues; 782 u32 mrqc, rxcsum; 783 u32 rss_key[10]; 784 785 netdev_rss_key_fill(rss_key, sizeof(rss_key)); 786 for (j = 0; j < 10; j++) 787 wr32(IGC_RSSRK(j), rss_key[j]); 788 789 num_rx_queues = adapter->rss_queues; 790 791 if (adapter->rss_indir_tbl_init != num_rx_queues) { 792 for (j = 0; j < IGC_RETA_SIZE; j++) 793 adapter->rss_indir_tbl[j] = 794 (j * num_rx_queues) / IGC_RETA_SIZE; 795 adapter->rss_indir_tbl_init = num_rx_queues; 796 } 797 igc_write_rss_indir_tbl(adapter); 798 799 /* Disable raw packet checksumming so that RSS hash is placed in 800 * descriptor on writeback. No need to enable TCP/UDP/IP checksum 801 * offloads as they are enabled by default 802 */ 803 rxcsum = rd32(IGC_RXCSUM); 804 rxcsum |= IGC_RXCSUM_PCSD; 805 806 /* Enable Receive Checksum Offload for SCTP */ 807 rxcsum |= IGC_RXCSUM_CRCOFL; 808 809 /* Don't need to set TUOFL or IPOFL, they default to 1 */ 810 wr32(IGC_RXCSUM, rxcsum); 811 812 /* Generate RSS hash based on packet types, TCP/UDP 813 * port numbers and/or IPv4/v6 src and dst addresses 814 */ 815 mrqc = IGC_MRQC_RSS_FIELD_IPV4 | 816 IGC_MRQC_RSS_FIELD_IPV4_TCP | 817 IGC_MRQC_RSS_FIELD_IPV6 | 818 IGC_MRQC_RSS_FIELD_IPV6_TCP | 819 IGC_MRQC_RSS_FIELD_IPV6_TCP_EX; 820 821 if (adapter->flags & IGC_FLAG_RSS_FIELD_IPV4_UDP) 822 mrqc |= IGC_MRQC_RSS_FIELD_IPV4_UDP; 823 if (adapter->flags & IGC_FLAG_RSS_FIELD_IPV6_UDP) 824 mrqc |= IGC_MRQC_RSS_FIELD_IPV6_UDP; 825 826 mrqc |= IGC_MRQC_ENABLE_RSS_MQ; 827 828 wr32(IGC_MRQC, mrqc); 829 } 830 831 /** 832 * igc_setup_rctl - configure the receive control registers 833 * @adapter: Board private structure 834 */ 835 static void igc_setup_rctl(struct igc_adapter *adapter) 836 { 837 struct igc_hw *hw = &adapter->hw; 838 u32 rctl; 839 840 rctl = rd32(IGC_RCTL); 841 842 rctl &= ~(3 << IGC_RCTL_MO_SHIFT); 843 rctl &= ~(IGC_RCTL_LBM_TCVR | IGC_RCTL_LBM_MAC); 844 845 rctl |= IGC_RCTL_EN | IGC_RCTL_BAM | IGC_RCTL_RDMTS_HALF | 846 (hw->mac.mc_filter_type << IGC_RCTL_MO_SHIFT); 847 848 /* enable stripping of CRC. Newer features require 849 * that the HW strips the CRC. 850 */ 851 rctl |= IGC_RCTL_SECRC; 852 853 /* disable store bad packets and clear size bits. */ 854 rctl &= ~(IGC_RCTL_SBP | IGC_RCTL_SZ_256); 855 856 /* enable LPE to allow for reception of jumbo frames */ 857 rctl |= IGC_RCTL_LPE; 858 859 /* disable queue 0 to prevent tail write w/o re-config */ 860 wr32(IGC_RXDCTL(0), 0); 861 862 /* This is useful for sniffing bad packets. */ 863 if (adapter->netdev->features & NETIF_F_RXALL) { 864 /* UPE and MPE will be handled by normal PROMISC logic 865 * in set_rx_mode 866 */ 867 rctl |= (IGC_RCTL_SBP | /* Receive bad packets */ 868 IGC_RCTL_BAM | /* RX All Bcast Pkts */ 869 IGC_RCTL_PMCF); /* RX All MAC Ctrl Pkts */ 870 871 rctl &= ~(IGC_RCTL_DPF | /* Allow filtered pause */ 872 IGC_RCTL_CFIEN); /* Disable VLAN CFIEN Filter */ 873 } 874 875 wr32(IGC_RCTL, rctl); 876 } 877 878 /** 879 * igc_setup_tctl - configure the transmit control registers 880 * @adapter: Board private structure 881 */ 882 static void igc_setup_tctl(struct igc_adapter *adapter) 883 { 884 struct igc_hw *hw = &adapter->hw; 885 u32 tctl; 886 887 /* disable queue 0 which icould be enabled by default */ 888 wr32(IGC_TXDCTL(0), 0); 889 890 /* Program the Transmit Control Register */ 891 tctl = rd32(IGC_TCTL); 892 tctl &= ~IGC_TCTL_CT; 893 tctl |= IGC_TCTL_PSP | IGC_TCTL_RTLC | 894 (IGC_COLLISION_THRESHOLD << IGC_CT_SHIFT); 895 896 /* Enable transmits */ 897 tctl |= IGC_TCTL_EN; 898 899 wr32(IGC_TCTL, tctl); 900 } 901 902 /** 903 * igc_set_mac_filter_hw() - Set MAC address filter in hardware 904 * @adapter: Pointer to adapter where the filter should be set 905 * @index: Filter index 906 * @type: MAC address filter type (source or destination) 907 * @addr: MAC address 908 * @queue: If non-negative, queue assignment feature is enabled and frames 909 * matching the filter are enqueued onto 'queue'. Otherwise, queue 910 * assignment is disabled. 911 */ 912 static void igc_set_mac_filter_hw(struct igc_adapter *adapter, int index, 913 enum igc_mac_filter_type type, 914 const u8 *addr, int queue) 915 { 916 struct net_device *dev = adapter->netdev; 917 struct igc_hw *hw = &adapter->hw; 918 u32 ral, rah; 919 920 if (WARN_ON(index >= hw->mac.rar_entry_count)) 921 return; 922 923 ral = le32_to_cpup((__le32 *)(addr)); 924 rah = le16_to_cpup((__le16 *)(addr + 4)); 925 926 if (type == IGC_MAC_FILTER_TYPE_SRC) { 927 rah &= ~IGC_RAH_ASEL_MASK; 928 rah |= IGC_RAH_ASEL_SRC_ADDR; 929 } 930 931 if (queue >= 0) { 932 rah &= ~IGC_RAH_QSEL_MASK; 933 rah |= (queue << IGC_RAH_QSEL_SHIFT); 934 rah |= IGC_RAH_QSEL_ENABLE; 935 } 936 937 rah |= IGC_RAH_AV; 938 939 wr32(IGC_RAL(index), ral); 940 wr32(IGC_RAH(index), rah); 941 942 netdev_dbg(dev, "MAC address filter set in HW: index %d", index); 943 } 944 945 /** 946 * igc_clear_mac_filter_hw() - Clear MAC address filter in hardware 947 * @adapter: Pointer to adapter where the filter should be cleared 948 * @index: Filter index 949 */ 950 static void igc_clear_mac_filter_hw(struct igc_adapter *adapter, int index) 951 { 952 struct net_device *dev = adapter->netdev; 953 struct igc_hw *hw = &adapter->hw; 954 955 if (WARN_ON(index >= hw->mac.rar_entry_count)) 956 return; 957 958 wr32(IGC_RAL(index), 0); 959 wr32(IGC_RAH(index), 0); 960 961 netdev_dbg(dev, "MAC address filter cleared in HW: index %d", index); 962 } 963 964 /* Set default MAC address for the PF in the first RAR entry */ 965 static void igc_set_default_mac_filter(struct igc_adapter *adapter) 966 { 967 struct net_device *dev = adapter->netdev; 968 u8 *addr = adapter->hw.mac.addr; 969 970 netdev_dbg(dev, "Set default MAC address filter: address %pM", addr); 971 972 igc_set_mac_filter_hw(adapter, 0, IGC_MAC_FILTER_TYPE_DST, addr, -1); 973 } 974 975 /** 976 * igc_set_mac - Change the Ethernet Address of the NIC 977 * @netdev: network interface device structure 978 * @p: pointer to an address structure 979 * 980 * Returns 0 on success, negative on failure 981 */ 982 static int igc_set_mac(struct net_device *netdev, void *p) 983 { 984 struct igc_adapter *adapter = netdev_priv(netdev); 985 struct igc_hw *hw = &adapter->hw; 986 struct sockaddr *addr = p; 987 988 if (!is_valid_ether_addr(addr->sa_data)) 989 return -EADDRNOTAVAIL; 990 991 eth_hw_addr_set(netdev, addr->sa_data); 992 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len); 993 994 /* set the correct pool for the new PF MAC address in entry 0 */ 995 igc_set_default_mac_filter(adapter); 996 997 return 0; 998 } 999 1000 /** 1001 * igc_write_mc_addr_list - write multicast addresses to MTA 1002 * @netdev: network interface device structure 1003 * 1004 * Writes multicast address list to the MTA hash table. 1005 * Returns: -ENOMEM on failure 1006 * 0 on no addresses written 1007 * X on writing X addresses to MTA 1008 **/ 1009 static int igc_write_mc_addr_list(struct net_device *netdev) 1010 { 1011 struct igc_adapter *adapter = netdev_priv(netdev); 1012 struct igc_hw *hw = &adapter->hw; 1013 struct netdev_hw_addr *ha; 1014 u8 *mta_list; 1015 int i; 1016 1017 if (netdev_mc_empty(netdev)) { 1018 /* nothing to program, so clear mc list */ 1019 igc_update_mc_addr_list(hw, NULL, 0); 1020 return 0; 1021 } 1022 1023 mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC); 1024 if (!mta_list) 1025 return -ENOMEM; 1026 1027 /* The shared function expects a packed array of only addresses. */ 1028 i = 0; 1029 netdev_for_each_mc_addr(ha, netdev) 1030 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN); 1031 1032 igc_update_mc_addr_list(hw, mta_list, i); 1033 kfree(mta_list); 1034 1035 return netdev_mc_count(netdev); 1036 } 1037 1038 static __le32 igc_tx_launchtime(struct igc_ring *ring, ktime_t txtime, 1039 bool *first_flag, bool *insert_empty) 1040 { 1041 struct igc_adapter *adapter = netdev_priv(ring->netdev); 1042 ktime_t cycle_time = adapter->cycle_time; 1043 ktime_t base_time = adapter->base_time; 1044 ktime_t now = ktime_get_clocktai(); 1045 ktime_t baset_est, end_of_cycle; 1046 s32 launchtime; 1047 s64 n; 1048 1049 n = div64_s64(ktime_sub_ns(now, base_time), cycle_time); 1050 1051 baset_est = ktime_add_ns(base_time, cycle_time * (n)); 1052 end_of_cycle = ktime_add_ns(baset_est, cycle_time); 1053 1054 if (ktime_compare(txtime, end_of_cycle) >= 0) { 1055 if (baset_est != ring->last_ff_cycle) { 1056 *first_flag = true; 1057 ring->last_ff_cycle = baset_est; 1058 1059 if (ktime_compare(end_of_cycle, ring->last_tx_cycle) > 0) 1060 *insert_empty = true; 1061 } 1062 } 1063 1064 /* Introducing a window at end of cycle on which packets 1065 * potentially not honor launchtime. Window of 5us chosen 1066 * considering software update the tail pointer and packets 1067 * are dma'ed to packet buffer. 1068 */ 1069 if ((ktime_sub_ns(end_of_cycle, now) < 5 * NSEC_PER_USEC)) 1070 netdev_warn(ring->netdev, "Packet with txtime=%llu may not be honoured\n", 1071 txtime); 1072 1073 ring->last_tx_cycle = end_of_cycle; 1074 1075 launchtime = ktime_sub_ns(txtime, baset_est); 1076 if (launchtime > 0) 1077 div_s64_rem(launchtime, cycle_time, &launchtime); 1078 else 1079 launchtime = 0; 1080 1081 return cpu_to_le32(launchtime); 1082 } 1083 1084 static int igc_init_empty_frame(struct igc_ring *ring, 1085 struct igc_tx_buffer *buffer, 1086 struct sk_buff *skb) 1087 { 1088 unsigned int size; 1089 dma_addr_t dma; 1090 1091 size = skb_headlen(skb); 1092 1093 dma = dma_map_single(ring->dev, skb->data, size, DMA_TO_DEVICE); 1094 if (dma_mapping_error(ring->dev, dma)) { 1095 netdev_err_once(ring->netdev, "Failed to map DMA for TX\n"); 1096 return -ENOMEM; 1097 } 1098 1099 buffer->skb = skb; 1100 buffer->protocol = 0; 1101 buffer->bytecount = skb->len; 1102 buffer->gso_segs = 1; 1103 buffer->time_stamp = jiffies; 1104 dma_unmap_len_set(buffer, len, skb->len); 1105 dma_unmap_addr_set(buffer, dma, dma); 1106 1107 return 0; 1108 } 1109 1110 static int igc_init_tx_empty_descriptor(struct igc_ring *ring, 1111 struct sk_buff *skb, 1112 struct igc_tx_buffer *first) 1113 { 1114 union igc_adv_tx_desc *desc; 1115 u32 cmd_type, olinfo_status; 1116 int err; 1117 1118 if (!igc_desc_unused(ring)) 1119 return -EBUSY; 1120 1121 err = igc_init_empty_frame(ring, first, skb); 1122 if (err) 1123 return err; 1124 1125 cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT | 1126 IGC_ADVTXD_DCMD_IFCS | IGC_TXD_DCMD | 1127 first->bytecount; 1128 olinfo_status = first->bytecount << IGC_ADVTXD_PAYLEN_SHIFT; 1129 1130 desc = IGC_TX_DESC(ring, ring->next_to_use); 1131 desc->read.cmd_type_len = cpu_to_le32(cmd_type); 1132 desc->read.olinfo_status = cpu_to_le32(olinfo_status); 1133 desc->read.buffer_addr = cpu_to_le64(dma_unmap_addr(first, dma)); 1134 1135 netdev_tx_sent_queue(txring_txq(ring), skb->len); 1136 1137 first->next_to_watch = desc; 1138 1139 ring->next_to_use++; 1140 if (ring->next_to_use == ring->count) 1141 ring->next_to_use = 0; 1142 1143 return 0; 1144 } 1145 1146 #define IGC_EMPTY_FRAME_SIZE 60 1147 1148 static void igc_tx_ctxtdesc(struct igc_ring *tx_ring, 1149 __le32 launch_time, bool first_flag, 1150 u32 vlan_macip_lens, u32 type_tucmd, 1151 u32 mss_l4len_idx) 1152 { 1153 struct igc_adv_tx_context_desc *context_desc; 1154 u16 i = tx_ring->next_to_use; 1155 1156 context_desc = IGC_TX_CTXTDESC(tx_ring, i); 1157 1158 i++; 1159 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; 1160 1161 /* set bits to identify this as an advanced context descriptor */ 1162 type_tucmd |= IGC_TXD_CMD_DEXT | IGC_ADVTXD_DTYP_CTXT; 1163 1164 /* For i225, context index must be unique per ring. */ 1165 if (test_bit(IGC_RING_FLAG_TX_CTX_IDX, &tx_ring->flags)) 1166 mss_l4len_idx |= tx_ring->reg_idx << 4; 1167 1168 if (first_flag) 1169 mss_l4len_idx |= IGC_ADVTXD_TSN_CNTX_FIRST; 1170 1171 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens); 1172 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd); 1173 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx); 1174 context_desc->launch_time = launch_time; 1175 } 1176 1177 static void igc_tx_csum(struct igc_ring *tx_ring, struct igc_tx_buffer *first, 1178 __le32 launch_time, bool first_flag) 1179 { 1180 struct sk_buff *skb = first->skb; 1181 u32 vlan_macip_lens = 0; 1182 u32 type_tucmd = 0; 1183 1184 if (skb->ip_summed != CHECKSUM_PARTIAL) { 1185 csum_failed: 1186 if (!(first->tx_flags & IGC_TX_FLAGS_VLAN) && 1187 !tx_ring->launchtime_enable) 1188 return; 1189 goto no_csum; 1190 } 1191 1192 switch (skb->csum_offset) { 1193 case offsetof(struct tcphdr, check): 1194 type_tucmd = IGC_ADVTXD_TUCMD_L4T_TCP; 1195 fallthrough; 1196 case offsetof(struct udphdr, check): 1197 break; 1198 case offsetof(struct sctphdr, checksum): 1199 /* validate that this is actually an SCTP request */ 1200 if (skb_csum_is_sctp(skb)) { 1201 type_tucmd = IGC_ADVTXD_TUCMD_L4T_SCTP; 1202 break; 1203 } 1204 fallthrough; 1205 default: 1206 skb_checksum_help(skb); 1207 goto csum_failed; 1208 } 1209 1210 /* update TX checksum flag */ 1211 first->tx_flags |= IGC_TX_FLAGS_CSUM; 1212 vlan_macip_lens = skb_checksum_start_offset(skb) - 1213 skb_network_offset(skb); 1214 no_csum: 1215 vlan_macip_lens |= skb_network_offset(skb) << IGC_ADVTXD_MACLEN_SHIFT; 1216 vlan_macip_lens |= first->tx_flags & IGC_TX_FLAGS_VLAN_MASK; 1217 1218 igc_tx_ctxtdesc(tx_ring, launch_time, first_flag, 1219 vlan_macip_lens, type_tucmd, 0); 1220 } 1221 1222 static int __igc_maybe_stop_tx(struct igc_ring *tx_ring, const u16 size) 1223 { 1224 struct net_device *netdev = tx_ring->netdev; 1225 1226 netif_stop_subqueue(netdev, tx_ring->queue_index); 1227 1228 /* memory barriier comment */ 1229 smp_mb(); 1230 1231 /* We need to check again in a case another CPU has just 1232 * made room available. 1233 */ 1234 if (igc_desc_unused(tx_ring) < size) 1235 return -EBUSY; 1236 1237 /* A reprieve! */ 1238 netif_wake_subqueue(netdev, tx_ring->queue_index); 1239 1240 u64_stats_update_begin(&tx_ring->tx_syncp2); 1241 tx_ring->tx_stats.restart_queue2++; 1242 u64_stats_update_end(&tx_ring->tx_syncp2); 1243 1244 return 0; 1245 } 1246 1247 static inline int igc_maybe_stop_tx(struct igc_ring *tx_ring, const u16 size) 1248 { 1249 if (igc_desc_unused(tx_ring) >= size) 1250 return 0; 1251 return __igc_maybe_stop_tx(tx_ring, size); 1252 } 1253 1254 #define IGC_SET_FLAG(_input, _flag, _result) \ 1255 (((_flag) <= (_result)) ? \ 1256 ((u32)((_input) & (_flag)) * ((_result) / (_flag))) : \ 1257 ((u32)((_input) & (_flag)) / ((_flag) / (_result)))) 1258 1259 static u32 igc_tx_cmd_type(struct sk_buff *skb, u32 tx_flags) 1260 { 1261 /* set type for advanced descriptor with frame checksum insertion */ 1262 u32 cmd_type = IGC_ADVTXD_DTYP_DATA | 1263 IGC_ADVTXD_DCMD_DEXT | 1264 IGC_ADVTXD_DCMD_IFCS; 1265 1266 /* set HW vlan bit if vlan is present */ 1267 cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_VLAN, 1268 IGC_ADVTXD_DCMD_VLE); 1269 1270 /* set segmentation bits for TSO */ 1271 cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSO, 1272 (IGC_ADVTXD_DCMD_TSE)); 1273 1274 /* set timestamp bit if present, will select the register set 1275 * based on the _TSTAMP(_X) bit. 1276 */ 1277 cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP, 1278 (IGC_ADVTXD_MAC_TSTAMP)); 1279 1280 cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP_1, 1281 (IGC_ADVTXD_TSTAMP_REG_1)); 1282 1283 cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP_2, 1284 (IGC_ADVTXD_TSTAMP_REG_2)); 1285 1286 cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP_3, 1287 (IGC_ADVTXD_TSTAMP_REG_3)); 1288 1289 /* insert frame checksum */ 1290 cmd_type ^= IGC_SET_FLAG(skb->no_fcs, 1, IGC_ADVTXD_DCMD_IFCS); 1291 1292 return cmd_type; 1293 } 1294 1295 static void igc_tx_olinfo_status(struct igc_ring *tx_ring, 1296 union igc_adv_tx_desc *tx_desc, 1297 u32 tx_flags, unsigned int paylen) 1298 { 1299 u32 olinfo_status = paylen << IGC_ADVTXD_PAYLEN_SHIFT; 1300 1301 /* insert L4 checksum */ 1302 olinfo_status |= (tx_flags & IGC_TX_FLAGS_CSUM) * 1303 ((IGC_TXD_POPTS_TXSM << 8) / 1304 IGC_TX_FLAGS_CSUM); 1305 1306 /* insert IPv4 checksum */ 1307 olinfo_status |= (tx_flags & IGC_TX_FLAGS_IPV4) * 1308 (((IGC_TXD_POPTS_IXSM << 8)) / 1309 IGC_TX_FLAGS_IPV4); 1310 1311 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status); 1312 } 1313 1314 static int igc_tx_map(struct igc_ring *tx_ring, 1315 struct igc_tx_buffer *first, 1316 const u8 hdr_len) 1317 { 1318 struct sk_buff *skb = first->skb; 1319 struct igc_tx_buffer *tx_buffer; 1320 union igc_adv_tx_desc *tx_desc; 1321 u32 tx_flags = first->tx_flags; 1322 skb_frag_t *frag; 1323 u16 i = tx_ring->next_to_use; 1324 unsigned int data_len, size; 1325 dma_addr_t dma; 1326 u32 cmd_type; 1327 1328 cmd_type = igc_tx_cmd_type(skb, tx_flags); 1329 tx_desc = IGC_TX_DESC(tx_ring, i); 1330 1331 igc_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len); 1332 1333 size = skb_headlen(skb); 1334 data_len = skb->data_len; 1335 1336 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE); 1337 1338 tx_buffer = first; 1339 1340 for (frag = &skb_shinfo(skb)->frags[0];; frag++) { 1341 if (dma_mapping_error(tx_ring->dev, dma)) 1342 goto dma_error; 1343 1344 /* record length, and DMA address */ 1345 dma_unmap_len_set(tx_buffer, len, size); 1346 dma_unmap_addr_set(tx_buffer, dma, dma); 1347 1348 tx_desc->read.buffer_addr = cpu_to_le64(dma); 1349 1350 while (unlikely(size > IGC_MAX_DATA_PER_TXD)) { 1351 tx_desc->read.cmd_type_len = 1352 cpu_to_le32(cmd_type ^ IGC_MAX_DATA_PER_TXD); 1353 1354 i++; 1355 tx_desc++; 1356 if (i == tx_ring->count) { 1357 tx_desc = IGC_TX_DESC(tx_ring, 0); 1358 i = 0; 1359 } 1360 tx_desc->read.olinfo_status = 0; 1361 1362 dma += IGC_MAX_DATA_PER_TXD; 1363 size -= IGC_MAX_DATA_PER_TXD; 1364 1365 tx_desc->read.buffer_addr = cpu_to_le64(dma); 1366 } 1367 1368 if (likely(!data_len)) 1369 break; 1370 1371 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size); 1372 1373 i++; 1374 tx_desc++; 1375 if (i == tx_ring->count) { 1376 tx_desc = IGC_TX_DESC(tx_ring, 0); 1377 i = 0; 1378 } 1379 tx_desc->read.olinfo_status = 0; 1380 1381 size = skb_frag_size(frag); 1382 data_len -= size; 1383 1384 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, 1385 size, DMA_TO_DEVICE); 1386 1387 tx_buffer = &tx_ring->tx_buffer_info[i]; 1388 } 1389 1390 /* write last descriptor with RS and EOP bits */ 1391 cmd_type |= size | IGC_TXD_DCMD; 1392 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type); 1393 1394 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount); 1395 1396 /* set the timestamp */ 1397 first->time_stamp = jiffies; 1398 1399 skb_tx_timestamp(skb); 1400 1401 /* Force memory writes to complete before letting h/w know there 1402 * are new descriptors to fetch. (Only applicable for weak-ordered 1403 * memory model archs, such as IA-64). 1404 * 1405 * We also need this memory barrier to make certain all of the 1406 * status bits have been updated before next_to_watch is written. 1407 */ 1408 wmb(); 1409 1410 /* set next_to_watch value indicating a packet is present */ 1411 first->next_to_watch = tx_desc; 1412 1413 i++; 1414 if (i == tx_ring->count) 1415 i = 0; 1416 1417 tx_ring->next_to_use = i; 1418 1419 /* Make sure there is space in the ring for the next send. */ 1420 igc_maybe_stop_tx(tx_ring, DESC_NEEDED); 1421 1422 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) { 1423 writel(i, tx_ring->tail); 1424 } 1425 1426 return 0; 1427 dma_error: 1428 netdev_err(tx_ring->netdev, "TX DMA map failed\n"); 1429 tx_buffer = &tx_ring->tx_buffer_info[i]; 1430 1431 /* clear dma mappings for failed tx_buffer_info map */ 1432 while (tx_buffer != first) { 1433 if (dma_unmap_len(tx_buffer, len)) 1434 igc_unmap_tx_buffer(tx_ring->dev, tx_buffer); 1435 1436 if (i-- == 0) 1437 i += tx_ring->count; 1438 tx_buffer = &tx_ring->tx_buffer_info[i]; 1439 } 1440 1441 if (dma_unmap_len(tx_buffer, len)) 1442 igc_unmap_tx_buffer(tx_ring->dev, tx_buffer); 1443 1444 dev_kfree_skb_any(tx_buffer->skb); 1445 tx_buffer->skb = NULL; 1446 1447 tx_ring->next_to_use = i; 1448 1449 return -1; 1450 } 1451 1452 static int igc_tso(struct igc_ring *tx_ring, 1453 struct igc_tx_buffer *first, 1454 __le32 launch_time, bool first_flag, 1455 u8 *hdr_len) 1456 { 1457 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx; 1458 struct sk_buff *skb = first->skb; 1459 union { 1460 struct iphdr *v4; 1461 struct ipv6hdr *v6; 1462 unsigned char *hdr; 1463 } ip; 1464 union { 1465 struct tcphdr *tcp; 1466 struct udphdr *udp; 1467 unsigned char *hdr; 1468 } l4; 1469 u32 paylen, l4_offset; 1470 int err; 1471 1472 if (skb->ip_summed != CHECKSUM_PARTIAL) 1473 return 0; 1474 1475 if (!skb_is_gso(skb)) 1476 return 0; 1477 1478 err = skb_cow_head(skb, 0); 1479 if (err < 0) 1480 return err; 1481 1482 ip.hdr = skb_network_header(skb); 1483 l4.hdr = skb_checksum_start(skb); 1484 1485 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */ 1486 type_tucmd = IGC_ADVTXD_TUCMD_L4T_TCP; 1487 1488 /* initialize outer IP header fields */ 1489 if (ip.v4->version == 4) { 1490 unsigned char *csum_start = skb_checksum_start(skb); 1491 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4); 1492 1493 /* IP header will have to cancel out any data that 1494 * is not a part of the outer IP header 1495 */ 1496 ip.v4->check = csum_fold(csum_partial(trans_start, 1497 csum_start - trans_start, 1498 0)); 1499 type_tucmd |= IGC_ADVTXD_TUCMD_IPV4; 1500 1501 ip.v4->tot_len = 0; 1502 first->tx_flags |= IGC_TX_FLAGS_TSO | 1503 IGC_TX_FLAGS_CSUM | 1504 IGC_TX_FLAGS_IPV4; 1505 } else { 1506 ip.v6->payload_len = 0; 1507 first->tx_flags |= IGC_TX_FLAGS_TSO | 1508 IGC_TX_FLAGS_CSUM; 1509 } 1510 1511 /* determine offset of inner transport header */ 1512 l4_offset = l4.hdr - skb->data; 1513 1514 /* remove payload length from inner checksum */ 1515 paylen = skb->len - l4_offset; 1516 if (type_tucmd & IGC_ADVTXD_TUCMD_L4T_TCP) { 1517 /* compute length of segmentation header */ 1518 *hdr_len = (l4.tcp->doff * 4) + l4_offset; 1519 csum_replace_by_diff(&l4.tcp->check, 1520 (__force __wsum)htonl(paylen)); 1521 } else { 1522 /* compute length of segmentation header */ 1523 *hdr_len = sizeof(*l4.udp) + l4_offset; 1524 csum_replace_by_diff(&l4.udp->check, 1525 (__force __wsum)htonl(paylen)); 1526 } 1527 1528 /* update gso size and bytecount with header size */ 1529 first->gso_segs = skb_shinfo(skb)->gso_segs; 1530 first->bytecount += (first->gso_segs - 1) * *hdr_len; 1531 1532 /* MSS L4LEN IDX */ 1533 mss_l4len_idx = (*hdr_len - l4_offset) << IGC_ADVTXD_L4LEN_SHIFT; 1534 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IGC_ADVTXD_MSS_SHIFT; 1535 1536 /* VLAN MACLEN IPLEN */ 1537 vlan_macip_lens = l4.hdr - ip.hdr; 1538 vlan_macip_lens |= (ip.hdr - skb->data) << IGC_ADVTXD_MACLEN_SHIFT; 1539 vlan_macip_lens |= first->tx_flags & IGC_TX_FLAGS_VLAN_MASK; 1540 1541 igc_tx_ctxtdesc(tx_ring, launch_time, first_flag, 1542 vlan_macip_lens, type_tucmd, mss_l4len_idx); 1543 1544 return 1; 1545 } 1546 1547 static bool igc_request_tx_tstamp(struct igc_adapter *adapter, struct sk_buff *skb, u32 *flags) 1548 { 1549 int i; 1550 1551 for (i = 0; i < IGC_MAX_TX_TSTAMP_REGS; i++) { 1552 struct igc_tx_timestamp_request *tstamp = &adapter->tx_tstamp[i]; 1553 1554 if (tstamp->skb) 1555 continue; 1556 1557 tstamp->skb = skb_get(skb); 1558 tstamp->start = jiffies; 1559 *flags = tstamp->flags; 1560 1561 return true; 1562 } 1563 1564 return false; 1565 } 1566 1567 static netdev_tx_t igc_xmit_frame_ring(struct sk_buff *skb, 1568 struct igc_ring *tx_ring) 1569 { 1570 struct igc_adapter *adapter = netdev_priv(tx_ring->netdev); 1571 bool first_flag = false, insert_empty = false; 1572 u16 count = TXD_USE_COUNT(skb_headlen(skb)); 1573 __be16 protocol = vlan_get_protocol(skb); 1574 struct igc_tx_buffer *first; 1575 __le32 launch_time = 0; 1576 u32 tx_flags = 0; 1577 unsigned short f; 1578 ktime_t txtime; 1579 u8 hdr_len = 0; 1580 int tso = 0; 1581 1582 /* need: 1 descriptor per page * PAGE_SIZE/IGC_MAX_DATA_PER_TXD, 1583 * + 1 desc for skb_headlen/IGC_MAX_DATA_PER_TXD, 1584 * + 2 desc gap to keep tail from touching head, 1585 * + 1 desc for context descriptor, 1586 * otherwise try next time 1587 */ 1588 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) 1589 count += TXD_USE_COUNT(skb_frag_size( 1590 &skb_shinfo(skb)->frags[f])); 1591 1592 if (igc_maybe_stop_tx(tx_ring, count + 5)) { 1593 /* this is a hard error */ 1594 return NETDEV_TX_BUSY; 1595 } 1596 1597 if (!tx_ring->launchtime_enable) 1598 goto done; 1599 1600 txtime = skb->tstamp; 1601 skb->tstamp = ktime_set(0, 0); 1602 launch_time = igc_tx_launchtime(tx_ring, txtime, &first_flag, &insert_empty); 1603 1604 if (insert_empty) { 1605 struct igc_tx_buffer *empty_info; 1606 struct sk_buff *empty; 1607 void *data; 1608 1609 empty_info = &tx_ring->tx_buffer_info[tx_ring->next_to_use]; 1610 empty = alloc_skb(IGC_EMPTY_FRAME_SIZE, GFP_ATOMIC); 1611 if (!empty) 1612 goto done; 1613 1614 data = skb_put(empty, IGC_EMPTY_FRAME_SIZE); 1615 memset(data, 0, IGC_EMPTY_FRAME_SIZE); 1616 1617 igc_tx_ctxtdesc(tx_ring, 0, false, 0, 0, 0); 1618 1619 if (igc_init_tx_empty_descriptor(tx_ring, 1620 empty, 1621 empty_info) < 0) 1622 dev_kfree_skb_any(empty); 1623 } 1624 1625 done: 1626 /* record the location of the first descriptor for this packet */ 1627 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use]; 1628 first->type = IGC_TX_BUFFER_TYPE_SKB; 1629 first->skb = skb; 1630 first->bytecount = skb->len; 1631 first->gso_segs = 1; 1632 1633 if (adapter->qbv_transition || tx_ring->oper_gate_closed) 1634 goto out_drop; 1635 1636 if (tx_ring->max_sdu > 0 && first->bytecount > tx_ring->max_sdu) { 1637 adapter->stats.txdrop++; 1638 goto out_drop; 1639 } 1640 1641 if (unlikely(test_bit(IGC_RING_FLAG_TX_HWTSTAMP, &tx_ring->flags) && 1642 skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) { 1643 unsigned long flags; 1644 u32 tstamp_flags; 1645 1646 spin_lock_irqsave(&adapter->ptp_tx_lock, flags); 1647 if (igc_request_tx_tstamp(adapter, skb, &tstamp_flags)) { 1648 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS; 1649 tx_flags |= IGC_TX_FLAGS_TSTAMP | tstamp_flags; 1650 } else { 1651 adapter->tx_hwtstamp_skipped++; 1652 } 1653 1654 spin_unlock_irqrestore(&adapter->ptp_tx_lock, flags); 1655 } 1656 1657 if (skb_vlan_tag_present(skb)) { 1658 tx_flags |= IGC_TX_FLAGS_VLAN; 1659 tx_flags |= (skb_vlan_tag_get(skb) << IGC_TX_FLAGS_VLAN_SHIFT); 1660 } 1661 1662 /* record initial flags and protocol */ 1663 first->tx_flags = tx_flags; 1664 first->protocol = protocol; 1665 1666 tso = igc_tso(tx_ring, first, launch_time, first_flag, &hdr_len); 1667 if (tso < 0) 1668 goto out_drop; 1669 else if (!tso) 1670 igc_tx_csum(tx_ring, first, launch_time, first_flag); 1671 1672 igc_tx_map(tx_ring, first, hdr_len); 1673 1674 return NETDEV_TX_OK; 1675 1676 out_drop: 1677 dev_kfree_skb_any(first->skb); 1678 first->skb = NULL; 1679 1680 return NETDEV_TX_OK; 1681 } 1682 1683 static inline struct igc_ring *igc_tx_queue_mapping(struct igc_adapter *adapter, 1684 struct sk_buff *skb) 1685 { 1686 unsigned int r_idx = skb->queue_mapping; 1687 1688 if (r_idx >= adapter->num_tx_queues) 1689 r_idx = r_idx % adapter->num_tx_queues; 1690 1691 return adapter->tx_ring[r_idx]; 1692 } 1693 1694 static netdev_tx_t igc_xmit_frame(struct sk_buff *skb, 1695 struct net_device *netdev) 1696 { 1697 struct igc_adapter *adapter = netdev_priv(netdev); 1698 1699 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb 1700 * in order to meet this minimum size requirement. 