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/module.h> 7 #include <linux/types.h> 8 #include <linux/init.h> 9 #include <linux/pci.h> 10 #include <linux/vmalloc.h> 11 #include <linux/pagemap.h> 12 #include <linux/delay.h> 13 #include <linux/netdevice.h> 14 #include <linux/tcp.h> 15 #include <linux/ipv6.h> 16 #include <linux/slab.h> 17 #include <net/checksum.h> 18 #include <net/ip6_checksum.h> 19 #include <linux/mii.h> 20 #include <linux/ethtool.h> 21 #include <linux/if_vlan.h> 22 #include <linux/prefetch.h> 23 #include <linux/sctp.h> 24 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_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) 277 & E1000_RXDADV_HDRBUFLEN_MASK) >> 278 E1000_RXDADV_HDRBUFLEN_SHIFT; 279 if (hlen > adapter->rx_ps_hdr_size) 280 hlen = adapter->rx_ps_hdr_size; 281 282 length = le16_to_cpu(rx_desc->wb.upper.length); 283 cleaned = true; 284 cleaned_count++; 285 286 skb = buffer_info->skb; 287 prefetch(skb->data - NET_IP_ALIGN); 288 buffer_info->skb = NULL; 289 if (!adapter->rx_ps_hdr_size) { 290 dma_unmap_single(&pdev->dev, buffer_info->dma, 291 adapter->rx_buffer_len, 292 DMA_FROM_DEVICE); 293 buffer_info->dma = 0; 294 skb_put(skb, length); 295 goto send_up; 296 } 297 298 if (!skb_shinfo(skb)->nr_frags) { 299 dma_unmap_single(&pdev->dev, buffer_info->dma, 300 adapter->rx_ps_hdr_size, 301 DMA_FROM_DEVICE); 302 buffer_info->dma = 0; 303 skb_put(skb, hlen); 304 } 305 306 if (length) { 307 dma_unmap_page(&pdev->dev, buffer_info->page_dma, 308 PAGE_SIZE / 2, 309 DMA_FROM_DEVICE); 310 buffer_info->page_dma = 0; 311 312 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags, 313 buffer_info->page, 314 buffer_info->page_offset, 315 length); 316 317 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) || 318 (page_count(buffer_info->page) != 1)) 319 buffer_info->page = NULL; 320 else 321 get_page(buffer_info->page); 322 323 skb->len += length; 324 skb->data_len += length; 325 skb->truesize += PAGE_SIZE / 2; 326 } 327 send_up: 328 i++; 329 if (i == rx_ring->count) 330 i = 0; 331 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i); 332 prefetch(next_rxd); 333 next_buffer = &rx_ring->buffer_info[i]; 334 335 if (!(staterr & E1000_RXD_STAT_EOP)) { 336 buffer_info->skb = next_buffer->skb; 337 buffer_info->dma = next_buffer->dma; 338 next_buffer->skb = skb; 339 next_buffer->dma = 0; 340 goto next_desc; 341 } 342 343 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) { 344 dev_kfree_skb_irq(skb); 345 goto next_desc; 346 } 347 348 total_bytes += skb->len; 349 total_packets++; 350 351 igbvf_rx_checksum_adv(adapter, staterr, skb); 352 353 skb->protocol = eth_type_trans(skb, netdev); 354 355 igbvf_receive_skb(adapter, netdev, skb, staterr, 356 rx_desc->wb.upper.vlan); 357 358 next_desc: 359 rx_desc->wb.upper.status_error = 0; 360 361 /* return some buffers to hardware, one at a time is too slow */ 362 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) { 363 igbvf_alloc_rx_buffers(rx_ring, cleaned_count); 364 cleaned_count = 0; 365 } 366 367 /* use prefetched values */ 368 rx_desc = next_rxd; 369 buffer_info = next_buffer; 370 371 staterr = le32_to_cpu(rx_desc->wb.upper.status_error); 372 } 373 374 rx_ring->next_to_clean = i; 375 cleaned_count = igbvf_desc_unused(rx_ring); 376 377 if (cleaned_count) 378 igbvf_alloc_rx_buffers(rx_ring, cleaned_count); 379 380 adapter->total_rx_packets += total_packets; 381 adapter->total_rx_bytes += total_bytes; 382 netdev->stats.rx_bytes += total_bytes; 383 netdev->stats.rx_packets += total_packets; 384 return cleaned; 385 } 386 387 static void igbvf_put_txbuf(struct igbvf_adapter *adapter, 388 struct igbvf_buffer *buffer_info) 389 { 390 if (buffer_info->dma) { 391 if (buffer_info->mapped_as_page) 392 dma_unmap_page(&adapter->pdev->dev, 393 buffer_info->dma, 394 buffer_info->length, 395 DMA_TO_DEVICE); 396 else 397 dma_unmap_single(&adapter->pdev->dev, 398 buffer_info->dma, 399 buffer_info->length, 400 DMA_TO_DEVICE); 401 buffer_info->dma = 0; 402 } 403 if (buffer_info->skb) { 404 dev_kfree_skb_any(buffer_info->skb); 405 buffer_info->skb = NULL; 406 } 407 buffer_info->time_stamp = 0; 408 } 409 410 /** 411 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors) 412 * @adapter: board private structure 413 * @tx_ring: ring being initialized 414 * 415 * Return 0 on success, negative on failure 416 **/ 417 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter, 418 struct igbvf_ring *tx_ring) 419 { 420 struct pci_dev *pdev = adapter->pdev; 421 int size; 422 423 size = sizeof(struct igbvf_buffer) * tx_ring->count; 424 tx_ring->buffer_info = vzalloc(size); 425 if (!tx_ring->buffer_info) 426 goto err; 427 428 /* round up to nearest 4K */ 429 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc); 430 tx_ring->size = ALIGN(tx_ring->size, 4096); 431 432 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size, 433 &tx_ring->dma, GFP_KERNEL); 434 if (!tx_ring->desc) 435 goto err; 436 437 tx_ring->adapter = adapter; 438 tx_ring->next_to_use = 0; 439 tx_ring->next_to_clean = 0; 440 441 return 0; 442 err: 443 vfree(tx_ring->buffer_info); 444 dev_err(&adapter->pdev->dev, 445 "Unable to allocate memory for the transmit descriptor ring\n"); 446 return -ENOMEM; 447 } 448 449 /** 450 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors) 451 * @adapter: