1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright(c) 2009 - 2018 Intel Corporation. */ 3 4 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 5 6 #include <linux/bitfield.h> 7 #include <linux/delay.h> 8 #include <linux/ethtool.h> 9 #include <linux/if_vlan.h> 10 #include <linux/init.h> 11 #include <linux/ipv6.h> 12 #include <linux/mii.h> 13 #include <linux/module.h> 14 #include <linux/netdevice.h> 15 #include <linux/pagemap.h> 16 #include <linux/pci.h> 17 #include <linux/prefetch.h> 18 #include <linux/sctp.h> 19 #include <linux/slab.h> 20 #include <linux/tcp.h> 21 #include <linux/types.h> 22 #include <linux/vmalloc.h> 23 #include <net/checksum.h> 24 #include <net/ip6_checksum.h> 25 #include "igbvf.h" 26 27 char igbvf_driver_name[] = "igbvf"; 28 static const char igbvf_driver_string[] = 29 "Intel(R) Gigabit Virtual Function Network Driver"; 30 static const char igbvf_copyright[] = 31 "Copyright (c) 2009 - 2012 Intel Corporation."; 32 33 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK) 34 static int debug = -1; 35 module_param(debug, int, 0); 36 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)"); 37 38 static int igbvf_poll(struct napi_struct *napi, int budget); 39 static void igbvf_reset(struct igbvf_adapter *); 40 static void igbvf_set_interrupt_capability(struct igbvf_adapter *); 41 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *); 42 43 static struct igbvf_info igbvf_vf_info = { 44 .mac = e1000_vfadapt, 45 .flags = 0, 46 .pba = 10, 47 .init_ops = e1000_init_function_pointers_vf, 48 }; 49 50 static struct igbvf_info igbvf_i350_vf_info = { 51 .mac = e1000_vfadapt_i350, 52 .flags = 0, 53 .pba = 10, 54 .init_ops = e1000_init_function_pointers_vf, 55 }; 56 57 static const struct igbvf_info *igbvf_info_tbl[] = { 58 [board_vf] = &igbvf_vf_info, 59 [board_i350_vf] = &igbvf_i350_vf_info, 60 }; 61 62 /** 63 * igbvf_desc_unused - calculate if we have unused descriptors 64 * @ring: address of receive ring structure 65 **/ 66 static int igbvf_desc_unused(struct igbvf_ring *ring) 67 { 68 if (ring->next_to_clean > ring->next_to_use) 69 return ring->next_to_clean - ring->next_to_use - 1; 70 71 return ring->count + ring->next_to_clean - ring->next_to_use - 1; 72 } 73 74 /** 75 * igbvf_receive_skb - helper function to handle Rx indications 76 * @adapter: board private structure 77 * @netdev: pointer to netdev struct 78 * @skb: skb to indicate to stack 79 * @status: descriptor status field as written by hardware 80 * @vlan: descriptor vlan field as written by hardware (no le/be conversion) 81 * @skb: pointer to sk_buff to be indicated to stack 82 **/ 83 static void igbvf_receive_skb(struct igbvf_adapter *adapter, 84 struct net_device *netdev, 85 struct sk_buff *skb, 86 u32 status, __le16 vlan) 87 { 88 u16 vid; 89 90 if (status & E1000_RXD_STAT_VP) { 91 if ((adapter->flags & IGBVF_FLAG_RX_LB_VLAN_BSWAP) && 92 (status & E1000_RXDEXT_STATERR_LB)) 93 vid = be16_to_cpu((__force __be16)vlan) & E1000_RXD_SPC_VLAN_MASK; 94 else 95 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK; 96 if (test_bit(vid, adapter->active_vlans)) 97 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid); 98 } 99 100 napi_gro_receive(&adapter->rx_ring->napi, skb); 101 } 102 103 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter, 104 u32 status_err, struct sk_buff *skb) 105 { 106 skb_checksum_none_assert(skb); 107 108 /* Ignore Checksum bit is set or checksum is disabled through ethtool */ 109 if ((status_err & E1000_RXD_STAT_IXSM) || 110 (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED)) 111 return; 112 113 /* TCP/UDP checksum error bit is set */ 114 if (status_err & 115 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) { 116 /* let the stack verify checksum errors */ 117 adapter->hw_csum_err++; 118 return; 119 } 120 121 /* It must be a TCP or UDP packet with a valid checksum */ 122 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)) 123 skb->ip_summed = CHECKSUM_UNNECESSARY; 124 125 adapter->hw_csum_good++; 126 } 127 128 /** 129 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split 130 * @rx_ring: address of ring structure to repopulate 131 * @cleaned_count: number of buffers to repopulate 132 **/ 133 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring, 134 int cleaned_count) 135 { 136 struct igbvf_adapter *adapter = rx_ring->adapter; 137 struct net_device *netdev = adapter->netdev; 138 struct pci_dev *pdev = adapter->pdev; 139 union e1000_adv_rx_desc *rx_desc; 140 struct igbvf_buffer *buffer_info; 141 struct sk_buff *skb; 142 unsigned int i; 143 int bufsz; 144 145 i = rx_ring->next_to_use; 146 buffer_info = &rx_ring->buffer_info[i]; 147 148 if (adapter->rx_ps_hdr_size) 149 bufsz = adapter->rx_ps_hdr_size; 150 else 151 bufsz = adapter->rx_buffer_len; 152 153 while (cleaned_count--) { 154 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i); 155 156 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) { 157 if (!buffer_info->page) { 158 buffer_info->page = alloc_page(GFP_ATOMIC); 159 if (!buffer_info->page) { 160 adapter->alloc_rx_buff_failed++; 161 goto no_buffers; 162 } 163 buffer_info->page_offset = 0; 164 } else { 165 buffer_info->page_offset ^= PAGE_SIZE / 2; 166 } 167 buffer_info->page_dma = 168 dma_map_page(&pdev->dev, buffer_info->page, 169 buffer_info->page_offset, 170 PAGE_SIZE / 2, 171 DMA_FROM_DEVICE); 172 if (dma_mapping_error(&pdev->dev, 173 buffer_info->page_dma)) { 174 __free_page(buffer_info->page); 175 buffer_info->page = NULL; 176 dev_err(&pdev->dev, "RX DMA map failed\n"); 177 break; 178 } 179 } 180 181 if (!buffer_info->skb) { 182 skb = netdev_alloc_skb_ip_align(netdev, bufsz); 183 if (!skb) { 184 adapter->alloc_rx_buff_failed++; 185 goto no_buffers; 186 } 187 188 buffer_info->skb = skb; 189 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, 190 bufsz, 191 DMA_FROM_DEVICE); 192 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) { 193 dev_kfree_skb(buffer_info->skb); 194 buffer_info->skb = NULL; 195 dev_err(&pdev->dev, "RX DMA map failed\n"); 196 goto no_buffers; 197 } 198 } 199 /* Refresh the desc even if buffer_addrs didn't change because 200 * each write-back erases this info. 201 */ 202 if (adapter->rx_ps_hdr_size) { 203 rx_desc->read.pkt_addr = 204 cpu_to_le64(buffer_info->page_dma); 205 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma); 206 } else { 207 rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->dma); 208 rx_desc->read.hdr_addr = 0; 209 } 210 211 i++; 212 if (i == rx_ring->count) 213 i = 0; 214 buffer_info = &rx_ring->buffer_info[i]; 215 } 216 217 no_buffers: 218 if (rx_ring->next_to_use != i) { 219 rx_ring->next_to_use = i; 220 if (i == 0) 221 i = (rx_ring->count - 1); 222 else 223 i--; 224 225 /* Force memory writes to complete before letting h/w 226 * know there are new descriptors to fetch. (Only 227 * applicable for weak-ordered memory model archs, 228 * such as IA-64). 229 */ 230 wmb(); 231 writel(i, adapter->hw.hw_addr + rx_ring->tail); 232 } 233 } 234 235 /** 236 * igbvf_clean_rx_irq - Send received data up the network stack; legacy 237 * @adapter: board private structure 238 * @work_done: output parameter used to indicate completed work 239 * @work_to_do: input parameter setting limit of work 240 * 241 * the return value indicates whether actual cleaning was done, there 242 * is no guarantee that everything was cleaned 243 **/ 244 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter, 245 int *work_done, int work_to_do) 246 { 247 struct igbvf_ring *rx_ring = adapter->rx_ring; 248 struct net_device *netdev = adapter->netdev; 249 struct pci_dev *pdev = adapter->pdev; 250 union e1000_adv_rx_desc *rx_desc, *next_rxd; 251 struct igbvf_buffer *buffer_info, *next_buffer; 252 struct sk_buff *skb; 253 bool cleaned = false; 254 int cleaned_count = 0; 255 unsigned int total_bytes = 0, total_packets = 0; 256 unsigned int i; 257 u32 length, hlen, staterr; 258 259 i = rx_ring->next_to_clean; 260 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i); 261 staterr = le32_to_cpu(rx_desc->wb.upper.status_error); 262 263 while (staterr & E1000_RXD_STAT_DD) { 264 if (*work_done >= work_to_do) 265 break; 266 (*work_done)++; 267 rmb(); /* read descriptor and rx_buffer_info after status DD */ 268 269 buffer_info = &rx_ring->buffer_info[i]; 270 271 /* HW will not DMA in data larger than the given buffer, even 272 * if it parses the (NFS, of course) header to be larger. In 273 * that case, it fills the header buffer and spills the rest 274 * into the page. 275 */ 276 hlen = le16_get_bits(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info, 277 E1000_RXDADV_HDRBUFLEN_MASK); 278 if (hlen > adapter->rx_ps_hdr_size) 279 hlen = adapter->rx_ps_hdr_size; 280 281 length = le16_to_cpu(rx_desc->wb.upper.length); 282 cleaned = true; 283 cleaned_count++; 284 285 skb = buffer_info->skb; 286 prefetch(skb->data - NET_IP_ALIGN); 287 buffer_info->skb = NULL; 288 if (!adapter->rx_ps_hdr_size) { 289 dma_unmap_single(&pdev->dev, buffer_info->dma, 290 adapter->rx_buffer_len, 291 DMA_FROM_DEVICE); 292 buffer_info->dma = 0; 293 skb_put(skb, length); 294 goto send_up; 295 } 296 297 if (!skb_shinfo(skb)->nr_frags) { 298 dma_unmap_single(&pdev->dev, buffer_info->dma, 299 adapter->rx_ps_hdr_size, 300 DMA_FROM_DEVICE); 301 buffer_info->dma = 0; 302 skb_put(skb, hlen); 303 } 304 305 if (length) { 306 dma_unmap_page(&pdev->dev, buffer_info->page_dma, 307 PAGE_SIZE / 2, 308 DMA_FROM_DEVICE); 309 buffer_info->page_dma = 0; 310 311 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags, 312 buffer_info->page, 313 buffer_info->page_offset, 314 length); 315 316 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) || 317 (page_count(buffer_info->page) != 1)) 318 buffer_info->page = NULL; 319 else 320 get_page(buffer_info->page); 321 322 skb->len += length; 323 skb->data_len += length; 324 skb->truesize += PAGE_SIZE / 2; 325 } 326 send_up: 327 i++; 328 if (i == rx_ring->count) 329 i = 0; 330 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i); 331 prefetch(next_rxd); 332 next_buffer = &rx_ring->buffer_info[i]; 333 334 if (!