1701 */ 1702 if (skb->len < 17) { 1703 if (skb_padto(skb, 17)) 1704 return NETDEV_TX_OK; 1705 skb->len = 17; 1706 } 1707 1708 return igc_xmit_frame_ring(skb, igc_tx_queue_mapping(adapter, skb)); 1709 } 1710 1711 static void igc_rx_checksum(struct igc_ring *ring, 1712 union igc_adv_rx_desc *rx_desc, 1713 struct sk_buff *skb) 1714 { 1715 skb_checksum_none_assert(skb); 1716 1717 /* Ignore Checksum bit is set */ 1718 if (igc_test_staterr(rx_desc, IGC_RXD_STAT_IXSM)) 1719 return; 1720 1721 /* Rx checksum disabled via ethtool */ 1722 if (!(ring->netdev->features & NETIF_F_RXCSUM)) 1723 return; 1724 1725 /* TCP/UDP checksum error bit is set */ 1726 if (igc_test_staterr(rx_desc, 1727 IGC_RXDEXT_STATERR_L4E | 1728 IGC_RXDEXT_STATERR_IPE)) { 1729 /* work around errata with sctp packets where the TCPE aka 1730 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc) 1731 * packets (aka let the stack check the crc32c) 1732 */ 1733 if (!(skb->len == 60 && 1734 test_bit(IGC_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) { 1735 u64_stats_update_begin(&ring->rx_syncp); 1736 ring->rx_stats.csum_err++; 1737 u64_stats_update_end(&ring->rx_syncp); 1738 } 1739 /* let the stack verify checksum errors */ 1740 return; 1741 } 1742 /* It must be a TCP or UDP packet with a valid checksum */ 1743 if (igc_test_staterr(rx_desc, IGC_RXD_STAT_TCPCS | 1744 IGC_RXD_STAT_UDPCS)) 1745 skb->ip_summed = CHECKSUM_UNNECESSARY; 1746 1747 netdev_dbg(ring->netdev, "cksum success: bits %08X\n", 1748 le32_to_cpu(rx_desc->wb.upper.status_error)); 1749 } 1750 1751 /* Mapping HW RSS Type to enum pkt_hash_types */ 1752 static const enum pkt_hash_types igc_rss_type_table[IGC_RSS_TYPE_MAX_TABLE] = { 1753 [IGC_RSS_TYPE_NO_HASH] = PKT_HASH_TYPE_L2, 1754 [IGC_RSS_TYPE_HASH_TCP_IPV4] = PKT_HASH_TYPE_L4, 1755 [IGC_RSS_TYPE_HASH_IPV4] = PKT_HASH_TYPE_L3, 1756 [IGC_RSS_TYPE_HASH_TCP_IPV6] = PKT_HASH_TYPE_L4, 1757 [IGC_RSS_TYPE_HASH_IPV6_EX] = PKT_HASH_TYPE_L3, 1758 [IGC_RSS_TYPE_HASH_IPV6] = PKT_HASH_TYPE_L3, 1759 [IGC_RSS_TYPE_HASH_TCP_IPV6_EX] = PKT_HASH_TYPE_L4, 1760 [IGC_RSS_TYPE_HASH_UDP_IPV4] = PKT_HASH_TYPE_L4, 1761 [IGC_RSS_TYPE_HASH_UDP_IPV6] = PKT_HASH_TYPE_L4, 1762 [IGC_RSS_TYPE_HASH_UDP_IPV6_EX] = PKT_HASH_TYPE_L4, 1763 [10] = PKT_HASH_TYPE_NONE, /* RSS Type above 9 "Reserved" by HW */ 1764 [11] = PKT_HASH_TYPE_NONE, /* keep array sized for SW bit-mask */ 1765 [12] = PKT_HASH_TYPE_NONE, /* to handle future HW revisons */ 1766 [13] = PKT_HASH_TYPE_NONE, 1767 [14] = PKT_HASH_TYPE_NONE, 1768 [15] = PKT_HASH_TYPE_NONE, 1769 }; 1770 1771 static inline void igc_rx_hash(struct igc_ring *ring, 1772 union igc_adv_rx_desc *rx_desc, 1773 struct sk_buff *skb) 1774 { 1775 if (ring->netdev->features & NETIF_F_RXHASH) { 1776 u32 rss_hash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss); 1777 u32 rss_type = igc_rss_type(rx_desc); 1778 1779 skb_set_hash(skb, rss_hash, igc_rss_type_table[rss_type]); 1780 } 1781 } 1782 1783 static void igc_rx_vlan(struct igc_ring *rx_ring, 1784 union igc_adv_rx_desc *rx_desc, 1785 struct sk_buff *skb) 1786 { 1787 struct net_device *dev = rx_ring->netdev; 1788 u16 vid; 1789 1790 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) && 1791 igc_test_staterr(rx_desc, IGC_RXD_STAT_VP)) { 1792 if (igc_test_staterr(rx_desc, IGC_RXDEXT_STATERR_LB) && 1793 test_bit(IGC_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags)) 1794 vid = be16_to_cpu((__force __be16)rx_desc->wb.upper.vlan); 1795 else 1796 vid = le16_to_cpu(rx_desc->wb.upper.vlan); 1797 1798 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid); 1799 } 1800 } 1801 1802 /** 1803 * igc_process_skb_fields - Populate skb header fields from Rx descriptor 1804 * @rx_ring: rx descriptor ring packet is being transacted on 1805 * @rx_desc: pointer to the EOP Rx descriptor 1806 * @skb: pointer to current skb being populated 1807 * 1808 * This function checks the ring, descriptor, and packet information in order 1809 * to populate the hash, checksum, VLAN, protocol, and other fields within the 1810 * skb. 1811 */ 1812 static void igc_process_skb_fields(struct igc_ring *rx_ring, 1813 union igc_adv_rx_desc *rx_desc, 1814 struct sk_buff *skb) 1815 { 1816 igc_rx_hash(rx_ring, rx_desc, skb); 1817 1818 igc_rx_checksum(rx_ring, rx_desc, skb); 1819 1820 igc_rx_vlan(rx_ring, rx_desc, skb); 1821 1822 skb_record_rx_queue(skb, rx_ring->queue_index); 1823 1824 skb->protocol = eth_type_trans(skb, rx_ring->netdev); 1825 } 1826 1827 static void igc_vlan_mode(struct net_device *netdev, netdev_features_t features) 1828 { 1829 bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX); 1830 struct igc_adapter *adapter = netdev_priv(netdev); 1831 struct igc_hw *hw = &adapter->hw; 1832 u32 ctrl; 1833 1834 ctrl = rd32(IGC_CTRL); 1835 1836 if (enable) { 1837 /* enable VLAN tag insert/strip */ 1838 ctrl |= IGC_CTRL_VME; 1839 } else { 1840 /* disable VLAN tag insert/strip */ 1841 ctrl &= ~IGC_CTRL_VME; 1842 } 1843 wr32(IGC_CTRL, ctrl); 1844 } 1845 1846 static void igc_restore_vlan(struct igc_adapter *adapter) 1847 { 1848 igc_vlan_mode(adapter->netdev, adapter->netdev->features); 1849 } 1850 1851 static struct igc_rx_buffer *igc_get_rx_buffer(struct igc_ring *rx_ring, 1852 const unsigned int size, 1853 int *rx_buffer_pgcnt) 1854 { 1855 struct igc_rx_buffer *rx_buffer; 1856 1857 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean]; 1858 *rx_buffer_pgcnt = 1859 #if (PAGE_SIZE < 8192) 1860 page_count(rx_buffer->page); 1861 #else 1862 0; 1863 #endif 1864 prefetchw(rx_buffer->page); 1865 1866 /* we are reusing so sync this buffer for CPU use */ 1867 dma_sync_single_range_for_cpu(rx_ring->dev, 1868 rx_buffer->dma, 1869 rx_buffer->page_offset, 1870 size, 1871 DMA_FROM_DEVICE); 1872 1873 rx_buffer->pagecnt_bias--; 1874 1875 return rx_buffer; 1876 } 1877 1878 static void igc_rx_buffer_flip(struct igc_rx_buffer *buffer, 1879 unsigned int truesize) 1880 { 1881 #if (PAGE_SIZE < 8192) 1882 buffer->page_offset ^= truesize; 1883 #else 1884 buffer->page_offset += truesize; 1885 #endif 1886 } 1887 1888 static unsigned int igc_get_rx_frame_truesize(struct igc_ring *ring, 1889 unsigned int size) 1890 { 1891 unsigned int truesize; 1892 1893 #if (PAGE_SIZE < 8192) 1894 truesize = igc_rx_pg_size(ring) / 2; 1895 #else 1896 truesize = ring_uses_build_skb(ring) ? 1897 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) + 1898 SKB_DATA_ALIGN(IGC_SKB_PAD + size) : 1899 SKB_DATA_ALIGN(size); 1900 #endif 1901 return truesize; 1902 } 1903 1904 /** 1905 * igc_add_rx_frag - Add contents of Rx buffer to sk_buff 1906 * @rx_ring: rx descriptor ring to transact packets on 1907 * @rx_buffer: buffer containing page to add 1908 * @skb: sk_buff to place the data into 1909 * @size: size of buffer to be added 1910 * 1911 * This function will add the data contained in rx_buffer->page to the skb. 1912 */ 1913 static void igc_add_rx_frag(struct igc_ring *rx_ring, 1914 struct igc_rx_buffer *rx_buffer, 1915 struct sk_buff *skb, 1916 unsigned int size) 1917 { 1918 unsigned int truesize; 1919 1920 #if (PAGE_SIZE < 8192) 1921 truesize = igc_rx_pg_size(rx_ring) / 2; 1922 #else 1923 truesize = ring_uses_build_skb(rx_ring) ? 1924 SKB_DATA_ALIGN(IGC_SKB_PAD + size) : 1925 SKB_DATA_ALIGN(size); 1926 #endif 1927 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page, 1928 rx_buffer->page_offset, size, truesize); 1929 1930 igc_rx_buffer_flip(rx_buffer, truesize); 1931 } 1932 1933 static struct sk_buff *igc_build_skb(struct igc_ring *rx_ring, 1934 struct igc_rx_buffer *rx_buffer, 1935 struct xdp_buff *xdp) 1936 { 1937 unsigned int size = xdp->data_end - xdp->data; 1938 unsigned int truesize = igc_get_rx_frame_truesize(rx_ring, size); 1939 unsigned int metasize = xdp->data - xdp->data_meta; 1940 struct sk_buff *skb; 1941 1942 /* prefetch first cache line of first page */ 1943 net_prefetch(xdp->data_meta); 1944 1945 /* build an skb around the page buffer */ 1946 skb = napi_build_skb(xdp->data_hard_start, truesize); 1947 if (unlikely(!skb)) 1948 return NULL; 1949 1950 /* update pointers within the skb to store the data */ 1951 skb_reserve(skb, xdp->data - xdp->data_hard_start); 1952 __skb_put(skb, size); 1953 if (metasize) 1954 skb_metadata_set(skb, metasize); 1955 1956 igc_rx_buffer_flip(rx_buffer, truesize); 1957 return skb; 1958 } 1959 1960 static struct sk_buff *igc_construct_skb(struct igc_ring *rx_ring, 1961 struct igc_rx_buffer *rx_buffer, 1962 struct xdp_buff *xdp, 1963 ktime_t timestamp) 1964 { 1965 unsigned int metasize = xdp->data - xdp->data_meta; 1966 unsigned int size = xdp->data_end - xdp->data; 1967 unsigned int truesize = igc_get_rx_frame_truesize(rx_ring, size); 1968 void *va = xdp->data; 1969 unsigned int headlen; 1970 struct sk_buff *skb; 1971 1972 /* prefetch first cache line of first page */ 1973 net_prefetch(xdp->data_meta); 1974 1975 /* allocate a skb to store the frags */ 1976 skb = napi_alloc_skb(&rx_ring->q_vector->napi, 1977 IGC_RX_HDR_LEN + metasize); 1978 if (unlikely(!skb)) 1979 return NULL; 1980 1981 if (timestamp) 1982 skb_hwtstamps(skb)->hwtstamp = timestamp; 1983 1984 /* Determine available headroom for copy */ 1985 headlen = size; 1986 if (headlen > IGC_RX_HDR_LEN) 1987 headlen = eth_get_headlen(skb->dev, va, IGC_RX_HDR_LEN); 1988 1989 /* align pull length to size of long to optimize memcpy performance */ 1990 memcpy(__skb_put(skb, headlen + metasize), xdp->data_meta, 1991 ALIGN(headlen + metasize, sizeof(long))); 1992 1993 if (metasize) { 1994 skb_metadata_set(skb, metasize); 1995 __skb_pull(skb, metasize); 1996 } 1997 1998 /* update all of the pointers */ 1999 size -= headlen; 2000 if (size) { 2001 skb_add_rx_frag(skb, 0, rx_buffer->page, 2002 (va + headlen) - page_address(rx_buffer->page), 2003 size, truesize); 2004 igc_rx_buffer_flip(rx_buffer, truesize); 2005 } else { 2006 rx_buffer->pagecnt_bias++; 2007 } 2008 2009 return skb; 2010 } 2011 2012 /** 2013 * igc_reuse_rx_page - page flip buffer and store it back on the ring 2014 * @rx_ring: rx descriptor ring to store buffers on 2015 * @old_buff: donor buffer to have page reused 2016 * 2017 * Synchronizes page for reuse by the adapter 2018 */ 2019 static void igc_reuse_rx_page(struct igc_ring *rx_ring, 2020 struct igc_rx_buffer *old_buff) 2021 { 2022 u16 nta = rx_ring->next_to_alloc; 2023 struct igc_rx_buffer *new_buff; 2024 2025 new_buff = &rx_ring->rx_buffer_info[nta]; 2026 2027 /* update, and store next to alloc */ 2028 nta++; 2029 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0; 2030 2031 /* Transfer page from old buffer to new buffer. 2032 * Move each member individually to avoid possible store 2033 * forwarding stalls. 2034 */ 2035 new_buff->dma = old_buff->dma; 2036 new_buff->page = old_buff->page; 2037 new_buff->page_offset = old_buff->page_offset; 2038 new_buff->pagecnt_bias = old_buff->pagecnt_bias; 2039 } 2040 2041 static bool igc_can_reuse_rx_page(struct igc_rx_buffer *rx_buffer, 2042 int rx_buffer_pgcnt) 2043 { 2044 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias; 2045 struct page *page = rx_buffer->page; 2046 2047 /* avoid re-using remote and pfmemalloc pages */ 2048 if (!dev_page_is_reusable(page)) 2049 return false; 2050 2051 #if (PAGE_SIZE < 8192) 2052 /* if we are only owner of page we can reuse it */ 2053 if (unlikely((rx_buffer_pgcnt - pagecnt_bias) > 1)) 2054 return false; 2055 #else 2056 #define IGC_LAST_OFFSET \ 2057 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IGC_RXBUFFER_2048) 2058 2059 if (rx_buffer->page_offset > IGC_LAST_OFFSET) 2060 return false; 2061 #endif 2062 2063 /* If we have drained the page fragment pool we need to update 2064 * the pagecnt_bias and page count so that we fully restock the 2065 * number of references the driver holds. 2066 */ 2067 if (unlikely(pagecnt_bias == 1)) { 2068 page_ref_add(page, USHRT_MAX - 1); 2069 rx_buffer->pagecnt_bias = USHRT_MAX; 2070 } 2071 2072 return true; 2073 } 2074 2075 /** 2076 * igc_is_non_eop - process handling of non-EOP buffers 2077 * @rx_ring: Rx ring being processed 2078 * @rx_desc: Rx descriptor for current buffer 2079 * 2080 * This function updates next to clean. If the buffer is an EOP buffer 2081 * this function exits returning false, otherwise it will place the 2082 * sk_buff in the next buffer to be chained and return true indicating 2083 * that this is in fact a non-EOP buffer. 2084 */ 2085 static bool igc_is_non_eop(struct igc_ring *rx_ring, 2086 union igc_adv_rx_desc *rx_desc) 2087 { 2088 u32 ntc = rx_ring->next_to_clean + 1; 2089 2090 /* fetch, update, and store next to clean */ 2091 ntc = (ntc < rx_ring->count) ? ntc : 0; 2092 rx_ring->next_to_clean = ntc; 2093 2094 prefetch(IGC_RX_DESC(rx_ring, ntc)); 2095 2096 if (likely(igc_test_staterr(rx_desc, IGC_RXD_STAT_EOP))) 2097 return false; 2098 2099 return true; 2100 } 2101 2102 /** 2103 * igc_cleanup_headers - Correct corrupted or empty headers 2104 * @rx_ring: rx descriptor ring packet is being transacted on 2105 * @rx_desc: pointer to the EOP Rx descriptor 2106 * @skb: pointer to current skb being fixed 2107 * 2108 * Address the case where we are pulling data in on pages only 2109 * and as such no data is present in the skb header. 2110 * 2111 * In addition if skb is not at least 60 bytes we need to pad it so that 2112 * it is large enough to qualify as a valid Ethernet frame. 2113 * 2114 * Returns true if an error was encountered and skb was freed. 2115 */ 2116 static bool igc_cleanup_headers(struct igc_ring *rx_ring, 2117 union igc_adv_rx_desc *rx_desc, 2118 struct sk_buff *skb) 2119 { 2120 /* XDP packets use error pointer so abort at this point */ 2121 if (IS_ERR(skb)) 2122 return true; 2123 2124 if (unlikely(igc_test_staterr(rx_desc, IGC_RXDEXT_STATERR_RXE))) { 2125 struct net_device *netdev = rx_ring->netdev; 2126 2127 if (!(netdev->features & NETIF_F_RXALL)) { 2128 dev_kfree_skb_any(skb); 2129 return true; 2130 } 2131 } 2132 2133 /* if eth_skb_pad returns an error the skb was freed */ 2134 if (eth_skb_pad(skb)) 2135 return true; 2136 2137 return false; 2138 } 2139 2140 static void igc_put_rx_buffer(struct igc_ring *rx_ring, 2141 struct igc_rx_buffer *rx_buffer, 2142 int rx_buffer_pgcnt) 2143 { 2144 if (igc_can_reuse_rx_page(rx_buffer, rx_buffer_pgcnt)) { 2145 /* hand second half of page back to the ring */ 2146 igc_reuse_rx_page(rx_ring, rx_buffer); 2147 } else { 2148 /* We are not reusing the buffer so unmap it and free 2149 * any references we are holding to it 2150 */ 2151 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma, 2152 igc_rx_pg_size(rx_ring), DMA_FROM_DEVICE, 2153 IGC_RX_DMA_ATTR); 2154 __page_frag_cache_drain(rx_buffer->page, 2155 rx_buffer->pagecnt_bias); 2156 } 2157 2158 /* clear contents of rx_buffer */ 2159 rx_buffer->page = NULL; 2160 } 2161 2162 static inline unsigned int igc_rx_offset(struct igc_ring *rx_ring) 2163 { 2164 struct igc_adapter *adapter = rx_ring->q_vector->adapter; 2165 2166 if (ring_uses_build_skb(rx_ring)) 2167 return IGC_SKB_PAD; 2168 if (igc_xdp_is_enabled(adapter)) 2169 return XDP_PACKET_HEADROOM; 2170 2171 return 0; 2172 } 2173 2174 static bool igc_alloc_mapped_page(struct igc_ring *rx_ring, 2175 struct igc_rx_buffer *bi) 2176 { 2177 struct page *page = bi->page; 2178 dma_addr_t dma; 2179 2180 /* since we are recycling buffers we should seldom need to alloc */ 2181 if (likely(page)) 2182 return true; 2183 2184 /* alloc new page for storage */ 2185 page = dev_alloc_pages(igc_rx_pg_order(rx_ring)); 2186 if (unlikely(!page)) { 2187 rx_ring->rx_stats.alloc_failed++; 2188 return false; 2189 } 2190 2191 /* map page for use */ 2192 dma = dma_map_page_attrs(rx_ring->dev, page, 0, 2193 igc_rx_pg_size(rx_ring), 2194 DMA_FROM_DEVICE, 2195 IGC_RX_DMA_ATTR); 2196 2197 /* if mapping failed free memory back to system since 2198 * there isn't much point in holding memory we can't use 2199 */ 2200 if (dma_mapping_error(rx_ring->dev, dma)) { 2201 __free_page(page); 2202 2203 rx_ring->rx_stats.alloc_failed++; 2204 return false; 2205 } 2206 2207 bi->dma = dma; 2208 bi->page = page; 2209 bi->page_offset = igc_rx_offset(rx_ring); 2210 page_ref_add(page, USHRT_MAX - 1); 2211 bi->pagecnt_bias = USHRT_MAX; 2212 2213 return true; 2214 } 2215 2216 /** 2217 * igc_alloc_rx_buffers - Replace used receive buffers; packet split 2218 * @rx_ring: rx descriptor ring 2219 * @cleaned_count: number of buffers to clean 2220 */ 2221 static void igc_alloc_rx_buffers(struct igc_ring *rx_ring, u16 cleaned_count) 2222 { 2223 union igc_adv_rx_desc *rx_desc; 2224 u16 i = rx_ring->next_to_use; 2225 struct igc_rx_buffer *bi; 2226 u16 bufsz; 2227 2228 /* nothing to do */ 2229 if (!cleaned_count) 2230 return; 2231 2232 rx_desc = IGC_RX_DESC(rx_ring, i); 2233 bi = &rx_ring->rx_buffer_info[i]; 2234 i -= rx_ring->count; 2235 2236 bufsz = igc_rx_bufsz(rx_ring); 2237 2238 do { 2239 if (!igc_alloc_mapped_page(rx_ring, bi)) 2240 break; 2241 2242 /* sync the buffer for use by the device */ 2243 dma_sync_single_range_for_device(rx_ring->dev, bi->dma, 2244 bi->page_offset, bufsz, 2245 DMA_FROM_DEVICE); 2246 2247 /* Refresh the desc even if buffer_addrs didn't change 2248 * because each write-back erases this info. 2249 */ 2250 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset); 2251 2252 rx_desc++; 2253 bi++; 2254 i++; 2255 if (unlikely(!i)) { 2256 rx_desc = IGC_RX_DESC(rx_ring, 0); 2257 bi = rx_ring->rx_buffer_info; 2258 i -= rx_ring->count; 2259 } 2260 2261 /* clear the length for the next_to_use descriptor */ 2262 rx_desc->wb.upper.length = 0; 2263 2264 cleaned_count--; 2265 } while (cleaned_count); 2266 2267 i += rx_ring->count; 2268 2269 if (rx_ring->next_to_use != i) { 2270 /* record the next descriptor to use */ 2271 rx_ring->next_to_use = i; 2272 2273 /* update next to alloc since we have filled the ring */ 2274 rx_ring->next_to_alloc = i; 2275 2276 /* Force memory writes to complete before letting h/w 2277 * know there are new descriptors to fetch. (Only 2278 * applicable for weak-ordered memory model archs, 2279 * such as IA-64). 2280 */ 2281 wmb(); 2282 writel(i, rx_ring->tail); 2283 } 2284 } 2285 2286 static bool igc_alloc_rx_buffers_zc(struct igc_ring *ring, u16 count) 2287 { 2288 union igc_adv_rx_desc *desc; 2289 u16 i = ring->next_to_use; 2290 struct igc_rx_buffer *bi; 2291 dma_addr_t dma; 2292 bool ok = true; 2293 2294 if (!count) 2295 return ok; 2296 2297 XSK_CHECK_PRIV_TYPE(struct igc_xdp_buff); 2298 2299 desc = IGC_RX_DESC(ring, i); 2300 bi = &ring->rx_buffer_info[i]; 2301 i -= ring->count; 2302 2303 do { 2304 bi->xdp = xsk_buff_alloc(ring->xsk_pool); 2305 if (!bi->xdp) { 2306 ok = false; 2307 break; 2308 } 2309 2310 dma = xsk_buff_xdp_get_dma(bi->xdp); 2311 desc->read.pkt_addr = cpu_to_le64(dma); 2312 2313 desc++; 2314 bi++; 2315 i++; 2316 if (unlikely(!i)) { 2317 desc = IGC_RX_DESC(ring, 0); 2318 bi = ring->rx_buffer_info; 2319 i -= ring->count; 2320 } 2321 2322 /* Clear the length for the next_to_use descriptor. */ 2323 desc->wb.upper.length = 0; 2324 2325 count--; 2326 } while (count); 2327 2328 i += ring->count; 2329 2330 if (ring->next_to_use != i) { 2331 ring->next_to_use = i; 2332 2333 /* Force memory writes to complete before letting h/w 2334 * know there are new descriptors to fetch. (Only 2335 * applicable for weak-ordered memory model archs, 2336 * such as IA-64). 