board private structure 452 * @rx_ring: ring being initialized 453 * 454 * Returns 0 on success, negative on failure 455 **/ 456 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter, 457 struct igbvf_ring *rx_ring) 458 { 459 struct pci_dev *pdev = adapter->pdev; 460 int size, desc_len; 461 462 size = sizeof(struct igbvf_buffer) * rx_ring->count; 463 rx_ring->buffer_info = vzalloc(size); 464 if (!rx_ring->buffer_info) 465 goto err; 466 467 desc_len = sizeof(union e1000_adv_rx_desc); 468 469 /* Round up to nearest 4K */ 470 rx_ring->size = rx_ring->count * desc_len; 471 rx_ring->size = ALIGN(rx_ring->size, 4096); 472 473 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size, 474 &rx_ring->dma, GFP_KERNEL); 475 if (!rx_ring->desc) 476 goto err; 477 478 rx_ring->next_to_clean = 0; 479 rx_ring->next_to_use = 0; 480 481 rx_ring->adapter = adapter; 482 483 return 0; 484 485 err: 486 vfree(rx_ring->buffer_info); 487 rx_ring->buffer_info = NULL; 488 dev_err(&adapter->pdev->dev, 489 "Unable to allocate memory for the receive descriptor ring\n"); 490 return -ENOMEM; 491 } 492 493 /** 494 * igbvf_clean_tx_ring - Free Tx Buffers 495 * @tx_ring: ring to be cleaned 496 **/ 497 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring) 498 { 499 struct igbvf_adapter *adapter = tx_ring->adapter; 500 struct igbvf_buffer *buffer_info; 501 unsigned long size; 502 unsigned int i; 503 504 if (!tx_ring->buffer_info) 505 return; 506 507 /* Free all the Tx ring sk_buffs */ 508 for (i = 0; i < tx_ring->count; i++) { 509 buffer_info = &tx_ring->buffer_info[i]; 510 igbvf_put_txbuf(adapter, buffer_info); 511 } 512 513 size = sizeof(struct igbvf_buffer) * tx_ring->count; 514 memset(tx_ring->buffer_info, 0, size); 515 516 /* Zero out the descriptor ring */ 517 memset(tx_ring->desc, 0, tx_ring->size); 518 519 tx_ring->next_to_use = 0; 520 tx_ring->next_to_clean = 0; 521 522 writel(0, adapter->hw.hw_addr + tx_ring->head); 523 writel(0, adapter->hw.hw_addr + tx_ring->tail); 524 } 525 526 /** 527 * igbvf_free_tx_resources - Free Tx Resources per Queue 528 * @tx_ring: ring to free resources from 529 * 530 * Free all transmit software resources 531 **/ 532 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring) 533 { 534 struct pci_dev *pdev = tx_ring->adapter->pdev; 535 536 igbvf_clean_tx_ring(tx_ring); 537 538 vfree(tx_ring->buffer_info); 539 tx_ring->buffer_info = NULL; 540 541 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc, 542 tx_ring->dma); 543 544 tx_ring->desc = NULL; 545 } 546 547 /** 548 * igbvf_clean_rx_ring - Free Rx Buffers per Queue 549 * @rx_ring: ring structure pointer to free buffers from 550 **/ 551 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring) 552 { 553 struct igbvf_adapter *adapter = rx_ring->adapter; 554 struct igbvf_buffer *buffer_info; 555 struct pci_dev *pdev = adapter->pdev; 556 unsigned long size; 557 unsigned int i; 558 559 if (!rx_ring->buffer_info) 560 return; 561 562 /* Free all the Rx ring sk_buffs */ 563 for (i = 0; i < rx_ring->count; i++) { 564 buffer_info = &rx_ring->buffer_info[i]; 565 if (buffer_info->dma) { 566 if (adapter->rx_ps_hdr_size) { 567 dma_unmap_single(&pdev->dev, buffer_info->dma, 568 adapter->rx_ps_hdr_size, 569 DMA_FROM_DEVICE); 570 } else { 571 dma_unmap_single(&pdev->dev, buffer_info->dma, 572 adapter->rx_buffer_len, 573 DMA_FROM_DEVICE); 574 } 575 buffer_info->dma = 0; 576 } 577 578 if (buffer_info->skb) { 579 dev_kfree_skb(buffer_info->skb); 580 buffer_info->skb = NULL; 581 } 582 583 if (buffer_info->page) { 584 if (buffer_info->page_dma) 585 dma_unmap_page(&pdev->dev, 586 buffer_info->page_dma, 587 PAGE_SIZE / 2, 588 DMA_FROM_DEVICE); 589 put_page(buffer_info->page); 590 buffer_info->page = NULL; 591 buffer_info->page_dma = 0; 592 buffer_info->page_offset = 0; 593 } 594 } 595 596 size = sizeof(struct igbvf_buffer) * rx_ring->count; 597 memset(rx_ring->buffer_info, 0, size); 598 599 /* Zero out the descriptor ring */ 600 memset(rx_ring->desc, 0, rx_ring->size); 601 602 rx_ring->next_to_clean = 0; 603 rx_ring->next_to_use = 0; 604 605 writel(0, adapter->hw.hw_addr + rx_ring->head); 606 writel(0, adapter->hw.hw_addr + rx_ring->tail); 607 } 608 609 /** 610 * igbvf_free_rx_resources - Free Rx Resources 611 * @rx_ring: ring to clean the resources from 612 * 613 * Free all receive software resources 614 **/ 615 616 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring) 617 { 618 struct pci_dev *pdev = rx_ring->adapter->pdev; 619 620 igbvf_clean_rx_ring(rx_ring); 621 622 vfree(rx_ring->buffer_info); 623 rx_ring->buffer_info = NULL; 624 625 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc, 626 rx_ring->dma); 627 rx_ring->desc = NULL; 628 } 629 630 /** 631 * igbvf_update_itr - update the dynamic ITR value based on statistics 632 * @adapter: pointer to adapter 633 * @itr_setting: current adapter->itr 634 * @packets: the number of packets during this measurement interval 635 * @bytes: the number of bytes during this measurement interval 636 * 637 * Stores a new ITR value based on packets and byte counts during the last 638 * interrupt. The advantage of per interrupt computation is faster updates 639 * and more accurate ITR for the current traffic pattern. Constants in this 640 * function were computed based on theoretical maximum wire speed and thresholds 641 * were set based on testing data as well as attempting to minimize response 642 * time while increasing bulk throughput. 