(staterr & E1000_RXD_STAT_EOP)) { 335 buffer_info->skb = next_buffer->skb; 336 buffer_info->dma = next_buffer->dma; 337 next_buffer->skb = skb; 338 next_buffer->dma = 0; 339 goto next_desc; 340 } 341 342 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) { 343 dev_kfree_skb_irq(skb); 344 goto next_desc; 345 } 346 347 total_bytes += skb->len; 348 total_packets++; 349 350 igbvf_rx_checksum_adv(adapter, staterr, skb); 351 352 skb->protocol = eth_type_trans(skb, netdev); 353 354 igbvf_receive_skb(adapter, netdev, skb, staterr, 355 rx_desc->wb.upper.vlan); 356 357 next_desc: 358 rx_desc->wb.upper.status_error = 0; 359 360 /* return some buffers to hardware, one at a time is too slow */ 361 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) { 362 igbvf_alloc_rx_buffers(rx_ring, cleaned_count); 363 cleaned_count = 0; 364 } 365 366 /* use prefetched values */ 367 rx_desc = next_rxd; 368 buffer_info = next_buffer; 369 370 staterr = le32_to_cpu(rx_desc->wb.upper.status_error); 371 } 372 373 rx_ring->next_to_clean = i; 374 cleaned_count = igbvf_desc_unused(rx_ring); 375 376 if (cleaned_count) 377 igbvf_alloc_rx_buffers(rx_ring, cleaned_count); 378 379 adapter->total_rx_packets += total_packets; 380 adapter->total_rx_bytes += total_bytes; 381 netdev->stats.rx_bytes += total_bytes; 382 netdev->stats.rx_packets += total_packets; 383 return cleaned; 384 } 385 386 static void igbvf_put_txbuf(struct igbvf_adapter *adapter, 387 struct igbvf_buffer *buffer_info) 388 { 389 if (buffer_info->dma) { 390 if (buffer_info->mapped_as_page) 391 dma_unmap_page(&adapter->pdev->dev, 392 buffer_info->dma, 393 buffer_info->length, 394 DMA_TO_DEVICE); 395 else 396 dma_unmap_single(&adapter->pdev->dev, 397 buffer_info->dma, 398 buffer_info->length, 399 DMA_TO_DEVICE); 400 buffer_info->dma = 0; 401 } 402 if (buffer_info->skb) { 403 dev_kfree_skb_any(buffer_info->skb); 404 buffer_info->skb = NULL; 405 } 406 buffer_info->time_stamp = 0; 407 } 408 409 /** 410 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors) 411 * @adapter: board private structure 412 * @tx_ring: ring being initialized 413 * 414 * Return 0 on success, negative on failure 415 **/ 416 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter, 417 struct igbvf_ring *tx_ring) 418 { 419 struct pci_dev *pdev = adapter->pdev; 420 int size; 421 422 size = sizeof(struct igbvf_buffer) * tx_ring->count; 423 tx_ring->buffer_info = vzalloc(size); 424 if (!tx_ring->buffer_info) 425 goto err; 426 427 /* round up to nearest 4K */ 428 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc); 429 tx_ring->size = ALIGN(tx_ring->size, 4096); 430 431 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size, 432 &tx_ring->dma, GFP_KERNEL); 433 if (!tx_ring->desc) 434 goto err; 435 436 tx_ring->adapter = adapter; 437 tx_ring->next_to_use = 0; 438 tx_ring->next_to_clean = 0; 439 440 return 0; 441 err: 442 vfree(tx_ring->buffer_info); 443 dev_err(&adapter->pdev->dev, 444 "Unable to allocate memory for the transmit descriptor ring\n"); 445 return -ENOMEM; 446 } 447 448 /** 449 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors) 450 * @adapter: board private structure 451 * @rx_ring: ring being initialized 452 * 453 * Returns 0 on success, negative on failure 454 **/ 455 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter, 456 struct igbvf_ring *rx_ring) 457 { 458 struct pci_dev *pdev = adapter->pdev; 459 int size, desc_len; 460 461 size = sizeof(struct igbvf_buffer) * rx_ring->count; 462 rx_ring->buffer_info = vzalloc(size); 463 if (!rx_ring->buffer_info) 464 goto err; 465 466 desc_len = sizeof(union e1000_adv_rx_desc); 467 468 /* Round up to nearest 4K */ 469 rx_ring->size = rx_ring->count * desc_len; 470 rx_ring->size = ALIGN(rx_ring->size, 4096); 471 472 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size, 473 &rx_ring->dma, GFP_KERNEL); 474 if (!rx_ring->desc) 475 goto err; 476 477 rx_ring->next_to_clean = 0; 478 rx_ring->next_to_use = 0; 479 480 rx_ring->adapter = adapter; 481 482 return 0; 483 484 err: 485 vfree(rx_ring->buffer_info); 486 rx_ring->buffer_info = NULL; 487 dev_err(&adapter->pdev->dev, 488 "Unable to allocate memory for the receive descriptor ring\n"); 489 return -ENOMEM; 490 } 491 492 /** 493 * igbvf_clean_tx_ring - Free Tx Buffers 494 * @tx_ring: ring to be cleaned 495 **/ 496 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring) 497 { 498 struct igbvf_adapter *adapter = tx_ring->adapter; 499 struct igbvf_buffer *buffer_info; 500 unsigned long size; 501 unsigned int i; 502 503 if (!tx_ring->buffer_info) 504 return; 505 506 /* Free all the Tx ring sk_buffs */ 507 for (i = 0; i < tx_ring->count; i++) { 508 buffer_info = &tx_ring->buffer_info[i]; 509 igbvf_put_txbuf(adapter, buffer_info); 510 } 511 512 size = sizeof(struct igbvf_buffer) * tx_ring->count; 513 memset(tx_ring->buffer_info, 0, size); 514 515 /* Zero out the descriptor ring */ 516 memset(tx_ring->desc, 0, tx_ring->size); 517 518 tx_ring->next_to_use = 0; 519 tx_ring->next_to_clean = 0; 520 521 writel(0, adapter->hw.hw_addr + tx_ring->head); 522 writel(0, adapter->hw.hw_addr + tx_ring->tail); 523 } 524 525 /** 526 * igbvf_free_tx_resources - Free Tx Resources per Queue 527 * @tx_ring: ring to free resources from 528 * 529 * Free all transmit software resources 530 **/ 531 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring) 532 { 533 struct pci_dev *pdev = tx_ring->adapter->pdev; 534 535 igbvf_clean_tx_ring(tx_ring); 536 537 vfree(tx_ring->buffer_info); 538 tx_ring->buffer_info = NULL; 539 540 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc, 541 tx_ring->dma); 542 543 tx_ring->desc = NULL; 544 } 545 546 /** 547 * igbvf_clean_rx_ring - Free Rx Buffers per Queue 548 * @rx_ring: ring structure pointer to free buffers from 549 **/ 550 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring) 551 { 552 struct igbvf_adapter *adapter = rx_ring->adapter; 553 struct igbvf_buffer *buffer_info; 554 struct pci_dev *pdev = adapter->pdev; 555 unsigned long size; 556 unsigned int i; 557 558 if (!rx_ring->buffer_info) 559 return; 560 561 /* Free all the Rx ring sk_buffs */ 562 for (i = 0; i < rx_ring->count; i++) { 563 buffer_info = &rx_ring->buffer_info[i]; 564 if (buffer_info->dma) { 565 if (adapter->rx_ps_hdr_size) { 566 dma_unmap_single(&pdev->dev, buffer_info->dma, 567 adapter->rx_ps_hdr_size, 568 DMA_FROM_DEVICE); 569 } else { 570 dma_unmap_single(&pdev->dev, buffer_info->dma, 571 adapter->rx_buffer_len, 572 DMA_FROM_DEVICE); 573 } 574 buffer_info->dma = 0; 575 } 576 577 if (buffer_info->skb) { 578 dev_kfree_skb(buffer_info->skb); 579 buffer_info->skb = NULL; 580 } 581 582 if (buffer_info->page) { 583 if (buffer_info->page_dma) 584 dma_unmap_page(&pdev->dev, 585 buffer_info->page_dma, 586 PAGE_SIZE / 2, 587 DMA_FROM_DEVICE); 588 put_page(buffer_info->page); 589 buffer_info->page = NULL; 590 buffer_info->page_dma = 0; 591 buffer_info->page_offset = 0; 592 } 593 } 594 595 size = sizeof(struct igbvf_buffer) * rx_ring->count; 596 memset(rx_ring->buffer_info, 0, size); 597 598 /* Zero out the descriptor ring */ 599 memset(rx_ring->desc, 0, rx_ring->size); 600 601 rx_ring->next_to_clean = 0; 602 rx_ring->next_to_use = 0; 603 604 writel(0, adapter->hw.hw_addr + rx_ring->head); 605 writel(0, adapter->hw.hw_addr + rx_ring->tail); 606 } 607 608 /** 609 * igbvf_free_rx_resources - Free Rx Resources 610 * @rx_ring: ring to clean the resources from 611 * 612 * Free all receive software resources 613 **/ 614 615 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring) 616 { 617 struct pci_dev *pdev = rx_ring->adapter->pdev; 618 619 igbvf_clean_rx_ring(rx_ring); 620 621 vfree(rx_ring->buffer_info); 622 rx_ring->buffer_info = NULL; 623 624 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc, 625 rx_ring->dma); 626 rx_ring->desc = NULL; 627 } 628 629 /** 630 * igbvf_update_itr - update the dynamic ITR value based on statistics 631 * @adapter: pointer to adapter 632 * @itr_setting: current adapter->itr 633 * @packets: the number of packets during this measurement interval 634 * @bytes: the number of bytes during this measurement interval 635 * 636 * Stores a new ITR value based on packets and byte counts during the last 637 * interrupt. The advantage of per interrupt computation is faster updates 638 * and more accurate ITR for the current traffic pattern. Constants in this 639 * function were computed based on theoretical maximum wire speed and thresholds 640 * were set based on testing data as well as attempting to minimize response 641 * time while increasing bulk throughput. 642 **/ 643 static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter, 644 enum latency_range itr_setting, 645 int packets, int bytes) 646 { 647 enum latency_range retval = itr_setting; 648 649 if (packets == 0) 650 goto update_itr_done; 651 652 switch (itr_setting) { 653 case lowest_latency: 654 /* handle TSO and jumbo frames */ 655 if (bytes/packets > 8000) 656 retval = bulk_latency; 657 else if ((packets < 5) && (bytes > 512)) 658 retval = low_latency; 659 break; 660 case low_latency: /* 50 usec aka 20000 ints/s */ 661 if (bytes > 10000) { 662 /* this if handles the TSO accounting */ 663 if (bytes/packets > 8000) 664 retval = bulk_latency; 665 else if ((packets < 10) || ((bytes/packets) > 1200)) 666 retval = bulk_latency; 667 else if ((packets > 35)) 668 retval = lowest_latency; 669 } else if (bytes/packets > 2000) { 670 retval = bulk_latency; 671 } else if (packets <= 2 && bytes < 512) { 672 retval = lowest_latency; 673 } 674 break; 675 case bulk_latency: /* 250 usec aka 4000 ints/s */ 676 if (bytes > 25000) { 677 if (packets > 35) 678 retval = low_latency; 679 } else if (bytes < 6000) { 680 retval = low_latency; 681 } 682 break; 683 default: 684 break; 685 } 686 687 update_itr_done: 688 return retval; 689 } 690 691 static int igbvf_range_to_itr(enum latency_range current_range) 692 { 693 int new_itr; 694 695 switch (current_range) { 696 /* counts and packets in update_itr are dependent on these numbers */ 697 case lowest_latency: 698 new_itr = IGBVF_70K_ITR; 699 break; 700 case low_latency: 701 new_itr = IGBVF_20K_ITR; 702 break; 703 case bulk_latency: 704 new_itr = IGBVF_4K_ITR; 705 break; 706 default: 707 new_itr = IGBVF_START_ITR; 708 break; 709 } 710 return new_itr; 711 } 712 713 static void igbvf_set_itr(struct igbvf_adapter *adapter) 714 { 715 u32 new_itr; 716 717 adapter->tx_ring->itr_range = 718 igbvf_update_itr(adapter, 719 adapter->tx_ring->itr_val, 720 adapter->total_tx_packets, 721 adapter->total_tx_bytes); 722 723 /* conservative mode (itr 3) eliminates the lowest_latency setting */ 724 if (adapter->requested_itr == 3 && 725 adapter->tx_ring->itr_range == lowest_latency) 726 adapter->tx_ring->itr_range = low_latency; 727 728 new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range); 729 730 if (new_itr != adapter->tx_ring->itr_val) { 731 u32 current_itr = adapter->tx_ring->itr_val; 732 /* this attempts to bias the interrupt rate towards Bulk 733 * by adding intermediate steps when interrupt rate is 734 * increasing 735 */ 736 new_itr = new_itr > current_itr ? 737 min(current_itr + (new_itr >> 2), new_itr) : 738 new_itr; 739 adapter->tx_ring->itr_val = new_itr; 740 741 adapter->tx_ring->set_itr = 1; 742 } 743 744 adapter->rx_ring->itr_range = 745 igbvf_update_itr(adapter, adapter->rx_ring->itr_val, 746 adapter->total_rx_packets, 747 adapter->total_rx_bytes); 748 if (adapter->requested_itr == 3 && 749 adapter->rx_ring->itr_range == lowest_latency) 750 adapter->rx_ring->itr_range = low_latency; 751 752 new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range); 753 754 if (new_itr != adapter->rx_ring->itr_val) { 755 u32 current_itr = adapter->rx_ring->itr_val; 756 757 new_itr = new_itr > current_itr ? 