2337 */ 2338 wmb(); 2339 writel(i, ring->tail); 2340 } 2341 2342 return ok; 2343 } 2344 2345 /* This function requires __netif_tx_lock is held by the caller. */ 2346 static int igc_xdp_init_tx_descriptor(struct igc_ring *ring, 2347 struct xdp_frame *xdpf) 2348 { 2349 struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf); 2350 u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0; 2351 u16 count, index = ring->next_to_use; 2352 struct igc_tx_buffer *head = &ring->tx_buffer_info[index]; 2353 struct igc_tx_buffer *buffer = head; 2354 union igc_adv_tx_desc *desc = IGC_TX_DESC(ring, index); 2355 u32 olinfo_status, len = xdpf->len, cmd_type; 2356 void *data = xdpf->data; 2357 u16 i; 2358 2359 count = TXD_USE_COUNT(len); 2360 for (i = 0; i < nr_frags; i++) 2361 count += TXD_USE_COUNT(skb_frag_size(&sinfo->frags[i])); 2362 2363 if (igc_maybe_stop_tx(ring, count + 3)) { 2364 /* this is a hard error */ 2365 return -EBUSY; 2366 } 2367 2368 i = 0; 2369 head->bytecount = xdp_get_frame_len(xdpf); 2370 head->type = IGC_TX_BUFFER_TYPE_XDP; 2371 head->gso_segs = 1; 2372 head->xdpf = xdpf; 2373 2374 olinfo_status = head->bytecount << IGC_ADVTXD_PAYLEN_SHIFT; 2375 desc->read.olinfo_status = cpu_to_le32(olinfo_status); 2376 2377 for (;;) { 2378 dma_addr_t dma; 2379 2380 dma = dma_map_single(ring->dev, data, len, DMA_TO_DEVICE); 2381 if (dma_mapping_error(ring->dev, dma)) { 2382 netdev_err_once(ring->netdev, 2383 "Failed to map DMA for TX\n"); 2384 goto unmap; 2385 } 2386 2387 dma_unmap_len_set(buffer, len, len); 2388 dma_unmap_addr_set(buffer, dma, dma); 2389 2390 cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT | 2391 IGC_ADVTXD_DCMD_IFCS | len; 2392 2393 desc->read.cmd_type_len = cpu_to_le32(cmd_type); 2394 desc->read.buffer_addr = cpu_to_le64(dma); 2395 2396 buffer->protocol = 0; 2397 2398 if (++index == ring->count) 2399 index = 0; 2400 2401 if (i == nr_frags) 2402 break; 2403 2404 buffer = &ring->tx_buffer_info[index]; 2405 desc = IGC_TX_DESC(ring, index); 2406 desc->read.olinfo_status = 0; 2407 2408 data = skb_frag_address(&sinfo->frags[i]); 2409 len = skb_frag_size(&sinfo->frags[i]); 2410 i++; 2411 } 2412 desc->read.cmd_type_len |= cpu_to_le32(IGC_TXD_DCMD); 2413 2414 netdev_tx_sent_queue(txring_txq(ring), head->bytecount); 2415 /* set the timestamp */ 2416 head->time_stamp = jiffies; 2417 /* set next_to_watch value indicating a packet is present */ 2418 head->next_to_watch = desc; 2419 ring->next_to_use = index; 2420 2421 return 0; 2422 2423 unmap: 2424 for (;;) { 2425 buffer = &ring->tx_buffer_info[index]; 2426 if (dma_unmap_len(buffer, len)) 2427 dma_unmap_page(ring->dev, 2428 dma_unmap_addr(buffer, dma), 2429 dma_unmap_len(buffer, len), 2430 DMA_TO_DEVICE); 2431 dma_unmap_len_set(buffer, len, 0); 2432 if (buffer == head) 2433 break; 2434 2435 if (!index) 2436 index += ring->count; 2437 index--; 2438 } 2439 2440 return -ENOMEM; 2441 } 2442 2443 static struct igc_ring *igc_xdp_get_tx_ring(struct igc_adapter *adapter, 2444 int cpu) 2445 { 2446 int index = cpu; 2447 2448 if (unlikely(index < 0)) 2449 index = 0; 2450 2451 while (index >= adapter->num_tx_queues) 2452 index -= adapter->num_tx_queues; 2453 2454 return adapter->tx_ring[index]; 2455 } 2456 2457 static int igc_xdp_xmit_back(struct igc_adapter *adapter, struct xdp_buff *xdp) 2458 { 2459 struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp); 2460 int cpu = smp_processor_id(); 2461 struct netdev_queue *nq; 2462 struct igc_ring *ring; 2463 int res; 2464 2465 if (unlikely(!xdpf)) 2466 return -EFAULT; 2467 2468 ring = igc_xdp_get_tx_ring(adapter, cpu); 2469 nq = txring_txq(ring); 2470 2471 __netif_tx_lock(nq, cpu); 2472 /* Avoid transmit queue timeout since we share it with the slow path */ 2473 txq_trans_cond_update(nq); 2474 res = igc_xdp_init_tx_descriptor(ring, xdpf); 2475 __netif_tx_unlock(nq); 2476 return res; 2477 } 2478 2479 /* This function assumes rcu_read_lock() is held by the caller. */ 2480 static int __igc_xdp_run_prog(struct igc_adapter *adapter, 2481 struct bpf_prog *prog, 2482 struct xdp_buff *xdp) 2483 { 2484 u32 act = bpf_prog_run_xdp(prog, xdp); 2485 2486 switch (act) { 2487 case XDP_PASS: 2488 return IGC_XDP_PASS; 2489 case XDP_TX: 2490 if (igc_xdp_xmit_back(adapter, xdp) < 0) 2491 goto out_failure; 2492 return IGC_XDP_TX; 2493 case XDP_REDIRECT: 2494 if (xdp_do_redirect(adapter->netdev, xdp, prog) < 0) 2495 goto out_failure; 2496 return IGC_XDP_REDIRECT; 2497 break; 2498 default: 2499 bpf_warn_invalid_xdp_action(adapter->netdev, prog, act); 2500 fallthrough; 2501 case XDP_ABORTED: 2502 out_failure: 2503 trace_xdp_exception(adapter->netdev, prog, act); 2504 fallthrough; 2505 case XDP_DROP: 2506 return IGC_XDP_CONSUMED; 2507 } 2508 } 2509 2510 static struct sk_buff *igc_xdp_run_prog(struct igc_adapter *adapter, 2511 struct xdp_buff *xdp) 2512 { 2513 struct bpf_prog *prog; 2514 int res; 2515 2516 prog = READ_ONCE(adapter->xdp_prog); 2517 if (!prog) { 2518 res = IGC_XDP_PASS; 2519 goto out; 2520 } 2521 2522 res = __igc_xdp_run_prog(adapter, prog, xdp); 2523 2524 out: 2525 return ERR_PTR(-res); 2526 } 2527 2528 /* This function assumes __netif_tx_lock is held by the caller. */ 2529 static void igc_flush_tx_descriptors(struct igc_ring *ring) 2530 { 2531 /* Once tail pointer is updated, hardware can fetch the descriptors 2532 * any time so we issue a write membar here to ensure all memory 2533 * writes are complete before the tail pointer is updated. 2534 */ 2535 wmb(); 2536 writel(ring->next_to_use, ring->tail); 2537 } 2538 2539 static void igc_finalize_xdp(struct igc_adapter *adapter, int status) 2540 { 2541 int cpu = smp_processor_id(); 2542 struct netdev_queue *nq; 2543 struct igc_ring *ring; 2544 2545 if (status & IGC_XDP_TX) { 2546 ring = igc_xdp_get_tx_ring(adapter, cpu); 2547 nq = txring_txq(ring); 2548 2549 __netif_tx_lock(nq, cpu); 2550 igc_flush_tx_descriptors(ring); 2551 __netif_tx_unlock(nq); 2552 } 2553 2554 if (status & IGC_XDP_REDIRECT) 2555 xdp_do_flush(); 2556 } 2557 2558 static void igc_update_rx_stats(struct igc_q_vector *q_vector, 2559 unsigned int packets, unsigned int bytes) 2560 { 2561 struct igc_ring *ring = q_vector->rx.ring; 2562 2563 u64_stats_update_begin(&ring->rx_syncp); 2564 ring->rx_stats.packets += packets; 2565 ring->rx_stats.bytes += bytes; 2566 u64_stats_update_end(&ring->rx_syncp); 2567 2568 q_vector->rx.total_packets += packets; 2569 q_vector->rx.total_bytes += bytes; 2570 } 2571 2572 static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget) 2573 { 2574 unsigned int total_bytes = 0, total_packets = 0; 2575 struct igc_adapter *adapter = q_vector->adapter; 2576 struct igc_ring *rx_ring = q_vector->rx.ring; 2577 struct sk_buff *skb = rx_ring->skb; 2578 u16 cleaned_count = igc_desc_unused(rx_ring); 2579 int xdp_status = 0, rx_buffer_pgcnt; 2580 2581 while (likely(total_packets < budget)) { 2582 union igc_adv_rx_desc *rx_desc; 2583 struct igc_rx_buffer *rx_buffer; 2584 unsigned int size, truesize; 2585 struct igc_xdp_buff ctx; 2586 ktime_t timestamp = 0; 2587 int pkt_offset = 0; 2588 void *pktbuf; 2589 2590 /* return some buffers to hardware, one at a time is too slow */ 2591 if (cleaned_count >= IGC_RX_BUFFER_WRITE) { 2592 igc_alloc_rx_buffers(rx_ring, cleaned_count); 2593 cleaned_count = 0; 2594 } 2595 2596 rx_desc = IGC_RX_DESC(rx_ring, rx_ring->next_to_clean); 2597 size = le16_to_cpu(rx_desc->wb.upper.length); 2598 if (!size) 2599 break; 2600 2601 /* This memory barrier is needed to keep us from reading 2602 * any other fields out of the rx_desc until we know the 2603 * descriptor has been written back 2604 */ 2605 dma_rmb(); 2606 2607 rx_buffer = igc_get_rx_buffer(rx_ring, size, &rx_buffer_pgcnt); 2608 truesize = igc_get_rx_frame_truesize(rx_ring, size); 2609 2610 pktbuf = page_address(rx_buffer->page) + rx_buffer->page_offset; 2611 2612 if (igc_test_staterr(rx_desc, IGC_RXDADV_STAT_TSIP)) { 2613 timestamp = igc_ptp_rx_pktstamp(q_vector->adapter, 2614 pktbuf); 2615 ctx.rx_ts = timestamp; 2616 pkt_offset = IGC_TS_HDR_LEN; 2617 size -= IGC_TS_HDR_LEN; 2618 } 2619 2620 if (!skb) { 2621 xdp_init_buff(&ctx.xdp, truesize, &rx_ring->xdp_rxq); 2622 xdp_prepare_buff(&ctx.xdp, pktbuf - igc_rx_offset(rx_ring), 2623 igc_rx_offset(rx_ring) + pkt_offset, 2624 size, true); 2625 xdp_buff_clear_frags_flag(&ctx.xdp); 2626 ctx.rx_desc = rx_desc; 2627 2628 skb = igc_xdp_run_prog(adapter, &ctx.xdp); 2629 } 2630 2631 if (IS_ERR(skb)) { 2632 unsigned int xdp_res = -PTR_ERR(skb); 2633 2634 switch (xdp_res) { 2635 case IGC_XDP_CONSUMED: 2636 rx_buffer->pagecnt_bias++; 2637 break; 2638 case IGC_XDP_TX: 2639 case IGC_XDP_REDIRECT: 2640 igc_rx_buffer_flip(rx_buffer, truesize); 2641 xdp_status |= xdp_res; 2642 break; 2643 } 2644 2645 total_packets++; 2646 total_bytes += size; 2647 } else if (skb) 2648 igc_add_rx_frag(rx_ring, rx_buffer, skb, size); 2649 else if (ring_uses_build_skb(rx_ring)) 2650 skb = igc_build_skb(rx_ring, rx_buffer, &ctx.xdp); 2651 else 2652 skb = igc_construct_skb(rx_ring, rx_buffer, &ctx.xdp, 2653 timestamp); 2654 2655 /* exit if we failed to retrieve a buffer */ 2656 if (!skb) { 2657 rx_ring->rx_stats.alloc_failed++; 2658 rx_buffer->pagecnt_bias++; 2659 break; 2660 } 2661 2662 igc_put_rx_buffer(rx_ring, rx_buffer, rx_buffer_pgcnt); 2663 cleaned_count++; 2664 2665 /* fetch next buffer in frame if non-eop */ 2666 if (igc_is_non_eop(rx_ring, rx_desc)) 2667 continue; 2668 2669 /* verify the packet layout is correct */ 2670 if (igc_cleanup_headers(rx_ring, rx_desc, skb)) { 2671 skb = NULL; 2672 continue; 2673 } 2674 2675 /* probably a little skewed due to removing CRC */ 2676 total_bytes += skb->len; 2677 2678 /* populate checksum, VLAN, and protocol */ 2679 igc_process_skb_fields(rx_ring, rx_desc, skb); 2680 2681 napi_gro_receive(&q_vector->napi, skb); 2682 2683 /* reset skb pointer */ 2684 skb = NULL; 2685 2686 /* update budget accounting */ 2687 total_packets++; 2688 } 2689 2690 if (xdp_status) 2691 igc_finalize_xdp(adapter, xdp_status); 2692 2693 /* place incomplete frames back on ring for completion */ 2694 rx_ring->skb = skb; 2695 2696 igc_update_rx_stats(q_vector, total_packets, total_bytes); 2697 2698 if (cleaned_count) 2699 igc_alloc_rx_buffers(rx_ring, cleaned_count); 2700 2701 return total_packets; 2702 } 2703 2704 static struct sk_buff *igc_construct_skb_zc(struct igc_ring *ring, 2705 struct xdp_buff *xdp) 2706 { 2707 unsigned int totalsize = xdp->data_end - xdp->data_meta; 2708 unsigned int metasize = xdp->data - xdp->data_meta; 2709 struct sk_buff *skb; 2710 2711 net_prefetch(xdp->data_meta); 2712 2713 skb = __napi_alloc_skb(&ring->q_vector->napi, totalsize, 2714 GFP_ATOMIC | __GFP_NOWARN); 2715 if (unlikely(!skb)) 2716 return NULL; 2717 2718 memcpy(__skb_put(skb, totalsize), xdp->data_meta, 2719 ALIGN(totalsize, sizeof(long))); 2720 2721 if (metasize) { 2722 skb_metadata_set(skb, metasize); 2723 __skb_pull(skb, metasize); 2724 } 2725 2726 return skb; 2727 } 2728 2729 static void igc_dispatch_skb_zc(struct igc_q_vector *q_vector, 2730 union igc_adv_rx_desc *desc, 2731 struct xdp_buff *xdp, 2732 ktime_t timestamp) 2733 { 2734 struct igc_ring *ring = q_vector->rx.ring; 2735 struct sk_buff *skb; 2736 2737 skb = igc_construct_skb_zc(ring, xdp); 2738 if (!skb) { 2739 ring->rx_stats.alloc_failed++; 2740 return; 2741 } 2742 2743 if (timestamp) 2744 skb_hwtstamps(skb)->hwtstamp = timestamp; 2745 2746 if (igc_cleanup_headers(ring, desc, skb)) 2747 return; 2748 2749 igc_process_skb_fields(ring, desc, skb); 2750 napi_gro_receive(&q_vector->napi, skb); 2751 } 2752 2753 static struct igc_xdp_buff *xsk_buff_to_igc_ctx(struct xdp_buff *xdp) 2754 { 2755 /* xdp_buff pointer used by ZC code path is alloc as xdp_buff_xsk. The 2756 * igc_xdp_buff shares its layout with xdp_buff_xsk and private 2757 * igc_xdp_buff fields fall into xdp_buff_xsk->cb 2758 */ 2759 return (struct igc_xdp_buff *)xdp; 2760 } 2761 2762 static int igc_clean_rx_irq_zc(struct igc_q_vector *q_vector, const int budget) 2763 { 2764 struct igc_adapter *adapter = q_vector->adapter; 2765 struct igc_ring *ring = q_vector->rx.ring; 2766 u16 cleaned_count = igc_desc_unused(ring); 2767 int total_bytes = 0, total_packets = 0; 2768 u16 ntc = ring->next_to_clean; 2769 struct bpf_prog *prog; 2770 bool failure = false; 2771 int xdp_status = 0; 2772 2773 rcu_read_lock(); 2774 2775 prog = READ_ONCE(adapter->xdp_prog); 2776 2777 while (likely(total_packets < budget)) { 2778 union igc_adv_rx_desc *desc; 2779 struct igc_rx_buffer *bi; 2780 struct igc_xdp_buff *ctx; 2781 ktime_t timestamp = 0; 2782 unsigned int size; 2783 int res; 2784 2785 desc = IGC_RX_DESC(ring, ntc); 2786 size = le16_to_cpu(desc->wb.upper.length); 2787 if (!size) 2788 break; 2789 2790 /* This memory barrier is needed to keep us from reading 2791 * any other fields out of the rx_desc until we know the 2792 * descriptor has been written back 2793 */ 2794 dma_rmb(); 2795 2796 bi = &ring->rx_buffer_info[ntc]; 2797 2798 ctx = xsk_buff_to_igc_ctx(bi->xdp); 2799 ctx->rx_desc = desc; 2800 2801 if (igc_test_staterr(desc, IGC_RXDADV_STAT_TSIP)) { 2802 timestamp = igc_ptp_rx_pktstamp(q_vector->adapter, 2803 bi->xdp->data); 2804 ctx->rx_ts = timestamp; 2805 2806 bi->xdp->data += IGC_TS_HDR_LEN; 2807 2808 /* HW timestamp has been copied into local variable. Metadata 2809 * length when XDP program is called should be 0. 2810 */ 2811 bi->xdp->data_meta += IGC_TS_HDR_LEN; 2812 size -= IGC_TS_HDR_LEN; 2813 } 2814 2815 bi->xdp->data_end = bi->xdp->data + size; 2816 xsk_buff_dma_sync_for_cpu(bi->xdp, ring->xsk_pool); 2817 2818 res = __igc_xdp_run_prog(adapter, prog, bi->xdp); 2819 switch (res) { 2820 case IGC_XDP_PASS: 2821 igc_dispatch_skb_zc(q_vector, desc, bi->xdp, timestamp); 2822 fallthrough; 2823 case IGC_XDP_CONSUMED: 2824 xsk_buff_free(bi->xdp); 2825 break; 2826 case IGC_XDP_TX: 2827 case IGC_XDP_REDIRECT: 2828 xdp_status |= res; 2829 break; 2830 } 2831 2832 bi->xdp = NULL; 2833 total_bytes += size; 2834 total_packets++; 2835 cleaned_count++; 2836 ntc++; 2837 if (ntc == ring->count) 2838 ntc = 0; 2839 } 2840 2841 ring->next_to_clean = ntc; 2842 rcu_read_unlock(); 2843 2844 if (cleaned_count >= IGC_RX_BUFFER_WRITE) 2845 failure = !igc_alloc_rx_buffers_zc(ring, cleaned_count); 2846 2847 if (xdp_status) 2848 igc_finalize_xdp(adapter, xdp_status); 2849 2850 igc_update_rx_stats(q_vector, total_packets, total_bytes); 2851 2852 if (xsk_uses_need_wakeup(ring->xsk_pool)) { 2853 if (failure || ring->next_to_clean == ring->next_to_use) 2854 xsk_set_rx_need_wakeup(ring->xsk_pool); 2855 else 2856 xsk_clear_rx_need_wakeup(ring->xsk_pool); 2857 return total_packets; 2858 } 2859 2860 return failure ? budget : total_packets; 2861 } 2862 2863 static void igc_update_tx_stats(struct igc_q_vector *q_vector, 2864 unsigned int packets, unsigned int bytes) 2865 { 2866 struct igc_ring *ring = q_vector->tx.ring; 2867 2868 u64_stats_update_begin(&ring->tx_syncp); 2869 ring->tx_stats.bytes += bytes; 2870 ring->tx_stats.packets += packets; 2871 u64_stats_update_end(&ring->tx_syncp); 2872 2873 q_vector->tx.total_bytes += bytes; 2874 q_vector->tx.total_packets += packets; 2875 } 2876 2877 static void igc_xdp_xmit_zc(struct igc_ring *ring) 2878 { 2879 struct xsk_buff_pool *pool = ring->xsk_pool; 2880 struct netdev_queue *nq = txring_txq(ring); 2881 union igc_adv_tx_desc *tx_desc = NULL; 2882 int cpu = smp_processor_id(); 2883 struct xdp_desc xdp_desc; 2884 u16 budget, ntu; 2885 2886 if (!netif_carrier_ok(ring->netdev)) 2887 return; 2888 2889 __netif_tx_lock(nq, cpu); 2890 2891 /* Avoid transmit queue timeout since we share it with the slow path */ 2892 txq_trans_cond_update(nq); 2893 2894 ntu = ring->next_to_use; 2895 budget = igc_desc_unused(ring); 2896 2897 while (xsk_tx_peek_desc(pool, &xdp_desc) && budget--) { 2898 u32 cmd_type, olinfo_status; 2899 struct igc_tx_buffer *bi; 2900 dma_addr_t dma; 2901 2902 cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT | 2903 IGC_ADVTXD_DCMD_IFCS | IGC_TXD_DCMD | 2904 xdp_desc.len; 2905 olinfo_status = xdp_desc.len << IGC_ADVTXD_PAYLEN_SHIFT; 2906 2907 dma = xsk_buff_raw_get_dma(pool, xdp_desc.addr); 2908 xsk_buff_raw_dma_sync_for_device(pool, dma, xdp_desc.len); 2909 2910 tx_desc = IGC_TX_DESC(ring, ntu); 2911 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type); 2912 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status); 2913 tx_desc->read.buffer_addr = cpu_to_le64(dma); 2914 2915 bi = &ring->tx_buffer_info[ntu]; 2916 bi->type = IGC_TX_BUFFER_TYPE_XSK; 2917 bi->protocol = 0; 2918 bi->bytecount = xdp_desc.len; 2919 bi->gso_segs = 1; 2920 bi->time_stamp = jiffies; 2921 bi->next_to_watch = tx_desc; 2922 2923 netdev_tx_sent_queue(txring_txq(ring), xdp_desc.len); 2924 2925 ntu++; 2926 if (ntu == ring->count) 2927 ntu = 0; 2928 } 2929 2930 ring->next_to_use = ntu; 2931 if (tx_desc) { 2932 igc_flush_tx_descriptors(ring); 2933 xsk_tx_release(pool); 2934 } 2935 2936 __netif_tx_unlock(nq); 2937 } 2938 2939 /** 2940 * igc_clean_tx_irq - Reclaim resources after transmit completes 2941 * @q_vector: pointer to q_vector containing needed info 2942 * @napi_budget: Used to determine if we are in netpoll 2943 * 2944 * returns true if ring is completely cleaned 2945 */ 2946 static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget) 2947 { 2948 struct igc_adapter *adapter = q_vector->adapter; 2949 unsigned int total_bytes = 0, total_packets = 0; 2950 unsigned int budget = q_vector->tx.work_limit; 2951 struct igc_ring *tx_ring = q_vector->tx.ring; 2952 unsigned int i = tx_ring->next_to_clean; 2953 struct igc_tx_buffer *tx_buffer; 2954 union igc_adv_tx_desc *tx_desc; 2955 u32 xsk_frames = 0; 2956 2957 if (test_bit(__IGC_DOWN, &adapter->state)) 2958 return true; 2959 2960 tx_buffer = &tx_ring->tx_buffer_info[i]; 2961 tx_desc = IGC_TX_DESC(tx_ring, i); 2962 i -= tx_ring->count; 2963 2964 do { 2965 union igc_adv_tx_desc *eop_desc = tx_buffer->next_to_watch; 2966 2967 /* if next_to_watch is not set then there is no work pending */ 2968 if (!eop_desc) 2969 break; 2970 2971 /* prevent any other reads prior to eop_desc */ 2972 smp_rmb(); 2973 2974 /* if DD is not set pending work has not been completed */ 2975 if (!(eop_desc->wb.status & cpu_to_le32(IGC_TXD_STAT_DD))) 2976 break; 2977 2978 /* clear next_to_watch to prevent false hangs */ 2979 tx_buffer->next_to_watch = NULL; 2980 2981 /* update the statistics for this packet */ 2982 total_bytes += tx_buffer->bytecount; 2983 total_packets += tx_buffer->gso_segs; 2984 2985 switch (tx_buffer->type) { 2986 case IGC_TX_BUFFER_TYPE_XSK: 2987 xsk_frames++; 2988 break; 2989 case IGC_TX_BUFFER_TYPE_XDP: 2990 xdp_return_frame(tx_buffer->xdpf); 2991 igc_unmap_tx_buffer(tx_ring->dev, tx_buffer); 2992 break; 2993 case IGC_TX_BUFFER_TYPE_SKB: 2994 napi_consume_skb(tx_buffer->skb, napi_budget); 2995 igc_unmap_tx_buffer(tx_ring->dev, tx_buffer); 2996 break; 2997 default: 2998 netdev_warn_once(tx_ring->netdev, "Unknown Tx buffer type\n"); 2999 break; 3000 } 3001 3002 /* clear last DMA location and unmap remaining buffers */ 3003 while (tx_desc != eop_desc) { 3004 tx_buffer++; 3005 tx_desc++; 3006 i++; 3007 if (unlikely(!i)) { 3008 i -= tx_ring->count; 3009 tx_buffer = tx_ring->tx_buffer_info; 3010 tx_desc = IGC_TX_DESC(tx_ring, 0); 3011 } 3012 3013 /* unmap any remaining paged data */ 3014 if (dma_unmap_len(tx_buffer, len)) 3015 igc_unmap_tx_buffer(tx_ring->dev, tx_buffer); 3016 } 3017 3018 /* move us one more past the eop_desc for start of next pkt */ 3019 tx_buffer++; 3020 tx_desc++; 3021 i++; 3022 if (unlikely(!i)) { 3023 i -= tx_ring->count; 3024 tx_buffer = tx_ring->tx_buffer_info; 3025 tx_desc = IGC_TX_DESC(tx_ring, 0); 3026 } 3027 3028 /* issue prefetch for next Tx descriptor */ 3029 prefetch(tx_desc); 3030 3031 /* update budget accounting */ 3032 budget--; 3033 } while (likely(budget)); 3034 3035 netdev_tx_completed_queue(txring_txq(tx_ring), 3036 total_packets, total_bytes); 3037 3038 i += tx_ring->count; 3039 tx_ring->next_to_clean = i; 3040 3041 igc_update_tx_stats(q_vector, total_packets, total_bytes); 3042 3043 if (tx_ring->xsk_pool) { 3044 if (xsk_frames) 3045 xsk_tx_completed(tx_ring->xsk_pool, xsk_frames); 3046 if (xsk_uses_need_wakeup(tx_ring->xsk_pool)) 3047 xsk_set_tx_need_wakeup(tx_ring->xsk_pool); 3048 igc_xdp_xmit_zc(tx_ring); 3049 } 3050 3051 if (test_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) { 3052 struct igc_hw *hw = &adapter->hw; 3053 3054 /* Detect a transmit hang in hardware, this serializes the 3055 * check with the clearing of time_stamp and movement of i 3056 */ 3057 clear_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags); 3058 if (tx_buffer->next_to_watch && 3059 time_after(jiffies, tx_buffer->time_stamp + 3060 (adapter->tx_timeout_factor * HZ)) && 3061 !(rd32(IGC_STATUS) & IGC_STATUS_TXOFF) && 3062 (rd32(IGC_TDH(tx_ring->reg_idx)) != readl(tx_ring->tail)) && 3063 !tx_ring->oper_gate_closed) { 3064 /* detected Tx unit hang */ 3065 netdev_err(tx_ring->netdev, 3066 "Detected Tx Unit Hang\n" 3067 " Tx Queue <%d>\n" 3068 " TDH <%x>\n" 3069 " TDT <%x>\n" 3070 " next_to_use <%x>\n" 3071 " next_to_clean <%x>\n" 3072 "buffer_info[next_to_clean]\n" 3073 " time_stamp <%lx>\n" 3074 " next_to_watch <%p>\n" 3075 " jiffies <%lx>\n" 3076 " desc.status <%x>\n", 3077 tx_ring->queue_index, 3078 rd32(IGC_TDH(tx_ring->reg_idx)), 3079 readl(tx_ring->tail), 3080 tx_ring->next_to_use, 3081 tx_ring->next_to_clean, 3082 tx_buffer->time_stamp, 3083 tx_buffer->next_to_watch, 3084 jiffies, 3085 tx_buffer->next_to_watch->wb.status); 3086 netif_stop_subqueue(tx_ring->netdev, 3087 tx_ring->queue_index); 3088 3089 /* we are about to reset, no point in enabling stuff */ 3090 return true; 3091 } 3092 } 3093 3094 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2) 3095 if (unlikely(total_packets && 3096 netif_carrier_ok(tx_ring->netdev) && 3097 igc_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) { 3098 /* Make sure that anybody stopping the queue after this 3099 * sees the new next_to_clean. 3100 */ 3101 smp_mb(); 3102 if (__netif_subqueue_stopped(tx_ring->netdev, 3103 tx_ring->queue_index) && 3104 !(test_bit(__IGC_DOWN, &adapter->state))) { 3105 netif_wake_subqueue(tx_ring->netdev, 3106 tx_ring->queue_index); 3107 3108 u64_stats_update_begin(&tx_ring->tx_syncp); 3109 tx_ring->tx_stats.restart_queue++; 3110 u64_stats_update_end(&tx_ring->tx_syncp); 3111 } 3112 } 3113 3114 return !!budget; 3115 } 3116 3117 static int igc_find_mac_filter(struct igc_adapter *adapter, 3118 enum igc_mac_filter_type type, const u8 *addr) 3119 { 3120 struct igc_hw *hw = &adapter->hw; 3121 int max_entries = hw->mac.rar_entry_count; 3122 u32 ral, rah; 3123 int i; 3124 3125 for (i = 0; i < max_entries; i++) { 3126 ral = rd32(IGC_RAL(i)); 3127 rah = rd32(IGC_RAH(i)); 3128 3129 if (!(rah & IGC_RAH_AV)) 3130 continue; 3131 if (!!(rah & IGC_RAH_ASEL_SRC_ADDR) != type) 3132 continue; 3133 if ((rah & IGC_RAH_RAH_MASK) != 3134 le16_to_cpup((__le16 *)(addr + 4))) 3135 continue; 3136 if (ral != le32_to_cpup((__le32 *)(addr))) 3137 continue; 3138 3139 return i; 3140 } 3141 3142 return -1; 3143 } 3144 3145 static int igc_get_avail_mac_filter_slot(struct igc_adapter *adapter) 3146 { 3147 struct igc_hw *hw = &adapter->hw; 3148 int max_entries = hw->mac.rar_entry_count; 3149 u32 rah; 3150 int i; 3151 3152 for (i = 0; i < max_entries; i++) { 3153 rah = rd32(IGC_RAH(i)); 3154 3155 if (!(rah & IGC_RAH_AV)) 3156 return i; 3157 } 3158 3159 return -1; 3160 } 3161 3162 /** 3163 * igc_add_mac_filter() - Add MAC address filter 3164 * @adapter: Pointer to adapter where the filter should be added 3165 * @type: MAC address filter type (source or destination) 3166 * @addr: MAC address 3167 * @queue: If non-negative, queue assignment feature is enabled and frames 3168 * matching the filter are enqueued onto 'queue'. Otherwise, queue 3169 * assignment is disabled. 3170 * 3171 * Return: 0 in case of success, negative errno code otherwise. 3172 */ 3173 static int igc_add_mac_filter(struct igc_adapter *adapter, 3174 enum igc_mac_filter_type type, const u8 *addr, 3175 int queue) 3176 { 3177 struct net_device *dev = adapter->netdev; 3178 int index; 3179 3180 index = igc_find_mac_filter(adapter, type, addr); 3181 if (index >= 0) 3182 goto update_filter; 3183 3184 index = igc_get_avail_mac_filter_slot(adapter); 3185 if (index < 0) 3186 return -ENOSPC; 3187 3188 netdev_dbg(dev, "Add MAC address filter: index %d type %s address %pM queue %d\n", 3189 index, type == IGC_MAC_FILTER_TYPE_DST ? "dst" : "src", 3190 addr, queue); 3191 3192 update_filter: 3193 igc_set_mac_filter_hw(adapter, index, type, addr, queue); 3194 return 0; 3195 } 3196 3197 /** 3198 * igc_del_mac_filter() - Delete MAC address filter 3199 * @adapter: Pointer to adapter where the filter should be deleted from 3200 * @type: MAC address filter type (source or destination) 3201 * @addr: MAC address 3202 */ 3203 static void igc_del_mac_filter(struct igc_adapter *adapter, 3204 enum igc_mac_filter_type type, const u8 *addr) 3205 { 3206 struct net_device *dev = adapter->netdev; 3207 int index; 3208 3209 index = igc_find_mac_filter(adapter, type, addr); 3210 if (index < 0) 3211 return; 3212 3213 if (index == 0) { 3214 /* If this is the default filter, we don't actually delete it. 3215 * We just reset to its default value i.e. disable queue 3216 * assignment. 3217 */ 3218 netdev_dbg(dev, "Disable default MAC filter queue assignment"); 3219 3220 igc_set_mac_filter_hw(adapter, 0, type, addr, -1); 3221 } else { 3222 netdev_dbg(dev, "Delete MAC address filter: index %d type %s address %pM\n", 3223 index, 3224 type == IGC_MAC_FILTER_TYPE_DST ? "dst" : "src", 3225 addr); 3226 3227 igc_clear_mac_filter_hw(adapter, index); 3228 } 3229 } 3230 3231 /** 3232 * igc_add_vlan_prio_filter() - Add VLAN priority filter 3233 * @adapter: Pointer to adapter where the filter should be added 3234 * @prio: VLAN priority value 3235 * @queue: Queue number which matching frames are assigned to 3236 * 3237 * Return: 0 in case of success, negative errno code otherwise. 3238 */ 3239 static int igc_add_vlan_prio_filter(struct igc_adapter *adapter, int prio, 3240 int queue) 3241 { 3242 struct net_device *dev = adapter->netdev; 3243 struct igc_hw *hw = &adapter->hw; 3244 u32 vlanpqf; 3245 3246 vlanpqf = rd32(IGC_VLANPQF); 3247 3248 if (vlanpqf & IGC_VLANPQF_VALID(prio)) { 3249 netdev_dbg(dev, "VLAN priority filter already in use\n"); 3250 return -EEXIST; 3251 } 3252 3253 vlanpqf |= IGC_VLANPQF_QSEL(prio, queue); 3254 vlanpqf |= IGC_VLANPQF_VALID(prio); 3255 3256 wr32(IGC_VLANPQF, vlanpqf); 3257 3258 netdev_dbg(dev, "Add VLAN priority filter: prio %d queue %d\n", 3259 prio, queue); 3260 return 0; 3261 } 3262 3263 /** 3264 * igc_del_vlan_prio_filter() - Delete VLAN priority filter 3265 * @adapter: Pointer to adapter where the filter should be deleted from 3266 * @prio: VLAN priority value 3267 */ 3268 static void igc_del_vlan_prio_filter(struct igc_adapter *adapter, int prio) 3269 { 3270 struct igc_hw *hw = &adapter->hw; 3271 u32 vlanpqf; 3272 3273 vlanpqf = rd32(IGC_VLANPQF); 3274 3275 vlanpqf &= ~IGC_VLANPQF_VALID(prio); 3276 vlanpqf &= ~IGC_VLANPQF_QSEL(prio, IGC_VLANPQF_QUEUE_MASK); 3277 3278 wr32(IGC_VLANPQF, vlanpqf); 3279 3280 netdev_dbg(adapter->netdev, "Delete VLAN priority filter: prio %d\n", 3281 prio); 3282 } 3283 3284 static int igc_get_avail_etype_filter_slot(struct igc_adapter *adapter) 3285 { 3286 struct igc_hw *hw = &adapter->hw; 3287 int i; 3288 3289 for (i = 0; i < MAX_ETYPE_FILTER; i++) { 3290 u32 etqf = rd32(IGC_ETQF(i)); 3291 3292 if (!