643 **/ 644 static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter, 645 enum latency_range itr_setting, 646 int packets, int bytes) 647 { 648 enum latency_range retval = itr_setting; 649 650 if (packets == 0) 651 goto update_itr_done; 652 653 switch (itr_setting) { 654 case lowest_latency: 655 /* handle TSO and jumbo frames */ 656 if (bytes/packets > 8000) 657 retval = bulk_latency; 658 else if ((packets < 5) && (bytes > 512)) 659 retval = low_latency; 660 break; 661 case low_latency: /* 50 usec aka 20000 ints/s */ 662 if (bytes > 10000) { 663 /* this if handles the TSO accounting */ 664 if (bytes/packets > 8000) 665 retval = bulk_latency; 666 else if ((packets < 10) || ((bytes/packets) > 1200)) 667 retval = bulk_latency; 668 else if ((packets > 35)) 669 retval = lowest_latency; 670 } else if (bytes/packets > 2000) { 671 retval = bulk_latency; 672 } else if (packets <= 2 && bytes < 512) { 673 retval = lowest_latency; 674 } 675 break; 676 case bulk_latency: /* 250 usec aka 4000 ints/s */ 677 if (bytes > 25000) { 678 if (packets > 35) 679 retval = low_latency; 680 } else if (bytes < 6000) { 681 retval = low_latency; 682 } 683 break; 684 default: 685 break; 686 } 687 688 update_itr_done: 689 return retval; 690 } 691 692 static int igbvf_range_to_itr(enum latency_range current_range) 693 { 694 int new_itr; 695 696 switch (current_range) { 697 /* counts and packets in update_itr are dependent on these numbers */ 698 case lowest_latency: 699 new_itr = IGBVF_70K_ITR; 700 break; 701 case low_latency: 702 new_itr = IGBVF_20K_ITR; 703 break; 704 case bulk_latency: 705 new_itr = IGBVF_4K_ITR; 706 break; 707 default: 708 new_itr = IGBVF_START_ITR; 709 break; 710 } 711 return new_itr; 712 } 713 714 static void igbvf_set_itr(struct igbvf_adapter *adapter) 715 { 716 u32 new_itr; 717 718 adapter->tx_ring->itr_range = 719 igbvf_update_itr(adapter, 720 adapter->tx_ring->itr_val, 721 adapter->total_tx_packets, 722 adapter->total_tx_bytes); 723 724 /* conservative mode (itr 3) eliminates the lowest_latency setting */ 725 if (adapter->requested_itr == 3 && 726 adapter->tx_ring->itr_range == lowest_latency) 727 adapter->tx_ring->itr_range = low_latency; 728 729 new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range); 730 731 if (new_itr != adapter->tx_ring->itr_val) { 732 u32 current_itr = adapter->tx_ring->itr_val; 733 /* this attempts to bias the interrupt rate towards Bulk 734 * by adding intermediate steps when interrupt rate is 735 * increasing 736 */ 737 new_itr = new_itr > current_itr ? 738 min(current_itr + (new_itr >> 2), new_itr) : 739 new_itr; 740 adapter->tx_ring->itr_val = new_itr; 741 742 adapter->tx_ring->set_itr = 1; 743 } 744 745 adapter->rx_ring->itr_range = 746 igbvf_update_itr(adapter, adapter->rx_ring->itr_val, 747 adapter->total_rx_packets, 748 adapter->total_rx_bytes); 749 if (adapter->requested_itr == 3 && 750 adapter->rx_ring->itr_range == lowest_latency) 751 adapter->rx_ring->itr_range = low_latency; 752 753 new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range); 754 755 if (new_itr != adapter->rx_ring->itr_val) { 756 u32 current_itr = adapter->rx_ring->itr_val; 757 758 new_itr = new_itr > current_itr ? 759 min(current_itr + (new_itr >> 2), new_itr) : 760 new_itr; 761 adapter->rx_ring->itr_val = new_itr; 762 763 adapter->rx_ring->set_itr = 1; 764 } 765 } 766 767 /** 768 * igbvf_clean_tx_irq - Reclaim resources after transmit completes 769 * @tx_ring: ring structure to clean descriptors from 770 * 771 * returns true if ring is completely cleaned 772 **/ 773 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring) 774 { 775 struct igbvf_adapter *adapter = tx_ring->adapter; 776 struct net_device *netdev = adapter->netdev; 777 struct igbvf_buffer *buffer_info; 778 struct sk_buff *skb; 779 union e1000_adv_tx_desc *tx_desc, *eop_desc; 780 unsigned int total_bytes = 0, total_packets = 0; 781 unsigned int i, count = 0; 782 bool cleaned = false; 783 784 i = tx_ring->next_to_clean; 785 buffer_info = &tx_ring->buffer_info[i]; 786 eop_desc = buffer_info->next_to_watch; 787 788 do { 789 /* if next_to_watch is not set then there is no work pending */ 790 if (!eop_desc) 791 break; 792 793 /* prevent any other reads prior to eop_desc */ 794 smp_rmb(); 795 796 /* if DD is not set pending work has not been completed */ 797 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD))) 798 break; 799 800 /* clear next_to_watch to prevent false hangs */ 801 buffer_info->next_to_watch = NULL; 802 803 for (cleaned = false; !cleaned; count++) { 804 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i); 805 cleaned = (tx_desc == eop_desc); 806 skb = buffer_info->skb; 807 808 if (skb) { 809 unsigned int segs, bytecount; 810 811 /* gso_segs is currently only valid for tcp */ 812 segs = skb_shinfo(skb)->gso_segs ?: 1; 813 /* multiply data chunks by size of headers */ 814 bytecount = ((segs - 1) * skb_headlen(skb)) + 815 skb->len; 816 total_packets += segs; 817 total_bytes += bytecount; 818 } 819 820 igbvf_put_txbuf(adapter, buffer_info); 821 tx_desc->wb.status = 0; 822 823 i++; 824 if (i == tx_ring->count) 825 i = 0; 826 827 buffer_info = &tx_ring->buffer_info[i]; 828 } 829 830 eop_desc = buffer_info->next_to_watch; 831 } while (count < tx_ring->count); 832 833 tx_ring->next_to_clean = i; 834 835 if (unlikely(count && netif_carrier_ok(netdev) && 836 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) { 837 /* Make sure that anybody stopping the queue after this 838 * sees the new next_to_clean. 