758 min(current_itr + (new_itr >> 2), new_itr) : 759 new_itr; 760 adapter->rx_ring->itr_val = new_itr; 761 762 adapter->rx_ring->set_itr = 1; 763 } 764 } 765 766 /** 767 * igbvf_clean_tx_irq - Reclaim resources after transmit completes 768 * @tx_ring: ring structure to clean descriptors from 769 * 770 * returns true if ring is completely cleaned 771 **/ 772 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring) 773 { 774 struct igbvf_adapter *adapter = tx_ring->adapter; 775 struct net_device *netdev = adapter->netdev; 776 struct igbvf_buffer *buffer_info; 777 struct sk_buff *skb; 778 union e1000_adv_tx_desc *tx_desc, *eop_desc; 779 unsigned int total_bytes = 0, total_packets = 0; 780 unsigned int i, count = 0; 781 bool cleaned = false; 782 783 i = tx_ring->next_to_clean; 784 buffer_info = &tx_ring->buffer_info[i]; 785 eop_desc = buffer_info->next_to_watch; 786 787 do { 788 /* if next_to_watch is not set then there is no work pending */ 789 if (!eop_desc) 790 break; 791 792 /* prevent any other reads prior to eop_desc */ 793 smp_rmb(); 794 795 /* if DD is not set pending work has not been completed */ 796 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD))) 797 break; 798 799 /* clear next_to_watch to prevent false hangs */ 800 buffer_info->next_to_watch = NULL; 801 802 for (cleaned = false; !cleaned; count++) { 803 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i); 804 cleaned = (tx_desc == eop_desc); 805 skb = buffer_info->skb; 806 807 if (skb) { 808 unsigned int segs, bytecount; 809 810 /* gso_segs is currently only valid for tcp */ 811 segs = skb_shinfo(skb)->gso_segs ?: 1; 812 /* multiply data chunks by size of headers */ 813 bytecount = ((segs - 1) * skb_headlen(skb)) + 814 skb->len; 815 total_packets += segs; 816 total_bytes += bytecount; 817 } 818 819 igbvf_put_txbuf(adapter, buffer_info); 820 tx_desc->wb.status = 0; 821 822 i++; 823 if (i == tx_ring->count) 824 i = 0; 825 826 buffer_info = &tx_ring->buffer_info[i]; 827 } 828 829 eop_desc = buffer_info->next_to_watch; 830 } while (count < tx_ring->count); 831 832 tx_ring->next_to_clean = i; 833 834 if (unlikely(count && netif_carrier_ok(netdev) && 835 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) { 836 /* Make sure that anybody stopping the queue after this 837 * sees the new next_to_clean. 838 */ 839 smp_mb(); 840 if (netif_queue_stopped(netdev) && 841 !(test_bit(__IGBVF_DOWN, &adapter->state))) { 842 netif_wake_queue(netdev); 843 ++adapter->restart_queue; 844 } 845 } 846 847 netdev->stats.tx_bytes += total_bytes; 848 netdev->stats.tx_packets += total_packets; 849 return count < tx_ring->count; 850 } 851 852 static irqreturn_t igbvf_msix_other(int irq, void *data) 853 { 854 struct net_device *netdev = data; 855 struct igbvf_adapter *adapter = netdev_priv(netdev); 856 struct e1000_hw *hw = &adapter->hw; 857 858 adapter->int_counter1++; 859 860 hw->mac.get_link_status = 1; 861 if (!test_bit(__IGBVF_DOWN, &adapter->state)) 862 mod_timer(&adapter->watchdog_timer, jiffies + 1); 863 864 ew32(EIMS, adapter->eims_other); 865 866 return IRQ_HANDLED; 867 } 868 869 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data) 870 { 871 struct net_device *netdev = data; 872 struct igbvf_adapter *adapter = netdev_priv(netdev); 873 struct e1000_hw *hw = &adapter->hw; 874 struct igbvf_ring *tx_ring = adapter->tx_ring; 875 876 if (tx_ring->set_itr) { 877 writel(tx_ring->itr_val, 878 adapter->hw.hw_addr + tx_ring->itr_register); 879 adapter->tx_ring->set_itr = 0; 880 } 881 882 adapter->total_tx_bytes = 0; 883 adapter->total_tx_packets = 0; 884 885 /* auto mask will automatically re-enable the interrupt when we write 886 * EICS 887 */ 888 if (!igbvf_clean_tx_irq(tx_ring)) 889 /* Ring was not completely cleaned, so fire another interrupt */ 890 ew32(EICS, tx_ring->eims_value); 891 else 892 ew32(EIMS, tx_ring->eims_value); 893 894 return IRQ_HANDLED; 895 } 896 897 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data) 898 { 899 struct net_device *netdev = data; 900 struct igbvf_adapter *adapter = netdev_priv(netdev); 901 902 adapter->int_counter0++; 903 904 /* Write the ITR value calculated at the end of the 905 * previous interrupt. 906 */ 907 if (adapter->rx_ring->set_itr) { 908 writel(adapter->rx_ring->itr_val, 909 adapter->hw.hw_addr + adapter->rx_ring->itr_register); 910 adapter->rx_ring->set_itr = 0; 911 } 912 913 if (napi_schedule_prep(&adapter->rx_ring->napi)) { 914 adapter->total_rx_bytes = 0; 915 adapter->total_rx_packets = 0; 916 __napi_schedule(&adapter->rx_ring->napi); 917 } 918 919 return IRQ_HANDLED; 920 } 921 922 #define IGBVF_NO_QUEUE -1 923 924 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue, 925 int tx_queue, int msix_vector) 926 { 927 struct e1000_hw *hw = &adapter->hw; 928 u32 ivar, index; 929 930 /* 82576 uses a table-based method for assigning vectors. 931 * Each queue has a single entry in the table to which we write 932 * a vector number along with a "valid" bit. Sadly, the layout 933 * of the table is somewhat counterintuitive. 934 */ 935 if (rx_queue > IGBVF_NO_QUEUE) { 936 index = (rx_queue >> 1); 937 ivar = array_er32(IVAR0, index); 938 if (rx_queue & 0x1) { 939 /* vector goes into third byte of register */ 940 ivar = ivar & 0xFF00FFFF; 941 ivar |= (msix_vector | E1000_IVAR_VALID) << 16; 942 } else { 943 /* vector goes into low byte of register */ 944 ivar = ivar & 0xFFFFFF00; 945 ivar |= msix_vector | E1000_IVAR_VALID; 946 } 947 adapter->rx_ring[rx_queue].eims_value = BIT(msix_vector); 948 array_ew32(IVAR0, index, ivar); 949 } 950 if (tx_queue > IGBVF_NO_QUEUE) { 951 index = (tx_queue >> 1); 952 ivar = array_er32(IVAR0, index); 953 if (tx_queue & 0x1) { 954 /* vector goes into high byte of register */ 955 ivar = ivar & 0x00FFFFFF; 956 ivar |= (msix_vector | E1000_IVAR_VALID) << 24; 957 } else { 958 /* vector goes into second byte of register */ 959 ivar = ivar & 0xFFFF00FF; 960 ivar |= (msix_vector | E1000_IVAR_VALID) << 8; 961 } 962 adapter->tx_ring[tx_queue].eims_value = BIT(msix_vector); 963 array_ew32(IVAR0, index, ivar); 964 } 965 } 966 967 /** 968 * igbvf_configure_msix - Configure MSI-X hardware 969 * @adapter: board private structure 970 * 971 * igbvf_configure_msix sets up the hardware to properly 972 * generate MSI-X interrupts. 973 **/ 974 static void igbvf_configure_msix(struct igbvf_adapter *adapter) 975 { 976 u32 tmp; 977 struct e1000_hw *hw = &adapter->hw; 978 struct igbvf_ring *tx_ring = adapter->tx_ring; 979 struct igbvf_ring *rx_ring = adapter->rx_ring; 980 int vector = 0; 981 982 adapter->eims_enable_mask = 0; 983 984 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++); 985 adapter->eims_enable_mask |= tx_ring->eims_value; 986 writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register); 987 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++); 988 adapter->eims_enable_mask |= rx_ring->eims_value; 989 writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register); 990 991 /* set vector for other causes, i.e. link changes */ 992 993 tmp = (vector++ | E1000_IVAR_VALID); 994 995 ew32(IVAR_MISC, tmp); 996 997 adapter->eims_enable_mask = GENMASK(vector - 1, 0); 998 adapter->eims_other = BIT(vector - 1); 999 e1e_flush(); 1000 } 1001 1002 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter) 1003 { 1004 if (adapter->msix_entries) { 1005 pci_disable_msix(adapter->pdev); 1006 kfree(adapter->msix_entries); 1007 adapter->msix_entries = NULL; 1008 } 1009 } 1010 1011 /** 1012 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported 1013 * @adapter: board private structure 1014 * 1015 * Attempt to configure interrupts using the best available 1016 * capabilities of the hardware and kernel. 1017 **/ 1018 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter) 1019 { 1020 int err = -ENOMEM; 1021 int i; 1022 1023 /* we allocate 3 vectors, 1 for Tx, 1 for Rx, one for PF messages */ 1024 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry), 1025 GFP_KERNEL); 1026 if (adapter->msix_entries) { 1027 for (i = 0; i < 3; i++) 1028 adapter->msix_entries[i].entry = i; 1029 1030 err = pci_enable_msix_range(adapter->pdev, 1031 adapter->msix_entries, 3, 3); 1032 } 1033 1034 if (err < 0) { 1035 /* MSI-X failed */ 1036 dev_err(&adapter->pdev->dev, 1037 "Failed to initialize MSI-X interrupts.\n"); 1038 igbvf_reset_interrupt_capability(adapter); 1039 } 1040 } 1041 1042 /** 1043 * igbvf_request_msix - Initialize MSI-X interrupts 1044 * @adapter: board private structure 1045 * 1046 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the 1047 * kernel. 1048 **/ 1049 static int igbvf_request_msix(struct igbvf_adapter *adapter) 1050 { 1051 struct net_device *netdev = adapter->netdev; 1052 int err = 0, vector = 0; 1053 1054 if (strlen(netdev->name) < (IFNAMSIZ - 5)) { 1055 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name); 1056 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name); 1057 } else { 1058 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ); 1059 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ); 1060 } 1061 1062 err = request_irq(adapter->msix_entries[vector].vector, 1063 igbvf_intr_msix_tx, 0, adapter->tx_ring->name, 1064 netdev); 1065 if (err) 1066 goto out; 1067 1068 adapter->tx_ring->itr_register = E1000_EITR(vector); 1069 adapter->tx_ring->itr_val = adapter->current_itr; 1070 vector++; 1071 1072 err = request_irq(adapter->msix_entries[vector].vector, 1073 igbvf_intr_msix_rx, 0, adapter->rx_ring->name, 1074 netdev); 1075 if (err) 1076 goto free_irq_tx; 1077 1078 adapter->rx_ring->itr_register = E1000_EITR(vector); 1079 adapter->rx_ring->itr_val = adapter->current_itr; 1080 vector++; 1081 1082 err = request_irq(adapter->msix_entries[vector].vector, 1083 igbvf_msix_other, 0, netdev->name, netdev); 1084 if (err) 1085 goto free_irq_rx; 1086 1087 igbvf_configure_msix(adapter); 1088 return 0; 1089 free_irq_rx: 1090 free_irq(adapter->msix_entries[--vector].vector, netdev); 1091 free_irq_tx: 1092 free_irq(adapter->msix_entries[--vector].vector, netdev); 1093 out: 1094 return err; 1095 } 1096 1097 /** 1098 * igbvf_alloc_queues - Allocate memory for all rings 1099 * @adapter: board private structure to initialize 1100 **/ 1101 static int igbvf_alloc_queues(struct igbvf_adapter *adapter) 1102 { 1103 struct net_device *netdev = adapter->netdev; 1104 1105 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL); 1106 if (!adapter->tx_ring) 1107 return -ENOMEM; 1108 1109 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL); 1110 if (!adapter->rx_ring) { 1111 kfree(adapter->tx_ring); 1112 return -ENOMEM; 1113 } 1114 1115 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll); 1116 1117 return 0; 1118 } 1119 1120 /** 1121 * igbvf_request_irq - initialize interrupts 1122 * @adapter: board private structure 1123 * 1124 * Attempts to configure interrupts using the best available 1125 * capabilities of the hardware and kernel. 