(etqf & IGC_ETQF_FILTER_ENABLE)) 3293 return i; 3294 } 3295 3296 return -1; 3297 } 3298 3299 /** 3300 * igc_add_etype_filter() - Add ethertype filter 3301 * @adapter: Pointer to adapter where the filter should be added 3302 * @etype: Ethertype value 3303 * @queue: If non-negative, queue assignment feature is enabled and frames 3304 * matching the filter are enqueued onto 'queue'. Otherwise, queue 3305 * assignment is disabled. 3306 * 3307 * Return: 0 in case of success, negative errno code otherwise. 3308 */ 3309 static int igc_add_etype_filter(struct igc_adapter *adapter, u16 etype, 3310 int queue) 3311 { 3312 struct igc_hw *hw = &adapter->hw; 3313 int index; 3314 u32 etqf; 3315 3316 index = igc_get_avail_etype_filter_slot(adapter); 3317 if (index < 0) 3318 return -ENOSPC; 3319 3320 etqf = rd32(IGC_ETQF(index)); 3321 3322 etqf &= ~IGC_ETQF_ETYPE_MASK; 3323 etqf |= etype; 3324 3325 if (queue >= 0) { 3326 etqf &= ~IGC_ETQF_QUEUE_MASK; 3327 etqf |= (queue << IGC_ETQF_QUEUE_SHIFT); 3328 etqf |= IGC_ETQF_QUEUE_ENABLE; 3329 } 3330 3331 etqf |= IGC_ETQF_FILTER_ENABLE; 3332 3333 wr32(IGC_ETQF(index), etqf); 3334 3335 netdev_dbg(adapter->netdev, "Add ethertype filter: etype %04x queue %d\n", 3336 etype, queue); 3337 return 0; 3338 } 3339 3340 static int igc_find_etype_filter(struct igc_adapter *adapter, u16 etype) 3341 { 3342 struct igc_hw *hw = &adapter->hw; 3343 int i; 3344 3345 for (i = 0; i < MAX_ETYPE_FILTER; i++) { 3346 u32 etqf = rd32(IGC_ETQF(i)); 3347 3348 if ((etqf & IGC_ETQF_ETYPE_MASK) == etype) 3349 return i; 3350 } 3351 3352 return -1; 3353 } 3354 3355 /** 3356 * igc_del_etype_filter() - Delete ethertype filter 3357 * @adapter: Pointer to adapter where the filter should be deleted from 3358 * @etype: Ethertype value 3359 */ 3360 static void igc_del_etype_filter(struct igc_adapter *adapter, u16 etype) 3361 { 3362 struct igc_hw *hw = &adapter->hw; 3363 int index; 3364 3365 index = igc_find_etype_filter(adapter, etype); 3366 if (index < 0) 3367 return; 3368 3369 wr32(IGC_ETQF(index), 0); 3370 3371 netdev_dbg(adapter->netdev, "Delete ethertype filter: etype %04x\n", 3372 etype); 3373 } 3374 3375 static int igc_flex_filter_select(struct igc_adapter *adapter, 3376 struct igc_flex_filter *input, 3377 u32 *fhft) 3378 { 3379 struct igc_hw *hw = &adapter->hw; 3380 u8 fhft_index; 3381 u32 fhftsl; 3382 3383 if (input->index >= MAX_FLEX_FILTER) { 3384 dev_err(&adapter->pdev->dev, "Wrong Flex Filter index selected!\n"); 3385 return -EINVAL; 3386 } 3387 3388 /* Indirect table select register */ 3389 fhftsl = rd32(IGC_FHFTSL); 3390 fhftsl &= ~IGC_FHFTSL_FTSL_MASK; 3391 switch (input->index) { 3392 case 0 ... 7: 3393 fhftsl |= 0x00; 3394 break; 3395 case 8 ... 15: 3396 fhftsl |= 0x01; 3397 break; 3398 case 16 ... 23: 3399 fhftsl |= 0x02; 3400 break; 3401 case 24 ... 31: 3402 fhftsl |= 0x03; 3403 break; 3404 } 3405 wr32(IGC_FHFTSL, fhftsl); 3406 3407 /* Normalize index down to host table register */ 3408 fhft_index = input->index % 8; 3409 3410 *fhft = (fhft_index < 4) ? IGC_FHFT(fhft_index) : 3411 IGC_FHFT_EXT(fhft_index - 4); 3412 3413 return 0; 3414 } 3415 3416 static int igc_write_flex_filter_ll(struct igc_adapter *adapter, 3417 struct igc_flex_filter *input) 3418 { 3419 struct device *dev = &adapter->pdev->dev; 3420 struct igc_hw *hw = &adapter->hw; 3421 u8 *data = input->data; 3422 u8 *mask = input->mask; 3423 u32 queuing; 3424 u32 fhft; 3425 u32 wufc; 3426 int ret; 3427 int i; 3428 3429 /* Length has to be aligned to 8. Otherwise the filter will fail. Bail 3430 * out early to avoid surprises later. 3431 */ 3432 if (input->length % 8 != 0) { 3433 dev_err(dev, "The length of a flex filter has to be 8 byte aligned!\n"); 3434 return -EINVAL; 3435 } 3436 3437 /* Select corresponding flex filter register and get base for host table. */ 3438 ret = igc_flex_filter_select(adapter, input, &fhft); 3439 if (ret) 3440 return ret; 3441 3442 /* When adding a filter globally disable flex filter feature. That is 3443 * recommended within the datasheet. 3444 */ 3445 wufc = rd32(IGC_WUFC); 3446 wufc &= ~IGC_WUFC_FLEX_HQ; 3447 wr32(IGC_WUFC, wufc); 3448 3449 /* Configure filter */ 3450 queuing = input->length & IGC_FHFT_LENGTH_MASK; 3451 queuing |= (input->rx_queue << IGC_FHFT_QUEUE_SHIFT) & IGC_FHFT_QUEUE_MASK; 3452 queuing |= (input->prio << IGC_FHFT_PRIO_SHIFT) & IGC_FHFT_PRIO_MASK; 3453 3454 if (input->immediate_irq) 3455 queuing |= IGC_FHFT_IMM_INT; 3456 3457 if (input->drop) 3458 queuing |= IGC_FHFT_DROP; 3459 3460 wr32(fhft + 0xFC, queuing); 3461 3462 /* Write data (128 byte) and mask (128 bit) */ 3463 for (i = 0; i < 16; ++i) { 3464 const size_t data_idx = i * 8; 3465 const size_t row_idx = i * 16; 3466 u32 dw0 = 3467 (data[data_idx + 0] << 0) | 3468 (data[data_idx + 1] << 8) | 3469 (data[data_idx + 2] << 16) | 3470 (data[data_idx + 3] << 24); 3471 u32 dw1 = 3472 (data[data_idx + 4] << 0) | 3473 (data[data_idx + 5] << 8) | 3474 (data[data_idx + 6] << 16) | 3475 (data[data_idx + 7] << 24); 3476 u32 tmp; 3477 3478 /* Write row: dw0, dw1 and mask */ 3479 wr32(fhft + row_idx, dw0); 3480 wr32(fhft + row_idx + 4, dw1); 3481 3482 /* mask is only valid for MASK(7, 0) */ 3483 tmp = rd32(fhft + row_idx + 8); 3484 tmp &= ~GENMASK(7, 0); 3485 tmp |= mask[i]; 3486 wr32(fhft + row_idx + 8, tmp); 3487 } 3488 3489 /* Enable filter. */ 3490 wufc |= IGC_WUFC_FLEX_HQ; 3491 if (input->index > 8) { 3492 /* Filter 0-7 are enabled via WUFC. The other 24 filters are not. */ 3493 u32 wufc_ext = rd32(IGC_WUFC_EXT); 3494 3495 wufc_ext |= (IGC_WUFC_EXT_FLX8 << (input->index - 8)); 3496 3497 wr32(IGC_WUFC_EXT, wufc_ext); 3498 } else { 3499 wufc |= (IGC_WUFC_FLX0 << input->index); 3500 } 3501 wr32(IGC_WUFC, wufc); 3502 3503 dev_dbg(&adapter->pdev->dev, "Added flex filter %u to HW.\n", 3504 input->index); 3505 3506 return 0; 3507 } 3508 3509 static void igc_flex_filter_add_field(struct igc_flex_filter *flex, 3510 const void *src, unsigned int offset, 3511 size_t len, const void *mask) 3512 { 3513 int i; 3514 3515 /* data */ 3516 memcpy(&flex->data[offset], src, len); 3517 3518 /* mask */ 3519 for (i = 0; i < len; ++i) { 3520 const unsigned int idx = i + offset; 3521 const u8 *ptr = mask; 3522 3523 if (mask) { 3524 if (ptr[i] & 0xff) 3525 flex->mask[idx / 8] |= BIT(idx % 8); 3526 3527 continue; 3528 } 3529 3530 flex->mask[idx / 8] |= BIT(idx % 8); 3531 } 3532 } 3533 3534 static int igc_find_avail_flex_filter_slot(struct igc_adapter *adapter) 3535 { 3536 struct igc_hw *hw = &adapter->hw; 3537 u32 wufc, wufc_ext; 3538 int i; 3539 3540 wufc = rd32(IGC_WUFC); 3541 wufc_ext = rd32(IGC_WUFC_EXT); 3542 3543 for (i = 0; i < MAX_FLEX_FILTER; i++) { 3544 if (i < 8) { 3545 if (!(wufc & (IGC_WUFC_FLX0 << i))) 3546 return i; 3547 } else { 3548 if (!(wufc_ext & (IGC_WUFC_EXT_FLX8 << (i - 8)))) 3549 return i; 3550 } 3551 } 3552 3553 return -ENOSPC; 3554 } 3555 3556 static bool igc_flex_filter_in_use(struct igc_adapter *adapter) 3557 { 3558 struct igc_hw *hw = &adapter->hw; 3559 u32 wufc, wufc_ext; 3560 3561 wufc = rd32(IGC_WUFC); 3562 wufc_ext = rd32(IGC_WUFC_EXT); 3563 3564 if (wufc & IGC_WUFC_FILTER_MASK) 3565 return true; 3566 3567 if (wufc_ext & IGC_WUFC_EXT_FILTER_MASK) 3568 return true; 3569 3570 return false; 3571 } 3572 3573 static int igc_add_flex_filter(struct igc_adapter *adapter, 3574 struct igc_nfc_rule *rule) 3575 { 3576 struct igc_flex_filter flex = { }; 3577 struct igc_nfc_filter *filter = &rule->filter; 3578 unsigned int eth_offset, user_offset; 3579 int ret, index; 3580 bool vlan; 3581 3582 index = igc_find_avail_flex_filter_slot(adapter); 3583 if (index < 0) 3584 return -ENOSPC; 3585 3586 /* Construct the flex filter: 3587 * -> dest_mac [6] 3588 * -> src_mac [6] 3589 * -> tpid [2] 3590 * -> vlan tci [2] 3591 * -> ether type [2] 3592 * -> user data [8] 3593 * -> = 26 bytes => 32 length 3594 */ 3595 flex.index = index; 3596 flex.length = 32; 3597 flex.rx_queue = rule->action; 3598 3599 vlan = rule->filter.vlan_tci || rule->filter.vlan_etype; 3600 eth_offset = vlan ? 16 : 12; 3601 user_offset = vlan ? 18 : 14; 3602 3603 /* Add destination MAC */ 3604 if (rule->filter.match_flags & IGC_FILTER_FLAG_DST_MAC_ADDR) 3605 igc_flex_filter_add_field(&flex, &filter->dst_addr, 0, 3606 ETH_ALEN, NULL); 3607 3608 /* Add source MAC */ 3609 if (rule->filter.match_flags & IGC_FILTER_FLAG_SRC_MAC_ADDR) 3610 igc_flex_filter_add_field(&flex, &filter->src_addr, 6, 3611 ETH_ALEN, NULL); 3612 3613 /* Add VLAN etype */ 3614 if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_ETYPE) 3615 igc_flex_filter_add_field(&flex, &filter->vlan_etype, 12, 3616 sizeof(filter->vlan_etype), 3617 NULL); 3618 3619 /* Add VLAN TCI */ 3620 if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_TCI) 3621 igc_flex_filter_add_field(&flex, &filter->vlan_tci, 14, 3622 sizeof(filter->vlan_tci), NULL); 3623 3624 /* Add Ether type */ 3625 if (rule->filter.match_flags & IGC_FILTER_FLAG_ETHER_TYPE) { 3626 __be16 etype = cpu_to_be16(filter->etype); 3627 3628 igc_flex_filter_add_field(&flex, &etype, eth_offset, 3629 sizeof(etype), NULL); 3630 } 3631 3632 /* Add user data */ 3633 if (rule->filter.match_flags & IGC_FILTER_FLAG_USER_DATA) 3634 igc_flex_filter_add_field(&flex, &filter->user_data, 3635 user_offset, 3636 sizeof(filter->user_data), 3637 filter->user_mask); 3638 3639 /* Add it down to the hardware and enable it. */ 3640 ret = igc_write_flex_filter_ll(adapter, &flex); 3641 if (ret) 3642 return ret; 3643 3644 filter->flex_index = index; 3645 3646 return 0; 3647 } 3648 3649 static void igc_del_flex_filter(struct igc_adapter *adapter, 3650 u16 reg_index) 3651 { 3652 struct igc_hw *hw = &adapter->hw; 3653 u32 wufc; 3654 3655 /* Just disable the filter. The filter table itself is kept 3656 * intact. Another flex_filter_add() should override the "old" data 3657 * then. 3658 */ 3659 if (reg_index > 8) { 3660 u32 wufc_ext = rd32(IGC_WUFC_EXT); 3661 3662 wufc_ext &= ~(IGC_WUFC_EXT_FLX8 << (reg_index - 8)); 3663 wr32(IGC_WUFC_EXT, wufc_ext); 3664 } else { 3665 wufc = rd32(IGC_WUFC); 3666 3667 wufc &= ~(IGC_WUFC_FLX0 << reg_index); 3668 wr32(IGC_WUFC, wufc); 3669 } 3670 3671 if (igc_flex_filter_in_use(adapter)) 3672 return; 3673 3674 /* No filters are in use, we may disable flex filters */ 3675 wufc = rd32(IGC_WUFC); 3676 wufc &= ~IGC_WUFC_FLEX_HQ; 3677 wr32(IGC_WUFC, wufc); 3678 } 3679 3680 static int igc_enable_nfc_rule(struct igc_adapter *adapter, 3681 struct igc_nfc_rule *rule) 3682 { 3683 int err; 3684 3685 if (rule->flex) { 3686 return igc_add_flex_filter(adapter, rule); 3687 } 3688 3689 if (rule->filter.match_flags & IGC_FILTER_FLAG_ETHER_TYPE) { 3690 err = igc_add_etype_filter(adapter, rule->filter.etype, 3691 rule->action); 3692 if (err) 3693 return err; 3694 } 3695 3696 if (rule->filter.match_flags & IGC_FILTER_FLAG_SRC_MAC_ADDR) { 3697 err = igc_add_mac_filter(adapter, IGC_MAC_FILTER_TYPE_SRC, 3698 rule->filter.src_addr, rule->action); 3699 if (err) 3700 return err; 3701 } 3702 3703 if (rule->filter.match_flags & IGC_FILTER_FLAG_DST_MAC_ADDR) { 3704 err = igc_add_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST, 3705 rule->filter.dst_addr, rule->action); 3706 if (err) 3707 return err; 3708 } 3709 3710 if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_TCI) { 3711 int prio = FIELD_GET(VLAN_PRIO_MASK, rule->filter.vlan_tci); 3712 3713 err = igc_add_vlan_prio_filter(adapter, prio, rule->action); 3714 if (err) 3715 return err; 3716 } 3717 3718 return 0; 3719 } 3720 3721 static void igc_disable_nfc_rule(struct igc_adapter *adapter, 3722 const struct igc_nfc_rule *rule) 3723 { 3724 if (rule->flex) { 3725 igc_del_flex_filter(adapter, rule->filter.flex_index); 3726 return; 3727 } 3728 3729 if (rule->filter.match_flags & IGC_FILTER_FLAG_ETHER_TYPE) 3730 igc_del_etype_filter(adapter, rule->filter.etype); 3731 3732 if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_TCI) { 3733 int prio = FIELD_GET(VLAN_PRIO_MASK, rule->filter.vlan_tci); 3734 3735 igc_del_vlan_prio_filter(adapter, prio); 3736 } 3737 3738 if (rule->filter.match_flags & IGC_FILTER_FLAG_SRC_MAC_ADDR) 3739 igc_del_mac_filter(adapter, IGC_MAC_FILTER_TYPE_SRC, 3740 rule->filter.src_addr); 3741 3742 if (rule->filter.match_flags & IGC_FILTER_FLAG_DST_MAC_ADDR) 3743 igc_del_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST, 3744 rule->filter.dst_addr); 3745 } 3746 3747 /** 3748 * igc_get_nfc_rule() - Get NFC rule 3749 * @adapter: Pointer to adapter 3750 * @location: Rule location 3751 * 3752 * Context: Expects adapter->nfc_rule_lock to be held by caller. 3753 * 3754 * Return: Pointer to NFC rule at @location. If not found, NULL. 3755 */ 3756 struct igc_nfc_rule *igc_get_nfc_rule(struct igc_adapter *adapter, 3757 u32 location) 3758 { 3759 struct igc_nfc_rule *rule; 3760 3761 list_for_each_entry(rule, &adapter->nfc_rule_list, list) { 3762 if (rule->location == location) 3763 return rule; 3764 if (rule->location > location) 3765 break; 3766 } 3767 3768 return NULL; 3769 } 3770 3771 /** 3772 * igc_del_nfc_rule() - Delete NFC rule 3773 * @adapter: Pointer to adapter 3774 * @rule: Pointer to rule to be deleted 3775 * 3776 * Disable NFC rule in hardware and delete it from adapter. 3777 * 3778 * Context: Expects adapter->nfc_rule_lock to be held by caller. 3779 */ 3780 void igc_del_nfc_rule(struct igc_adapter *adapter, struct igc_nfc_rule *rule) 3781 { 3782 igc_disable_nfc_rule(adapter, rule); 3783 3784 list_del(&rule->list); 3785 adapter->nfc_rule_count--; 3786 3787 kfree(rule); 3788 } 3789 3790 static void igc_flush_nfc_rules(struct igc_adapter *adapter) 3791 { 3792 struct igc_nfc_rule *rule, *tmp; 3793 3794 mutex_lock(&adapter->nfc_rule_lock); 3795 3796 list_for_each_entry_safe(rule, tmp, &adapter->nfc_rule_list, list) 3797 igc_del_nfc_rule(adapter, rule); 3798 3799 mutex_unlock(&adapter->nfc_rule_lock); 3800 } 3801 3802 /** 3803 * igc_add_nfc_rule() - Add NFC rule 3804 * @adapter: Pointer to adapter 3805 * @rule: Pointer to rule to be added 3806 * 3807 * Enable NFC rule in hardware and add it to adapter. 3808 * 3809 * Context: Expects adapter->nfc_rule_lock to be held by caller. 3810 * 3811 * Return: 0 on success, negative errno on failure. 3812 */ 3813 int igc_add_nfc_rule(struct igc_adapter *adapter, struct igc_nfc_rule *rule) 3814 { 3815 struct igc_nfc_rule *pred, *cur; 3816 int err; 3817 3818 err = igc_enable_nfc_rule(adapter, rule); 3819 if (err) 3820 return err; 3821 3822 pred = NULL; 3823 list_for_each_entry(cur, &adapter->nfc_rule_list, list) { 3824 if (cur->location >= rule->location) 3825 break; 3826 pred = cur; 3827 } 3828 3829 list_add(&rule->list, pred ? &pred->list : &adapter->nfc_rule_list); 3830 adapter->nfc_rule_count++; 3831 return 0; 3832 } 3833 3834 static void igc_restore_nfc_rules(struct igc_adapter *adapter) 3835 { 3836 struct igc_nfc_rule *rule; 3837 3838 mutex_lock(&adapter->nfc_rule_lock); 3839 3840 list_for_each_entry_reverse(rule, &adapter->nfc_rule_list, list) 3841 igc_enable_nfc_rule(adapter, rule); 3842 3843 mutex_unlock(&adapter->nfc_rule_lock); 3844 } 3845 3846 static int igc_uc_sync(struct net_device *netdev, const unsigned char *addr) 3847 { 3848 struct igc_adapter *adapter = netdev_priv(netdev); 3849 3850 return igc_add_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST, addr, -1); 3851 } 3852 3853 static int igc_uc_unsync(struct net_device *netdev, const unsigned char *addr) 3854 { 3855 struct igc_adapter *adapter = netdev_priv(netdev); 3856 3857 igc_del_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST, addr); 3858 return 0; 3859 } 3860 3861 /** 3862 * igc_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set 3863 * @netdev: network interface device structure 3864 * 3865 * The set_rx_mode entry point is called whenever the unicast or multicast 3866 * address lists or the network interface flags are updated. This routine is 3867 * responsible for configuring the hardware for proper unicast, multicast, 3868 * promiscuous mode, and all-multi behavior. 3869 */ 3870 static void igc_set_rx_mode(struct net_device *netdev) 3871 { 3872 struct igc_adapter *adapter = netdev_priv(netdev); 3873 struct igc_hw *hw = &adapter->hw; 3874 u32 rctl = 0, rlpml = MAX_JUMBO_FRAME_SIZE; 3875 int count; 3876 3877 /* Check for Promiscuous and All Multicast modes */ 3878 if (netdev->flags & IFF_PROMISC) { 3879 rctl |= IGC_RCTL_UPE | IGC_RCTL_MPE; 3880 } else { 3881 if (netdev->flags & IFF_ALLMULTI) { 3882 rctl |= IGC_RCTL_MPE; 3883 } else { 3884 /* Write addresses to the MTA, if the attempt fails 3885 * then we should just turn on promiscuous mode so 3886 * that we can at least receive multicast traffic 3887 */ 3888 count = igc_write_mc_addr_list(netdev); 3889 if (count < 0) 3890 rctl |= IGC_RCTL_MPE; 3891 } 3892 } 3893 3894 /* Write addresses to available RAR registers, if there is not 3895 * sufficient space to store all the addresses then enable 3896 * unicast promiscuous mode 3897 */ 3898 if (__dev_uc_sync(netdev, igc_uc_sync, igc_uc_unsync)) 3899 rctl |= IGC_RCTL_UPE; 3900 3901 /* update state of unicast and multicast */ 3902 rctl |= rd32(IGC_RCTL) & ~(IGC_RCTL_UPE | IGC_RCTL_MPE); 3903 wr32(IGC_RCTL, rctl); 3904 3905 #if (PAGE_SIZE < 8192) 3906 if (adapter->max_frame_size <= IGC_MAX_FRAME_BUILD_SKB) 3907 rlpml = IGC_MAX_FRAME_BUILD_SKB; 3908 #endif 3909 wr32(IGC_RLPML, rlpml); 3910 } 3911 3912 /** 3913 * igc_configure - configure the hardware for RX and TX 3914 * @adapter: private board structure 3915 */ 3916 static void igc_configure(struct igc_adapter *adapter) 3917 { 3918 struct net_device *netdev = adapter->netdev; 3919 int i = 0; 3920 3921 igc_get_hw_control(adapter); 3922 igc_set_rx_mode(netdev); 3923 3924 igc_restore_vlan(adapter); 3925 3926 igc_setup_tctl(adapter); 3927 igc_setup_mrqc(adapter); 3928 igc_setup_rctl(adapter); 3929 3930 igc_set_default_mac_filter(adapter); 3931 igc_restore_nfc_rules(adapter); 3932 3933 igc_configure_tx(adapter); 3934 igc_configure_rx(adapter); 3935 3936 igc_rx_fifo_flush_base(&adapter->hw); 3937 3938 /* call igc_desc_unused which always leaves 3939 * at least 1 descriptor unused to make sure 3940 * next_to_use != next_to_clean 3941 */ 3942 for (i = 0; i < adapter->num_rx_queues; i++) { 3943 struct igc_ring *ring = adapter->rx_ring[i]; 3944 3945 if (ring->xsk_pool) 3946 igc_alloc_rx_buffers_zc(ring, igc_desc_unused(ring)); 3947 else 3948 igc_alloc_rx_buffers(ring, igc_desc_unused(ring)); 3949 } 3950 } 3951 3952 /** 3953 * igc_write_ivar - configure ivar for given MSI-X vector 3954 * @hw: pointer to the HW structure 3955 * @msix_vector: vector number we are allocating to a given ring 3956 * @index: row index of IVAR register to write within IVAR table 3957 * @offset: column offset of in IVAR, should be multiple of 8 3958 * 3959 * The IVAR table consists of 2 columns, 3960 * each containing an cause allocation for an Rx and Tx ring, and a 3961 * variable number of rows depending on the number of queues supported. 3962 */ 3963 static void igc_write_ivar(struct igc_hw *hw, int msix_vector, 3964 int index, int offset) 3965 { 3966 u32 ivar = array_rd32(IGC_IVAR0, index); 3967 3968 /* clear any bits that are currently set */ 3969 ivar &= ~((u32)0xFF << offset); 3970 3971 /* write vector and valid bit */ 3972 ivar |= (msix_vector | IGC_IVAR_VALID) << offset; 3973 3974 array_wr32(IGC_IVAR0, index, ivar); 3975 } 3976 3977 static void igc_assign_vector(struct igc_q_vector *q_vector, int msix_vector) 3978 { 3979 struct igc_adapter *adapter = q_vector->adapter; 3980 struct igc_hw *hw = &adapter->hw; 3981 int rx_queue = IGC_N0_QUEUE; 3982 int tx_queue = IGC_N0_QUEUE; 3983 3984 if (q_vector->rx.ring) 3985 rx_queue = q_vector->rx.ring->reg_idx; 3986 if (q_vector->tx.ring) 3987 tx_queue = q_vector->tx.ring->reg_idx; 3988 3989 switch (hw->mac.type) { 3990 case igc_i225: 3991 if (rx_queue > IGC_N0_QUEUE) 3992 igc_write_ivar(hw, msix_vector, 3993 rx_queue >> 1, 3994 (rx_queue & 0x1) << 4); 3995 if (tx_queue > IGC_N0_QUEUE) 3996 igc_write_ivar(hw, msix_vector, 3997 tx_queue >> 1, 3998 ((tx_queue & 0x1) << 4) + 8); 3999 q_vector->eims_value = BIT(msix_vector); 4000 break; 4001 default: 4002 WARN_ONCE(hw->mac.type != igc_i225, "Wrong MAC type\n"); 4003 break; 4004 } 4005 4006 /* add q_vector eims value to global eims_enable_mask */ 4007 adapter->eims_enable_mask |= q_vector->eims_value; 4008 4009 /* configure q_vector to set itr on first interrupt */ 4010 q_vector->set_itr = 1; 4011 } 4012 4013 /** 4014 * igc_configure_msix - Configure MSI-X hardware 4015 * @adapter: Pointer to adapter structure 4016 * 4017 * igc_configure_msix sets up the hardware to properly 4018 * generate MSI-X interrupts. 4019 */ 4020 static void igc_configure_msix(struct igc_adapter *adapter) 4021 { 4022 struct igc_hw *hw = &adapter->hw; 4023 int i, vector = 0; 4024 u32 tmp; 4025 4026 adapter->eims_enable_mask = 0; 4027 4028 /* set vector for other causes, i.e. link changes */ 4029 switch (hw->mac.type) { 4030 case igc_i225: 4031 /* Turn on MSI-X capability first, or our settings 4032 * won't stick. And it will take days to debug. 4033 */ 4034 wr32(IGC_GPIE, IGC_GPIE_MSIX_MODE | 4035 IGC_GPIE_PBA | IGC_GPIE_EIAME | 4036 IGC_GPIE_NSICR); 4037 4038 /* enable msix_other interrupt */ 4039 adapter->eims_other = BIT(vector); 4040 tmp = (vector++ | IGC_IVAR_VALID) << 8; 4041 4042 wr32(IGC_IVAR_MISC, tmp); 4043 break; 4044 default: 4045 /* do nothing, since nothing else supports MSI-X */ 4046 break; 4047 } /* switch (hw->mac.type) */ 4048 4049 adapter->eims_enable_mask |= adapter->eims_other; 4050 4051 for (i = 0; i < adapter->num_q_vectors; i++) 4052 igc_assign_vector(adapter->q_vector[i], vector++); 4053 4054 wrfl(); 4055 } 4056 4057 /** 4058 * igc_irq_enable - Enable default interrupt generation settings 4059 * @adapter: board private structure 4060 */ 4061 static void igc_irq_enable(struct igc_adapter *adapter) 4062 { 4063 struct igc_hw *hw = &adapter->hw; 4064 4065 if (adapter->msix_entries) { 4066 u32 ims = IGC_IMS_LSC | IGC_IMS_DOUTSYNC | IGC_IMS_DRSTA; 4067 u32 regval = rd32(IGC_EIAC); 4068 4069 wr32(IGC_EIAC, regval | adapter->eims_enable_mask); 4070 regval = rd32(IGC_EIAM); 4071 wr32(IGC_EIAM, regval | adapter->eims_enable_mask); 4072 wr32(IGC_EIMS, adapter->eims_enable_mask); 4073 wr32(IGC_IMS, ims); 4074 } else { 4075 wr32(IGC_IMS, IMS_ENABLE_MASK | IGC_IMS_DRSTA); 4076 wr32(IGC_IAM, IMS_ENABLE_MASK | IGC_IMS_DRSTA); 4077 } 4078 } 4079 4080 /** 4081 * igc_irq_disable - Mask off interrupt generation on the NIC 4082 * @adapter: board private structure 4083 */ 4084 static void igc_irq_disable(struct igc_adapter *adapter) 4085 { 4086 struct igc_hw *hw = &adapter->hw; 4087 4088 if (adapter->msix_entries) { 4089 u32 regval = rd32(IGC_EIAM); 4090 4091 wr32(IGC_EIAM, regval & ~adapter->eims_enable_mask); 4092 wr32(IGC_EIMC, adapter->eims_enable_mask); 4093 regval = rd32(IGC_EIAC); 4094 wr32(IGC_EIAC, regval & ~adapter->eims_enable_mask); 4095 } 4096 4097 wr32(IGC_IAM, 0); 4098 wr32(IGC_IMC, ~0); 4099 wrfl(); 4100 4101 if (adapter->msix_entries) { 4102 int vector = 0, i; 4103 4104 synchronize_irq(adapter->msix_entries[vector++].vector); 4105 4106 for (i = 0; i < adapter->num_q_vectors; i++) 4107 synchronize_irq(adapter->msix_entries[vector++].vector); 4108 } else { 4109 synchronize_irq(adapter->pdev->irq); 4110 } 4111 } 4112 4113 void igc_set_flag_queue_pairs(struct igc_adapter *adapter, 4114 const u32 max_rss_queues) 4115 { 4116 /* Determine if we need to pair queues. */ 4117 /* If rss_queues > half of max_rss_queues, pair the queues in 4118 * order to conserve interrupts due to limited supply. 