839 */ 840 smp_mb(); 841 if (netif_queue_stopped(netdev) && 842 !(test_bit(__IGBVF_DOWN, &adapter->state))) { 843 netif_wake_queue(netdev); 844 ++adapter->restart_queue; 845 } 846 } 847 848 netdev->stats.tx_bytes += total_bytes; 849 netdev->stats.tx_packets += total_packets; 850 return count < tx_ring->count; 851 } 852 853 static irqreturn_t igbvf_msix_other(int irq, void *data) 854 { 855 struct net_device *netdev = data; 856 struct igbvf_adapter *adapter = netdev_priv(netdev); 857 struct e1000_hw *hw = &adapter->hw; 858 859 adapter->int_counter1++; 860 861 hw->mac.get_link_status = 1; 862 if (!test_bit(__IGBVF_DOWN, &adapter->state)) 863 mod_timer(&adapter->watchdog_timer, jiffies + 1); 864 865 ew32(EIMS, adapter->eims_other); 866 867 return IRQ_HANDLED; 868 } 869 870 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data) 871 { 872 struct net_device *netdev = data; 873 struct igbvf_adapter *adapter = netdev_priv(netdev); 874 struct e1000_hw *hw = &adapter->hw; 875 struct igbvf_ring *tx_ring = adapter->tx_ring; 876 877 if (tx_ring->set_itr) { 878 writel(tx_ring->itr_val, 879 adapter->hw.hw_addr + tx_ring->itr_register); 880 adapter->tx_ring->set_itr = 0; 881 } 882 883 adapter->total_tx_bytes = 0; 884 adapter->total_tx_packets = 0; 885 886 /* auto mask will automatically re-enable the interrupt when we write 887 * EICS 888 */ 889 if (!igbvf_clean_tx_irq(tx_ring)) 890 /* Ring was not completely cleaned, so fire another interrupt */ 891 ew32(EICS, tx_ring->eims_value); 892 else 893 ew32(EIMS, tx_ring->eims_value); 894 895 return IRQ_HANDLED; 896 } 897 898 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data) 899 { 900 struct net_device *netdev = data; 901 struct igbvf_adapter *adapter = netdev_priv(netdev); 902 903 adapter->int_counter0++; 904 905 /* Write the ITR value calculated at the end of the 906 * previous interrupt. 907 */ 908 if (adapter->rx_ring->set_itr) { 909 writel(adapter->rx_ring->itr_val, 910 adapter->hw.hw_addr + adapter->rx_ring->itr_register); 911 adapter->rx_ring->set_itr = 0; 912 } 913 914 if (napi_schedule_prep(&adapter->rx_ring->napi)) { 915 adapter->total_rx_bytes = 0; 916 adapter->total_rx_packets = 0; 917 __napi_schedule(&adapter->rx_ring->napi); 918 } 919 920 return IRQ_HANDLED; 921 } 922 923 #define IGBVF_NO_QUEUE -1 924 925 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue, 926 int tx_queue, int msix_vector) 927 { 928 struct e1000_hw *hw = &adapter->hw; 929 u32 ivar, index; 930 931 /* 82576 uses a table-based method for assigning vectors. 932 * Each queue has a single entry in the table to which we write 933 * a vector number along with a "valid" bit. Sadly, the layout 934 * of the table is somewhat counterintuitive. 935 */ 936 if (rx_queue > IGBVF_NO_QUEUE) { 937 index = (rx_queue >> 1); 938 ivar = array_er32(IVAR0, index); 939 if (rx_queue & 0x1) { 940 /* vector goes into third byte of register */ 941 ivar = ivar & 0xFF00FFFF; 942 ivar |= (msix_vector | E1000_IVAR_VALID) << 16; 943 } else { 944 /* vector goes into low byte of register */ 945 ivar = ivar & 0xFFFFFF00; 946 ivar |= msix_vector | E1000_IVAR_VALID; 947 } 948 adapter->rx_ring[rx_queue].eims_value = BIT(msix_vector); 949 array_ew32(IVAR0, index, ivar); 950 } 951 if (tx_queue > IGBVF_NO_QUEUE) { 952 index = (tx_queue >> 1); 953 ivar = array_er32(IVAR0, index); 954 if (tx_queue & 0x1) { 955 /* vector goes into high byte of register */ 956 ivar = ivar & 0x00FFFFFF; 957 ivar |= (msix_vector | E1000_IVAR_VALID) << 24; 958 } else { 959 /* vector goes into second byte of register */ 960 ivar = ivar & 0xFFFF00FF; 961 ivar |= (msix_vector | E1000_IVAR_VALID) << 8; 962 } 963 adapter->tx_ring[tx_queue].eims_value = BIT(msix_vector); 964 array_ew32(IVAR0, index, ivar); 965 } 966 } 967 968 /** 969 * igbvf_configure_msix - Configure MSI-X hardware 970 * @adapter: board private structure 971 * 972 * igbvf_configure_msix sets up the hardware to properly 973 * generate MSI-X interrupts. 974 **/ 975 static void igbvf_configure_msix(struct igbvf_adapter *adapter) 976 { 977 u32 tmp; 978 struct e1000_hw *hw = &adapter->hw; 979 struct igbvf_ring *tx_ring = adapter->tx_ring; 980 struct igbvf_ring *rx_ring = adapter->rx_ring; 981 int vector = 0; 982 983 adapter->eims_enable_mask = 0; 984 985 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++); 986 adapter->eims_enable_mask |= tx_ring->eims_value; 987 writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register); 988 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++); 989 adapter->eims_enable_mask |= rx_ring->eims_value; 990 writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register); 991 992 /* set vector for other causes, i.e. link changes */ 993 994 tmp = (vector++ | E1000_IVAR_VALID); 995 996 ew32(IVAR_MISC, tmp); 997 998 adapter->eims_enable_mask = GENMASK(vector - 1, 0); 999 adapter->eims_other = BIT(vector - 1); 1000 e1e_flush(); 1001 } 1002 1003 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter) 1004 { 1005 if (adapter->msix_entries) { 1006 pci_disable_msix(adapter->pdev); 1007 kfree(adapter->msix_entries); 1008 adapter->msix_entries = NULL; 1009 } 1010 } 1011 1012 /** 1013 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported 1014 * @adapter: board private structure 1015 * 1016 * Attempt to configure interrupts using the best available 1017 * capabilities of the hardware and kernel. 