1126 **/ 1127 static int igbvf_request_irq(struct igbvf_adapter *adapter) 1128 { 1129 int err = -1; 1130 1131 /* igbvf supports msi-x only */ 1132 if (adapter->msix_entries) 1133 err = igbvf_request_msix(adapter); 1134 1135 if (!err) 1136 return err; 1137 1138 dev_err(&adapter->pdev->dev, 1139 "Unable to allocate interrupt, Error: %d\n", err); 1140 1141 return err; 1142 } 1143 1144 static void igbvf_free_irq(struct igbvf_adapter *adapter) 1145 { 1146 struct net_device *netdev = adapter->netdev; 1147 int vector; 1148 1149 if (adapter->msix_entries) { 1150 for (vector = 0; vector < 3; vector++) 1151 free_irq(adapter->msix_entries[vector].vector, netdev); 1152 } 1153 } 1154 1155 /** 1156 * igbvf_irq_disable - Mask off interrupt generation on the NIC 1157 * @adapter: board private structure 1158 **/ 1159 static void igbvf_irq_disable(struct igbvf_adapter *adapter) 1160 { 1161 struct e1000_hw *hw = &adapter->hw; 1162 1163 ew32(EIMC, ~0); 1164 1165 if (adapter->msix_entries) 1166 ew32(EIAC, 0); 1167 } 1168 1169 /** 1170 * igbvf_irq_enable - Enable default interrupt generation settings 1171 * @adapter: board private structure 1172 **/ 1173 static void igbvf_irq_enable(struct igbvf_adapter *adapter) 1174 { 1175 struct e1000_hw *hw = &adapter->hw; 1176 1177 ew32(EIAC, adapter->eims_enable_mask); 1178 ew32(EIAM, adapter->eims_enable_mask); 1179 ew32(EIMS, adapter->eims_enable_mask); 1180 } 1181 1182 /** 1183 * igbvf_poll - NAPI Rx polling callback 1184 * @napi: struct associated with this polling callback 1185 * @budget: amount of packets driver is allowed to process this poll 1186 **/ 1187 static int igbvf_poll(struct napi_struct *napi, int budget) 1188 { 1189 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi); 1190 struct igbvf_adapter *adapter = rx_ring->adapter; 1191 struct e1000_hw *hw = &adapter->hw; 1192 int work_done = 0; 1193 1194 igbvf_clean_rx_irq(adapter, &work_done, budget); 1195 1196 if (work_done == budget) 1197 return budget; 1198 1199 /* Exit the polling mode, but don't re-enable interrupts if stack might 1200 * poll us due to busy-polling 1201 */ 1202 if (likely(napi_complete_done(napi, work_done))) { 1203 if (adapter->requested_itr & 3) 1204 igbvf_set_itr(adapter); 1205 1206 if (!test_bit(__IGBVF_DOWN, &adapter->state)) 1207 ew32(EIMS, adapter->rx_ring->eims_value); 1208 } 1209 1210 return work_done; 1211 } 1212 1213 /** 1214 * igbvf_set_rlpml - set receive large packet maximum length 1215 * @adapter: board private structure 1216 * 1217 * Configure the maximum size of packets that will be received 1218 */ 1219 static void igbvf_set_rlpml(struct igbvf_adapter *adapter) 1220 { 1221 int max_frame_size; 1222 struct e1000_hw *hw = &adapter->hw; 1223 1224 max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE; 1225 1226 spin_lock_bh(&hw->mbx_lock); 1227 1228 e1000_rlpml_set_vf(hw, max_frame_size); 1229 1230 spin_unlock_bh(&hw->mbx_lock); 1231 } 1232 1233 static int igbvf_vlan_rx_add_vid(struct net_device *netdev, 1234 __be16 proto, u16 vid) 1235 { 1236 struct igbvf_adapter *adapter = netdev_priv(netdev); 1237 struct e1000_hw *hw = &adapter->hw; 1238 1239 spin_lock_bh(&hw->mbx_lock); 1240 1241 if (hw->mac.ops.set_vfta(hw, vid, true)) { 1242 dev_warn(&adapter->pdev->dev, "Vlan id %d\n is not added", vid); 1243 spin_unlock_bh(&hw->mbx_lock); 1244 return -EINVAL; 1245 } 1246 1247 spin_unlock_bh(&hw->mbx_lock); 1248 1249 set_bit(vid, adapter->active_vlans); 1250 return 0; 1251 } 1252 1253 static int igbvf_vlan_rx_kill_vid(struct net_device *netdev, 1254 __be16 proto, u16 vid) 1255 { 1256 struct igbvf_adapter *adapter = netdev_priv(netdev); 1257 struct e1000_hw *hw = &adapter->hw; 1258 1259 spin_lock_bh(&hw->mbx_lock); 1260 1261 if (hw->mac.ops.set_vfta(hw, vid, false)) { 1262 dev_err(&adapter->pdev->dev, 1263 "Failed to remove vlan id %d\n", vid); 1264 spin_unlock_bh(&hw->mbx_lock); 1265 return -EINVAL; 1266 } 1267 1268 spin_unlock_bh(&hw->mbx_lock); 1269 1270 clear_bit(vid, adapter->active_vlans); 1271 return 0; 1272 } 1273 1274 static void igbvf_restore_vlan(struct igbvf_adapter *adapter) 1275 { 1276 u16 vid; 1277 1278 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID) 1279 igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid); 1280 } 1281 1282 /** 1283 * igbvf_configure_tx - Configure Transmit Unit after Reset 1284 * @adapter: board private structure 1285 * 1286 * Configure the Tx unit of the MAC after a reset. 1287 **/ 1288 static void igbvf_configure_tx(struct igbvf_adapter *adapter) 1289 { 1290 struct e1000_hw *hw = &adapter->hw; 1291 struct igbvf_ring *tx_ring = adapter->tx_ring; 1292 u64 tdba; 1293 u32 txdctl, dca_txctrl; 1294 1295 /* disable transmits */ 1296 txdctl = er32(TXDCTL(0)); 1297 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE); 1298 e1e_flush(); 1299 msleep(10); 1300 1301 /* Setup the HW Tx Head and Tail descriptor pointers */ 1302 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc)); 1303 tdba = tx_ring->dma; 1304 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32))); 1305 ew32(TDBAH(0), (tdba >> 32)); 1306 ew32(TDH(0), 0); 1307 ew32(TDT(0), 0); 1308 tx_ring->head = E1000_TDH(0); 1309 tx_ring->tail = E1000_TDT(0); 1310 1311 /* Turn off Relaxed Ordering on head write-backs. The writebacks 1312 * MUST be delivered in order or it will completely screw up 1313 * our bookkeeping. 1314 */ 1315 dca_txctrl = er32(DCA_TXCTRL(0)); 1316 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN; 1317 ew32(DCA_TXCTRL(0), dca_txctrl); 1318 1319 /* enable transmits */ 1320 txdctl |= E1000_TXDCTL_QUEUE_ENABLE; 1321 ew32(TXDCTL(0), txdctl); 1322 1323 /* Setup Transmit Descriptor Settings for eop descriptor */ 1324 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS; 1325 1326 /* enable Report Status bit */ 1327 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS; 1328 } 1329 1330 /** 1331 * igbvf_setup_srrctl - configure the receive control registers 1332 * @adapter: Board private structure 1333 **/ 1334 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter) 1335 { 1336 struct e1000_hw *hw = &adapter->hw; 1337 u32 srrctl = 0; 1338 1339 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK | 1340 E1000_SRRCTL_BSIZEHDR_MASK | 1341 E1000_SRRCTL_BSIZEPKT_MASK); 1342 1343 /* Enable queue drop to avoid head of line blocking */ 1344 srrctl |= E1000_SRRCTL_DROP_EN; 1345 1346 /* Setup buffer sizes */ 1347 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >> 1348 E1000_SRRCTL_BSIZEPKT_SHIFT; 1349 1350 if (adapter->rx_buffer_len < 2048) { 1351 adapter->rx_ps_hdr_size = 0; 1352 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF; 1353 } else { 1354 adapter->rx_ps_hdr_size = 128; 1355 srrctl |= adapter->rx_ps_hdr_size << 1356 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT; 1357 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS; 1358 } 1359 1360 ew32(SRRCTL(0), srrctl); 1361 } 1362 1363 /** 1364 * igbvf_configure_rx - Configure Receive Unit after Reset 1365 * @adapter: board private structure 1366 * 1367 * Configure the Rx unit of the MAC after a reset. 1368 **/ 1369 static void igbvf_configure_rx(struct igbvf_adapter *adapter) 1370 { 1371 struct e1000_hw *hw = &adapter->hw; 1372 struct igbvf_ring *rx_ring = adapter->rx_ring; 1373 u64 rdba; 1374 u32 rxdctl; 1375 1376 /* disable receives */ 1377 rxdctl = er32(RXDCTL(0)); 1378 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE); 1379 e1e_flush(); 1380 msleep(10); 1381 1382 /* Setup the HW Rx Head and Tail Descriptor Pointers and 1383 * the Base and Length of the Rx Descriptor Ring 1384 */ 1385 rdba = rx_ring->dma; 1386 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32))); 1387 ew32(RDBAH(0), (rdba >> 32)); 1388 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc)); 1389 rx_ring->head = E1000_RDH(0); 1390 rx_ring->tail = E1000_RDT(0); 1391 ew32(RDH(0), 0); 1392 ew32(RDT(0), 0); 1393 1394 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE; 1395 rxdctl &= 0xFFF00000; 1396 rxdctl |= IGBVF_RX_PTHRESH; 1397 rxdctl |= IGBVF_RX_HTHRESH << 8; 1398 rxdctl |= IGBVF_RX_WTHRESH << 16; 1399 1400 igbvf_set_rlpml(adapter); 1401 1402 /* enable receives */ 1403 ew32(RXDCTL(0), rxdctl); 1404 } 1405 1406 /** 1407 * igbvf_set_multi - Multicast and Promiscuous mode set 1408 * @netdev: network interface device structure 1409 * 1410 * The set_multi entry point is called whenever the multicast address 1411 * list or the network interface flags are updated. This routine is 1412 * responsible for configuring the hardware for proper multicast, 1413 * promiscuous mode, and all-multi behavior. 1414 **/ 1415 static void igbvf_set_multi(struct net_device *netdev) 1416 { 1417 struct igbvf_adapter *adapter = netdev_priv(netdev); 1418 struct e1000_hw *hw = &adapter->hw; 1419 struct netdev_hw_addr *ha; 1420 u8 *mta_list = NULL; 1421 int i; 1422 1423 if (!netdev_mc_empty(netdev)) { 1424 mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN, 1425 GFP_ATOMIC); 1426 if (!mta_list) 1427 return; 1428 } 1429 1430 /* prepare a packed array of only addresses. */ 1431 i = 0; 1432 netdev_for_each_mc_addr(ha, netdev) 1433 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN); 1434 1435 spin_lock_bh(&hw->mbx_lock); 1436 1437 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0); 1438 1439 spin_unlock_bh(&hw->mbx_lock); 1440 kfree(mta_list); 1441 } 1442 1443 /** 1444 * igbvf_set_uni - Configure unicast MAC filters 1445 * @netdev: network interface device structure 1446 * 1447 * This routine is responsible for configuring the hardware for proper 1448 * unicast filters. 1449 **/ 1450 static int igbvf_set_uni(struct net_device *netdev) 1451 { 1452 struct igbvf_adapter *adapter = netdev_priv(netdev); 1453 struct e1000_hw *hw = &adapter->hw; 1454 1455 if (netdev_uc_count(netdev) > IGBVF_MAX_MAC_FILTERS) { 1456 pr_err("Too many unicast filters - No Space\n"); 1457 return -ENOSPC; 1458 } 1459 1460 spin_lock_bh(&hw->mbx_lock); 1461 1462 /* Clear all unicast MAC filters */ 1463 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_CLR, NULL); 1464 1465 spin_unlock_bh(&hw->mbx_lock); 1466 1467 if (!netdev_uc_empty(netdev)) { 1468 struct netdev_hw_addr *ha; 1469 1470 /* Add MAC filters one by one */ 1471 netdev_for_each_uc_addr(ha, netdev) { 1472 spin_lock_bh(&hw->mbx_lock); 1473 1474 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_ADD, 1475 ha->addr); 1476 1477 spin_unlock_bh(&hw->mbx_lock); 1478 udelay(200); 1479 } 1480 } 1481 1482 return 0; 1483 } 1484 1485 static void igbvf_set_rx_mode(struct net_device *netdev) 1486 { 1487 igbvf_set_multi(netdev); 1488 igbvf_set_uni(netdev); 1489 } 1490 1491 /** 1492 * igbvf_configure - configure the hardware for Rx and Tx 1493 * @adapter: private board structure 1494 **/ 1495 static void igbvf_configure(struct igbvf_adapter *adapter) 1496 { 1497 igbvf_set_rx_mode(adapter->netdev); 1498 1499 igbvf_restore_vlan(adapter); 1500 1501 igbvf_configure_tx(adapter); 1502 igbvf_setup_srrctl(adapter); 1503 igbvf_configure_rx(adapter); 1504 igbvf_alloc_rx_buffers(adapter->rx_ring, 1505 igbvf_desc_unused(adapter->rx_ring)); 1506 } 1507 1508 /* igbvf_reset - bring the hardware into a known good state 1509 * @adapter: private board structure 1510 * 1511 * This function boots the hardware and enables some settings that 1512 * require a configuration cycle of the hardware - those cannot be 1513 * set/changed during runtime. After reset the device needs to be 1514 * properly configured for Rx, Tx etc. 1515 */ 1516 static void igbvf_reset(struct igbvf_adapter *adapter) 1517 { 1518 struct e1000_mac_info *mac = &adapter->hw.mac; 1519 struct net_device *netdev = adapter->netdev; 1520 struct e1000_hw *hw = &adapter->hw; 1521 1522 spin_lock_bh(&hw->mbx_lock); 1523 1524 /* Allow time for pending master requests to run */ 1525 if (mac->ops.reset_hw(hw)) 1526 dev_info(&adapter->pdev->dev, "PF still resetting\n"); 1527 1528 mac->ops.init_hw(hw); 1529 1530 spin_unlock_bh(&hw->mbx_lock); 1531 1532 if (is_valid_ether_addr(adapter->hw.mac.addr)) { 1533 eth_hw_addr_set(netdev, adapter->hw.mac.addr); 1534 memcpy(netdev->perm_addr, adapter->hw.mac.addr, 1535 netdev->addr_len); 1536 } 1537 1538 adapter->last_reset = jiffies; 1539 } 1540 1541 int igbvf_up(struct igbvf_adapter *adapter) 1542 { 1543 struct e1000_hw *hw = &adapter->hw; 1544 1545 /* hardware has been reset, we need to reload some things */ 1546 igbvf_configure(adapter); 1547 1548 clear_bit(__IGBVF_DOWN, &adapter->state); 1549 1550 napi_enable(&adapter->rx_ring->napi); 1551 if (adapter->msix_entries) 1552 igbvf_configure_msix(adapter); 1553 1554 /* Clear any pending interrupts. */ 1555 er32(EICR); 1556 igbvf_irq_enable(adapter); 1557 1558 /* start the watchdog */ 1559 hw->mac.get_link_status = 1; 1560 mod_timer(&adapter->watchdog_timer, jiffies + 1); 1561 1562 return 0; 1563 } 1564 1565 void igbvf_down(struct igbvf_adapter *adapter) 1566 { 1567 struct net_device *netdev = adapter->netdev; 1568 struct e1000_hw *hw = &adapter->hw; 1569 u32 rxdctl, txdctl; 1570 1571 /* signal that we're down so the interrupt handler does not 1572 * reschedule our watchdog timer 1573 */ 1574 set_bit(__IGBVF_DOWN, &adapter->state); 1575 1576 /* disable receives in the hardware */ 1577 rxdctl = er32(RXDCTL(0)); 1578 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE); 1579 1580 netif_carrier_off(netdev); 1581 netif_stop_queue(netdev); 1582 1583 /* disable transmits in the hardware */ 1584 txdctl = er32(TXDCTL(0)); 1585 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE); 1586 1587 /* flush both disables and wait for them to finish */ 1588 e1e_flush(); 1589 msleep(10); 1590 1591 napi_disable(&adapter->rx_ring->napi); 1592 1593 igbvf_irq_disable(adapter); 1594 1595 del_timer_sync(&adapter->watchdog_timer); 1596 1597 /* record the stats before reset*/ 1598 igbvf_update_stats(adapter); 1599 1600 adapter->link_speed = 0; 1601 adapter->link_duplex = 0; 1602 1603 igbvf_reset(adapter); 1604 igbvf_clean_tx_ring(adapter->tx_ring); 1605 igbvf_clean_rx_ring(adapter->rx_ring); 1606 } 1607 1608 void igbvf_reinit_locked(struct igbvf_adapter *adapter) 1609 { 1610 might_sleep(); 1611 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state)) 1612 usleep_range(1000, 2000); 1613 igbvf_down(adapter); 1614 igbvf_up(adapter); 1615 clear_bit(__IGBVF_RESETTING, &adapter->state); 1616 } 1617 1618 /** 1619 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter) 1620 * @adapter: board private structure to initialize 1621 * 1622 * igbvf_sw_init initializes the Adapter private data structure. 1623 * Fields are initialized based on PCI device information and 1624 * OS network device settings (MTU size). 1625 **/ 1626 static int igbvf_sw_init(struct igbvf_adapter *adapter) 1627 { 1628 struct net_device *netdev = adapter->netdev; 1629 s32 rc; 1630 1631 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN; 1632 adapter->rx_ps_hdr_size = 0; 1633 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN; 1634 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN; 1635 1636 adapter->tx_int_delay = 8; 1637 adapter->tx_abs_int_delay = 32; 1638 adapter->rx_int_delay = 0; 1639 adapter->rx_abs_int_delay = 8; 1640 adapter->requested_itr = 3; 1641 adapter->current_itr = IGBVF_START_ITR; 1642 1643 /* Set various function pointers */ 1644 adapter->ei->init_ops(&adapter->hw); 1645 1646 rc = adapter->hw.mac.ops.init_params(&adapter->hw); 1647 if (rc) 1648 return rc; 1649 1650 rc = adapter->hw.mbx.ops.init_params(&adapter->hw); 1651 if (rc) 1652 return rc; 1653 1654 igbvf_set_interrupt_capability(adapter); 1655 1656 if (igbvf_alloc_queues(adapter)) 1657 return -ENOMEM; 1658 1659 spin_lock_init(&adapter->tx_queue_lock); 1660 1661 /* Explicitly disable IRQ since the NIC can be in any state. */ 1662 igbvf_irq_disable(adapter); 1663 1664 spin_lock_init(&adapter->stats_lock); 1665 spin_lock_init(&adapter->hw.mbx_lock); 1666 1667 set_bit(__IGBVF_DOWN, &adapter->state); 1668 return 0; 1669 } 1670 1671 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter) 1672 { 1673 struct e1000_hw *hw = &adapter->hw; 1674 1675 adapter->stats.last_gprc = er32(VFGPRC); 1676 adapter->stats.last_gorc = er32(VFGORC); 1677 adapter->stats.last_gptc = er32(VFGPTC); 1678 adapter->stats.last_gotc = er32(VFGOTC); 1679 adapter->stats.last_mprc = er32(VFMPRC); 1680 adapter->stats.last_gotlbc = er32(VFGOTLBC); 1681 adapter->stats.last_gptlbc = er32(VFGPTLBC); 1682 adapter->stats.last_gorlbc = er32(VFGORLBC); 1683 adapter->stats.last_gprlbc = er32(VFGPRLBC); 1684 1685 adapter->stats.base_gprc = er32(VFGPRC); 1686 adapter->stats.base_gorc = er32(VFGORC); 1687 adapter->stats.base_gptc = er32(VFGPTC); 1688 adapter->stats.base_gotc = er32(VFGOTC); 1689 adapter->stats.base_mprc = er32(VFMPRC); 1690 adapter->stats.base_gotlbc = er32(VFGOTLBC); 1691 adapter->stats.base_gptlbc = er32(VFGPTLBC); 1692 adapter->stats.base_gorlbc = er32(VFGORLBC); 1693 adapter->stats.base_gprlbc = er32(VFGPRLBC); 1694 } 1695 1696 /** 1697 * igbvf_open - Called when a network interface is made active 1698 * @netdev: network interface device structure 1699 * 1700 * Returns 0 on success, negative value on failure 1701 * 1702 * The open entry point is called when a network interface is made 1703 * active by the system (IFF_UP). At this point all resources needed 1704 * for transmit and receive operations are allocated, the interrupt 1705 * handler is registered with the OS, the watchdog timer is started, 1706 * and the stack is notified that the interface is ready. 1707 **/ 1708 static int igbvf_open(struct net_device *netdev) 1709 { 1710 struct igbvf_adapter *adapter = netdev_priv(netdev); 1711 struct e1000_hw *hw = &adapter->hw; 1712 int err; 1713 1714 /* disallow open during test */ 1715 if (test_bit(__IGBVF_TESTING, &adapter->state)) 1716 return -EBUSY; 1717 1718 /* allocate transmit descriptors */ 1719 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring); 1720 if (err) 1721 goto err_setup_tx; 1722 1723 /* allocate receive descriptors */ 1724 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring); 1725 if (err) 1726 goto err_setup_rx; 1727 1728 /* before we allocate an interrupt, we must be ready to handle it. 1729 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt 1730 * as soon as we call pci_request_irq, so we have to setup our 1731 * clean_rx handler before we do so. 1732 */ 1733 igbvf_configure(adapter); 1734 1735 err = igbvf_request_irq(adapter); 1736 if (err) 1737 goto err_req_irq; 1738 1739 /* From here on the code is the same as igbvf_up() */ 1740 clear_bit(__IGBVF_DOWN, &adapter->state); 1741 1742 napi_enable(&adapter->rx_ring->napi); 1743 1744 /* clear any pending interrupts */ 1745 er32(EICR); 1746 1747 igbvf_irq_enable(adapter); 1748 1749 /* start the watchdog */ 1750 hw->mac.get_link_status = 1; 1751 mod_timer(&adapter->watchdog_timer, jiffies + 1); 1752 1753 return 0; 1754 1755 err_req_irq: 1756 igbvf_free_rx_resources(adapter->rx_ring); 1757 err_setup_rx: 1758 igbvf_free_tx_resources(adapter->tx_ring); 1759 err_setup_tx: 1760 igbvf_reset(adapter); 1761 1762 return err; 1763 } 1764 1765 /** 1766 * igbvf_close - Disables a network interface 1767 * @netdev: network interface device structure 1768 * 1769 * Returns 0, this is not allowed to fail 1770 * 1771 * The close entry point is called when an interface is de-activated 1772 * by the OS. The hardware is still under the drivers control, but 1773 * needs to be disabled. A global MAC reset is issued to stop the 1774 * hardware, and all transmit and receive resources are freed. 1775 **/ 1776 static int igbvf_close(struct net_device *netdev) 1777 { 1778 struct igbvf_adapter *adapter = netdev_priv(netdev); 1779 1780 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state)); 1781 igbvf_down(adapter); 1782 1783 igbvf_free_irq(adapter); 1784 1785 igbvf_free_tx_resources(adapter->tx_ring); 1786 igbvf_free_rx_resources(adapter->rx_ring); 1787 1788 return 0; 1789 } 1790 1791 /** 1792 * igbvf_set_mac - Change the Ethernet Address of the NIC 1793 * @netdev: network interface device structure 1794 * @p: pointer to an address structure 1795 * 1796 * Returns 0 on success, negative on failure 1797 **/ 1798 static int igbvf_set_mac(struct net_device *netdev, void *p) 1799 { 1800 struct igbvf_adapter *adapter = netdev_priv(netdev); 1801 struct e1000_hw *hw = &adapter->hw; 1802 struct sockaddr *addr = p; 1803 1804 if (!is_valid_ether_addr(addr->sa_data)) 1805 return -EADDRNOTAVAIL; 1806 1807 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len); 1808 1809 spin_lock_bh(&hw->mbx_lock); 1810 1811 hw->mac.ops.rar_set(hw, hw->mac.addr, 0); 1812 1813 spin_unlock_bh(&hw->mbx_lock); 1814 1815 if (!ether_addr_equal(addr->sa_data, hw->mac.addr)) 1816 return -EADDRNOTAVAIL; 1817 1818 eth_hw_addr_set(netdev, addr->sa_data); 1819 1820 return 0; 1821 } 1822 1823 #define UPDATE_VF_COUNTER(reg, name) \ 1824 { \ 1825 u32 current_counter = er32(reg); \ 1826 if (current_counter < adapter->stats.last_##name) \ 1827 adapter->stats.name += 0x100000000LL; \ 1828 adapter->stats.last_##name = current_counter; \ 1829 adapter->stats.name &= 0xFFFFFFFF00000000LL; \ 1830 adapter->stats.name |= current_counter; \ 1831 } 1832 1833 /** 1834 * igbvf_update_stats - Update the board statistics counters 1835 * @adapter: board private structure 1836 **/ 1837 void igbvf_update_stats(struct igbvf_adapter *adapter) 1838 { 1839 struct e1000_hw *hw = &adapter->hw; 1840 struct pci_dev *pdev = adapter->pdev; 1841 1842 /* Prevent stats update while adapter is being reset, link is down 1843 * or if the pci connection is down. 1844 */ 1845 if (adapter->link_speed == 0) 1846 return; 1847 1848 if (test_bit(__IGBVF_RESETTING, &adapter->state)) 1849 return; 1850 1851 if (pci_channel_offline(pdev)) 1852 return; 1853 1854 UPDATE_VF_COUNTER(VFGPRC, gprc); 1855 UPDATE_VF_COUNTER(VFGORC, gorc); 1856 UPDATE_VF_COUNTER(VFGPTC, gptc); 1857 UPDATE_VF_COUNTER(VFGOTC, gotc); 1858 UPDATE_VF_COUNTER(VFMPRC, mprc); 1859 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc); 1860 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc); 1861 UPDATE_VF_COUNTER(VFGORLBC, gorlbc); 1862 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc); 1863 1864 /* Fill out the OS statistics structure */ 1865 adapter->netdev->stats.multicast = adapter->stats.mprc; 1866 } 1867 1868 static void igbvf_print_link_info(struct igbvf_adapter *adapter) 1869 { 1870 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n", 1871 adapter->link_speed, 1872 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half"); 1873 } 1874 1875 static bool igbvf_has_link(struct igbvf_adapter *adapter) 1876 { 1877 struct e1000_hw *hw = &adapter->hw; 1878 s32 ret_val = E1000_SUCCESS; 1879 bool link_active; 1880 1881 /* If interface is down, stay link down */ 1882 if (test_bit(__IGBVF_DOWN, &adapter->state)) 1883 return false; 1884 1885 spin_lock_bh(&hw->mbx_lock); 1886 1887 ret_val = hw->mac.ops.check_for_link(hw); 1888 1889 spin_unlock_bh(&hw->mbx_lock); 1890 1891 link_active = !hw->mac.get_link_status; 1892 1893 /* if check for link returns error we will need to reset */ 1894 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ))) 1895 