4119 */ 4120 if (adapter->rss_queues > (max_rss_queues / 2)) 4121 adapter->flags |= IGC_FLAG_QUEUE_PAIRS; 4122 else 4123 adapter->flags &= ~IGC_FLAG_QUEUE_PAIRS; 4124 } 4125 4126 unsigned int igc_get_max_rss_queues(struct igc_adapter *adapter) 4127 { 4128 return IGC_MAX_RX_QUEUES; 4129 } 4130 4131 static void igc_init_queue_configuration(struct igc_adapter *adapter) 4132 { 4133 u32 max_rss_queues; 4134 4135 max_rss_queues = igc_get_max_rss_queues(adapter); 4136 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus()); 4137 4138 igc_set_flag_queue_pairs(adapter, max_rss_queues); 4139 } 4140 4141 /** 4142 * igc_reset_q_vector - Reset config for interrupt vector 4143 * @adapter: board private structure to initialize 4144 * @v_idx: Index of vector to be reset 4145 * 4146 * If NAPI is enabled it will delete any references to the 4147 * NAPI struct. This is preparation for igc_free_q_vector. 4148 */ 4149 static void igc_reset_q_vector(struct igc_adapter *adapter, int v_idx) 4150 { 4151 struct igc_q_vector *q_vector = adapter->q_vector[v_idx]; 4152 4153 /* if we're coming from igc_set_interrupt_capability, the vectors are 4154 * not yet allocated 4155 */ 4156 if (!q_vector) 4157 return; 4158 4159 if (q_vector->tx.ring) 4160 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL; 4161 4162 if (q_vector->rx.ring) 4163 adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL; 4164 4165 netif_napi_del(&q_vector->napi); 4166 } 4167 4168 /** 4169 * igc_free_q_vector - Free memory allocated for specific interrupt vector 4170 * @adapter: board private structure to initialize 4171 * @v_idx: Index of vector to be freed 4172 * 4173 * This function frees the memory allocated to the q_vector. 4174 */ 4175 static void igc_free_q_vector(struct igc_adapter *adapter, int v_idx) 4176 { 4177 struct igc_q_vector *q_vector = adapter->q_vector[v_idx]; 4178 4179 adapter->q_vector[v_idx] = NULL; 4180 4181 /* igc_get_stats64() might access the rings on this vector, 4182 * we must wait a grace period before freeing it. 4183 */ 4184 if (q_vector) 4185 kfree_rcu(q_vector, rcu); 4186 } 4187 4188 /** 4189 * igc_free_q_vectors - Free memory allocated for interrupt vectors 4190 * @adapter: board private structure to initialize 4191 * 4192 * This function frees the memory allocated to the q_vectors. In addition if 4193 * NAPI is enabled it will delete any references to the NAPI struct prior 4194 * to freeing the q_vector. 4195 */ 4196 static void igc_free_q_vectors(struct igc_adapter *adapter) 4197 { 4198 int v_idx = adapter->num_q_vectors; 4199 4200 adapter->num_tx_queues = 0; 4201 adapter->num_rx_queues = 0; 4202 adapter->num_q_vectors = 0; 4203 4204 while (v_idx--) { 4205 igc_reset_q_vector(adapter, v_idx); 4206 igc_free_q_vector(adapter, v_idx); 4207 } 4208 } 4209 4210 /** 4211 * igc_update_itr - update the dynamic ITR value based on statistics 4212 * @q_vector: pointer to q_vector 4213 * @ring_container: ring info to update the itr for 4214 * 4215 * Stores a new ITR value based on packets and byte 4216 * counts during the last interrupt. The advantage of per interrupt 4217 * computation is faster updates and more accurate ITR for the current 4218 * traffic pattern. Constants in this function were computed 4219 * based on theoretical maximum wire speed and thresholds were set based 4220 * on testing data as well as attempting to minimize response time 4221 * while increasing bulk throughput. 4222 * NOTE: These calculations are only valid when operating in a single- 4223 * queue environment. 4224 */ 4225 static void igc_update_itr(struct igc_q_vector *q_vector, 4226 struct igc_ring_container *ring_container) 4227 { 4228 unsigned int packets = ring_container->total_packets; 4229 unsigned int bytes = ring_container->total_bytes; 4230 u8 itrval = ring_container->itr; 4231 4232 /* no packets, exit with status unchanged */ 4233 if (packets == 0) 4234 return; 4235 4236 switch (itrval) { 4237 case lowest_latency: 4238 /* handle TSO and jumbo frames */ 4239 if (bytes / packets > 8000) 4240 itrval = bulk_latency; 4241 else if ((packets < 5) && (bytes > 512)) 4242 itrval = low_latency; 4243 break; 4244 case low_latency: /* 50 usec aka 20000 ints/s */ 4245 if (bytes > 10000) { 4246 /* this if handles the TSO accounting */ 4247 if (bytes / packets > 8000) 4248 itrval = bulk_latency; 4249 else if ((packets < 10) || ((bytes / packets) > 1200)) 4250 itrval = bulk_latency; 4251 else if ((packets > 35)) 4252 itrval = lowest_latency; 4253 } else if (bytes / packets > 2000) { 4254 itrval = bulk_latency; 4255 } else if (packets <= 2 && bytes < 512) { 4256 itrval = lowest_latency; 4257 } 4258 break; 4259 case bulk_latency: /* 250 usec aka 4000 ints/s */ 4260 if (bytes > 25000) { 4261 if (packets > 35) 4262 itrval = low_latency; 4263 } else if (bytes < 1500) { 4264 itrval = low_latency; 4265 } 4266 break; 4267 } 4268 4269 /* clear work counters since we have the values we need */ 4270 ring_container->total_bytes = 0; 4271 ring_container->total_packets = 0; 4272 4273 /* write updated itr to ring container */ 4274 ring_container->itr = itrval; 4275 } 4276 4277 static void igc_set_itr(struct igc_q_vector *q_vector) 4278 { 4279 struct igc_adapter *adapter = q_vector->adapter; 4280 u32 new_itr = q_vector->itr_val; 4281 u8 current_itr = 0; 4282 4283 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */ 4284 switch (adapter->link_speed) { 4285 case SPEED_10: 4286 case SPEED_100: 4287 current_itr = 0; 4288 new_itr = IGC_4K_ITR; 4289 goto set_itr_now; 4290 default: 4291 break; 4292 } 4293 4294 igc_update_itr(q_vector, &q_vector->tx); 4295 igc_update_itr(q_vector, &q_vector->rx); 4296 4297 current_itr = max(q_vector->rx.itr, q_vector->tx.itr); 4298 4299 /* conservative mode (itr 3) eliminates the lowest_latency setting */ 4300 if (current_itr == lowest_latency && 4301 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) || 4302 (!q_vector->rx.ring && adapter->tx_itr_setting == 3))) 4303 current_itr = low_latency; 4304 4305 switch (current_itr) { 4306 /* counts and packets in update_itr are dependent on these numbers */ 4307 case lowest_latency: 4308 new_itr = IGC_70K_ITR; /* 70,000 ints/sec */ 4309 break; 4310 case low_latency: 4311 new_itr = IGC_20K_ITR; /* 20,000 ints/sec */ 4312 break; 4313 case bulk_latency: 4314 new_itr = IGC_4K_ITR; /* 4,000 ints/sec */ 4315 break; 4316 default: 4317 break; 4318 } 4319 4320 set_itr_now: 4321 if (new_itr != q_vector->itr_val) { 4322 /* this attempts to bias the interrupt rate towards Bulk 4323 * by adding intermediate steps when interrupt rate is 4324 * increasing 4325 */ 4326 new_itr = new_itr > q_vector->itr_val ? 4327 max((new_itr * q_vector->itr_val) / 4328 (new_itr + (q_vector->itr_val >> 2)), 4329 new_itr) : new_itr; 4330 /* Don't write the value here; it resets the adapter's 4331 * internal timer, and causes us to delay far longer than 4332 * we should between interrupts. Instead, we write the ITR 4333 * value at the beginning of the next interrupt so the timing 4334 * ends up being correct. 4335 */ 4336 q_vector->itr_val = new_itr; 4337 q_vector->set_itr = 1; 4338 } 4339 } 4340 4341 static void igc_reset_interrupt_capability(struct igc_adapter *adapter) 4342 { 4343 int v_idx = adapter->num_q_vectors; 4344 4345 if (adapter->msix_entries) { 4346 pci_disable_msix(adapter->pdev); 4347 kfree(adapter->msix_entries); 4348 adapter->msix_entries = NULL; 4349 } else if (adapter->flags & IGC_FLAG_HAS_MSI) { 4350 pci_disable_msi(adapter->pdev); 4351 } 4352 4353 while (v_idx--) 4354 igc_reset_q_vector(adapter, v_idx); 4355 } 4356 4357 /** 4358 * igc_set_interrupt_capability - set MSI or MSI-X if supported 4359 * @adapter: Pointer to adapter structure 4360 * @msix: boolean value for MSI-X capability 4361 * 4362 * Attempt to configure interrupts using the best available 4363 * capabilities of the hardware and kernel. 4364 */ 4365 static void igc_set_interrupt_capability(struct igc_adapter *adapter, 4366 bool msix) 4367 { 4368 int numvecs, i; 4369 int err; 4370 4371 if (!msix) 4372 goto msi_only; 4373 adapter->flags |= IGC_FLAG_HAS_MSIX; 4374 4375 /* Number of supported queues. */ 4376 adapter->num_rx_queues = adapter->rss_queues; 4377 4378 adapter->num_tx_queues = adapter->rss_queues; 4379 4380 /* start with one vector for every Rx queue */ 4381 numvecs = adapter->num_rx_queues; 4382 4383 /* if Tx handler is separate add 1 for every Tx queue */ 4384 if (!(adapter->flags & IGC_FLAG_QUEUE_PAIRS)) 4385 numvecs += adapter->num_tx_queues; 4386 4387 /* store the number of vectors reserved for queues */ 4388 adapter->num_q_vectors = numvecs; 4389 4390 /* add 1 vector for link status interrupts */ 4391 numvecs++; 4392 4393 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry), 4394 GFP_KERNEL); 4395 4396 if (!adapter->msix_entries) 4397 return; 4398 4399 /* populate entry values */ 4400 for (i = 0; i < numvecs; i++) 4401 adapter->msix_entries[i].entry = i; 4402 4403 err = pci_enable_msix_range(adapter->pdev, 4404 adapter->msix_entries, 4405 numvecs, 4406 numvecs); 4407 if (err > 0) 4408 return; 4409 4410 kfree(adapter->msix_entries); 4411 adapter->msix_entries = NULL; 4412 4413 igc_reset_interrupt_capability(adapter); 4414 4415 msi_only: 4416 adapter->flags &= ~IGC_FLAG_HAS_MSIX; 4417 4418 adapter->rss_queues = 1; 4419 adapter->flags |= IGC_FLAG_QUEUE_PAIRS; 4420 adapter->num_rx_queues = 1; 4421 adapter->num_tx_queues = 1; 4422 adapter->num_q_vectors = 1; 4423 if (!pci_enable_msi(adapter->pdev)) 4424 adapter->flags |= IGC_FLAG_HAS_MSI; 4425 } 4426 4427 /** 4428 * igc_update_ring_itr - update the dynamic ITR value based on packet size 4429 * @q_vector: pointer to q_vector 4430 * 4431 * Stores a new ITR value based on strictly on packet size. This 4432 * algorithm is less sophisticated than that used in igc_update_itr, 4433 * due to the difficulty of synchronizing statistics across multiple 4434 * receive rings. The divisors and thresholds used by this function 4435 * were determined based on theoretical maximum wire speed and testing 4436 * data, in order to minimize response time while increasing bulk 4437 * throughput. 4438 * NOTE: This function is called only when operating in a multiqueue 4439 * receive environment. 4440 */ 4441 static void igc_update_ring_itr(struct igc_q_vector *q_vector) 4442 { 4443 struct igc_adapter *adapter = q_vector->adapter; 4444 int new_val = q_vector->itr_val; 4445 int avg_wire_size = 0; 4446 unsigned int packets; 4447 4448 /* For non-gigabit speeds, just fix the interrupt rate at 4000 4449 * ints/sec - ITR timer value of 120 ticks. 4450 */ 4451 switch (adapter->link_speed) { 4452 case SPEED_10: 4453 case SPEED_100: 4454 new_val = IGC_4K_ITR; 4455 goto set_itr_val; 4456 default: 4457 break; 4458 } 4459 4460 packets = q_vector->rx.total_packets; 4461 if (packets) 4462 avg_wire_size = q_vector->rx.total_bytes / packets; 4463 4464 packets = q_vector->tx.total_packets; 4465 if (packets) 4466 avg_wire_size = max_t(u32, avg_wire_size, 4467 q_vector->tx.total_bytes / packets); 4468 4469 /* if avg_wire_size isn't set no work was done */ 4470 if (!avg_wire_size) 4471 goto clear_counts; 4472 4473 /* Add 24 bytes to size to account for CRC, preamble, and gap */ 4474 avg_wire_size += 24; 4475 4476 /* Don't starve jumbo frames */ 4477 avg_wire_size = min(avg_wire_size, 3000); 4478 4479 /* Give a little boost to mid-size frames */ 4480 if (avg_wire_size > 300 && avg_wire_size < 1200) 4481 new_val = avg_wire_size / 3; 4482 else 4483 new_val = avg_wire_size / 2; 4484 4485 /* conservative mode (itr 3) eliminates the lowest_latency setting */ 4486 if (new_val < IGC_20K_ITR && 4487 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) || 4488 (!q_vector->rx.ring && adapter->tx_itr_setting == 3))) 4489 new_val = IGC_20K_ITR; 4490 4491 set_itr_val: 4492 if (new_val != q_vector->itr_val) { 4493 q_vector->itr_val = new_val; 4494 q_vector->set_itr = 1; 4495 } 4496 clear_counts: 4497 q_vector->rx.total_bytes = 0; 4498 q_vector->rx.total_packets = 0; 4499 q_vector->tx.total_bytes = 0; 4500 q_vector->tx.total_packets = 0; 4501 } 4502 4503 static void igc_ring_irq_enable(struct igc_q_vector *q_vector) 4504 { 4505 struct igc_adapter *adapter = q_vector->adapter; 4506 struct igc_hw *hw = &adapter->hw; 4507 4508 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) || 4509 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) { 4510 if (adapter->num_q_vectors == 1) 4511 igc_set_itr(q_vector); 4512 else 4513 igc_update_ring_itr(q_vector); 4514 } 4515 4516 if (!test_bit(__IGC_DOWN, &adapter->state)) { 4517 if (adapter->msix_entries) 4518 wr32(IGC_EIMS, q_vector->eims_value); 4519 else 4520 igc_irq_enable(adapter); 4521 } 4522 } 4523 4524 static void igc_add_ring(struct igc_ring *ring, 4525 struct igc_ring_container *head) 4526 { 4527 head->ring = ring; 4528 head->count++; 4529 } 4530 4531 /** 4532 * igc_cache_ring_register - Descriptor ring to register mapping 4533 * @adapter: board private structure to initialize 4534 * 4535 * Once we know the feature-set enabled for the device, we'll cache 4536 * the register offset the descriptor ring is assigned to. 4537 */ 4538 static void igc_cache_ring_register(struct igc_adapter *adapter) 4539 { 4540 int i = 0, j = 0; 4541 4542 switch (adapter->hw.mac.type) { 4543 case igc_i225: 4544 default: 4545 for (; i < adapter->num_rx_queues; i++) 4546 adapter->rx_ring[i]->reg_idx = i; 4547 for (; j < adapter->num_tx_queues; j++) 4548 adapter->tx_ring[j]->reg_idx = j; 4549 break; 4550 } 4551 } 4552 4553 /** 4554 * igc_poll - NAPI Rx polling callback 4555 * @napi: napi polling structure 4556 * @budget: count of how many packets we should handle 4557 */ 4558 static int igc_poll(struct napi_struct *napi, int budget) 4559 { 4560 struct igc_q_vector *q_vector = container_of(napi, 4561 struct igc_q_vector, 4562 napi); 4563 struct igc_ring *rx_ring = q_vector->rx.ring; 4564 bool clean_complete = true; 4565 int work_done = 0; 4566 4567 if (q_vector->tx.ring) 4568 clean_complete = igc_clean_tx_irq(q_vector, budget); 4569 4570 if (rx_ring) { 4571 int cleaned = rx_ring->xsk_pool ? 4572 igc_clean_rx_irq_zc(q_vector, budget) : 4573 igc_clean_rx_irq(q_vector, budget); 4574 4575 work_done += cleaned; 4576 if (cleaned >= budget) 4577 clean_complete = false; 4578 } 4579 4580 /* If all work not completed, return budget and keep polling */ 4581 if (!clean_complete) 4582 return budget; 4583 4584 /* Exit the polling mode, but don't re-enable interrupts if stack might 4585 * poll us due to busy-polling 4586 */ 4587 if (likely(napi_complete_done(napi, work_done))) 4588 igc_ring_irq_enable(q_vector); 4589 4590 return min(work_done, budget - 1); 4591 } 4592 4593 /** 4594 * igc_alloc_q_vector - Allocate memory for a single interrupt vector 4595 * @adapter: board private structure to initialize 4596 * @v_count: q_vectors allocated on adapter, used for ring interleaving 4597 * @v_idx: index of vector in adapter struct 4598 * @txr_count: total number of Tx rings to allocate 4599 * @txr_idx: index of first Tx ring to allocate 4600 * @rxr_count: total number of Rx rings to allocate 4601 * @rxr_idx: index of first Rx ring to allocate 4602 * 4603 * We allocate one q_vector. If allocation fails we return -ENOMEM. 4604 */ 4605 static int igc_alloc_q_vector(struct igc_adapter *adapter, 4606 unsigned int v_count, unsigned int v_idx, 4607 unsigned int txr_count, unsigned int txr_idx, 4608 unsigned int rxr_count, unsigned int rxr_idx) 4609 { 4610 struct igc_q_vector *q_vector; 4611 struct igc_ring *ring; 4612 int ring_count; 4613 4614 /* igc only supports 1 Tx and/or 1 Rx queue per vector */ 4615 if (txr_count > 1 || rxr_count > 1) 4616 return -ENOMEM; 4617 4618 ring_count = txr_count + rxr_count; 4619 4620 /* allocate q_vector and rings */ 4621 q_vector = adapter->q_vector[v_idx]; 4622 if (!q_vector) 4623 q_vector = kzalloc(struct_size(q_vector, ring, ring_count), 4624 GFP_KERNEL); 4625 else 4626 memset(q_vector, 0, struct_size(q_vector, ring, ring_count)); 4627 if (!q_vector) 4628 return -ENOMEM; 4629 4630 /* initialize NAPI */ 4631 netif_napi_add(adapter->netdev, &q_vector->napi, igc_poll); 4632 4633 /* tie q_vector and adapter together */ 4634 adapter->q_vector[v_idx] = q_vector; 4635 q_vector->adapter = adapter; 4636 4637 /* initialize work limits */ 4638 q_vector->tx.work_limit = adapter->tx_work_limit; 4639 4640 /* initialize ITR configuration */ 4641 q_vector->itr_register = adapter->io_addr + IGC_EITR(0); 4642 q_vector->itr_val = IGC_START_ITR; 4643 4644 /* initialize pointer to rings */ 4645 ring = q_vector->ring; 4646 4647 /* initialize ITR */ 4648 if (rxr_count) { 4649 /* rx or rx/tx vector */ 4650 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3) 4651 q_vector->itr_val = adapter->rx_itr_setting; 4652 } else { 4653 /* tx only vector */ 4654 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3) 4655 q_vector->itr_val = adapter->tx_itr_setting; 4656 } 4657 4658 if (txr_count) { 4659 /* assign generic ring traits */ 4660 ring->dev = &adapter->pdev->dev; 4661 ring->netdev = adapter->netdev; 4662 4663 /* configure backlink on ring */ 4664 ring->q_vector = q_vector; 4665 4666 /* update q_vector Tx values */ 4667 igc_add_ring(ring, &q_vector->tx); 4668 4669 /* apply Tx specific ring traits */ 4670 ring->count = adapter->tx_ring_count; 4671 ring->queue_index = txr_idx; 4672 4673 /* assign ring to adapter */ 4674 adapter->tx_ring[txr_idx] = ring; 4675 4676 /* push pointer to next ring */ 4677 ring++; 4678 } 4679 4680 if (rxr_count) { 4681 /* assign generic ring traits */ 4682 ring->dev = &adapter->pdev->dev; 4683 ring->netdev = adapter->netdev; 4684 4685 /* configure backlink on ring */ 4686 ring->q_vector = q_vector; 4687 4688 /* update q_vector Rx values */ 4689 igc_add_ring(ring, &q_vector->rx); 4690 4691 /* apply Rx specific ring traits */ 4692 ring->count = adapter->rx_ring_count; 4693 ring->queue_index = rxr_idx; 4694 4695 /* assign ring to adapter */ 4696 adapter->rx_ring[rxr_idx] = ring; 4697 } 4698 4699 return 0; 4700 } 4701 4702 /** 4703 * igc_alloc_q_vectors - Allocate memory for interrupt vectors 4704 * @adapter: board private structure to initialize 4705 * 4706 * We allocate one q_vector per queue interrupt. If allocation fails we 4707 * return -ENOMEM. 4708 */ 4709 static int igc_alloc_q_vectors(struct igc_adapter *adapter) 4710 { 4711 int rxr_remaining = adapter->num_rx_queues; 4712 int txr_remaining = adapter->num_tx_queues; 4713 int rxr_idx = 0, txr_idx = 0, v_idx = 0; 4714 int q_vectors = adapter->num_q_vectors; 4715 int err; 4716 4717 if (q_vectors >= (rxr_remaining + txr_remaining)) { 4718 for (; rxr_remaining; v_idx++) { 4719 err = igc_alloc_q_vector(adapter, q_vectors, v_idx, 4720 0, 0, 1, rxr_idx); 4721 4722 if (err) 4723 goto err_out; 4724 4725 /* update counts and index */ 4726 rxr_remaining--; 4727 rxr_idx++; 4728 } 4729 } 4730 4731 for (; v_idx < q_vectors; v_idx++) { 4732 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx); 4733 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx); 4734 4735 err = igc_alloc_q_vector(adapter, q_vectors, v_idx, 4736 tqpv, txr_idx, rqpv, rxr_idx); 4737 4738 if (err) 4739 goto err_out; 4740 4741 /* update counts and index */ 4742 rxr_remaining -= rqpv; 4743 txr_remaining -= tqpv; 4744 rxr_idx++; 4745 txr_idx++; 4746 } 4747 4748 return 0; 4749 4750 err_out: 4751 adapter->num_tx_queues = 0; 4752 adapter->num_rx_queues = 0; 4753 adapter->num_q_vectors = 0; 4754 4755 while (v_idx--) 4756 igc_free_q_vector(adapter, v_idx); 4757 4758 return -ENOMEM; 4759 } 4760 4761 /** 4762 * igc_init_interrupt_scheme - initialize interrupts, allocate queues/vectors 4763 * @adapter: Pointer to adapter structure 4764 * @msix: boolean for MSI-X capability 4765 * 4766 * This function initializes the interrupts and allocates all of the queues. 4767 */ 4768 static int igc_init_interrupt_scheme(struct igc_adapter *adapter, bool msix) 4769 { 4770 struct net_device *dev = adapter->netdev; 4771 int err = 0; 4772 4773 igc_set_interrupt_capability(adapter, msix); 4774 4775 err = igc_alloc_q_vectors(adapter); 4776 if (err) { 4777 netdev_err(dev, "Unable to allocate memory for vectors\n"); 4778 goto err_alloc_q_vectors; 4779 } 4780 4781 igc_cache_ring_register(adapter); 4782 4783 return 0; 4784 4785 err_alloc_q_vectors: 4786 igc_reset_interrupt_capability(adapter); 4787 return err; 4788 } 4789 4790 /** 4791 * igc_sw_init - Initialize general software structures (struct igc_adapter) 4792 * @adapter: board private structure to initialize 4793 * 4794 * igc_sw_init initializes the Adapter private data structure. 4795 * Fields are initialized based on PCI device information and 4796 * OS network device settings (MTU size). 4797 */ 4798 static int igc_sw_init(struct igc_adapter *adapter) 4799 { 4800 struct net_device *netdev = adapter->netdev; 4801 struct pci_dev *pdev = adapter->pdev; 4802 struct igc_hw *hw = &adapter->hw; 4803 4804 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word); 4805 4806 /* set default ring sizes */ 4807 adapter->tx_ring_count = IGC_DEFAULT_TXD; 4808 adapter->rx_ring_count = IGC_DEFAULT_RXD; 4809 4810 /* set default ITR values */ 4811 adapter->rx_itr_setting = IGC_DEFAULT_ITR; 4812 adapter->tx_itr_setting = IGC_DEFAULT_ITR; 4813 4814 /* set default work limits */ 4815 adapter->tx_work_limit = IGC_DEFAULT_TX_WORK; 4816 4817 /* adjust max frame to be at least the size of a standard frame */ 4818 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN + 4819 VLAN_HLEN; 4820 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN; 4821 4822 mutex_init(&adapter->nfc_rule_lock); 4823 INIT_LIST_HEAD(&adapter->nfc_rule_list); 4824 adapter->nfc_rule_count = 0; 4825 4826 spin_lock_init(&adapter->stats64_lock); 4827 spin_lock_init(&adapter->qbv_tx_lock); 4828 /* Assume MSI-X interrupts, will be checked during IRQ allocation */ 4829 adapter->flags |= IGC_FLAG_HAS_MSIX; 4830 4831 igc_init_queue_configuration(adapter); 4832 4833 /* This call may decrease the number of queues */ 4834 if (igc_init_interrupt_scheme(adapter, true)) { 4835 netdev_err(netdev, "Unable to allocate memory for queues\n"); 4836 return -ENOMEM; 4837 } 4838 4839 /* Explicitly disable IRQ since the NIC can be in any state. */ 4840 igc_irq_disable(adapter); 4841 4842 set_bit(__IGC_DOWN, &adapter->state); 4843 4844 return 0; 4845 } 4846 4847 /** 4848 * igc_up - Open the interface and prepare it to handle traffic 4849 * @adapter: board private structure 4850 */ 4851 void igc_up(struct igc_adapter *adapter) 4852 { 4853 struct igc_hw *hw = &adapter->hw; 4854 int i = 0; 4855 4856 /* hardware has been reset, we need to reload some things */ 4857 igc_configure(adapter); 4858 4859 clear_bit(__IGC_DOWN, &adapter->state); 4860 4861 for (i = 0; i < adapter->num_q_vectors; i++) 4862 napi_enable(&adapter->q_vector[i]->napi); 4863 4864 if (adapter->msix_entries) 4865 igc_configure_msix(adapter); 4866 else 4867 igc_assign_vector(adapter->q_vector[0], 0); 4868 4869 /* Clear any pending interrupts. */ 4870 rd32(IGC_ICR); 4871 igc_irq_enable(adapter); 4872 4873 netif_tx_start_all_queues(adapter->netdev); 4874 4875 /* start the watchdog. */ 4876 hw->mac.get_link_status = true; 4877 schedule_work(&adapter->watchdog_task); 4878 } 4879 4880 /** 4881 * igc_update_stats - Update the board statistics counters 4882 * @adapter: board private structure 4883 */ 4884 void igc_update_stats(struct igc_adapter *adapter) 4885 { 4886 struct rtnl_link_stats64 *net_stats = &adapter->stats64; 4887 struct pci_dev *pdev = adapter->pdev; 4888 struct igc_hw *hw = &adapter->hw; 4889 u64 _bytes, _packets; 4890 u64 bytes, packets; 4891 unsigned int start; 4892 u32 mpc; 4893 int i; 4894 4895 /* Prevent stats update while adapter is being reset, or if the pci 4896 * connection is down. 4897 */ 4898 if (adapter->link_speed == 0) 4899 return; 4900 if (pci_channel_offline(pdev)) 4901 return; 4902 4903 packets = 0; 4904 bytes = 0; 4905 4906 rcu_read_lock(); 4907 for (i = 0; i < adapter->num_rx_queues; i++) { 4908 struct igc_ring *ring = adapter->rx_ring[i]; 4909 u32 rqdpc = rd32(IGC_RQDPC(i)); 4910 4911 if (hw->mac.type >= igc_i225) 4912 wr32(IGC_RQDPC(i), 0); 4913 4914 if (rqdpc) { 4915 ring->rx_stats.drops += rqdpc; 4916 net_stats->rx_fifo_errors += rqdpc; 4917 } 4918 4919 do { 4920 start = u64_stats_fetch_begin(&ring->rx_syncp); 4921 _bytes = ring->rx_stats.bytes; 4922 _packets = ring->rx_stats.packets; 4923 } while (u64_stats_fetch_retry(&ring->rx_syncp, start)); 4924 bytes += _bytes; 4925 packets += _packets; 4926 } 4927 4928 net_stats->rx_bytes = bytes; 4929 net_stats->rx_packets = packets; 4930 4931 packets = 0; 4932 bytes = 0; 4933 for (i = 0; i < adapter->num_tx_queues; i++) { 4934 struct igc_ring *ring = adapter->tx_ring[i]; 