1018 **/ 1019 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter) 1020 { 1021 int err = -ENOMEM; 1022 int i; 1023 1024 /* we allocate 3 vectors, 1 for Tx, 1 for Rx, one for PF messages */ 1025 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry), 1026 GFP_KERNEL); 1027 if (adapter->msix_entries) { 1028 for (i = 0; i < 3; i++) 1029 adapter->msix_entries[i].entry = i; 1030 1031 err = pci_enable_msix_range(adapter->pdev, 1032 adapter->msix_entries, 3, 3); 1033 } 1034 1035 if (err < 0) { 1036 /* MSI-X failed */ 1037 dev_err(&adapter->pdev->dev, 1038 "Failed to initialize MSI-X interrupts.\n"); 1039 igbvf_reset_interrupt_capability(adapter); 1040 } 1041 } 1042 1043 /** 1044 * igbvf_request_msix - Initialize MSI-X interrupts 1045 * @adapter: board private structure 1046 * 1047 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the 1048 * kernel. 1049 **/ 1050 static int igbvf_request_msix(struct igbvf_adapter *adapter) 1051 { 1052 struct net_device *netdev = adapter->netdev; 1053 int err = 0, vector = 0; 1054 1055 if (strlen(netdev->name) < (IFNAMSIZ - 5)) { 1056 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name); 1057 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name); 1058 } else { 1059 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ); 1060 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ); 1061 } 1062 1063 err = request_irq(adapter->msix_entries[vector].vector, 1064 igbvf_intr_msix_tx, 0, adapter->tx_ring->name, 1065 netdev); 1066 if (err) 1067 goto out; 1068 1069 adapter->tx_ring->itr_register = E1000_EITR(vector); 1070 adapter->tx_ring->itr_val = adapter->current_itr; 1071 vector++; 1072 1073 err = request_irq(adapter->msix_entries[vector].vector, 1074 igbvf_intr_msix_rx, 0, adapter->rx_ring->name, 1075 netdev); 1076 if (err) 1077 goto free_irq_tx; 1078 1079 adapter->rx_ring->itr_register = E1000_EITR(vector); 1080 adapter->rx_ring->itr_val = adapter->current_itr; 1081 vector++; 1082 1083 err = request_irq(adapter->msix_entries[vector].vector, 1084 igbvf_msix_other, 0, netdev->name, netdev); 1085 if (err) 1086 goto free_irq_rx; 1087 1088 igbvf_configure_msix(adapter); 1089 return 0; 1090 free_irq_rx: 1091 free_irq(adapter->msix_entries[--vector].vector, netdev); 1092 free_irq_tx: 1093 free_irq(adapter->msix_entries[--vector].vector, netdev); 1094 out: 1095 return err; 1096 } 1097 1098 /** 1099 * igbvf_alloc_queues - Allocate memory for all rings 1100 * @adapter: board private structure to initialize 1101 **/ 1102 static int igbvf_alloc_queues(struct igbvf_adapter *adapter) 1103 { 1104 struct net_device *netdev = adapter->netdev; 1105 1106 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL); 1107 if (!adapter->tx_ring) 1108 return -ENOMEM; 1109 1110 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL); 1111 if (!adapter->rx_ring) { 1112 kfree(adapter->tx_ring); 1113 return -ENOMEM; 1114 } 1115 1116 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll); 1117 1118 return 0; 1119 } 1120 1121 /** 1122 * igbvf_request_irq - initialize interrupts 1123 * @adapter: board private structure 1124 * 1125 * Attempts to configure interrupts using the best available 1126 * capabilities of the hardware and kernel. 1127 **/ 1128 static int igbvf_request_irq(struct igbvf_adapter *adapter) 1129 { 1130 int err = -1; 1131 1132 /* igbvf supports msi-x only */ 1133 if (adapter->msix_entries) 1134 err = igbvf_request_msix(adapter); 1135 1136 if (!err) 1137 return err; 1138 1139 dev_err(&adapter->pdev->dev, 1140 "Unable to allocate interrupt, Error: %d\n", err); 1141 1142 return err; 1143 } 1144 1145 static void igbvf_free_irq(struct igbvf_adapter *adapter) 1146 { 1147 struct net_device *netdev = adapter->netdev; 1148 int vector; 1149 1150 if (adapter->msix_entries) { 1151 for (vector = 0; vector < 3; vector++) 1152 free_irq(adapter->msix_entries[vector].vector, netdev); 1153 } 1154 } 1155 1156 /** 1157 * igbvf_irq_disable - Mask off interrupt generation on the NIC 1158 * @adapter: board private structure 1159 **/ 1160 static void igbvf_irq_disable(struct igbvf_adapter *adapter) 1161 { 1162 struct e1000_hw *hw = &adapter->hw; 1163 1164 ew32(EIMC, ~0); 1165 1166 if (adapter->msix_entries) 1167 ew32(EIAC, 0); 1168 } 1169 1170 /** 1171 * igbvf_irq_enable - Enable default interrupt generation settings 1172 * @adapter: board private structure 1173 **/ 1174 static void igbvf_irq_enable(struct igbvf_adapter *adapter) 1175 { 1176 struct e1000_hw *hw = &adapter->hw; 1177 1178 ew32(EIAC, adapter->eims_enable_mask); 1179 ew32(EIAM, adapter->eims_enable_mask); 1180 ew32(EIMS, adapter->eims_enable_mask); 1181 } 1182 1183 /** 1184 * igbvf_poll - NAPI Rx polling callback 1185 * @napi: struct associated with this polling callback 1186 * @budget: amount of packets driver is allowed to process this poll 1187 **/ 1188 static int igbvf_poll(struct napi_struct *napi, int budget) 1189 { 1190 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi); 