schedule_work(&adapter->reset_task); 1896 1897 return link_active; 1898 } 1899 1900 /** 1901 * igbvf_watchdog - Timer Call-back 1902 * @t: timer list pointer containing private struct 1903 **/ 1904 static void igbvf_watchdog(struct timer_list *t) 1905 { 1906 struct igbvf_adapter *adapter = from_timer(adapter, t, watchdog_timer); 1907 1908 /* Do the rest outside of interrupt context */ 1909 schedule_work(&adapter->watchdog_task); 1910 } 1911 1912 static void igbvf_watchdog_task(struct work_struct *work) 1913 { 1914 struct igbvf_adapter *adapter = container_of(work, 1915 struct igbvf_adapter, 1916 watchdog_task); 1917 struct net_device *netdev = adapter->netdev; 1918 struct e1000_mac_info *mac = &adapter->hw.mac; 1919 struct igbvf_ring *tx_ring = adapter->tx_ring; 1920 struct e1000_hw *hw = &adapter->hw; 1921 u32 link; 1922 int tx_pending = 0; 1923 1924 link = igbvf_has_link(adapter); 1925 1926 if (link) { 1927 if (!netif_carrier_ok(netdev)) { 1928 mac->ops.get_link_up_info(&adapter->hw, 1929 &adapter->link_speed, 1930 &adapter->link_duplex); 1931 igbvf_print_link_info(adapter); 1932 1933 netif_carrier_on(netdev); 1934 netif_wake_queue(netdev); 1935 } 1936 } else { 1937 if (netif_carrier_ok(netdev)) { 1938 adapter->link_speed = 0; 1939 adapter->link_duplex = 0; 1940 dev_info(&adapter->pdev->dev, "Link is Down\n"); 1941 netif_carrier_off(netdev); 1942 netif_stop_queue(netdev); 1943 } 1944 } 1945 1946 if (netif_carrier_ok(netdev)) { 1947 igbvf_update_stats(adapter); 1948 } else { 1949 tx_pending = (igbvf_desc_unused(tx_ring) + 1 < 1950 tx_ring->count); 1951 if (tx_pending) { 1952 /* We've lost link, so the controller stops DMA, 1953 * but we've got queued Tx work that's never going 1954 * to get done, so reset controller to flush Tx. 1955 * (Do the reset outside of interrupt context). 1956 */ 1957 adapter->tx_timeout_count++; 1958 schedule_work(&adapter->reset_task); 1959 } 1960 } 1961 1962 /* Cause software interrupt to ensure Rx ring is cleaned */ 1963 ew32(EICS, adapter->rx_ring->eims_value); 1964 1965 /* Reset the timer */ 1966 if (!test_bit(__IGBVF_DOWN, &adapter->state)) 1967 mod_timer(&adapter->watchdog_timer, 1968 round_jiffies(jiffies + (2 * HZ))); 1969 } 1970 1971 #define IGBVF_TX_FLAGS_CSUM 0x00000001 1972 #define IGBVF_TX_FLAGS_VLAN 0x00000002 1973 #define IGBVF_TX_FLAGS_TSO 0x00000004 1974 #define IGBVF_TX_FLAGS_IPV4 0x00000008 1975 #define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000 1976 #define IGBVF_TX_FLAGS_VLAN_SHIFT 16 1977 1978 static void igbvf_tx_ctxtdesc(struct igbvf_ring *tx_ring, u32 vlan_macip_lens, 1979 u32 type_tucmd, u32 mss_l4len_idx) 1980 { 1981 struct e1000_adv_tx_context_desc *context_desc; 1982 struct igbvf_buffer *buffer_info; 1983 u16 i = tx_ring->next_to_use; 1984 1985 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i); 1986 buffer_info = &tx_ring->buffer_info[i]; 1987 1988 i++; 1989 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; 1990 1991 /* set bits to identify this as an advanced context descriptor */ 1992 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT; 1993 1994 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens); 1995 context_desc->seqnum_seed = 0; 1996 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd); 1997 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx); 1998 1999 buffer_info->time_stamp = jiffies; 2000 buffer_info->dma = 0; 2001 } 2002 2003 static int igbvf_tso(struct igbvf_ring *tx_ring, 2004 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len) 2005 { 2006 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx; 2007 union { 2008 struct iphdr *v4; 2009 struct ipv6hdr *v6; 2010 unsigned char *hdr; 2011 } ip; 2012 union { 2013 struct tcphdr *tcp; 2014 unsigned char *hdr; 2015 } l4; 2016 u32 paylen, l4_offset; 2017 int err; 2018 2019 if (skb->ip_summed != CHECKSUM_PARTIAL) 2020 return 0; 2021 2022 if (!skb_is_gso(skb)) 2023 return 0; 2024 2025 err = skb_cow_head(skb, 0); 2026 if (err < 0) 2027 return err; 2028 2029 ip.hdr = skb_network_header(skb); 2030 l4.hdr = skb_checksum_start(skb); 2031 2032 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */ 2033 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP; 2034 2035 /* initialize outer IP header fields */ 2036 if (ip.v4->version == 4) { 2037 unsigned char *csum_start = skb_checksum_start(skb); 2038 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4); 2039 2040 /* IP header will have to cancel out any data that 2041 * is not a part of the outer IP header 2042 */ 2043 ip.v4->check = csum_fold(csum_partial(trans_start, 2044 csum_start - trans_start, 2045 0)); 2046 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4; 2047 2048 ip.v4->tot_len = 0; 2049 } else { 2050 ip.v6->payload_len = 0; 2051 } 2052 2053 /* determine offset of inner transport header */ 2054 l4_offset = l4.hdr - skb->data; 2055 2056 /* compute length of segmentation header */ 2057 *hdr_len = (l4.tcp->doff * 4) + l4_offset; 2058 2059 /* remove payload length from inner checksum */ 2060 paylen = skb->len - l4_offset; 2061 csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen)); 2062 2063 /* MSS L4LEN IDX */ 2064 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT; 2065 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT; 2066 2067 /* VLAN MACLEN IPLEN */ 2068 vlan_macip_lens = l4.hdr - ip.hdr; 2069 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT; 2070 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK; 2071 2072 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx); 2073 2074 return 1; 2075 } 2076 2077 static bool igbvf_tx_csum(struct igbvf_ring *tx_ring, struct sk_buff *skb, 2078 u32 tx_flags, __be16 protocol) 2079 { 2080 u32 vlan_macip_lens = 0; 2081 u32 type_tucmd = 0; 2082 2083 if (skb->ip_summed != CHECKSUM_PARTIAL) { 2084 csum_failed: 2085 if (!(tx_flags & IGBVF_TX_FLAGS_VLAN)) 2086 return false; 2087 goto no_csum; 2088 } 2089 2090 switch (skb->csum_offset) { 2091 case offsetof(struct tcphdr, check): 2092 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP; 2093 fallthrough; 2094 case offsetof(struct udphdr, check): 2095 break; 2096 case offsetof(struct sctphdr, checksum): 2097 /* validate that this is actually an SCTP request */ 2098 if (skb_csum_is_sctp(skb)) { 2099 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP; 2100 break; 2101 } 2102 fallthrough; 2103 default: 2104 skb_checksum_help(skb); 2105 goto csum_failed; 2106 } 2107 2108 vlan_macip_lens = skb_checksum_start_offset(skb) - 2109 skb_network_offset(skb); 2110 no_csum: 2111 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT; 2112 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK; 2113 2114 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0); 2115 return true; 2116 } 2117 2118 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size) 2119 { 2120 struct igbvf_adapter *adapter = netdev_priv(netdev); 2121 2122 /* there is enough descriptors then we don't need to worry */ 2123 if (igbvf_desc_unused(adapter->tx_ring) >= size) 2124 return 0; 2125 2126 netif_stop_queue(netdev); 2127 2128 /* Herbert's original patch had: 2129 * smp_mb__after_netif_stop_queue(); 2130 * but since that doesn't exist yet, just open code it. 2131 */ 2132 smp_mb(); 2133 2134 /* We need to check again just in case room has been made available */ 2135 if (igbvf_desc_unused(adapter->tx_ring) < size) 2136 return -EBUSY; 2137 2138 netif_wake_queue(netdev); 2139 2140 ++adapter->restart_queue; 2141 return 0; 2142 } 2143 2144 #define IGBVF_MAX_TXD_PWR 16 2145 #define IGBVF_MAX_DATA_PER_TXD (1u << IGBVF_MAX_TXD_PWR) 2146 2147 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter, 2148 struct igbvf_ring *tx_ring, 2149 struct sk_buff *skb) 2150 { 2151 struct igbvf_buffer *buffer_info; 2152 struct pci_dev *pdev = adapter->pdev; 2153 unsigned int len = skb_headlen(skb); 2154 unsigned int count = 0, i; 2155 unsigned int f; 2156 2157 i = tx_ring->next_to_use; 2158 2159 buffer_info = &tx_ring->buffer_info[i]; 2160 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD); 2161 buffer_info->length = len; 2162 /* set time_stamp *before* dma to help avoid a possible race */ 2163 buffer_info->time_stamp = jiffies; 2164 buffer_info->mapped_as_page = false; 2165 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len, 2166 DMA_TO_DEVICE); 2167 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) 2168 goto dma_error; 2169 2170 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) { 2171 const skb_frag_t *frag; 2172 2173 count++; 2174 i++; 2175 if (i == tx_ring->count) 2176 i = 0; 2177 2178 frag = &skb_shinfo(skb)->frags[f]; 2179 len = skb_frag_size(frag); 2180 2181 buffer_info = &tx_ring->buffer_info[i]; 2182 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD); 2183 buffer_info->length = len; 2184 buffer_info->time_stamp = jiffies; 2185 buffer_info->mapped_as_page = true; 2186 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len, 2187 DMA_TO_DEVICE); 2188 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) 2189 goto dma_error; 2190 } 2191 2192 tx_ring->buffer_info[i].skb = skb; 2193 2194 return ++count; 2195 2196 dma_error: 2197 dev_err(&pdev->dev, "TX DMA map failed\n"); 2198 2199 /* clear timestamp and dma mappings for failed buffer_info mapping */ 2200 buffer_info->dma = 0; 2201 buffer_info->time_stamp = 0; 2202 buffer_info->length = 0; 2203 buffer_info->mapped_as_page = false; 2204 if (count) 2205 count--; 2206 2207 /* clear timestamp and dma mappings for remaining portion of packet */ 2208 while (count--) { 2209 if (i == 0) 2210 i += tx_ring->count; 2211 i--; 2212 buffer_info = &tx_ring->buffer_info[i]; 2213 igbvf_put_txbuf(adapter, buffer_info); 2214 } 2215 2216 return 0; 2217 } 2218 2219 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter, 2220 struct igbvf_ring *tx_ring, 2221 int tx_flags, int count, 2222 unsigned int first, u32 paylen, 2223 u8 hdr_len) 2224 { 2225 union e1000_adv_tx_desc *tx_desc = NULL; 2226 struct igbvf_buffer *buffer_info; 2227 u32 olinfo_status = 0, cmd_type_len; 2228 unsigned int i; 2229 2230 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS | 2231 E1000_ADVTXD_DCMD_DEXT); 2232 2233 if (tx_flags & IGBVF_TX_FLAGS_VLAN) 2234 cmd_type_len |= E1000_ADVTXD_DCMD_VLE; 2235 2236 if (tx_flags & IGBVF_TX_FLAGS_TSO) { 2237 cmd_type_len |= E1000_ADVTXD_DCMD_TSE; 2238 2239 /* insert tcp checksum */ 2240 olinfo_status |= E1000_TXD_POPTS_TXSM << 8; 2241 2242 /* insert ip checksum */ 2243 if (tx_flags & IGBVF_TX_FLAGS_IPV4) 2244 olinfo_status |= E1000_TXD_POPTS_IXSM << 8; 2245 2246 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) { 2247 olinfo_status |= E1000_TXD_POPTS_TXSM << 8; 2248 } 2249 2250 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT); 2251 2252 i = tx_ring->next_to_use; 2253 while (count--) { 2254 buffer_info = &tx_ring->buffer_info[i]; 2255 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i); 2256 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma); 2257 tx_desc->read.cmd_type_len = 2258 cpu_to_le32(cmd_type_len | buffer_info->length); 2259 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status); 2260 i++; 2261 if (i == tx_ring->count) 2262 i = 0; 2263 } 2264 2265 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd); 2266 /* Force memory writes to complete before letting h/w 2267 * know there are new descriptors to fetch. (Only 2268 * applicable for weak-ordered memory model archs, 2269 * such as IA-64). 