4935 4936 do { 4937 start = u64_stats_fetch_begin(&ring->tx_syncp); 4938 _bytes = ring->tx_stats.bytes; 4939 _packets = ring->tx_stats.packets; 4940 } while (u64_stats_fetch_retry(&ring->tx_syncp, start)); 4941 bytes += _bytes; 4942 packets += _packets; 4943 } 4944 net_stats->tx_bytes = bytes; 4945 net_stats->tx_packets = packets; 4946 rcu_read_unlock(); 4947 4948 /* read stats registers */ 4949 adapter->stats.crcerrs += rd32(IGC_CRCERRS); 4950 adapter->stats.gprc += rd32(IGC_GPRC); 4951 adapter->stats.gorc += rd32(IGC_GORCL); 4952 rd32(IGC_GORCH); /* clear GORCL */ 4953 adapter->stats.bprc += rd32(IGC_BPRC); 4954 adapter->stats.mprc += rd32(IGC_MPRC); 4955 adapter->stats.roc += rd32(IGC_ROC); 4956 4957 adapter->stats.prc64 += rd32(IGC_PRC64); 4958 adapter->stats.prc127 += rd32(IGC_PRC127); 4959 adapter->stats.prc255 += rd32(IGC_PRC255); 4960 adapter->stats.prc511 += rd32(IGC_PRC511); 4961 adapter->stats.prc1023 += rd32(IGC_PRC1023); 4962 adapter->stats.prc1522 += rd32(IGC_PRC1522); 4963 adapter->stats.tlpic += rd32(IGC_TLPIC); 4964 adapter->stats.rlpic += rd32(IGC_RLPIC); 4965 adapter->stats.hgptc += rd32(IGC_HGPTC); 4966 4967 mpc = rd32(IGC_MPC); 4968 adapter->stats.mpc += mpc; 4969 net_stats->rx_fifo_errors += mpc; 4970 adapter->stats.scc += rd32(IGC_SCC); 4971 adapter->stats.ecol += rd32(IGC_ECOL); 4972 adapter->stats.mcc += rd32(IGC_MCC); 4973 adapter->stats.latecol += rd32(IGC_LATECOL); 4974 adapter->stats.dc += rd32(IGC_DC); 4975 adapter->stats.rlec += rd32(IGC_RLEC); 4976 adapter->stats.xonrxc += rd32(IGC_XONRXC); 4977 adapter->stats.xontxc += rd32(IGC_XONTXC); 4978 adapter->stats.xoffrxc += rd32(IGC_XOFFRXC); 4979 adapter->stats.xofftxc += rd32(IGC_XOFFTXC); 4980 adapter->stats.fcruc += rd32(IGC_FCRUC); 4981 adapter->stats.gptc += rd32(IGC_GPTC); 4982 adapter->stats.gotc += rd32(IGC_GOTCL); 4983 rd32(IGC_GOTCH); /* clear GOTCL */ 4984 adapter->stats.rnbc += rd32(IGC_RNBC); 4985 adapter->stats.ruc += rd32(IGC_RUC); 4986 adapter->stats.rfc += rd32(IGC_RFC); 4987 adapter->stats.rjc += rd32(IGC_RJC); 4988 adapter->stats.tor += rd32(IGC_TORH); 4989 adapter->stats.tot += rd32(IGC_TOTH); 4990 adapter->stats.tpr += rd32(IGC_TPR); 4991 4992 adapter->stats.ptc64 += rd32(IGC_PTC64); 4993 adapter->stats.ptc127 += rd32(IGC_PTC127); 4994 adapter->stats.ptc255 += rd32(IGC_PTC255); 4995 adapter->stats.ptc511 += rd32(IGC_PTC511); 4996 adapter->stats.ptc1023 += rd32(IGC_PTC1023); 4997 adapter->stats.ptc1522 += rd32(IGC_PTC1522); 4998 4999 adapter->stats.mptc += rd32(IGC_MPTC); 5000 adapter->stats.bptc += rd32(IGC_BPTC); 5001 5002 adapter->stats.tpt += rd32(IGC_TPT); 5003 adapter->stats.colc += rd32(IGC_COLC); 5004 adapter->stats.colc += rd32(IGC_RERC); 5005 5006 adapter->stats.algnerrc += rd32(IGC_ALGNERRC); 5007 5008 adapter->stats.tsctc += rd32(IGC_TSCTC); 5009 5010 adapter->stats.iac += rd32(IGC_IAC); 5011 5012 /* Fill out the OS statistics structure */ 5013 net_stats->multicast = adapter->stats.mprc; 5014 net_stats->collisions = adapter->stats.colc; 5015 5016 /* Rx Errors */ 5017 5018 /* RLEC on some newer hardware can be incorrect so build 5019 * our own version based on RUC and ROC 5020 */ 5021 net_stats->rx_errors = adapter->stats.rxerrc + 5022 adapter->stats.crcerrs + adapter->stats.algnerrc + 5023 adapter->stats.ruc + adapter->stats.roc + 5024 adapter->stats.cexterr; 5025 net_stats->rx_length_errors = adapter->stats.ruc + 5026 adapter->stats.roc; 5027 net_stats->rx_crc_errors = adapter->stats.crcerrs; 5028 net_stats->rx_frame_errors = adapter->stats.algnerrc; 5029 net_stats->rx_missed_errors = adapter->stats.mpc; 5030 5031 /* Tx Errors */ 5032 net_stats->tx_errors = adapter->stats.ecol + 5033 adapter->stats.latecol; 5034 net_stats->tx_aborted_errors = adapter->stats.ecol; 5035 net_stats->tx_window_errors = adapter->stats.latecol; 5036 net_stats->tx_carrier_errors = adapter->stats.tncrs; 5037 5038 /* Tx Dropped */ 5039 net_stats->tx_dropped = adapter->stats.txdrop; 5040 5041 /* Management Stats */ 5042 adapter->stats.mgptc += rd32(IGC_MGTPTC); 5043 adapter->stats.mgprc += rd32(IGC_MGTPRC); 5044 adapter->stats.mgpdc += rd32(IGC_MGTPDC); 5045 } 5046 5047 /** 5048 * igc_down - Close the interface 5049 * @adapter: board private structure 5050 */ 5051 void igc_down(struct igc_adapter *adapter) 5052 { 5053 struct net_device *netdev = adapter->netdev; 5054 struct igc_hw *hw = &adapter->hw; 5055 u32 tctl, rctl; 5056 int i = 0; 5057 5058 set_bit(__IGC_DOWN, &adapter->state); 5059 5060 igc_ptp_suspend(adapter); 5061 5062 if (pci_device_is_present(adapter->pdev)) { 5063 /* disable receives in the hardware */ 5064 rctl = rd32(IGC_RCTL); 5065 wr32(IGC_RCTL, rctl & ~IGC_RCTL_EN); 5066 /* flush and sleep below */ 5067 } 5068 /* set trans_start so we don't get spurious watchdogs during reset */ 5069 netif_trans_update(netdev); 5070 5071 netif_carrier_off(netdev); 5072 netif_tx_stop_all_queues(netdev); 5073 5074 if (pci_device_is_present(adapter->pdev)) { 5075 /* disable transmits in the hardware */ 5076 tctl = rd32(IGC_TCTL); 5077 tctl &= ~IGC_TCTL_EN; 5078 wr32(IGC_TCTL, tctl); 5079 /* flush both disables and wait for them to finish */ 5080 wrfl(); 5081 usleep_range(10000, 20000); 5082 5083 igc_irq_disable(adapter); 5084 } 5085 5086 adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE; 5087 5088 for (i = 0; i < adapter->num_q_vectors; i++) { 5089 if (adapter->q_vector[i]) { 5090 napi_synchronize(&adapter->q_vector[i]->napi); 5091 napi_disable(&adapter->q_vector[i]->napi); 5092 } 5093 } 5094 5095 del_timer_sync(&adapter->watchdog_timer); 5096 del_timer_sync(&adapter->phy_info_timer); 5097 5098 /* record the stats before reset*/ 5099 spin_lock(&adapter->stats64_lock); 5100 igc_update_stats(adapter); 5101 spin_unlock(&adapter->stats64_lock); 5102 5103 adapter->link_speed = 0; 5104 adapter->link_duplex = 0; 5105 5106 if (!pci_channel_offline(adapter->pdev)) 5107 igc_reset(adapter); 5108 5109 /* clear VLAN promisc flag so VFTA will be updated if necessary */ 5110 adapter->flags &= ~IGC_FLAG_VLAN_PROMISC; 5111 5112 igc_disable_all_tx_rings_hw(adapter); 5113 igc_clean_all_tx_rings(adapter); 5114 igc_clean_all_rx_rings(adapter); 5115 } 5116 5117 void igc_reinit_locked(struct igc_adapter *adapter) 5118 { 5119 while (test_and_set_bit(__IGC_RESETTING, &adapter->state)) 5120 usleep_range(1000, 2000); 5121 igc_down(adapter); 5122 igc_up(adapter); 5123 clear_bit(__IGC_RESETTING, &adapter->state); 5124 } 5125 5126 static void igc_reset_task(struct work_struct *work) 5127 { 5128 struct igc_adapter *adapter; 5129 5130 adapter = container_of(work, struct igc_adapter, reset_task); 5131 5132 rtnl_lock(); 5133 /* If we're already down or resetting, just bail */ 5134 if (test_bit(__IGC_DOWN, &adapter->state) || 5135 test_bit(__IGC_RESETTING, &adapter->state)) { 5136 rtnl_unlock(); 5137 return; 5138 } 5139 5140 igc_rings_dump(adapter); 5141 igc_regs_dump(adapter); 5142 netdev_err(adapter->netdev, "Reset adapter\n"); 5143 igc_reinit_locked(adapter); 5144 rtnl_unlock(); 5145 } 5146 5147 /** 5148 * igc_change_mtu - Change the Maximum Transfer Unit 5149 * @netdev: network interface device structure 5150 * @new_mtu: new value for maximum frame size 5151 * 5152 * Returns 0 on success, negative on failure 5153 */ 5154 static int igc_change_mtu(struct net_device *netdev, int new_mtu) 5155 { 5156 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN; 5157 struct igc_adapter *adapter = netdev_priv(netdev); 5158 5159 if (igc_xdp_is_enabled(adapter) && new_mtu > ETH_DATA_LEN) { 5160 netdev_dbg(netdev, "Jumbo frames not supported with XDP"); 5161 return -EINVAL; 5162 } 5163 5164 /* adjust max frame to be at least the size of a standard frame */ 5165 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN)) 5166 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN; 5167 5168 while (test_and_set_bit(__IGC_RESETTING, &adapter->state)) 5169 usleep_range(1000, 2000); 5170 5171 /* igc_down has a dependency on max_frame_size */ 5172 adapter->max_frame_size = max_frame; 5173 5174 if (netif_running(netdev)) 5175 igc_down(adapter); 5176 5177 netdev_dbg(netdev, "changing MTU from %d to %d\n", netdev->mtu, new_mtu); 5178 netdev->mtu = new_mtu; 5179 5180 if (netif_running(netdev)) 5181 igc_up(adapter); 5182 else 5183 igc_reset(adapter); 5184 5185 clear_bit(__IGC_RESETTING, &adapter->state); 5186 5187 return 0; 5188 } 5189 5190 /** 5191 * igc_tx_timeout - Respond to a Tx Hang 5192 * @netdev: network interface device structure 5193 * @txqueue: queue number that timed out 5194 **/ 5195 static void igc_tx_timeout(struct net_device *netdev, 5196 unsigned int __always_unused txqueue) 5197 { 5198 struct igc_adapter *adapter = netdev_priv(netdev); 5199 struct igc_hw *hw = &adapter->hw; 5200 5201 /* Do the reset outside of interrupt context */ 5202 adapter->tx_timeout_count++; 5203 schedule_work(&adapter->reset_task); 5204 wr32(IGC_EICS, 5205 (adapter->eims_enable_mask & ~adapter->eims_other)); 5206 } 5207 5208 /** 5209 * igc_get_stats64 - Get System Network Statistics 5210 * @netdev: network interface device structure 5211 * @stats: rtnl_link_stats64 pointer 5212 * 5213 * Returns the address of the device statistics structure. 5214 * The statistics are updated here and also from the timer callback. 5215 */ 5216 static void igc_get_stats64(struct net_device *netdev, 5217 struct rtnl_link_stats64 *stats) 5218 { 5219 struct igc_adapter *adapter = netdev_priv(netdev); 5220 5221 spin_lock(&adapter->stats64_lock); 5222 if (!test_bit(__IGC_RESETTING, &adapter->state)) 5223 igc_update_stats(adapter); 5224 memcpy(stats, &adapter->stats64, sizeof(*stats)); 5225 spin_unlock(&adapter->stats64_lock); 5226 } 5227 5228 static netdev_features_t igc_fix_features(struct net_device *netdev, 5229 netdev_features_t features) 5230 { 5231 /* Since there is no support for separate Rx/Tx vlan accel 5232 * enable/disable make sure Tx flag is always in same state as Rx. 5233 */ 5234 if (features & NETIF_F_HW_VLAN_CTAG_RX) 5235 features |= NETIF_F_HW_VLAN_CTAG_TX; 5236 else 5237 features &= ~NETIF_F_HW_VLAN_CTAG_TX; 5238 5239 return features; 5240 } 5241 5242 static int igc_set_features(struct net_device *netdev, 5243 netdev_features_t features) 5244 { 5245 netdev_features_t changed = netdev->features ^ features; 5246 struct igc_adapter *adapter = netdev_priv(netdev); 5247 5248 if (changed & NETIF_F_HW_VLAN_CTAG_RX) 5249 igc_vlan_mode(netdev, features); 5250 5251 /* Add VLAN support */ 5252 if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE))) 5253 return 0; 5254 5255 if (!(features & NETIF_F_NTUPLE)) 5256 igc_flush_nfc_rules(adapter); 5257 5258 netdev->features = features; 5259 5260 if (netif_running(netdev)) 5261 igc_reinit_locked(adapter); 5262 else 5263 igc_reset(adapter); 5264 5265 return 1; 5266 } 5267 5268 static netdev_features_t 5269 igc_features_check(struct sk_buff *skb, struct net_device *dev, 5270 netdev_features_t features) 5271 { 5272 unsigned int network_hdr_len, mac_hdr_len; 5273 5274 /* Make certain the headers can be described by a context descriptor */ 5275 mac_hdr_len = skb_network_header(skb) - skb->data; 5276 if (unlikely(mac_hdr_len > IGC_MAX_MAC_HDR_LEN)) 5277 return features & ~(NETIF_F_HW_CSUM | 5278 NETIF_F_SCTP_CRC | 5279 NETIF_F_HW_VLAN_CTAG_TX | 5280 NETIF_F_TSO | 5281 NETIF_F_TSO6); 5282 5283 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb); 5284 if (unlikely(network_hdr_len > IGC_MAX_NETWORK_HDR_LEN)) 5285 return features & ~(NETIF_F_HW_CSUM | 5286 NETIF_F_SCTP_CRC | 5287 NETIF_F_TSO | 5288 NETIF_F_TSO6); 5289 5290 /* We can only support IPv4 TSO in tunnels if we can mangle the 5291 * inner IP ID field, so strip TSO if MANGLEID is not supported. 5292 */ 5293 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) 5294 features &= ~NETIF_F_TSO; 5295 5296 return features; 5297 } 5298 5299 static void igc_tsync_interrupt(struct igc_adapter *adapter) 5300 { 5301 struct igc_hw *hw = &adapter->hw; 5302 u32 tsauxc, sec, nsec, tsicr; 5303 struct ptp_clock_event event; 5304 struct timespec64 ts; 5305 5306 tsicr = rd32(IGC_TSICR); 5307 5308 if (tsicr & IGC_TSICR_SYS_WRAP) { 5309 event.type = PTP_CLOCK_PPS; 5310 if (adapter->ptp_caps.pps) 5311 ptp_clock_event(adapter->ptp_clock, &event); 5312 } 5313 5314 if (tsicr & IGC_TSICR_TXTS) { 5315 /* retrieve hardware timestamp */ 5316 igc_ptp_tx_tstamp_event(adapter); 5317 } 5318 5319 if (tsicr & IGC_TSICR_TT0) { 5320 spin_lock(&adapter->tmreg_lock); 5321 ts = timespec64_add(adapter->perout[0].start, 5322 adapter->perout[0].period); 5323 wr32(IGC_TRGTTIML0, ts.tv_nsec | IGC_TT_IO_TIMER_SEL_SYSTIM0); 5324 wr32(IGC_TRGTTIMH0, (u32)ts.tv_sec); 5325 tsauxc = rd32(IGC_TSAUXC); 5326 tsauxc |= IGC_TSAUXC_EN_TT0; 5327 wr32(IGC_TSAUXC, tsauxc); 5328 adapter->perout[0].start = ts; 5329 spin_unlock(&adapter->tmreg_lock); 5330 } 5331 5332 if (tsicr & IGC_TSICR_TT1) { 5333 spin_lock(&adapter->tmreg_lock); 5334 ts = timespec64_add(adapter->perout[1].start, 5335 adapter->perout[1].period); 5336 wr32(IGC_TRGTTIML1, ts.tv_nsec | IGC_TT_IO_TIMER_SEL_SYSTIM0); 5337 wr32(IGC_TRGTTIMH1, (u32)ts.tv_sec); 5338 tsauxc = rd32(IGC_TSAUXC); 5339 tsauxc |= IGC_TSAUXC_EN_TT1; 5340 wr32(IGC_TSAUXC, tsauxc); 5341 adapter->perout[1].start = ts; 5342 spin_unlock(&adapter->tmreg_lock); 5343 } 5344 5345 if (tsicr & IGC_TSICR_AUTT0) { 5346 nsec = rd32(IGC_AUXSTMPL0); 5347 sec = rd32(IGC_AUXSTMPH0); 5348 event.type = PTP_CLOCK_EXTTS; 5349 event.index = 0; 5350 event.timestamp = sec * NSEC_PER_SEC + nsec; 5351 ptp_clock_event(adapter->ptp_clock, &event); 5352 } 5353 5354 if (tsicr & IGC_TSICR_AUTT1) { 5355 nsec = rd32(IGC_AUXSTMPL1); 5356 sec = rd32(IGC_AUXSTMPH1); 5357 event.type = PTP_CLOCK_EXTTS; 5358 event.index = 1; 5359 event.timestamp = sec * NSEC_PER_SEC + nsec; 5360 ptp_clock_event(adapter->ptp_clock, &event); 5361 } 5362 } 5363 5364 /** 5365 * igc_msix_other - msix other interrupt handler 5366 * @irq: interrupt number 5367 * @data: pointer to a q_vector 5368 */ 5369 static irqreturn_t igc_msix_other(int irq, void *data) 5370 { 5371 struct igc_adapter *adapter = data; 5372 struct igc_hw *hw = &adapter->hw; 5373 u32 icr = rd32(IGC_ICR); 5374 5375 /* reading ICR causes bit 31 of EICR to be cleared */ 5376 if (icr & IGC_ICR_DRSTA) 5377 schedule_work(&adapter->reset_task); 5378 5379 if (icr & IGC_ICR_DOUTSYNC) { 5380 /* HW is reporting DMA is out of sync */ 5381 adapter->stats.doosync++; 5382 } 5383 5384 if (icr & IGC_ICR_LSC) { 5385 hw->mac.get_link_status = true; 5386 /* guard against interrupt when we're going down */ 5387 if (!test_bit(__IGC_DOWN, &adapter->state)) 5388 mod_timer(&adapter->watchdog_timer, jiffies + 1); 5389 } 5390 5391 if (icr & IGC_ICR_TS) 5392 igc_tsync_interrupt(adapter); 5393 5394 wr32(IGC_EIMS, adapter->eims_other); 5395 5396 return IRQ_HANDLED; 5397 } 5398 5399 static void igc_write_itr(struct igc_q_vector *q_vector) 5400 { 5401 u32 itr_val = q_vector->itr_val & IGC_QVECTOR_MASK; 5402 5403 if (!q_vector->set_itr) 5404 return; 5405 5406 if (!itr_val) 5407 itr_val = IGC_ITR_VAL_MASK; 5408 5409 itr_val |= IGC_EITR_CNT_IGNR; 5410 5411 writel(itr_val, q_vector->itr_register); 5412 q_vector->set_itr = 0; 5413 } 5414 5415 static irqreturn_t igc_msix_ring(int irq, void *data) 5416 { 5417 struct igc_q_vector *q_vector = data; 5418 5419 /* Write the ITR value calculated from the previous interrupt. */ 5420 igc_write_itr(q_vector); 5421 5422 napi_schedule(&q_vector->napi); 5423 5424 return IRQ_HANDLED; 5425 } 5426 5427 /** 5428 * igc_request_msix - Initialize MSI-X interrupts 5429 * @adapter: Pointer to adapter structure 5430 * 5431 * igc_request_msix allocates MSI-X vectors and requests interrupts from the 5432 * kernel. 5433 */ 5434 static int igc_request_msix(struct igc_adapter *adapter) 5435 { 5436 unsigned int num_q_vectors = adapter->num_q_vectors; 5437 int i = 0, err = 0, vector = 0, free_vector = 0; 5438 struct net_device *netdev = adapter->netdev; 5439 5440 err = request_irq(adapter->msix_entries[vector].vector, 5441 &igc_msix_other, 0, netdev->name, adapter); 5442 if (err) 5443 goto err_out; 5444 5445 if (num_q_vectors > MAX_Q_VECTORS) { 5446 num_q_vectors = MAX_Q_VECTORS; 5447 dev_warn(&adapter->pdev->dev, 5448 "The number of queue vectors (%d) is higher than max allowed (%d)\n", 5449 adapter->num_q_vectors, MAX_Q_VECTORS); 5450 } 5451 for (i = 0; i < num_q_vectors; i++) { 5452 struct igc_q_vector *q_vector = adapter->q_vector[i]; 5453 5454 vector++; 5455 5456 q_vector->itr_register = adapter->io_addr + IGC_EITR(vector); 5457 5458 if (q_vector->rx.ring && q_vector->tx.ring) 5459 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name, 5460 q_vector->rx.ring->queue_index); 5461 else if (q_vector->tx.ring) 5462 sprintf(q_vector->name, "%s-tx-%u", netdev->name, 5463 q_vector->tx.ring->queue_index); 5464 else if (q_vector->rx.ring) 5465 sprintf(q_vector->name, "%s-rx-%u", netdev->name, 5466 q_vector->rx.ring->queue_index); 5467 else 5468 sprintf(q_vector->name, "%s-unused", netdev->name); 5469 5470 err = request_irq(adapter->msix_entries[vector].vector, 5471 igc_msix_ring, 0, q_vector->name, 5472 q_vector); 5473 if (err) 5474 goto err_free; 5475 } 5476 5477 igc_configure_msix(adapter); 5478 return 0; 5479 5480 err_free: 5481 /* free already assigned IRQs */ 5482 free_irq(adapter->msix_entries[free_vector++].vector, adapter); 5483 5484 vector--; 5485 for (i = 0; i < vector; i++) { 5486 free_irq(adapter->msix_entries[free_vector++].vector, 5487 adapter->q_vector[i]); 5488 } 5489 err_out: 5490 return err; 5491 } 5492 5493 /** 5494 * igc_clear_interrupt_scheme - reset the device to a state of no interrupts 5495 * @adapter: Pointer to adapter structure 5496 * 5497 * This function resets the device so that it has 0 rx queues, tx queues, and 5498 * MSI-X interrupts allocated. 5499 */ 5500 static void igc_clear_interrupt_scheme(struct igc_adapter *adapter) 5501 { 5502 igc_free_q_vectors(adapter); 5503 igc_reset_interrupt_capability(adapter); 5504 } 5505 5506 /* Need to wait a few seconds after link up to get diagnostic information from 5507 * the phy 5508 */ 5509 static void igc_update_phy_info(struct timer_list *t) 5510 { 5511 struct igc_adapter *adapter = from_timer(adapter, t, phy_info_timer); 5512 5513 igc_get_phy_info(&adapter->hw); 5514 } 5515 5516 /** 5517 * igc_has_link - check shared code for link and determine up/down 5518 * @adapter: pointer to driver private info 5519 */ 5520 bool igc_has_link(struct igc_adapter *adapter) 5521 { 5522 struct igc_hw *hw = &adapter->hw; 5523 bool link_active = false; 5524 5525 /* get_link_status is set on LSC (link status) interrupt or 5526 * rx sequence error interrupt. get_link_status will stay 5527 * false until the igc_check_for_link establishes link 5528 * for copper adapters ONLY 5529 */ 5530 if (!hw->mac.get_link_status) 5531 return true; 5532 hw->mac.ops.check_for_link(hw); 5533 link_active = !hw->mac.get_link_status; 5534 5535 if (hw->mac.type == igc_i225) { 5536 if (!netif_carrier_ok(adapter->netdev)) { 5537 adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE; 5538 } else if (!(adapter->flags & IGC_FLAG_NEED_LINK_UPDATE)) { 5539 adapter->flags |= IGC_FLAG_NEED_LINK_UPDATE; 5540 adapter->link_check_timeout = jiffies; 5541 } 5542 } 5543 5544 return link_active; 5545 } 5546 5547 /** 5548 * igc_watchdog - Timer Call-back 5549 * @t: timer for the watchdog 5550 */ 5551 static void igc_watchdog(struct timer_list *t) 5552 { 5553 struct igc_adapter *adapter = from_timer(adapter, t, watchdog_timer); 5554 /* Do the rest outside of interrupt context */ 5555 schedule_work(&adapter->watchdog_task); 5556 } 5557 5558 static void igc_watchdog_task(struct work_struct *work) 5559 { 5560 struct igc_adapter *adapter = container_of(work, 5561 struct igc_adapter, 5562 watchdog_task); 5563 struct net_device *netdev = adapter->netdev; 5564 struct igc_hw *hw = &adapter->hw; 5565 struct igc_phy_info *phy = &hw->phy; 5566 u16 phy_data, retry_count = 20; 5567 u32 link; 5568 int i; 5569 5570 link = igc_has_link(adapter); 5571 5572 if (adapter->flags & IGC_FLAG_NEED_LINK_UPDATE) { 5573 if (time_after(jiffies, (adapter->link_check_timeout + HZ))) 5574 adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE; 5575 else 5576 link = false; 5577 } 5578 5579 if (link) { 5580 /* Cancel scheduled suspend requests. */ 5581 pm_runtime_resume(netdev->dev.parent); 5582 5583 if (!netif_carrier_ok(netdev)) { 5584 u32 ctrl; 5585 5586 hw->mac.ops.get_speed_and_duplex(hw, 5587 &adapter->link_speed, 5588 &adapter->link_duplex); 5589 5590 ctrl = rd32(IGC_CTRL); 5591 /* Link status message must follow this format */ 5592 netdev_info(netdev, 5593 "NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n", 5594 adapter->link_speed, 5595 adapter->link_duplex == FULL_DUPLEX ? 5596 "Full" : "Half", 5597 (ctrl & IGC_CTRL_TFCE) && 5598 (ctrl & IGC_CTRL_RFCE) ? "RX/TX" : 5599 (ctrl & IGC_CTRL_RFCE) ? "RX" : 5600 (ctrl & IGC_CTRL_TFCE) ? "TX" : "None"); 5601 5602 /* disable EEE if enabled */ 5603 if ((adapter->flags & IGC_FLAG_EEE) && 5604 adapter->link_duplex == HALF_DUPLEX) { 5605 netdev_info(netdev, 5606 "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex\n"); 5607 adapter->hw.dev_spec._base.eee_enable = false; 5608 adapter->flags &= ~IGC_FLAG_EEE; 5609 } 5610 5611 /* check if SmartSpeed worked */ 5612 igc_check_downshift(hw); 5613 if (phy->speed_downgraded) 5614 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n"); 5615 5616 /* adjust timeout factor according to speed/duplex */ 5617 adapter->tx_timeout_factor = 1; 5618 switch (adapter->link_speed) { 5619 case SPEED_10: 5620 adapter->tx_timeout_factor = 14; 5621 break; 5622 case SPEED_100: 5623 case SPEED_1000: 5624 case SPEED_2500: 5625 adapter->tx_timeout_factor = 1; 5626 break; 5627 } 5628 5629 /* Once the launch time has been set on the wire, there 5630 * is a delay before the link speed can be determined 5631 * based on link-up activity. Write into the register 5632 * as soon as we know the correct link speed. 5633 */ 5634 igc_tsn_adjust_txtime_offset(adapter); 5635 5636 if (adapter->link_speed != SPEED_1000) 5637 goto no_wait; 5638 5639 /* wait for Remote receiver status OK */ 5640 retry_read_status: 5641 if (!igc_read_phy_reg(hw, PHY_1000T_STATUS, 5642 &phy_data)) { 5643 if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) && 5644 retry_count) { 5645 msleep(100); 5646 retry_count--; 5647 goto retry_read_status; 5648 } else if (!retry_count) { 5649 netdev_err(netdev, "exceed max 2 second\n"); 5650 } 5651 } else { 5652 netdev_err(netdev, "read 1000Base-T Status Reg\n"); 5653 } 5654 no_wait: 5655 netif_carrier_on(netdev); 5656 5657 /* link state has changed, schedule phy info update */ 5658 if (!test_bit(__IGC_DOWN, &adapter->state)) 5659 mod_timer(&adapter->phy_info_timer, 5660 round_jiffies(jiffies + 2 * HZ)); 5661 } 5662 } else { 5663 if (netif_carrier_ok(netdev)) { 5664 adapter->link_speed = 0; 5665 adapter->link_duplex = 0; 5666 5667 /* Links status message must follow this format */ 5668 netdev_info(netdev, "NIC Link is Down\n"); 5669 netif_carrier_off(netdev); 5670 5671 /* link state has changed, schedule phy info update */ 5672 if (!test_bit(__IGC_DOWN, &adapter->state)) 5673 mod_timer(&adapter->phy_info_timer, 5674 round_jiffies(jiffies + 2 * HZ)); 5675 5676 pm_schedule_suspend(netdev->dev.parent, 5677 MSEC_PER_SEC * 5); 5678 } 5679 } 5680 5681 spin_lock(&adapter->stats64_lock); 5682 igc_update_stats(adapter); 5683 spin_unlock(&adapter->stats64_lock); 5684 5685 for (i = 0; i < adapter->num_tx_queues; i++) { 5686 struct igc_ring *tx_ring = adapter->tx_ring[i]; 5687 5688 if (!netif_carrier_ok(netdev)) { 5689 /* We've lost link, so the controller stops DMA, 5690 * but we've got queued Tx work that's never going 5691 * to get done, so reset controller to flush Tx. 5692 * (Do the reset outside of interrupt context). 