1191 struct igbvf_adapter *adapter = rx_ring->adapter; 1192 struct e1000_hw *hw = &adapter->hw; 1193 int work_done = 0; 1194 1195 igbvf_clean_rx_irq(adapter, &work_done, budget); 1196 1197 if (work_done == budget) 1198 return budget; 1199 1200 /* Exit the polling mode, but don't re-enable interrupts if stack might 1201 * poll us due to busy-polling 1202 */ 1203 if (likely(napi_complete_done(napi, work_done))) { 1204 if (adapter->requested_itr & 3) 1205 igbvf_set_itr(adapter); 1206 1207 if (!test_bit(__IGBVF_DOWN, &adapter->state)) 1208 ew32(EIMS, adapter->rx_ring->eims_value); 1209 } 1210 1211 return work_done; 1212 } 1213 1214 /** 1215 * igbvf_set_rlpml - set receive large packet maximum length 1216 * @adapter: board private structure 1217 * 1218 * Configure the maximum size of packets that will be received 1219 */ 1220 static void igbvf_set_rlpml(struct igbvf_adapter *adapter) 1221 { 1222 int max_frame_size; 1223 struct e1000_hw *hw = &adapter->hw; 1224 1225 max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE; 1226 1227 spin_lock_bh(&hw->mbx_lock); 1228 1229 e1000_rlpml_set_vf(hw, max_frame_size); 1230 1231 spin_unlock_bh(&hw->mbx_lock); 1232 } 1233 1234 static int igbvf_vlan_rx_add_vid(struct net_device *netdev, 1235 __be16 proto, u16 vid) 1236 { 1237 struct igbvf_adapter *adapter = netdev_priv(netdev); 1238 struct e1000_hw *hw = &adapter->hw; 1239 1240 spin_lock_bh(&hw->mbx_lock); 1241 1242 if (hw->mac.ops.set_vfta(hw, vid, true)) { 1243 dev_warn(&adapter->pdev->dev, "Vlan id %d\n is not added", vid); 1244 spin_unlock_bh(&hw->mbx_lock); 1245 return -EINVAL; 1246 } 1247 1248 spin_unlock_bh(&hw->mbx_lock); 1249 1250 set_bit(vid, adapter->active_vlans); 1251 return 0; 1252 } 1253 1254 static int igbvf_vlan_rx_kill_vid(struct net_device *netdev, 1255 __be16 proto, u16 vid) 1256 { 1257 struct igbvf_adapter *adapter = netdev_priv(netdev); 1258 struct e1000_hw *hw = &adapter->hw; 1259 1260 spin_lock_bh(&hw->mbx_lock); 1261 1262 if (hw->mac.ops.set_vfta(hw, vid, false)) { 1263 dev_err(&adapter->pdev->dev, 1264 "Failed to remove vlan id %d\n", vid); 1265 spin_unlock_bh(&hw->mbx_lock); 1266 return -EINVAL; 1267 } 1268 1269 spin_unlock_bh(&hw->mbx_lock); 1270 1271 clear_bit(vid, adapter->active_vlans); 1272 return 0; 1273 } 1274 1275 static void igbvf_restore_vlan(struct igbvf_adapter *adapter) 1276 { 1277 u16 vid; 1278 1279 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID) 1280 igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid); 1281 } 1282 1283 /** 1284 * igbvf_configure_tx - Configure Transmit Unit after Reset 1285 * @adapter: board private structure 1286 * 1287 * Configure the Tx unit of the MAC after a reset. 1288 **/ 1289 static void igbvf_configure_tx(struct igbvf_adapter *adapter) 1290 { 1291 struct e1000_hw *hw = &adapter->hw; 1292 struct igbvf_ring *tx_ring = adapter->tx_ring; 1293 u64 tdba; 1294 u32 txdctl, dca_txctrl; 1295 1296 /* disable transmits */ 1297 txdctl = er32(TXDCTL(0)); 1298 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE); 1299 e1e_flush(); 1300 msleep(10); 1301 1302 /* Setup the HW Tx Head and Tail descriptor pointers */ 1303 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc)); 1304 tdba = tx_ring->dma; 1305 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32))); 1306 ew32(TDBAH(0), (tdba >> 32)); 1307 ew32(TDH(0), 0); 1308 ew32(TDT(0), 0); 1309 tx_ring->head = E1000_TDH(0); 1310 tx_ring->tail = E1000_TDT(0); 1311 1312 /* Turn off Relaxed Ordering on head write-backs. The writebacks 1313 * MUST be delivered in order or it will completely screw up 1314 * our bookkeeping. 1315 */ 1316 dca_txctrl = er32(DCA_TXCTRL(0)); 1317 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN; 1318 ew32(DCA_TXCTRL(0), dca_txctrl); 1319 1320 /* enable transmits */ 1321 txdctl |= E1000_TXDCTL_QUEUE_ENABLE; 1322 ew32(TXDCTL(0), txdctl); 1323 1324 /* Setup Transmit Descriptor Settings for eop descriptor */ 1325 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS; 1326 1327 /* enable Report Status bit */ 1328 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS; 1329 } 1330 1331 /** 1332 * igbvf_setup_srrctl - configure the receive control registers 1333 * @adapter: Board private structure 1334 **/ 1335 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter) 1336 { 1337 struct e1000_hw *hw = &adapter->hw; 1338 u32 srrctl = 0; 1339 1340 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK | 1341 E1000_SRRCTL_BSIZEHDR_MASK | 1342 E1000_SRRCTL_BSIZEPKT_MASK); 1343 1344 /* Enable queue drop to avoid head of line blocking */ 1345 srrctl |= E1000_SRRCTL_DROP_EN; 1346 1347 /* Setup buffer sizes */ 1348 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >> 1349 E1000_SRRCTL_BSIZEPKT_SHIFT; 1350 1351 if (adapter->rx_buffer_len < 2048) { 1352 adapter->rx_ps_hdr_size = 0; 1353 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF; 1354 } else { 1355 adapter->rx_ps_hdr_size = 128; 1356 srrctl |= adapter->rx_ps_hdr_size << 1357 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT; 1358 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS; 1359 } 1360 1361 ew32(SRRCTL(0), srrctl); 1362 } 1363 1364 /** 1365 * igbvf_configure_rx - Configure Receive Unit after Reset 1366 * @adapter: board private structure 1367 * 1368 * Configure the Rx unit of the MAC after a reset. 