2270 */ 2271 wmb(); 2272 2273 tx_ring->buffer_info[first].next_to_watch = tx_desc; 2274 tx_ring->next_to_use = i; 2275 writel(i, adapter->hw.hw_addr + tx_ring->tail); 2276 } 2277 2278 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb, 2279 struct net_device *netdev, 2280 struct igbvf_ring *tx_ring) 2281 { 2282 struct igbvf_adapter *adapter = netdev_priv(netdev); 2283 unsigned int first, tx_flags = 0; 2284 u8 hdr_len = 0; 2285 int count = 0; 2286 int tso = 0; 2287 __be16 protocol = vlan_get_protocol(skb); 2288 2289 if (test_bit(__IGBVF_DOWN, &adapter->state)) { 2290 dev_kfree_skb_any(skb); 2291 return NETDEV_TX_OK; 2292 } 2293 2294 if (skb->len <= 0) { 2295 dev_kfree_skb_any(skb); 2296 return NETDEV_TX_OK; 2297 } 2298 2299 /* need: count + 4 desc gap to keep tail from touching 2300 * + 2 desc gap to keep tail from touching head, 2301 * + 1 desc for skb->data, 2302 * + 1 desc for context descriptor, 2303 * head, otherwise try next time 2304 */ 2305 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) { 2306 /* this is a hard error */ 2307 return NETDEV_TX_BUSY; 2308 } 2309 2310 if (skb_vlan_tag_present(skb)) { 2311 tx_flags |= IGBVF_TX_FLAGS_VLAN; 2312 tx_flags |= (skb_vlan_tag_get(skb) << 2313 IGBVF_TX_FLAGS_VLAN_SHIFT); 2314 } 2315 2316 if (protocol == htons(ETH_P_IP)) 2317 tx_flags |= IGBVF_TX_FLAGS_IPV4; 2318 2319 first = tx_ring->next_to_use; 2320 2321 tso = igbvf_tso(tx_ring, skb, tx_flags, &hdr_len); 2322 if (unlikely(tso < 0)) { 2323 dev_kfree_skb_any(skb); 2324 return NETDEV_TX_OK; 2325 } 2326 2327 if (tso) 2328 tx_flags |= IGBVF_TX_FLAGS_TSO; 2329 else if (igbvf_tx_csum(tx_ring, skb, tx_flags, protocol) && 2330 (skb->ip_summed == CHECKSUM_PARTIAL)) 2331 tx_flags |= IGBVF_TX_FLAGS_CSUM; 2332 2333 /* count reflects descriptors mapped, if 0 then mapping error 2334 * has occurred and we need to rewind the descriptor queue 2335 */ 2336 count = igbvf_tx_map_adv(adapter, tx_ring, skb); 2337 2338 if (count) { 2339 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count, 2340 first, skb->len, hdr_len); 2341 /* Make sure there is space in the ring for the next send. */ 2342 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4); 2343 } else { 2344 dev_kfree_skb_any(skb); 2345 tx_ring->buffer_info[first].time_stamp = 0; 2346 tx_ring->next_to_use = first; 2347 } 2348 2349 return NETDEV_TX_OK; 2350 } 2351 2352 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb, 2353 struct net_device *netdev) 2354 { 2355 struct igbvf_adapter *adapter = netdev_priv(netdev); 2356 struct igbvf_ring *tx_ring; 2357 2358 if (test_bit(__IGBVF_DOWN, &adapter->state)) { 2359 dev_kfree_skb_any(skb); 2360 return NETDEV_TX_OK; 2361 } 2362 2363 tx_ring = &adapter->tx_ring[0]; 2364 2365 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring); 2366 } 2367 2368 /** 2369 * igbvf_tx_timeout - Respond to a Tx Hang 2370 * @netdev: network interface device structure 2371 * @txqueue: queue timing out (unused) 2372 **/ 2373 static void igbvf_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue) 2374 { 2375 struct igbvf_adapter *adapter = netdev_priv(netdev); 2376 2377 /* Do the reset outside of interrupt context */ 2378 adapter->tx_timeout_count++; 2379 schedule_work(&adapter->reset_task); 2380 } 2381 2382 static void igbvf_reset_task(struct work_struct *work) 2383 { 2384 struct igbvf_adapter *adapter; 2385 2386 adapter = container_of(work, struct igbvf_adapter, reset_task); 2387 2388 igbvf_reinit_locked(adapter); 2389 } 2390 2391 /** 2392 * igbvf_change_mtu - Change the Maximum Transfer Unit 2393 * @netdev: network interface device structure 2394 * @new_mtu: new value for maximum frame size 2395 * 2396 * Returns 0 on success, negative on failure 2397 **/ 2398 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu) 2399 { 2400 struct igbvf_adapter *adapter = netdev_priv(netdev); 2401 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN; 2402 2403 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state)) 2404 usleep_range(1000, 2000); 2405 /* igbvf_down has a dependency on max_frame_size */ 2406 adapter->max_frame_size = max_frame; 2407 if (netif_running(netdev)) 2408 igbvf_down(adapter); 2409 2410 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN 2411 * means we reserve 2 more, this pushes us to allocate from the next 2412 * larger slab size. 2413 * i.e. RXBUFFER_2048 --> size-4096 slab 2414 * However with the new *_jumbo_rx* routines, jumbo receives will use 2415 * fragmented skbs 2416 */ 2417 2418 if (max_frame <= 1024) 2419 adapter->rx_buffer_len = 1024; 2420 else if (max_frame <= 2048) 2421 adapter->rx_buffer_len = 2048; 2422 else 2423 #if (PAGE_SIZE / 2) > 16384 2424 adapter->rx_buffer_len = 16384; 2425 #else 2426 adapter->rx_buffer_len = PAGE_SIZE / 2; 2427 #endif 2428 2429 /* adjust allocation if LPE protects us, and we aren't using SBP */ 2430 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) || 2431 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN)) 2432 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + 2433 ETH_FCS_LEN; 2434 2435 netdev_dbg(netdev, "changing MTU from %d to %d\n", 2436 netdev->mtu, new_mtu); 2437 WRITE_ONCE(netdev->mtu, new_mtu); 2438 2439 if (netif_running(netdev)) 2440 igbvf_up(adapter); 2441 else 2442 igbvf_reset(adapter); 2443 2444 clear_bit(__IGBVF_RESETTING, &adapter->state); 2445 2446 return 0; 2447 } 2448 2449 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd) 2450 { 2451 switch (cmd) { 2452 default: 2453 return -EOPNOTSUPP; 2454 } 2455 } 2456 2457 static int igbvf_suspend(struct device *dev_d) 2458 { 2459 struct net_device *netdev = dev_get_drvdata(dev_d); 2460 struct igbvf_adapter *adapter = netdev_priv(netdev); 2461 2462 netif_device_detach(netdev); 2463 2464 if (netif_running(netdev)) { 2465 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state)); 2466 igbvf_down(adapter); 2467 igbvf_free_irq(adapter); 2468 } 2469 2470 return 0; 2471 } 2472 2473 static int igbvf_resume(struct device *dev_d) 2474 { 2475 struct pci_dev *pdev = to_pci_dev(dev_d); 2476 struct net_device *netdev = pci_get_drvdata(pdev); 2477 struct igbvf_adapter *adapter = netdev_priv(netdev); 2478 u32 err; 2479 2480 pci_set_master(pdev); 2481 2482 if (netif_running(netdev)) { 2483 err = igbvf_request_irq(adapter); 2484 if (err) 2485 return err; 2486 } 2487 2488 igbvf_reset(adapter); 2489 2490 if (netif_running(netdev)) 2491 igbvf_up(adapter); 2492 2493 netif_device_attach(netdev); 2494 2495 return 0; 2496 } 2497 2498 static void igbvf_shutdown(struct pci_dev *pdev) 2499 { 2500 igbvf_suspend(&pdev->dev); 2501 } 2502 2503 #ifdef CONFIG_NET_POLL_CONTROLLER 2504 /* Polling 'interrupt' - used by things like netconsole to send skbs 2505 * without having to re-enable interrupts. It's not called while 2506 * the interrupt routine is executing. 2507 */ 2508 static void igbvf_netpoll(struct net_device *netdev) 2509 { 2510 struct igbvf_adapter *adapter = netdev_priv(netdev); 2511 2512 disable_irq(adapter->pdev->irq); 2513 2514 igbvf_clean_tx_irq(adapter->tx_ring); 2515 2516 enable_irq(adapter->pdev->irq); 2517 } 2518 #endif 2519 2520 /** 2521 * igbvf_io_error_detected - called when PCI error is detected 2522 * @pdev: Pointer to PCI device 2523 * @state: The current pci connection state 2524 * 2525 * This function is called after a PCI bus error affecting 2526 * this device has been detected. 2527 */ 2528 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev, 2529 pci_channel_state_t state) 2530 { 2531 struct net_device *netdev = pci_get_drvdata(pdev); 2532 struct igbvf_adapter *adapter = netdev_priv(netdev); 2533 2534 netif_device_detach(netdev); 2535 2536 if (state == pci_channel_io_perm_failure) 2537 return PCI_ERS_RESULT_DISCONNECT; 2538 2539 if (netif_running(netdev)) 2540 igbvf_down(adapter); 2541 pci_disable_device(pdev); 2542 2543 /* Request a slot reset. */ 2544 return PCI_ERS_RESULT_NEED_RESET; 2545 } 2546 2547 /** 2548 * igbvf_io_slot_reset - called after the pci bus has been reset. 2549 * @pdev: Pointer to PCI device 2550 * 2551 * Restart the card from scratch, as if from a cold-boot. Implementation 2552 * resembles the first-half of the igbvf_resume routine. 2553 */ 2554 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev) 2555 { 2556 struct net_device *netdev = pci_get_drvdata(pdev); 2557 struct igbvf_adapter *adapter = netdev_priv(netdev); 2558 2559 if (pci_enable_device_mem(pdev)) { 2560 dev_err(&pdev->dev, 2561 "Cannot re-enable PCI device after reset.\n"); 2562 return PCI_ERS_RESULT_DISCONNECT; 2563 } 2564 pci_set_master(pdev); 2565 2566 igbvf_reset(adapter); 2567 2568 return PCI_ERS_RESULT_RECOVERED; 2569 } 2570 2571 /** 2572 * igbvf_io_resume - called when traffic can start flowing again. 2573 * @pdev: Pointer to PCI device 2574 * 2575 * This callback is called when the error recovery driver tells us that 2576 * its OK to resume normal operation. Implementation resembles the 2577 * second-half of the igbvf_resume routine. 2578 */ 2579 static void igbvf_io_resume(struct pci_dev *pdev) 2580 { 2581 struct net_device *netdev = pci_get_drvdata(pdev); 2582 struct igbvf_adapter *adapter = netdev_priv(netdev); 2583 2584 if (netif_running(netdev)) { 2585 if (igbvf_up(adapter)) { 2586 dev_err(&pdev->dev, 2587 "can't bring device back up after reset\n"); 2588 return; 2589 } 2590 } 2591 2592 netif_device_attach(netdev); 2593 } 2594 2595 /** 2596 * igbvf_io_prepare - prepare device driver for PCI reset 2597 * @pdev: PCI device information struct 2598 */ 2599 static void igbvf_io_prepare(struct pci_dev *pdev) 2600 { 2601 struct net_device *netdev = pci_get_drvdata(pdev); 2602 struct igbvf_adapter *adapter = netdev_priv(netdev); 2603 2604 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state)) 2605 usleep_range(1000, 2000); 2606 igbvf_down(adapter); 2607 } 2608 2609 /** 2610 * igbvf_io_reset_done - PCI reset done, device driver reset can begin 2611 * @pdev: PCI device information struct 2612 */ 2613 static void igbvf_io_reset_done(struct pci_dev *pdev) 2614 { 2615 struct net_device *netdev = pci_get_drvdata(pdev); 2616 struct igbvf_adapter *adapter = netdev_priv(netdev); 2617 2618 igbvf_up(adapter); 2619 clear_bit(__IGBVF_RESETTING, &adapter->state); 2620 } 2621 2622 static void igbvf_print_device_info(struct igbvf_adapter *adapter) 2623 { 2624 struct e1000_hw *hw = &adapter->hw; 2625 struct net_device *netdev = adapter->netdev; 2626 struct pci_dev *pdev = adapter->pdev; 2627 2628 if (hw->mac.type == e1000_vfadapt_i350) 2629 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n"); 2630 else 2631 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n"); 2632 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr); 2633 } 2634 2635 static int igbvf_set_features(struct net_device *netdev, 2636 netdev_features_t features) 2637 { 2638 struct igbvf_adapter *adapter = netdev_priv(netdev); 2639 2640 if (features & NETIF_F_RXCSUM) 2641 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED; 2642 else 2643 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED; 2644 2645 return 0; 2646 } 2647 2648 #define IGBVF_MAX_MAC_HDR_LEN 127 2649 #define IGBVF_MAX_NETWORK_HDR_LEN 511 2650 2651 static netdev_features_t 2652 igbvf_features_check(struct sk_buff *skb, struct net_device *dev, 2653 netdev_features_t features) 2654 { 2655 unsigned int network_hdr_len, mac_hdr_len; 2656 2657 /* Make certain the headers can be described by a context descriptor */ 2658 mac_hdr_len = skb_network_offset(skb); 2659 if (unlikely(mac_hdr_len > IGBVF_MAX_MAC_HDR_LEN)) 2660 return features & ~(NETIF_F_HW_CSUM | 2661 NETIF_F_SCTP_CRC | 2662 NETIF_F_HW_VLAN_CTAG_TX | 2663 NETIF_F_TSO | 2664 NETIF_F_TSO6); 2665 2666 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb); 2667 if (unlikely(network_hdr_len > IGBVF_MAX_NETWORK_HDR_LEN)) 2668 return features & ~(NETIF_F_HW_CSUM | 2669 NETIF_F_SCTP_CRC | 2670 NETIF_F_TSO | 2671 NETIF_F_TSO6); 2672 2673 /* We can only support IPV4 TSO in tunnels if we can mangle the 2674 * inner IP ID field, so strip TSO if MANGLEID is not supported. 