5693 */ 5694 if (igc_desc_unused(tx_ring) + 1 < tx_ring->count) { 5695 adapter->tx_timeout_count++; 5696 schedule_work(&adapter->reset_task); 5697 /* return immediately since reset is imminent */ 5698 return; 5699 } 5700 } 5701 5702 /* Force detection of hung controller every watchdog period */ 5703 set_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags); 5704 } 5705 5706 /* Cause software interrupt to ensure Rx ring is cleaned */ 5707 if (adapter->flags & IGC_FLAG_HAS_MSIX) { 5708 u32 eics = 0; 5709 5710 for (i = 0; i < adapter->num_q_vectors; i++) 5711 eics |= adapter->q_vector[i]->eims_value; 5712 wr32(IGC_EICS, eics); 5713 } else { 5714 wr32(IGC_ICS, IGC_ICS_RXDMT0); 5715 } 5716 5717 igc_ptp_tx_hang(adapter); 5718 5719 /* Reset the timer */ 5720 if (!test_bit(__IGC_DOWN, &adapter->state)) { 5721 if (adapter->flags & IGC_FLAG_NEED_LINK_UPDATE) 5722 mod_timer(&adapter->watchdog_timer, 5723 round_jiffies(jiffies + HZ)); 5724 else 5725 mod_timer(&adapter->watchdog_timer, 5726 round_jiffies(jiffies + 2 * HZ)); 5727 } 5728 } 5729 5730 /** 5731 * igc_intr_msi - Interrupt Handler 5732 * @irq: interrupt number 5733 * @data: pointer to a network interface device structure 5734 */ 5735 static irqreturn_t igc_intr_msi(int irq, void *data) 5736 { 5737 struct igc_adapter *adapter = data; 5738 struct igc_q_vector *q_vector = adapter->q_vector[0]; 5739 struct igc_hw *hw = &adapter->hw; 5740 /* read ICR disables interrupts using IAM */ 5741 u32 icr = rd32(IGC_ICR); 5742 5743 igc_write_itr(q_vector); 5744 5745 if (icr & IGC_ICR_DRSTA) 5746 schedule_work(&adapter->reset_task); 5747 5748 if (icr & IGC_ICR_DOUTSYNC) { 5749 /* HW is reporting DMA is out of sync */ 5750 adapter->stats.doosync++; 5751 } 5752 5753 if (icr & (IGC_ICR_RXSEQ | IGC_ICR_LSC)) { 5754 hw->mac.get_link_status = true; 5755 if (!test_bit(__IGC_DOWN, &adapter->state)) 5756 mod_timer(&adapter->watchdog_timer, jiffies + 1); 5757 } 5758 5759 if (icr & IGC_ICR_TS) 5760 igc_tsync_interrupt(adapter); 5761 5762 napi_schedule(&q_vector->napi); 5763 5764 return IRQ_HANDLED; 5765 } 5766 5767 /** 5768 * igc_intr - Legacy Interrupt Handler 5769 * @irq: interrupt number 5770 * @data: pointer to a network interface device structure 5771 */ 5772 static irqreturn_t igc_intr(int irq, void *data) 5773 { 5774 struct igc_adapter *adapter = data; 5775 struct igc_q_vector *q_vector = adapter->q_vector[0]; 5776 struct igc_hw *hw = &adapter->hw; 5777 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No 5778 * need for the IMC write 5779 */ 5780 u32 icr = rd32(IGC_ICR); 5781 5782 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is 5783 * not set, then the adapter didn't send an interrupt 5784 */ 5785 if (!(icr & IGC_ICR_INT_ASSERTED)) 5786 return IRQ_NONE; 5787 5788 igc_write_itr(q_vector); 5789 5790 if (icr & IGC_ICR_DRSTA) 5791 schedule_work(&adapter->reset_task); 5792 5793 if (icr & IGC_ICR_DOUTSYNC) { 5794 /* HW is reporting DMA is out of sync */ 5795 adapter->stats.doosync++; 5796 } 5797 5798 if (icr & (IGC_ICR_RXSEQ | IGC_ICR_LSC)) { 5799 hw->mac.get_link_status = true; 5800 /* guard against interrupt when we're going down */ 5801 if (!test_bit(__IGC_DOWN, &adapter->state)) 5802 mod_timer(&adapter->watchdog_timer, jiffies + 1); 5803 } 5804 5805 if (icr & IGC_ICR_TS) 5806 igc_tsync_interrupt(adapter); 5807 5808 napi_schedule(&q_vector->napi); 5809 5810 return IRQ_HANDLED; 5811 } 5812 5813 static void igc_free_irq(struct igc_adapter *adapter) 5814 { 5815 if (adapter->msix_entries) { 5816 int vector = 0, i; 5817 5818 free_irq(adapter->msix_entries[vector++].vector, adapter); 5819 5820 for (i = 0; i < adapter->num_q_vectors; i++) 5821 free_irq(adapter->msix_entries[vector++].vector, 5822 adapter->q_vector[i]); 5823 } else { 5824 free_irq(adapter->pdev->irq, adapter); 5825 } 5826 } 5827 5828 /** 5829 * igc_request_irq - initialize interrupts 5830 * @adapter: Pointer to adapter structure 5831 * 5832 * Attempts to configure interrupts using the best available 5833 * capabilities of the hardware and kernel. 5834 */ 5835 static int igc_request_irq(struct igc_adapter *adapter) 5836 { 5837 struct net_device *netdev = adapter->netdev; 5838 struct pci_dev *pdev = adapter->pdev; 5839 int err = 0; 5840 5841 if (adapter->flags & IGC_FLAG_HAS_MSIX) { 5842 err = igc_request_msix(adapter); 5843 if (!err) 5844 goto request_done; 5845 /* fall back to MSI */ 5846 igc_free_all_tx_resources(adapter); 5847 igc_free_all_rx_resources(adapter); 5848 5849 igc_clear_interrupt_scheme(adapter); 5850 err = igc_init_interrupt_scheme(adapter, false); 5851 if (err) 5852 goto request_done; 5853 igc_setup_all_tx_resources(adapter); 5854 igc_setup_all_rx_resources(adapter); 5855 igc_configure(adapter); 5856 } 5857 5858 igc_assign_vector(adapter->q_vector[0], 0); 5859 5860 if (adapter->flags & IGC_FLAG_HAS_MSI) { 5861 err = request_irq(pdev->irq, &igc_intr_msi, 0, 5862 netdev->name, adapter); 5863 if (!err) 5864 goto request_done; 5865 5866 /* fall back to legacy interrupts */ 5867 igc_reset_interrupt_capability(adapter); 5868 adapter->flags &= ~IGC_FLAG_HAS_MSI; 5869 } 5870 5871 err = request_irq(pdev->irq, &igc_intr, IRQF_SHARED, 5872 netdev->name, adapter); 5873 5874 if (err) 5875 netdev_err(netdev, "Error %d getting interrupt\n", err); 5876 5877 request_done: 5878 return err; 5879 } 5880 5881 /** 5882 * __igc_open - Called when a network interface is made active 5883 * @netdev: network interface device structure 5884 * @resuming: boolean indicating if the device is resuming 5885 * 5886 * Returns 0 on success, negative value on failure 5887 * 5888 * The open entry point is called when a network interface is made 5889 * active by the system (IFF_UP). At this point all resources needed 5890 * for transmit and receive operations are allocated, the interrupt 5891 * handler is registered with the OS, the watchdog timer is started, 5892 * and the stack is notified that the interface is ready. 5893 */ 5894 static int __igc_open(struct net_device *netdev, bool resuming) 5895 { 5896 struct igc_adapter *adapter = netdev_priv(netdev); 5897 struct pci_dev *pdev = adapter->pdev; 5898 struct igc_hw *hw = &adapter->hw; 5899 int err = 0; 5900 int i = 0; 5901 5902 /* disallow open during test */ 5903 5904 if (test_bit(__IGC_TESTING, &adapter->state)) { 5905 WARN_ON(resuming); 5906 return -EBUSY; 5907 } 5908 5909 if (!resuming) 5910 pm_runtime_get_sync(&pdev->dev); 5911 5912 netif_carrier_off(netdev); 5913 5914 /* allocate transmit descriptors */ 5915 err = igc_setup_all_tx_resources(adapter); 5916 if (err) 5917 goto err_setup_tx; 5918 5919 /* allocate receive descriptors */ 5920 err = igc_setup_all_rx_resources(adapter); 5921 if (err) 5922 goto err_setup_rx; 5923 5924 igc_power_up_link(adapter); 5925 5926 igc_configure(adapter); 5927 5928 err = igc_request_irq(adapter); 5929 if (err) 5930 goto err_req_irq; 5931 5932 /* Notify the stack of the actual queue counts. */ 5933 err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues); 5934 if (err) 5935 goto err_set_queues; 5936 5937 err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues); 5938 if (err) 5939 goto err_set_queues; 5940 5941 clear_bit(__IGC_DOWN, &adapter->state); 5942 5943 for (i = 0; i < adapter->num_q_vectors; i++) 5944 napi_enable(&adapter->q_vector[i]->napi); 5945 5946 /* Clear any pending interrupts. */ 5947 rd32(IGC_ICR); 5948 igc_irq_enable(adapter); 5949 5950 if (!resuming) 5951 pm_runtime_put(&pdev->dev); 5952 5953 netif_tx_start_all_queues(netdev); 5954 5955 /* start the watchdog. */ 5956 hw->mac.get_link_status = true; 5957 schedule_work(&adapter->watchdog_task); 5958 5959 return IGC_SUCCESS; 5960 5961 err_set_queues: 5962 igc_free_irq(adapter); 5963 err_req_irq: 5964 igc_release_hw_control(adapter); 5965 igc_power_down_phy_copper_base(&adapter->hw); 5966 igc_free_all_rx_resources(adapter); 5967 err_setup_rx: 5968 igc_free_all_tx_resources(adapter); 5969 err_setup_tx: 5970 igc_reset(adapter); 5971 if (!resuming) 5972 pm_runtime_put(&pdev->dev); 5973 5974 return err; 5975 } 5976 5977 int igc_open(struct net_device *netdev) 5978 { 5979 return __igc_open(netdev, false); 5980 } 5981 5982 /** 5983 * __igc_close - Disables a network interface 5984 * @netdev: network interface device structure 5985 * @suspending: boolean indicating the device is suspending 5986 * 5987 * Returns 0, this is not allowed to fail 5988 * 5989 * The close entry point is called when an interface is de-activated 5990 * by the OS. The hardware is still under the driver's control, but 5991 * needs to be disabled. A global MAC reset is issued to stop the 5992 * hardware, and all transmit and receive resources are freed. 5993 */ 5994 static int __igc_close(struct net_device *netdev, bool suspending) 5995 { 5996 struct igc_adapter *adapter = netdev_priv(netdev); 5997 struct pci_dev *pdev = adapter->pdev; 5998 5999 WARN_ON(test_bit(__IGC_RESETTING, &adapter->state)); 6000 6001 if (!suspending) 6002 pm_runtime_get_sync(&pdev->dev); 6003 6004 igc_down(adapter); 6005 6006 igc_release_hw_control(adapter); 6007 6008 igc_free_irq(adapter); 6009 6010 igc_free_all_tx_resources(adapter); 6011 igc_free_all_rx_resources(adapter); 6012 6013 if (!suspending) 6014 pm_runtime_put_sync(&pdev->dev); 6015 6016 return 0; 6017 } 6018 6019 int igc_close(struct net_device *netdev) 6020 { 6021 if (netif_device_present(netdev) || netdev->dismantle) 6022 return __igc_close(netdev, false); 6023 return 0; 6024 } 6025 6026 /** 6027 * igc_ioctl - Access the hwtstamp interface 6028 * @netdev: network interface device structure 6029 * @ifr: interface request data 6030 * @cmd: ioctl command 6031 **/ 6032 static int igc_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd) 6033 { 6034 switch (cmd) { 6035 case SIOCGHWTSTAMP: 6036 return igc_ptp_get_ts_config(netdev, ifr); 6037 case SIOCSHWTSTAMP: 6038 return igc_ptp_set_ts_config(netdev, ifr); 6039 default: 6040 return -EOPNOTSUPP; 6041 } 6042 } 6043 6044 static int igc_save_launchtime_params(struct igc_adapter *adapter, int queue, 6045 bool enable) 6046 { 6047 struct igc_ring *ring; 6048 6049 if (queue < 0 || queue >= adapter->num_tx_queues) 6050 return -EINVAL; 6051 6052 ring = adapter->tx_ring[queue]; 6053 ring->launchtime_enable = enable; 6054 6055 return 0; 6056 } 6057 6058 static bool is_base_time_past(ktime_t base_time, const struct timespec64 *now) 6059 { 6060 struct timespec64 b; 6061 6062 b = ktime_to_timespec64(base_time); 6063 6064 return timespec64_compare(now, &b) > 0; 6065 } 6066 6067 static bool validate_schedule(struct igc_adapter *adapter, 6068 const struct tc_taprio_qopt_offload *qopt) 6069 { 6070 int queue_uses[IGC_MAX_TX_QUEUES] = { }; 6071 struct igc_hw *hw = &adapter->hw; 6072 struct timespec64 now; 6073 size_t n; 6074 6075 if (qopt->cycle_time_extension) 6076 return false; 6077 6078 igc_ptp_read(adapter, &now); 6079 6080 /* If we program the controller's BASET registers with a time 6081 * in the future, it will hold all the packets until that 6082 * time, causing a lot of TX Hangs, so to avoid that, we 6083 * reject schedules that would start in the future. 6084 * Note: Limitation above is no longer in i226. 6085 */ 6086 if (!is_base_time_past(qopt->base_time, &now) && 6087 igc_is_device_id_i225(hw)) 6088 return false; 6089 6090 for (n = 0; n < qopt->num_entries; n++) { 6091 const struct tc_taprio_sched_entry *e, *prev; 6092 int i; 6093 6094 prev = n ? &qopt->entries[n - 1] : NULL; 6095 e = &qopt->entries[n]; 6096 6097 /* i225 only supports "global" frame preemption 6098 * settings. 6099 */ 6100 if (e->command != TC_TAPRIO_CMD_SET_GATES) 6101 return false; 6102 6103 for (i = 0; i < adapter->num_tx_queues; i++) 6104 if (e->gate_mask & BIT(i)) { 6105 queue_uses[i]++; 6106 6107 /* There are limitations: A single queue cannot 6108 * be opened and closed multiple times per cycle 6109 * unless the gate stays open. Check for it. 6110 */ 6111 if (queue_uses[i] > 1 && 6112 !(prev->gate_mask & BIT(i))) 6113 return false; 6114 } 6115 } 6116 6117 return true; 6118 } 6119 6120 static int igc_tsn_enable_launchtime(struct igc_adapter *adapter, 6121 struct tc_etf_qopt_offload *qopt) 6122 { 6123 struct igc_hw *hw = &adapter->hw; 6124 int err; 6125 6126 if (hw->mac.type != igc_i225) 6127 return -EOPNOTSUPP; 6128 6129 err = igc_save_launchtime_params(adapter, qopt->queue, qopt->enable); 6130 if (err) 6131 return err; 6132 6133 return igc_tsn_offload_apply(adapter); 6134 } 6135 6136 static int igc_qbv_clear_schedule(struct igc_adapter *adapter) 6137 { 6138 unsigned long flags; 6139 int i; 6140 6141 adapter->base_time = 0; 6142 adapter->cycle_time = NSEC_PER_SEC; 6143 adapter->taprio_offload_enable = false; 6144 adapter->qbv_config_change_errors = 0; 6145 adapter->qbv_count = 0; 6146 6147 for (i = 0; i < adapter->num_tx_queues; i++) { 6148 struct igc_ring *ring = adapter->tx_ring[i]; 6149 6150 ring->start_time = 0; 6151 ring->end_time = NSEC_PER_SEC; 6152 ring->max_sdu = 0; 6153 } 6154 6155 spin_lock_irqsave(&adapter->qbv_tx_lock, flags); 6156 6157 adapter->qbv_transition = false; 6158 6159 for (i = 0; i < adapter->num_tx_queues; i++) { 6160 struct igc_ring *ring = adapter->tx_ring[i]; 6161 6162 ring->oper_gate_closed = false; 6163 ring->admin_gate_closed = false; 6164 } 6165 6166 spin_unlock_irqrestore(&adapter->qbv_tx_lock, flags); 6167 6168 return 0; 6169 } 6170 6171 static int igc_tsn_clear_schedule(struct igc_adapter *adapter) 6172 { 6173 igc_qbv_clear_schedule(adapter); 6174 6175 return 0; 6176 } 6177 6178 static void igc_taprio_stats(struct net_device *dev, 6179 struct tc_taprio_qopt_stats *stats) 6180 { 6181 /* When Strict_End is enabled, the tx_overruns counter 6182 * will always be zero. 6183 */ 6184 stats->tx_overruns = 0; 6185 } 6186 6187 static void igc_taprio_queue_stats(struct net_device *dev, 6188 struct tc_taprio_qopt_queue_stats *queue_stats) 6189 { 6190 struct tc_taprio_qopt_stats *stats = &queue_stats->stats; 6191 6192 /* When Strict_End is enabled, the tx_overruns counter 6193 * will always be zero. 6194 */ 6195 stats->tx_overruns = 0; 6196 } 6197 6198 static int igc_save_qbv_schedule(struct igc_adapter *adapter, 6199 struct tc_taprio_qopt_offload *qopt) 6200 { 6201 bool queue_configured[IGC_MAX_TX_QUEUES] = { }; 6202 struct igc_hw *hw = &adapter->hw; 6203 u32 start_time = 0, end_time = 0; 6204 struct timespec64 now; 6205 unsigned long flags; 6206 size_t n; 6207 int i; 6208 6209 if (qopt->base_time < 0) 6210 return -ERANGE; 6211 6212 if (igc_is_device_id_i225(hw) && adapter->taprio_offload_enable) 6213 return -EALREADY; 6214 6215 if (!validate_schedule(adapter, qopt)) 6216 return -EINVAL; 6217 6218 igc_ptp_read(adapter, &now); 6219 6220 if (igc_tsn_is_taprio_activated_by_user(adapter) && 6221 is_base_time_past(qopt->base_time, &now)) 6222 adapter->qbv_config_change_errors++; 6223 6224 adapter->cycle_time = qopt->cycle_time; 6225 adapter->base_time = qopt->base_time; 6226 adapter->taprio_offload_enable = true; 6227 6228 for (n = 0; n < qopt->num_entries; n++) { 6229 struct tc_taprio_sched_entry *e = &qopt->entries[n]; 6230 6231 end_time += e->interval; 6232 6233 /* If any of the conditions below are true, we need to manually 6234 * control the end time of the cycle. 6235 * 1. Qbv users can specify a cycle time that is not equal 6236 * to the total GCL intervals. Hence, recalculation is 6237 * necessary here to exclude the time interval that 6238 * exceeds the cycle time. 6239 * 2. According to IEEE Std. 802.1Q-2018 section 8.6.9.2, 6240 * once the end of the list is reached, it will switch 6241 * to the END_OF_CYCLE state and leave the gates in the 6242 * same state until the next cycle is started. 6243 */ 6244 if (end_time > adapter->cycle_time || 6245 n + 1 == qopt->num_entries) 6246 end_time = adapter->cycle_time; 6247 6248 for (i = 0; i < adapter->num_tx_queues; i++) { 6249 struct igc_ring *ring = adapter->tx_ring[i]; 6250 6251 if (!(e->gate_mask & BIT(i))) 6252 continue; 6253 6254 /* Check whether a queue stays open for more than one 6255 * entry. If so, keep the start and advance the end 6256 * time. 6257 */ 6258 if (!queue_configured[i]) 6259 ring->start_time = start_time; 6260 ring->end_time = end_time; 6261 6262 if (ring->start_time >= adapter->cycle_time) 6263 queue_configured[i] = false; 6264 else 6265 queue_configured[i] = true; 6266 } 6267 6268 start_time += e->interval; 6269 } 6270 6271 spin_lock_irqsave(&adapter->qbv_tx_lock, flags); 6272 6273 /* Check whether a queue gets configured. 6274 * If not, set the start and end time to be end time. 6275 */ 6276 for (i = 0; i < adapter->num_tx_queues; i++) { 6277 struct igc_ring *ring = adapter->tx_ring[i]; 6278 6279 if (!is_base_time_past(qopt->base_time, &now)) { 6280 ring->admin_gate_closed = false; 6281 } else { 6282 ring->oper_gate_closed = false; 6283 ring->admin_gate_closed = false; 6284 } 6285 6286 if (!queue_configured[i]) { 6287 if (!is_base_time_past(qopt->base_time, &now)) 6288 ring->admin_gate_closed = true; 6289 else 6290 ring->oper_gate_closed = true; 6291 6292 ring->start_time = end_time; 6293 ring->end_time = end_time; 6294 } 6295 } 6296 6297 spin_unlock_irqrestore(&adapter->qbv_tx_lock, flags); 6298 6299 for (i = 0; i < adapter->num_tx_queues; i++) { 6300 struct igc_ring *ring = adapter->tx_ring[i]; 6301 struct net_device *dev = adapter->netdev; 6302 6303 if (qopt->max_sdu[i]) 6304 ring->max_sdu = qopt->max_sdu[i] + dev->hard_header_len - ETH_TLEN; 6305 else 6306 ring->max_sdu = 0; 6307 } 6308 6309 return 0; 6310 } 6311 6312 static int igc_tsn_enable_qbv_scheduling(struct igc_adapter *adapter, 6313 struct tc_taprio_qopt_offload *qopt) 6314 { 6315 struct igc_hw *hw = &adapter->hw; 6316 int err; 6317 6318 if (hw->mac.type != igc_i225) 6319 return -EOPNOTSUPP; 6320 6321 switch (qopt->cmd) { 6322 case TAPRIO_CMD_REPLACE: 6323 err = igc_save_qbv_schedule(adapter, qopt); 6324 break; 6325 case TAPRIO_CMD_DESTROY: 6326 err = igc_tsn_clear_schedule(adapter); 6327 break; 6328 case TAPRIO_CMD_STATS: 6329 igc_taprio_stats(adapter->netdev, &qopt->stats); 6330 return 0; 6331 case TAPRIO_CMD_QUEUE_STATS: 6332 igc_taprio_queue_stats(adapter->netdev, &qopt->queue_stats); 6333 return 0; 6334 default: 6335 return -EOPNOTSUPP; 6336 } 6337 6338 if (err) 6339 return err; 6340 6341 return igc_tsn_offload_apply(adapter); 6342 } 6343 6344 static int igc_save_cbs_params(struct igc_adapter *adapter, int queue, 6345 bool enable, int idleslope, int sendslope, 6346 int hicredit, int locredit) 6347 { 6348 bool cbs_status[IGC_MAX_SR_QUEUES] = { false }; 6349 struct net_device *netdev = adapter->netdev; 6350 struct igc_ring *ring; 6351 int i; 6352 6353 /* i225 has two sets of credit-based shaper logic. 6354 * Supporting it only on the top two priority queues 6355 */ 6356 if (queue < 0 || queue > 1) 6357 return -EINVAL; 6358 6359 ring = adapter->tx_ring[queue]; 6360 6361 for (i = 0; i < IGC_MAX_SR_QUEUES; i++) 6362 if (adapter->tx_ring[i]) 6363 cbs_status[i] = adapter->tx_ring[i]->cbs_enable; 6364 6365 /* CBS should be enabled on the highest priority queue first in order 6366 * for the CBS algorithm to operate as intended. 6367 */ 6368 if (enable) { 6369 if (queue == 1 && !cbs_status[0]) { 6370 netdev_err(netdev, 6371 "Enabling CBS on queue1 before queue0\n"); 6372 return -EINVAL; 6373 } 6374 } else { 6375 if (queue == 0 && cbs_status[1]) { 6376 netdev_err(netdev, 6377 "Disabling CBS on queue0 before queue1\n"); 6378 return -EINVAL; 6379 } 6380 } 6381 6382 ring->cbs_enable = enable; 6383 ring->idleslope = idleslope; 6384 ring->sendslope = sendslope; 6385 ring->hicredit = hicredit; 6386 ring->locredit = locredit; 6387 6388 return 0; 6389 } 6390 6391 static int igc_tsn_enable_cbs(struct igc_adapter *adapter, 6392 struct tc_cbs_qopt_offload *qopt) 6393 { 6394 struct igc_hw *hw = &adapter->hw; 6395 int err; 6396 6397 if (hw->mac.type != igc_i225) 6398 return -EOPNOTSUPP; 6399 6400 if (qopt->queue < 0 || qopt->queue > 1) 6401 return -EINVAL; 6402 6403 err = igc_save_cbs_params(adapter, qopt->queue, qopt->enable, 6404 qopt->idleslope, qopt->sendslope, 6405 qopt->hicredit, qopt->locredit); 6406 if (err) 6407 return err; 6408 6409 return igc_tsn_offload_apply(adapter); 6410 } 6411 6412 static int igc_tc_query_caps(struct igc_adapter *adapter, 6413 struct tc_query_caps_base *base) 6414 { 6415 struct igc_hw *hw = &adapter->hw; 6416 6417 switch (base->type) { 6418 case TC_SETUP_QDISC_TAPRIO: { 6419 struct tc_taprio_caps *caps = base->caps; 6420 6421 caps->broken_mqprio = true; 6422 6423 if (hw->mac.type == igc_i225) { 6424 caps->supports_queue_max_sdu = true; 6425 caps->gate_mask_per_txq = true; 6426 } 6427 6428 return 0; 6429 } 6430 default: 6431 return -EOPNOTSUPP; 6432 } 6433 } 6434 6435 static int igc_setup_tc(struct net_device *dev, enum tc_setup_type type, 6436 void *type_data) 6437 { 6438 struct igc_adapter *adapter = netdev_priv(dev); 6439 6440 adapter->tc_setup_type = type; 6441 6442 switch (type) { 6443 case TC_QUERY_CAPS: 6444 return igc_tc_query_caps(adapter, type_data); 6445 case TC_SETUP_QDISC_TAPRIO: 6446 return igc_tsn_enable_qbv_scheduling(adapter, type_data); 6447 6448 case TC_SETUP_QDISC_ETF: 6449 return igc_tsn_enable_launchtime(adapter, type_data); 6450 6451 case TC_SETUP_QDISC_CBS: 6452 return igc_tsn_enable_cbs(adapter, type_data); 6453 6454 default: 6455 return -EOPNOTSUPP; 6456 } 6457 } 6458 6459 static int igc_bpf(struct net_device *dev, struct netdev_bpf *bpf) 6460 { 6461 struct igc_adapter *adapter = netdev_priv(dev); 6462 6463 switch (bpf->command) { 6464 case XDP_SETUP_PROG: 6465 return igc_xdp_set_prog(adapter, bpf->prog, bpf->extack); 6466 case XDP_SETUP_XSK_POOL: 6467 return igc_xdp_setup_pool(adapter, bpf->xsk.pool, 6468 bpf->xsk.queue_id); 6469 default: 6470 return -EOPNOTSUPP; 6471 } 6472 } 6473 6474 static int igc_xdp_xmit(struct net_device *dev, int num_frames, 6475 struct xdp_frame **frames, u32 flags) 6476 { 6477 struct igc_adapter *adapter = netdev_priv(dev); 6478 int cpu = smp_processor_id(); 6479 struct netdev_queue *nq; 6480 struct igc_ring *ring; 6481 int i, nxmit; 6482 6483 if (unlikely(!netif_carrier_ok(dev))) 6484 return -ENETDOWN; 6485 6486 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) 6487 return -EINVAL; 6488 6489 ring = igc_xdp_get_tx_ring(adapter, cpu); 6490 nq = txring_txq(ring); 6491 6492 __netif_tx_lock(nq, cpu); 6493 6494 /* Avoid transmit queue timeout since we share it with the slow path */ 6495 txq_trans_cond_update(nq); 6496 6497 nxmit = 0; 6498 for (i = 0; i < num_frames; i++) { 6499 int err; 6500 struct xdp_frame *xdpf = frames[i]; 6501 6502 err = igc_xdp_init_tx_descriptor(ring, xdpf); 6503 if (err) 6504 break; 6505 nxmit++; 6506 } 6507 6508 if (flags & XDP_XMIT_FLUSH) 6509 igc_flush_tx_descriptors(ring); 6510 6511 __netif_tx_unlock(nq); 6512 6513 return nxmit; 6514 } 6515 6516 static void igc_trigger_rxtxq_interrupt(struct igc_adapter *adapter, 6517 struct igc_q_vector *q_vector) 6518 { 6519 struct igc_hw *hw = &adapter->hw; 6520 u32 eics = 0; 6521 6522 eics |= q_vector->eims_value; 6523 wr32(IGC_EICS, eics); 6524 } 6525 6526 int igc_xsk_wakeup(struct net_device *dev, u32 queue_id, u32 flags) 6527 { 6528 struct igc_adapter *adapter = netdev_priv(dev); 6529 struct igc_q_vector *q_vector; 6530 struct igc_ring *ring; 6531 6532 if (test_bit(__IGC_DOWN, &adapter->state)) 6533 return -ENETDOWN; 6534 6535 if (!igc_xdp_is_enabled(adapter)) 6536 return -ENXIO; 6537 6538 if (queue_id >= adapter->num_rx_queues) 6539 return -EINVAL; 6540 6541 ring = adapter->rx_ring[queue_id]; 6542 6543 if (!ring->xsk_pool) 6544 return -ENXIO; 6545 6546 q_vector = adapter->q_vector[queue_id]; 6547 if (!napi_if_scheduled_mark_missed(&q_vector->napi)) 6548 igc_trigger_rxtxq_interrupt(adapter, q_vector); 6549 6550 return 0; 6551 } 6552 6553 static const struct net_device_ops igc_netdev_ops = { 6554 .ndo_open = igc_open, 6555 .ndo_stop = igc_close, 6556 .ndo_start_xmit = igc_xmit_frame, 6557 .ndo_set_rx_mode = igc_set_rx_mode, 6558 .ndo_set_mac_address = igc_set_mac, 6559 .ndo_change_mtu = igc_change_mtu, 6560 .ndo_tx_timeout = igc_tx_timeout, 6561 .ndo_get_stats64 = igc_get_stats64, 6562 .ndo_fix_features = igc_fix_features, 6563 .ndo_set_features = igc_set_features, 6564 .ndo_features_check = igc_features_check, 6565 .ndo_eth_ioctl = igc_ioctl, 6566 .ndo_setup_tc = igc_setup_tc, 6567 .ndo_bpf = igc_bpf, 6568 .ndo_xdp_xmit = igc_xdp_xmit, 6569 .ndo_xsk_wakeup = igc_xsk_wakeup, 6570 }; 6571 6572 /* PCIe configuration access */ 6573 void igc_read_pci_cfg(struct igc_hw *hw, u32 reg, u16 *value) 6574 { 6575 struct igc_adapter *adapter = hw->back; 6576 6577 pci_read_config_word(adapter->pdev, reg, value); 6578 } 6579 6580 void igc_write_pci_cfg(struct igc_hw *hw, u32 reg, u16 *value) 6581 { 6582 struct igc_adapter *adapter = hw->back; 6583 6584 pci_write_config_word(adapter->pdev, reg, *value); 6585 } 6586 6587 s32 igc_read_pcie_cap_reg(struct igc_hw *hw, u32 reg, u16 *value) 6588 { 6589 struct igc_adapter *adapter = hw->back; 6590 6591 if (!pci_is_pcie(adapter->pdev)) 6592 return -IGC_ERR_CONFIG; 6593 6594 pcie_capability_read_word(adapter->pdev, reg, value); 6595 6596 return IGC_SUCCESS; 6597 } 6598 6599 s32 igc_write_pcie_cap_reg(struct igc_hw *hw, u32 reg, u16 *value) 6600 { 6601 struct igc_adapter *adapter = hw->back; 6602 6603 if (!pci_is_pcie(adapter->pdev)) 6604 return -IGC_ERR_CONFIG; 6605 6606 pcie_capability_write_word(adapter->pdev, reg, *value); 6607 6608 return IGC_SUCCESS; 6609 } 6610 6611 u32 igc_rd32(struct igc_hw *hw, u32 reg) 6612 { 6613 struct igc_adapter *igc = container_of(hw, struct igc_adapter, hw); 6614 u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr); 6615 u32 value = 0; 6616 6617 if (IGC_REMOVED(hw_addr)) 6618 return ~value; 6619 6620 value = readl(&hw_addr[reg]); 6621 6622 /* reads should not return all F's */ 6623 if (!