1369 **/ 1370 static void igbvf_configure_rx(struct igbvf_adapter *adapter) 1371 { 1372 struct e1000_hw *hw = &adapter->hw; 1373 struct igbvf_ring *rx_ring = adapter->rx_ring; 1374 u64 rdba; 1375 u32 rxdctl; 1376 1377 /* disable receives */ 1378 rxdctl = er32(RXDCTL(0)); 1379 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE); 1380 e1e_flush(); 1381 msleep(10); 1382 1383 /* Setup the HW Rx Head and Tail Descriptor Pointers and 1384 * the Base and Length of the Rx Descriptor Ring 1385 */ 1386 rdba = rx_ring->dma; 1387 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32))); 1388 ew32(RDBAH(0), (rdba >> 32)); 1389 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc)); 1390 rx_ring->head = E1000_RDH(0); 1391 rx_ring->tail = E1000_RDT(0); 1392 ew32(RDH(0), 0); 1393 ew32(RDT(0), 0); 1394 1395 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE; 1396 rxdctl &= 0xFFF00000; 1397 rxdctl |= IGBVF_RX_PTHRESH; 1398 rxdctl |= IGBVF_RX_HTHRESH << 8; 1399 rxdctl |= IGBVF_RX_WTHRESH << 16; 1400 1401 igbvf_set_rlpml(adapter); 1402 1403 /* enable receives */ 1404 ew32(RXDCTL(0), rxdctl); 1405 } 1406 1407 /** 1408 * igbvf_set_multi - Multicast and Promiscuous mode set 1409 * @netdev: network interface device structure 1410 * 1411 * The set_multi entry point is called whenever the multicast address 1412 * list or the network interface flags are updated. This routine is 1413 * responsible for configuring the hardware for proper multicast, 1414 * promiscuous mode, and all-multi behavior. 1415 **/ 1416 static void igbvf_set_multi(struct net_device *netdev) 1417 { 1418 struct igbvf_adapter *adapter = netdev_priv(netdev); 1419 struct e1000_hw *hw = &adapter->hw; 1420 struct netdev_hw_addr *ha; 1421 u8 *mta_list = NULL; 1422 int i; 1423 1424 if (!netdev_mc_empty(netdev)) { 1425 mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN, 1426 GFP_ATOMIC); 1427 if (!mta_list) 1428 return; 1429 } 1430 1431 /* prepare a packed array of only addresses. */ 1432 i = 0; 1433 netdev_for_each_mc_addr(ha, netdev) 1434 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN); 1435 1436 spin_lock_bh(&hw->mbx_lock); 1437 1438 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0); 1439 1440 spin_unlock_bh(&hw->mbx_lock); 1441 kfree(mta_list); 1442 } 1443 1444 /** 1445 * igbvf_set_uni - Configure unicast MAC filters 1446 * @netdev: network interface device structure 1447 * 1448 * This routine is responsible for configuring the hardware for proper 1449 * unicast filters. 1450 **/ 1451 static int igbvf_set_uni(struct net_device *netdev) 1452 { 1453 struct igbvf_adapter *adapter = netdev_priv(netdev); 1454 struct e1000_hw *hw = &adapter->hw; 1455 1456 if (netdev_uc_count(netdev) > IGBVF_MAX_MAC_FILTERS) { 1457 pr_err("Too many unicast filters - No Space\n"); 1458 return -ENOSPC; 1459 } 1460 1461 spin_lock_bh(&hw->mbx_lock); 1462 1463 /* Clear all unicast MAC filters */ 1464 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_CLR, NULL); 1465 1466 spin_unlock_bh(&hw->mbx_lock); 1467 1468 if (!netdev_uc_empty(netdev)) { 1469 struct netdev_hw_addr *ha; 1470 1471 /* Add MAC filters one by one */ 1472 netdev_for_each_uc_addr(ha, netdev) { 1473 spin_lock_bh(&hw->mbx_lock); 1474 1475 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_ADD, 1476 ha->addr); 1477 1478 spin_unlock_bh(&hw->mbx_lock); 1479 udelay(200); 1480 } 1481 } 1482 1483 return 0; 1484 } 1485 1486 static void igbvf_set_rx_mode(struct net_device *netdev) 1487 { 1488 igbvf_set_multi(netdev); 1489 igbvf_set_uni(netdev); 1490 } 1491 1492 /** 1493 * igbvf_configure - configure the hardware for Rx and Tx 1494 * @adapter: private board structure 1495 **/ 1496 static void igbvf_configure(struct igbvf_adapter *adapter) 1497 { 1498 igbvf_set_rx_mode(adapter->netdev); 1499 1500 igbvf_restore_vlan(adapter); 1501 1502 igbvf_configure_tx(adapter); 1503 igbvf_setup_srrctl(adapter); 1504 igbvf_configure_rx(adapter); 1505 igbvf_alloc_rx_buffers(adapter->rx_ring, 1506 igbvf_desc_unused(adapter->rx_ring)); 1507 } 1508 1509 /* igbvf_reset - bring the hardware into a known good state 1510 * @adapter: private board structure 1511 * 1512 * This function boots the hardware and enables some settings that 1513 * require a configuration cycle of the hardware - those cannot be 1514 * set/changed during runtime. After reset the device needs to be 1515 * properly configured for Rx, Tx etc. 1516 */ 1517 static void igbvf_reset(struct igbvf_adapter *adapter) 1518 { 1519 struct e1000_mac_info *mac = &adapter->hw.mac; 1520 struct net_device *netdev = adapter->netdev; 1521 struct e1000_hw *hw = &adapter->hw; 1522 1523 spin_lock_bh(&hw->mbx_lock); 1524 1525 /* Allow time for pending master requests to run */ 1526 if (mac->ops.reset_hw(hw)) 1527 dev_info(&adapter->pdev->dev, "PF still resetting\n"); 1528 1529 mac->ops.init_hw(hw); 1530 1531 spin_unlock_bh(&hw->mbx_lock); 1532 1533 if (is_valid_ether_addr(adapter->hw.mac.addr)) { 1534 eth_hw_addr_set(netdev, adapter->hw.mac.addr); 1535 memcpy(netdev->perm_addr, adapter->hw.mac.addr, 1536 netdev->addr_len); 1537 } 1538 1539 adapter->last_reset = jiffies; 1540 } 1541 1542 int igbvf_up(struct igbvf_adapter *adapter) 1543 { 1544 struct e1000_hw *hw = &adapter->hw; 1545 1546 /* hardware has been reset, we need to reload some things */ 1547 igbvf_configure(adapter); 