2675 */ 2676 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) 2677 features &= ~NETIF_F_TSO; 2678 2679 return features; 2680 } 2681 2682 static const struct net_device_ops igbvf_netdev_ops = { 2683 .ndo_open = igbvf_open, 2684 .ndo_stop = igbvf_close, 2685 .ndo_start_xmit = igbvf_xmit_frame, 2686 .ndo_set_rx_mode = igbvf_set_rx_mode, 2687 .ndo_set_mac_address = igbvf_set_mac, 2688 .ndo_change_mtu = igbvf_change_mtu, 2689 .ndo_eth_ioctl = igbvf_ioctl, 2690 .ndo_tx_timeout = igbvf_tx_timeout, 2691 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid, 2692 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid, 2693 #ifdef CONFIG_NET_POLL_CONTROLLER 2694 .ndo_poll_controller = igbvf_netpoll, 2695 #endif 2696 .ndo_set_features = igbvf_set_features, 2697 .ndo_features_check = igbvf_features_check, 2698 }; 2699 2700 /** 2701 * igbvf_probe - Device Initialization Routine 2702 * @pdev: PCI device information struct 2703 * @ent: entry in igbvf_pci_tbl 2704 * 2705 * Returns 0 on success, negative on failure 2706 * 2707 * igbvf_probe initializes an adapter identified by a pci_dev structure. 2708 * The OS initialization, configuring of the adapter private structure, 2709 * and a hardware reset occur. 2710 **/ 2711 static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 2712 { 2713 struct net_device *netdev; 2714 struct igbvf_adapter *adapter; 2715 struct e1000_hw *hw; 2716 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data]; 2717 static int cards_found; 2718 int err; 2719 2720 err = pci_enable_device_mem(pdev); 2721 if (err) 2722 return err; 2723 2724 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); 2725 if (err) { 2726 dev_err(&pdev->dev, 2727 "No usable DMA configuration, aborting\n"); 2728 goto err_dma; 2729 } 2730 2731 err = pci_request_regions(pdev, igbvf_driver_name); 2732 if (err) 2733 goto err_pci_reg; 2734 2735 pci_set_master(pdev); 2736 2737 err = -ENOMEM; 2738 netdev = alloc_etherdev(sizeof(struct igbvf_adapter)); 2739 if (!netdev) 2740 goto err_alloc_etherdev; 2741 2742 SET_NETDEV_DEV(netdev, &pdev->dev); 2743 2744 pci_set_drvdata(pdev, netdev); 2745 adapter = netdev_priv(netdev); 2746 hw = &adapter->hw; 2747 adapter->netdev = netdev; 2748 adapter->pdev = pdev; 2749 adapter->ei = ei; 2750 adapter->pba = ei->pba; 2751 adapter->flags = ei->flags; 2752 adapter->hw.back = adapter; 2753 adapter->hw.mac.type = ei->mac; 2754 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE); 2755 2756 /* PCI config space info */ 2757 2758 hw->vendor_id = pdev->vendor; 2759 hw->device_id = pdev->device; 2760 hw->subsystem_vendor_id = pdev->subsystem_vendor; 2761 hw->subsystem_device_id = pdev->subsystem_device; 2762 hw->revision_id = pdev->revision; 2763 2764 err = -EIO; 2765 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0), 2766 pci_resource_len(pdev, 0)); 2767 2768 if (!adapter->hw.hw_addr) 2769 goto err_ioremap; 2770 2771 if (ei->get_variants) { 2772 err = ei->get_variants(adapter); 2773 if (err) 2774 goto err_get_variants; 2775 } 2776 2777 /* setup adapter struct */ 2778 err = igbvf_sw_init(adapter); 2779 if (err) 2780 goto err_sw_init; 2781 2782 /* construct the net_device struct */ 2783 netdev->netdev_ops = &igbvf_netdev_ops; 2784 2785 igbvf_set_ethtool_ops(netdev); 2786 netdev->watchdog_timeo = 5 * HZ; 2787 strscpy(netdev->name, pci_name(pdev), sizeof(netdev->name)); 2788 2789 adapter->bd_number = cards_found++; 2790 2791 netdev->hw_features = NETIF_F_SG | 2792 NETIF_F_TSO | 2793 NETIF_F_TSO6 | 2794 NETIF_F_RXCSUM | 2795 NETIF_F_HW_CSUM | 2796 NETIF_F_SCTP_CRC; 2797 2798 #define IGBVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \ 2799 NETIF_F_GSO_GRE_CSUM | \ 2800 NETIF_F_GSO_IPXIP4 | \ 2801 NETIF_F_GSO_IPXIP6 | \ 2802 NETIF_F_GSO_UDP_TUNNEL | \ 2803 NETIF_F_GSO_UDP_TUNNEL_CSUM) 2804 2805 netdev->gso_partial_features = IGBVF_GSO_PARTIAL_FEATURES; 2806 netdev->hw_features |= NETIF_F_GSO_PARTIAL | 2807 IGBVF_GSO_PARTIAL_FEATURES; 2808 2809 netdev->features = netdev->hw_features | NETIF_F_HIGHDMA; 2810 2811 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID; 2812 netdev->mpls_features |= NETIF_F_HW_CSUM; 2813 netdev->hw_enc_features |= netdev->vlan_features; 2814 2815 /* set this bit last since it cannot be part of vlan_features */ 2816 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER | 2817 NETIF_F_HW_VLAN_CTAG_RX | 2818 NETIF_F_HW_VLAN_CTAG_TX; 2819 2820 /* MTU range: 68 - 9216 */ 2821 netdev->min_mtu = ETH_MIN_MTU; 2822 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE; 2823 2824 spin_lock_bh(&hw->mbx_lock); 2825 2826 /*reset the controller to put the device in a known good state */ 2827 err = hw->mac.ops.reset_hw(hw); 2828 if (err) { 2829 dev_info(&pdev->dev, 2830 "PF still in reset state. Is the PF interface up?\n"); 2831 } else { 2832 err = hw->mac.ops.read_mac_addr(hw); 2833 if (err) 2834 dev_info(&pdev->dev, "Error reading MAC address.\n"); 2835 else if (is_zero_ether_addr(adapter->hw.mac.addr)) 2836 dev_info(&pdev->dev, 2837 "MAC address not assigned by administrator.\n"); 2838 eth_hw_addr_set(netdev, adapter->hw.mac.addr); 2839 } 2840 2841 spin_unlock_bh(&hw->mbx_lock); 2842 2843 if (!is_valid_ether_addr(netdev->dev_addr)) { 2844 dev_info(&pdev->dev, "Assigning random MAC address.\n"); 2845 eth_hw_addr_random(netdev); 2846 memcpy(adapter->hw.mac.addr, netdev->dev_addr, 2847 netdev->addr_len); 2848 } 2849 2850 timer_setup(&adapter->watchdog_timer, igbvf_watchdog, 0); 2851 2852 INIT_WORK(&adapter->reset_task, igbvf_reset_task); 2853 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task); 2854 2855 /* ring size defaults */ 2856 adapter->rx_ring->count = 1024; 2857 adapter->tx_ring->count = 1024; 2858 2859 /* reset the hardware with the new settings */ 2860 igbvf_reset(adapter); 2861 2862 /* set hardware-specific flags */ 2863 if (adapter->hw.mac.type == e1000_vfadapt_i350) 2864 adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP; 2865 2866 strcpy(netdev->name, "eth%d"); 2867 err = register_netdev(netdev); 2868 if (err) 2869 goto err_hw_init; 2870 2871 /* tell the stack to leave us alone until igbvf_open() is called */ 2872 netif_carrier_off(netdev); 2873 netif_stop_queue(netdev); 2874 2875 igbvf_print_device_info(adapter); 2876 2877 igbvf_initialize_last_counter_stats(adapter); 2878 2879 return 0; 2880 2881 err_hw_init: 2882 netif_napi_del(&adapter->rx_ring->napi); 2883 kfree(adapter->tx_ring); 2884 kfree(adapter->rx_ring); 2885 err_sw_init: 2886 igbvf_reset_interrupt_capability(adapter); 2887 err_get_variants: 2888 iounmap(adapter->hw.hw_addr); 2889 err_ioremap: 2890 free_netdev(netdev); 2891 err_alloc_etherdev: 2892 pci_release_regions(pdev); 2893 err_pci_reg: 2894 err_dma: 2895 pci_disable_device(pdev); 2896 return err; 2897 } 2898 2899 /** 2900 * igbvf_remove - Device Removal Routine 2901 * @pdev: PCI device information struct 2902 * 2903 * igbvf_remove is called by the PCI subsystem to alert the driver 2904 * that it should release a PCI device. The could be caused by a 2905 * Hot-Plug event, or because the driver is going to be removed from 2906 * memory. 2907 **/ 2908 static void igbvf_remove(struct pci_dev *pdev) 2909 { 2910 struct net_device *netdev = pci_get_drvdata(pdev); 2911 struct igbvf_adapter *adapter = netdev_priv(netdev); 2912 struct e1000_hw *hw = &adapter->hw; 2913 2914 /* The watchdog timer may be rescheduled, so explicitly 2915 * disable it from being rescheduled. 2916 */ 2917 set_bit(__IGBVF_DOWN, &adapter->state); 2918 del_timer_sync(&adapter->watchdog_timer); 2919 2920 cancel_work_sync(&adapter->reset_task); 2921 cancel_work_sync(&adapter->watchdog_task); 2922 2923 unregister_netdev(netdev); 2924 2925 igbvf_reset_interrupt_capability(adapter); 2926 2927 /* it is important to delete the NAPI struct prior to freeing the 2928 * Rx ring so that you do not end up with null pointer refs 2929 */ 2930 netif_napi_del(&adapter->rx_ring->napi); 2931 kfree(adapter->tx_ring); 2932 kfree(adapter->rx_ring); 2933 2934 iounmap(hw->hw_addr); 2935 if (hw->flash_address) 2936 iounmap(hw->flash_address); 2937 pci_release_regions(pdev); 2938 2939 free_netdev(netdev); 2940 2941 pci_disable_device(pdev); 2942 } 2943 2944 /* PCI Error Recovery (ERS) */ 2945 static const struct pci_error_handlers igbvf_err_handler = { 2946 .error_detected = igbvf_io_error_detected, 2947 .slot_reset = igbvf_io_slot_reset, 2948 .resume = igbvf_io_resume, 2949 .reset_prepare = igbvf_io_prepare, 2950 .reset_done = igbvf_io_reset_done, 2951 }; 2952 2953 static const struct pci_device_id igbvf_pci_tbl[] = { 2954 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf }, 2955 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf }, 2956 { } /* terminate list */ 2957 }; 2958 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl); 2959 2960 static DEFINE_SIMPLE_DEV_PM_OPS(igbvf_pm_ops, igbvf_suspend, igbvf_resume); 2961 2962 /* PCI Device API Driver */ 2963 static struct pci_driver igbvf_driver = { 2964 .name = igbvf_driver_name, 2965 .id_table = igbvf_pci_tbl, 2966 .probe = igbvf_probe, 2967 .remove = igbvf_remove, 2968 .driver.pm = pm_sleep_ptr(&igbvf_pm_ops), 2969 .shutdown = igbvf_shutdown, 2970 .err_handler = &igbvf_err_handler 2971 }; 2972 2973 /** 2974 * igbvf_init_module - Driver Registration Routine 2975 * 2976 * igbvf_init_module is the first routine called when the driver is 2977 * loaded. All it does is register with the PCI subsystem. 2978 **/ 2979 static int __init igbvf_init_module(void) 2980 { 2981 int ret; 2982 2983 pr_info("%s\n", igbvf_driver_string); 2984 pr_info("%s\n", igbvf_copyright); 2985 2986 ret = pci_register_driver(&igbvf_driver); 2987 2988 return ret; 2989 } 2990 module_init(igbvf_init_module); 2991 2992 /** 2993 * igbvf_exit_module - Driver Exit Cleanup Routine 2994 * 2995 * igbvf_exit_module is called just before the driver is removed 2996 * from memory. 2997 **/ 2998 static void __exit igbvf_exit_module(void) 2999 { 3000 pci_unregister_driver(&igbvf_driver); 3001 } 3002 module_exit(igbvf_exit_module); 3003 3004 MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver"); 3005 MODULE_LICENSE("GPL v2"); 3006 3007 /* netdev.c */ 3008