(~value) && (!reg || !(~readl(hw_addr)))) { 6624 struct net_device *netdev = igc->netdev; 6625 6626 hw->hw_addr = NULL; 6627 netif_device_detach(netdev); 6628 netdev_err(netdev, "PCIe link lost, device now detached\n"); 6629 WARN(pci_device_is_present(igc->pdev), 6630 "igc: Failed to read reg 0x%x!\n", reg); 6631 } 6632 6633 return value; 6634 } 6635 6636 /* Mapping HW RSS Type to enum xdp_rss_hash_type */ 6637 static enum xdp_rss_hash_type igc_xdp_rss_type[IGC_RSS_TYPE_MAX_TABLE] = { 6638 [IGC_RSS_TYPE_NO_HASH] = XDP_RSS_TYPE_L2, 6639 [IGC_RSS_TYPE_HASH_TCP_IPV4] = XDP_RSS_TYPE_L4_IPV4_TCP, 6640 [IGC_RSS_TYPE_HASH_IPV4] = XDP_RSS_TYPE_L3_IPV4, 6641 [IGC_RSS_TYPE_HASH_TCP_IPV6] = XDP_RSS_TYPE_L4_IPV6_TCP, 6642 [IGC_RSS_TYPE_HASH_IPV6_EX] = XDP_RSS_TYPE_L3_IPV6_EX, 6643 [IGC_RSS_TYPE_HASH_IPV6] = XDP_RSS_TYPE_L3_IPV6, 6644 [IGC_RSS_TYPE_HASH_TCP_IPV6_EX] = XDP_RSS_TYPE_L4_IPV6_TCP_EX, 6645 [IGC_RSS_TYPE_HASH_UDP_IPV4] = XDP_RSS_TYPE_L4_IPV4_UDP, 6646 [IGC_RSS_TYPE_HASH_UDP_IPV6] = XDP_RSS_TYPE_L4_IPV6_UDP, 6647 [IGC_RSS_TYPE_HASH_UDP_IPV6_EX] = XDP_RSS_TYPE_L4_IPV6_UDP_EX, 6648 [10] = XDP_RSS_TYPE_NONE, /* RSS Type above 9 "Reserved" by HW */ 6649 [11] = XDP_RSS_TYPE_NONE, /* keep array sized for SW bit-mask */ 6650 [12] = XDP_RSS_TYPE_NONE, /* to handle future HW revisons */ 6651 [13] = XDP_RSS_TYPE_NONE, 6652 [14] = XDP_RSS_TYPE_NONE, 6653 [15] = XDP_RSS_TYPE_NONE, 6654 }; 6655 6656 static int igc_xdp_rx_hash(const struct xdp_md *_ctx, u32 *hash, 6657 enum xdp_rss_hash_type *rss_type) 6658 { 6659 const struct igc_xdp_buff *ctx = (void *)_ctx; 6660 6661 if (!(ctx->xdp.rxq->dev->features & NETIF_F_RXHASH)) 6662 return -ENODATA; 6663 6664 *hash = le32_to_cpu(ctx->rx_desc->wb.lower.hi_dword.rss); 6665 *rss_type = igc_xdp_rss_type[igc_rss_type(ctx->rx_desc)]; 6666 6667 return 0; 6668 } 6669 6670 static int igc_xdp_rx_timestamp(const struct xdp_md *_ctx, u64 *timestamp) 6671 { 6672 const struct igc_xdp_buff *ctx = (void *)_ctx; 6673 6674 if (igc_test_staterr(ctx->rx_desc, IGC_RXDADV_STAT_TSIP)) { 6675 *timestamp = ctx->rx_ts; 6676 6677 return 0; 6678 } 6679 6680 return -ENODATA; 6681 } 6682 6683 static const struct xdp_metadata_ops igc_xdp_metadata_ops = { 6684 .xmo_rx_hash = igc_xdp_rx_hash, 6685 .xmo_rx_timestamp = igc_xdp_rx_timestamp, 6686 }; 6687 6688 static enum hrtimer_restart igc_qbv_scheduling_timer(struct hrtimer *timer) 6689 { 6690 struct igc_adapter *adapter = container_of(timer, struct igc_adapter, 6691 hrtimer); 6692 unsigned long flags; 6693 unsigned int i; 6694 6695 spin_lock_irqsave(&adapter->qbv_tx_lock, flags); 6696 6697 adapter->qbv_transition = true; 6698 for (i = 0; i < adapter->num_tx_queues; i++) { 6699 struct igc_ring *tx_ring = adapter->tx_ring[i]; 6700 6701 if (tx_ring->admin_gate_closed) { 6702 tx_ring->admin_gate_closed = false; 6703 tx_ring->oper_gate_closed = true; 6704 } else { 6705 tx_ring->oper_gate_closed = false; 6706 } 6707 } 6708 adapter->qbv_transition = false; 6709 6710 spin_unlock_irqrestore(&adapter->qbv_tx_lock, flags); 6711 6712 return HRTIMER_NORESTART; 6713 } 6714 6715 /** 6716 * igc_probe - Device Initialization Routine 6717 * @pdev: PCI device information struct 6718 * @ent: entry in igc_pci_tbl 6719 * 6720 * Returns 0 on success, negative on failure 6721 * 6722 * igc_probe initializes an adapter identified by a pci_dev structure. 6723 * The OS initialization, configuring the adapter private structure, 6724 * and a hardware reset occur. 6725 */ 6726 static int igc_probe(struct pci_dev *pdev, 6727 const struct pci_device_id *ent) 6728 { 6729 struct igc_adapter *adapter; 6730 struct net_device *netdev; 6731 struct igc_hw *hw; 6732 const struct igc_info *ei = igc_info_tbl[ent->driver_data]; 6733 int err; 6734 6735 err = pci_enable_device_mem(pdev); 6736 if (err) 6737 return err; 6738 6739 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); 6740 if (err) { 6741 dev_err(&pdev->dev, 6742 "No usable DMA configuration, aborting\n"); 6743 goto err_dma; 6744 } 6745 6746 err = pci_request_mem_regions(pdev, igc_driver_name); 6747 if (err) 6748 goto err_pci_reg; 6749 6750 err = pci_enable_ptm(pdev, NULL); 6751 if (err < 0) 6752 dev_info(&pdev->dev, "PCIe PTM not supported by PCIe bus/controller\n"); 6753 6754 pci_set_master(pdev); 6755 6756 err = -ENOMEM; 6757 netdev = alloc_etherdev_mq(sizeof(struct igc_adapter), 6758 IGC_MAX_TX_QUEUES); 6759 6760 if (!netdev) 6761 goto err_alloc_etherdev; 6762 6763 SET_NETDEV_DEV(netdev, &pdev->dev); 6764 6765 pci_set_drvdata(pdev, netdev); 6766 adapter = netdev_priv(netdev); 6767 adapter->netdev = netdev; 6768 adapter->pdev = pdev; 6769 hw = &adapter->hw; 6770 hw->back = adapter; 6771 adapter->port_num = hw->bus.func; 6772 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE); 6773 6774 err = pci_save_state(pdev); 6775 if (err) 6776 goto err_ioremap; 6777 6778 err = -EIO; 6779 adapter->io_addr = ioremap(pci_resource_start(pdev, 0), 6780 pci_resource_len(pdev, 0)); 6781 if (!adapter->io_addr) 6782 goto err_ioremap; 6783 6784 /* hw->hw_addr can be zeroed, so use adapter->io_addr for unmap */ 6785 hw->hw_addr = adapter->io_addr; 6786 6787 netdev->netdev_ops = &igc_netdev_ops; 6788 netdev->xdp_metadata_ops = &igc_xdp_metadata_ops; 6789 igc_ethtool_set_ops(netdev); 6790 netdev->watchdog_timeo = 5 * HZ; 6791 6792 netdev->mem_start = pci_resource_start(pdev, 0); 6793 netdev->mem_end = pci_resource_end(pdev, 0); 6794 6795 /* PCI config space info */ 6796 hw->vendor_id = pdev->vendor; 6797 hw->device_id = pdev->device; 6798 hw->revision_id = pdev->revision; 6799 hw->subsystem_vendor_id = pdev->subsystem_vendor; 6800 hw->subsystem_device_id = pdev->subsystem_device; 6801 6802 /* Copy the default MAC and PHY function pointers */ 6803 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops)); 6804 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops)); 6805 6806 /* Initialize skew-specific constants */ 6807 err = ei->get_invariants(hw); 6808 if (err) 6809 goto err_sw_init; 6810 6811 /* Add supported features to the features list*/ 6812 netdev->features |= NETIF_F_SG; 6813 netdev->features |= NETIF_F_TSO; 6814 netdev->features |= NETIF_F_TSO6; 6815 netdev->features |= NETIF_F_TSO_ECN; 6816 netdev->features |= NETIF_F_RXHASH; 6817 netdev->features |= NETIF_F_RXCSUM; 6818 netdev->features |= NETIF_F_HW_CSUM; 6819 netdev->features |= NETIF_F_SCTP_CRC; 6820 netdev->features |= NETIF_F_HW_TC; 6821 6822 #define IGC_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \ 6823 NETIF_F_GSO_GRE_CSUM | \ 6824 NETIF_F_GSO_IPXIP4 | \ 6825 NETIF_F_GSO_IPXIP6 | \ 6826 NETIF_F_GSO_UDP_TUNNEL | \ 6827 NETIF_F_GSO_UDP_TUNNEL_CSUM) 6828 6829 netdev->gso_partial_features = IGC_GSO_PARTIAL_FEATURES; 6830 netdev->features |= NETIF_F_GSO_PARTIAL | IGC_GSO_PARTIAL_FEATURES; 6831 6832 /* setup the private structure */ 6833 err = igc_sw_init(adapter); 6834 if (err) 6835 goto err_sw_init; 6836 6837 /* copy netdev features into list of user selectable features */ 6838 netdev->hw_features |= NETIF_F_NTUPLE; 6839 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX; 6840 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX; 6841 netdev->hw_features |= netdev->features; 6842 6843 netdev->features |= NETIF_F_HIGHDMA; 6844 6845 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID; 6846 netdev->mpls_features |= NETIF_F_HW_CSUM; 6847 netdev->hw_enc_features |= netdev->vlan_features; 6848 6849 netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT | 6850 NETDEV_XDP_ACT_XSK_ZEROCOPY; 6851 6852 /* MTU range: 68 - 9216 */ 6853 netdev->min_mtu = ETH_MIN_MTU; 6854 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE; 6855 6856 /* before reading the NVM, reset the controller to put the device in a 6857 * known good starting state 6858 */ 6859 hw->mac.ops.reset_hw(hw); 6860 6861 if (igc_get_flash_presence_i225(hw)) { 6862 if (hw->nvm.ops.validate(hw) < 0) { 6863 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n"); 6864 err = -EIO; 6865 goto err_eeprom; 6866 } 6867 } 6868 6869 if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) { 6870 /* copy the MAC address out of the NVM */ 6871 if (hw->mac.ops.read_mac_addr(hw)) 6872 dev_err(&pdev->dev, "NVM Read Error\n"); 6873 } 6874 6875 eth_hw_addr_set(netdev, hw->mac.addr); 6876 6877 if (!is_valid_ether_addr(netdev->dev_addr)) { 6878 dev_err(&pdev->dev, "Invalid MAC Address\n"); 6879 err = -EIO; 6880 goto err_eeprom; 6881 } 6882 6883 /* configure RXPBSIZE and TXPBSIZE */ 6884 wr32(IGC_RXPBS, I225_RXPBSIZE_DEFAULT); 6885 wr32(IGC_TXPBS, I225_TXPBSIZE_DEFAULT); 6886 6887 timer_setup(&adapter->watchdog_timer, igc_watchdog, 0); 6888 timer_setup(&adapter->phy_info_timer, igc_update_phy_info, 0); 6889 6890 INIT_WORK(&adapter->reset_task, igc_reset_task); 6891 INIT_WORK(&adapter->watchdog_task, igc_watchdog_task); 6892 6893 hrtimer_init(&adapter->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 6894 adapter->hrtimer.function = &igc_qbv_scheduling_timer; 6895 6896 /* Initialize link properties that are user-changeable */ 6897 adapter->fc_autoneg = true; 6898 hw->mac.autoneg = true; 6899 hw->phy.autoneg_advertised = 0xaf; 6900 6901 hw->fc.requested_mode = igc_fc_default; 6902 hw->fc.current_mode = igc_fc_default; 6903 6904 /* By default, support wake on port A */ 6905 adapter->flags |= IGC_FLAG_WOL_SUPPORTED; 6906 6907 /* initialize the wol settings based on the eeprom settings */ 6908 if (adapter->flags & IGC_FLAG_WOL_SUPPORTED) 6909 adapter->wol |= IGC_WUFC_MAG; 6910 6911 device_set_wakeup_enable(&adapter->pdev->dev, 6912 adapter->flags & IGC_FLAG_WOL_SUPPORTED); 6913 6914 igc_ptp_init(adapter); 6915 6916 igc_tsn_clear_schedule(adapter); 6917 6918 /* reset the hardware with the new settings */ 6919 igc_reset(adapter); 6920 6921 /* let the f/w know that the h/w is now under the control of the 6922 * driver. 6923 */ 6924 igc_get_hw_control(adapter); 6925 6926 strncpy(netdev->name, "eth%d", IFNAMSIZ); 6927 err = register_netdev(netdev); 6928 if (err) 6929 goto err_register; 6930 6931 /* carrier off reporting is important to ethtool even BEFORE open */ 6932 netif_carrier_off(netdev); 6933 6934 /* Check if Media Autosense is enabled */ 6935 adapter->ei = *ei; 6936 6937 /* print pcie link status and MAC address */ 6938 pcie_print_link_status(pdev); 6939 netdev_info(netdev, "MAC: %pM\n", netdev->dev_addr); 6940 6941 dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE); 6942 /* Disable EEE for internal PHY devices */ 6943 hw->dev_spec._base.eee_enable = false; 6944 adapter->flags &= ~IGC_FLAG_EEE; 6945 igc_set_eee_i225(hw, false, false, false); 6946 6947 pm_runtime_put_noidle(&pdev->dev); 6948 6949 return 0; 6950 6951 err_register: 6952 igc_release_hw_control(adapter); 6953 err_eeprom: 6954 if (!igc_check_reset_block(hw)) 6955 igc_reset_phy(hw); 6956 err_sw_init: 6957 igc_clear_interrupt_scheme(adapter); 6958 iounmap(adapter->io_addr); 6959 err_ioremap: 6960 free_netdev(netdev); 6961 err_alloc_etherdev: 6962 pci_release_mem_regions(pdev); 6963 err_pci_reg: 6964 err_dma: 6965 pci_disable_device(pdev); 6966 return err; 6967 } 6968 6969 /** 6970 * igc_remove - Device Removal Routine 6971 * @pdev: PCI device information struct 6972 * 6973 * igc_remove is called by the PCI subsystem to alert the driver 6974 * that it should release a PCI device. This could be caused by a 6975 * Hot-Plug event, or because the driver is going to be removed from 6976 * memory. 6977 */ 6978 static void igc_remove(struct pci_dev *pdev) 6979 { 6980 struct net_device *netdev = pci_get_drvdata(pdev); 6981 struct igc_adapter *adapter = netdev_priv(netdev); 6982 6983 pm_runtime_get_noresume(&pdev->dev); 6984 6985 igc_flush_nfc_rules(adapter); 6986 6987 igc_ptp_stop(adapter); 6988 6989 pci_disable_ptm(pdev); 6990 pci_clear_master(pdev); 6991 6992 set_bit(__IGC_DOWN, &adapter->state); 6993 6994 del_timer_sync(&adapter->watchdog_timer); 6995 del_timer_sync(&adapter->phy_info_timer); 6996 6997 cancel_work_sync(&adapter->reset_task); 6998 cancel_work_sync(&adapter->watchdog_task); 6999 hrtimer_cancel(&adapter->hrtimer); 7000 7001 /* Release control of h/w to f/w. If f/w is AMT enabled, this 7002 * would have already happened in close and is redundant. 7003 */ 7004 igc_release_hw_control(adapter); 7005 unregister_netdev(netdev); 7006 7007 igc_clear_interrupt_scheme(adapter); 7008 pci_iounmap(pdev, adapter->io_addr); 7009 pci_release_mem_regions(pdev); 7010 7011 free_netdev(netdev); 7012 7013 pci_disable_device(pdev); 7014 } 7015 7016 static int __igc_shutdown(struct pci_dev *pdev, bool *enable_wake, 7017 bool runtime) 7018 { 7019 struct net_device *netdev = pci_get_drvdata(pdev); 7020 struct igc_adapter *adapter = netdev_priv(netdev); 7021 u32 wufc = runtime ? IGC_WUFC_LNKC : adapter->wol; 7022 struct igc_hw *hw = &adapter->hw; 7023 u32 ctrl, rctl, status; 7024 bool wake; 7025 7026 rtnl_lock(); 7027 netif_device_detach(netdev); 7028 7029 if (netif_running(netdev)) 7030 __igc_close(netdev, true); 7031 7032 igc_ptp_suspend(adapter); 7033 7034 igc_clear_interrupt_scheme(adapter); 7035 rtnl_unlock(); 7036 7037 status = rd32(IGC_STATUS); 7038 if (status & IGC_STATUS_LU) 7039 wufc &= ~IGC_WUFC_LNKC; 7040 7041 if (wufc) { 7042 igc_setup_rctl(adapter); 7043 igc_set_rx_mode(netdev); 7044 7045 /* turn on all-multi mode if wake on multicast is enabled */ 7046 if (wufc & IGC_WUFC_MC) { 7047 rctl = rd32(IGC_RCTL); 7048 rctl |= IGC_RCTL_MPE; 7049 wr32(IGC_RCTL, rctl); 7050 } 7051 7052 ctrl = rd32(IGC_CTRL); 7053 ctrl |= IGC_CTRL_ADVD3WUC; 7054 wr32(IGC_CTRL, ctrl); 7055 7056 /* Allow time for pending master requests to run */ 7057 igc_disable_pcie_master(hw); 7058 7059 wr32(IGC_WUC, IGC_WUC_PME_EN); 7060 wr32(IGC_WUFC, wufc); 7061 } else { 7062 wr32(IGC_WUC, 0); 7063 wr32(IGC_WUFC, 0); 7064 } 7065 7066 wake = wufc || adapter->en_mng_pt; 7067 if (!wake) 7068 igc_power_down_phy_copper_base(&adapter->hw); 7069 else 7070 igc_power_up_link(adapter); 7071 7072 if (enable_wake) 7073 *enable_wake = wake; 7074 7075 /* Release control of h/w to f/w. If f/w is AMT enabled, this 7076 * would have already happened in close and is redundant. 7077 */ 7078 igc_release_hw_control(adapter); 7079 7080 pci_disable_device(pdev); 7081 7082 return 0; 7083 } 7084 7085 #ifdef CONFIG_PM 7086 static int __maybe_unused igc_runtime_suspend(struct device *dev) 7087 { 7088 return __igc_shutdown(to_pci_dev(dev), NULL, 1); 7089 } 7090 7091 static void igc_deliver_wake_packet(struct net_device *netdev) 7092 { 7093 struct igc_adapter *adapter = netdev_priv(netdev); 7094 struct igc_hw *hw = &adapter->hw; 7095 struct sk_buff *skb; 7096 u32 wupl; 7097 7098 wupl = rd32(IGC_WUPL) & IGC_WUPL_MASK; 7099 7100 /* WUPM stores only the first 128 bytes of the wake packet. 7101 * Read the packet only if we have the whole thing. 7102 */ 7103 if (wupl == 0 || wupl > IGC_WUPM_BYTES) 7104 return; 7105 7106 skb = netdev_alloc_skb_ip_align(netdev, IGC_WUPM_BYTES); 7107 if (!skb) 7108 return; 7109 7110 skb_put(skb, wupl); 7111 7112 /* Ensure reads are 32-bit aligned */ 7113 wupl = roundup(wupl, 4); 7114 7115 memcpy_fromio(skb->data, hw->hw_addr + IGC_WUPM_REG(0), wupl); 7116 7117 skb->protocol = eth_type_trans(skb, netdev); 7118 netif_rx(skb); 7119 } 7120 7121 static int __maybe_unused igc_resume(struct device *dev) 7122 { 7123 struct pci_dev *pdev = to_pci_dev(dev); 7124 struct net_device *netdev = pci_get_drvdata(pdev); 7125 struct igc_adapter *adapter = netdev_priv(netdev); 7126 struct igc_hw *hw = &adapter->hw; 7127 u32 err, val; 7128 7129 pci_set_power_state(pdev, PCI_D0); 7130 pci_restore_state(pdev); 7131 pci_save_state(pdev); 7132 7133 if (!pci_device_is_present(pdev)) 7134 return -ENODEV; 7135 err = pci_enable_device_mem(pdev); 7136 if (err) { 7137 netdev_err(netdev, "Cannot enable PCI device from suspend\n"); 7138 return err; 7139 } 7140 pci_set_master(pdev); 7141 7142 pci_enable_wake(pdev, PCI_D3hot, 0); 7143 pci_enable_wake(pdev, PCI_D3cold, 0); 7144 7145 if (igc_init_interrupt_scheme(adapter, true)) { 7146 netdev_err(netdev, "Unable to allocate memory for queues\n"); 7147 return -ENOMEM; 7148 } 7149 7150 igc_reset(adapter); 7151 7152 /* let the f/w know that the h/w is now under the control of the 7153 * driver. 7154 */ 7155 igc_get_hw_control(adapter); 7156 7157 val = rd32(IGC_WUS); 7158 if (val & WAKE_PKT_WUS) 7159 igc_deliver_wake_packet(netdev); 7160 7161 wr32(IGC_WUS, ~0); 7162 7163 rtnl_lock(); 7164 if (!err && netif_running(netdev)) 7165 err = __igc_open(netdev, true); 7166 7167 if (!err) 7168 netif_device_attach(netdev); 7169 rtnl_unlock(); 7170 7171 return err; 7172 } 7173 7174 static int __maybe_unused igc_runtime_resume(struct device *dev) 7175 { 7176 return igc_resume(dev); 7177 } 7178 7179 static int __maybe_unused igc_suspend(struct device *dev) 7180 { 7181 return __igc_shutdown(to_pci_dev(dev), NULL, 0); 7182 } 7183 7184 static int __maybe_unused igc_runtime_idle(struct device *dev) 7185 { 7186 struct net_device *netdev = dev_get_drvdata(dev); 7187 struct igc_adapter *adapter = netdev_priv(netdev); 7188 7189 if (!igc_has_link(adapter)) 7190 pm_schedule_suspend(dev, MSEC_PER_SEC * 5); 7191 7192 return -EBUSY; 7193 } 7194 #endif /* CONFIG_PM */ 7195 7196 static void igc_shutdown(struct pci_dev *pdev) 7197 { 7198 bool wake; 7199 7200 __igc_shutdown(pdev, &wake, 0); 7201 7202 if (system_state == SYSTEM_POWER_OFF) { 7203 pci_wake_from_d3(pdev, wake); 7204 pci_set_power_state(pdev, PCI_D3hot); 7205 } 7206 } 7207 7208 /** 7209 * igc_io_error_detected - called when PCI error is detected 7210 * @pdev: Pointer to PCI device 7211 * @state: The current PCI connection state 7212 * 7213 * This function is called after a PCI bus error affecting 7214 * this device has been detected. 7215 **/ 7216 static pci_ers_result_t igc_io_error_detected(struct pci_dev *pdev, 7217 pci_channel_state_t state) 7218 { 7219 struct net_device *netdev = pci_get_drvdata(pdev); 7220 struct igc_adapter *adapter = netdev_priv(netdev); 7221 7222 netif_device_detach(netdev); 7223 7224 if (state == pci_channel_io_perm_failure) 7225 return PCI_ERS_RESULT_DISCONNECT; 7226 7227 if (netif_running(netdev)) 7228 igc_down(adapter); 7229 pci_disable_device(pdev); 7230 7231 /* Request a slot reset. */ 7232 return PCI_ERS_RESULT_NEED_RESET; 7233 } 7234 7235 /** 7236 * igc_io_slot_reset - called after the PCI bus has been reset. 7237 * @pdev: Pointer to PCI device 7238 * 7239 * Restart the card from scratch, as if from a cold-boot. Implementation 7240 * resembles the first-half of the igc_resume routine. 7241 **/ 7242 static pci_ers_result_t igc_io_slot_reset(struct pci_dev *pdev) 7243 { 7244 struct net_device *netdev = pci_get_drvdata(pdev); 7245 struct igc_adapter *adapter = netdev_priv(netdev); 7246 struct igc_hw *hw = &adapter->hw; 7247 pci_ers_result_t result; 7248 7249 if (pci_enable_device_mem(pdev)) { 7250 netdev_err(netdev, "Could not re-enable PCI device after reset\n"); 7251 result = PCI_ERS_RESULT_DISCONNECT; 7252 } else { 7253 pci_set_master(pdev); 7254 pci_restore_state(pdev); 7255 pci_save_state(pdev); 7256 7257 pci_enable_wake(pdev, PCI_D3hot, 0); 7258 pci_enable_wake(pdev, PCI_D3cold, 0); 7259 7260 /* In case of PCI error, adapter loses its HW address 7261 * so we should re-assign it here. 7262 */ 7263 hw->hw_addr = adapter->io_addr; 7264 7265 igc_reset(adapter); 7266 wr32(IGC_WUS, ~0); 7267 result = PCI_ERS_RESULT_RECOVERED; 7268 } 7269 7270 return result; 7271 } 7272 7273 /** 7274 * igc_io_resume - called when traffic can start to flow again. 7275 * @pdev: Pointer to PCI device 7276 * 7277 * This callback is called when the error recovery driver tells us that 7278 * its OK to resume normal operation. Implementation resembles the 7279 * second-half of the igc_resume routine. 7280 */ 7281 static void igc_io_resume(struct pci_dev *pdev) 7282 { 7283 struct net_device *netdev = pci_get_drvdata(pdev); 7284 struct igc_adapter *adapter = netdev_priv(netdev); 7285 7286 rtnl_lock(); 7287 if (netif_running(netdev)) { 7288 if (igc_open(netdev)) { 7289 rtnl_unlock(); 7290 netdev_err(netdev, "igc_open failed after reset\n"); 7291 return; 7292 } 7293 } 7294 7295 netif_device_attach(netdev); 7296 7297 /* let the f/w know that the h/w is now under the control of the 7298 * driver. 7299 */ 7300 igc_get_hw_control(adapter); 7301 rtnl_unlock(); 7302 } 7303 7304 static const struct pci_error_handlers igc_err_handler = { 7305 .error_detected = igc_io_error_detected, 7306 .slot_reset = igc_io_slot_reset, 7307 .resume = igc_io_resume, 7308 }; 7309 7310 #ifdef CONFIG_PM 7311 static const struct dev_pm_ops igc_pm_ops = { 7312 SET_SYSTEM_SLEEP_PM_OPS(igc_suspend, igc_resume) 7313 SET_RUNTIME_PM_OPS(igc_runtime_suspend, igc_runtime_resume, 7314 igc_runtime_idle) 7315 }; 7316 #endif 7317 7318 static struct pci_driver igc_driver = { 7319 .name = igc_driver_name, 7320 .id_table = igc_pci_tbl, 7321 .probe = igc_probe, 7322 .remove = igc_remove, 7323 #ifdef CONFIG_PM 7324 .driver.pm = &igc_pm_ops, 7325 #endif 7326 .shutdown = igc_shutdown, 7327 .err_handler = &igc_err_handler, 7328 }; 7329 7330 /** 7331 * igc_reinit_queues - return error 7332 * @adapter: pointer to adapter structure 7333 */ 7334 int igc_reinit_queues(struct igc_adapter *adapter) 7335 { 7336 struct net_device *netdev = adapter->netdev; 7337 int err = 0; 7338 7339 if (netif_running(netdev)) 7340 igc_close(netdev); 7341 7342 igc_reset_interrupt_capability(adapter); 7343 7344 if (igc_init_interrupt_scheme(adapter, true)) { 7345 netdev_err(netdev, "Unable to allocate memory for queues\n"); 7346 return -ENOMEM; 7347 } 7348 7349 if (netif_running(netdev)) 7350 err = igc_open(netdev); 7351 7352 return err; 7353 } 7354 7355 /** 7356 * igc_get_hw_dev - return device 7357 * @hw: pointer to hardware structure 7358 * 7359 * used by hardware layer to print debugging information 7360 */ 7361 struct net_device *igc_get_hw_dev(struct igc_hw *hw) 7362 { 7363 struct igc_adapter *adapter = hw->back; 7364 7365 return adapter->netdev; 7366 } 7367 7368 static void igc_disable_rx_ring_hw(struct igc_ring *ring) 7369 { 7370 struct igc_hw *hw = &ring->q_vector->adapter->hw; 7371 u8 idx = ring->reg_idx; 7372 u32 rxdctl; 7373 7374 rxdctl = rd32(IGC_RXDCTL(idx)); 7375 rxdctl &= ~IGC_RXDCTL_QUEUE_ENABLE; 7376 rxdctl |= IGC_RXDCTL_SWFLUSH; 7377 wr32(IGC_RXDCTL(idx), rxdctl); 7378 } 7379 7380 void igc_disable_rx_ring(struct igc_ring *ring) 7381 { 7382 igc_disable_rx_ring_hw(ring); 7383 igc_clean_rx_ring(ring); 7384 } 7385 7386 void igc_enable_rx_ring(struct igc_ring *ring) 7387 { 7388 struct igc_adapter *adapter = ring->q_vector->adapter; 7389 7390 igc_configure_rx_ring(adapter, ring); 7391 7392 if (ring->xsk_pool) 7393 igc_alloc_rx_buffers_zc(ring, igc_desc_unused(ring)); 7394 else 7395 igc_alloc_rx_buffers(ring, igc_desc_unused(ring)); 7396 } 7397 7398 void igc_disable_tx_ring(struct igc_ring *ring) 7399 { 7400 igc_disable_tx_ring_hw(ring); 7401 igc_clean_tx_ring(ring); 7402 } 7403 7404 void igc_enable_tx_ring(struct igc_ring *ring) 7405 { 7406 struct igc_adapter *adapter = ring->q_vector->adapter; 7407 7408 igc_configure_tx_ring(adapter, ring); 7409 } 7410 7411 /** 7412 * igc_init_module - Driver Registration Routine 7413 * 7414 * igc_init_module is the first routine called when the driver is 7415 * loaded. All it does is register with the PCI subsystem. 7416 */ 7417 static int __init igc_init_module(void) 7418 { 7419 int ret; 7420 7421 pr_info("%s\n", igc_driver_string); 7422 pr_info("%s\n", igc_copyright); 7423 7424 ret = pci_register_driver(&igc_driver); 7425 return ret; 7426 } 7427 7428 module_init(igc_init_module); 7429 7430 /** 7431 * igc_exit_module - Driver Exit Cleanup Routine 7432 * 7433 * igc_exit_module is called just before the driver is removed 7434 * from memory. 7435 */ 7436 static void __exit igc_exit_module(void) 7437 { 7438 pci_unregister_driver(&igc_driver); 7439 } 7440 7441 module_exit(igc_exit_module); 7442 /* igc_main.c */ 7443