1548 1549 clear_bit(__IGBVF_DOWN, &adapter->state); 1550 1551 napi_enable(&adapter->rx_ring->napi); 1552 if (adapter->msix_entries) 1553 igbvf_configure_msix(adapter); 1554 1555 /* Clear any pending interrupts. */ 1556 er32(EICR); 1557 igbvf_irq_enable(adapter); 1558 1559 /* start the watchdog */ 1560 hw->mac.get_link_status = 1; 1561 mod_timer(&adapter->watchdog_timer, jiffies + 1); 1562 1563 return 0; 1564 } 1565 1566 void igbvf_down(struct igbvf_adapter *adapter) 1567 { 1568 struct net_device *netdev = adapter->netdev; 1569 struct e1000_hw *hw = &adapter->hw; 1570 u32 rxdctl, txdctl; 1571 1572 /* signal that we're down so the interrupt handler does not 1573 * reschedule our watchdog timer 1574 */ 1575 set_bit(__IGBVF_DOWN, &adapter->state); 1576 1577 /* disable receives in the hardware */ 1578 rxdctl = er32(RXDCTL(0)); 1579 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE); 1580 1581 netif_carrier_off(netdev); 1582 netif_stop_queue(netdev); 1583 1584 /* disable transmits in the hardware */ 1585 txdctl = er32(TXDCTL(0)); 1586 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE); 1587 1588 /* flush both disables and wait for them to finish */ 1589 e1e_flush(); 1590 msleep(10); 1591 1592 napi_disable(&adapter->rx_ring->napi); 1593 1594 igbvf_irq_disable(adapter); 1595 1596 del_timer_sync(&adapter->watchdog_timer); 1597 1598 /* record the stats before reset*/ 1599 igbvf_update_stats(adapter); 1600 1601 adapter->link_speed = 0; 1602 adapter->link_duplex = 0; 1603 1604 igbvf_reset(adapter); 1605 igbvf_clean_tx_ring(adapter->tx_ring); 1606 igbvf_clean_rx_ring(adapter->rx_ring); 1607 } 1608 1609 void igbvf_reinit_locked(struct igbvf_adapter *adapter) 1610 { 1611 might_sleep(); 1612 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state)) 1613 usleep_range(1000, 2000); 1614 igbvf_down(adapter); 1615 igbvf_up(adapter); 1616 clear_bit(__IGBVF_RESETTING, &adapter->state); 1617 } 1618 1619 /** 1620 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter) 1621 * @adapter: board private structure to initialize 1622 * 1623 * igbvf_sw_init initializes the Adapter private data structure. 1624 * Fields are initialized based on PCI device information and 1625 * OS network device settings (MTU size). 1626 **/ 1627 static int igbvf_sw_init(struct igbvf_adapter *adapter) 1628 { 1629 struct net_device *netdev = adapter->netdev; 1630 s32 rc; 1631 1632 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN; 1633 adapter->rx_ps_hdr_size = 0; 1634 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN; 1635 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN; 1636 1637 adapter->tx_int_delay = 8; 1638 adapter->tx_abs_int_delay = 32; 1639 adapter->rx_int_delay = 0; 1640 adapter->rx_abs_int_delay = 8; 1641 adapter->requested_itr = 3; 1642 adapter->current_itr = IGBVF_START_ITR; 1643 1644 /* Set various function pointers */ 1645 adapter->ei->init_ops(&adapter->hw); 1646 1647 rc = adapter->hw.mac.ops.init_params(&adapter->hw); 1648 if (rc) 1649 return rc; 1650 1651 rc = adapter->hw.mbx.ops.init_params(&adapter->hw); 1652 if (rc) 1653 return rc; 1654 1655 igbvf_set_interrupt_capability(adapter); 1656 1657 if (igbvf_alloc_queues(adapter)) 1658 return -ENOMEM; 1659 1660 spin_lock_init(&adapter->tx_queue_lock); 1661 1662 /* Explicitly disable IRQ since the NIC can be in any state. */ 1663 igbvf_irq_disable(adapter); 1664 1665 spin_lock_init(&adapter->stats_lock); 1666 spin_lock_init(&adapter->hw.mbx_lock); 1667 1668 set_bit(__IGBVF_DOWN, &adapter->state); 1669 return 0; 1670 } 1671 1672 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter) 1673 { 1674 struct e1000_hw *hw = &adapter->hw; 1675 1676 adapter->stats.last_gprc = er32(VFGPRC); 1677 adapter->stats.last_gorc = er32(VFGORC); 1678 adapter->stats.last_gptc = er32(VFGPTC); 1679 adapter->stats.last_gotc = er32(VFGOTC); 1680 adapter->stats.last_mprc = er32(VFMPRC); 1681 adapter->stats.last_gotlbc = er32(VFGOTLBC); 1682 adapter->stats.last_gptlbc = er32(VFGPTLBC); 1683 adapter->stats.last_gorlbc = er32(VFGORLBC); 1684 adapter->stats.last_gprlbc = er32(VFGPRLBC); 1685 1686 adapter->stats.base_gprc = er32(VFGPRC); 1687 adapter->stats.base_gorc = er32(VFGORC); 1688 adapter->stats.base_gptc = er32(VFGPTC); 1689 adapter->stats.base_gotc = er32(VFGOTC); 1690 adapter->stats.base_mprc = er32(VFMPRC); 1691 adapter->stats.base_gotlbc = er32(VFGOTLBC); 1692 adapter->stats.base_gptlbc = er32(VFGPTLBC); 1693 adapter->stats.base_gorlbc = er32(VFGORLBC); 1694 adapter->stats.base_gprlbc = er32(VFGPRLBC); 1695 } 1696 1697 /** 1698 * igbvf_open - Called when a network interface is made active 1699 * @netdev: network interface device structure 1700 * 1701 * Returns 0 on success, negative value on failure 1702 * 1703 * The open entry point is called when a network interface is made 1704 * active by the system (IFF_UP). At this point all resources needed 1705 * for transmit and receive operations are allocated, the interrupt 1706 * handler is registered with the OS, the watchdog timer is started, 1707 * and the stack is notified that the interface is ready. 1708 **/ 1709 static int igbvf_open(struct net_device *netdev) 1710 { 1711 struct igbvf_adapter *adapter = netdev_priv(netdev); 1712 struct e1000_hw *hw = &adapter->hw; 1713 int err; 1714 1715 /* disallow open during test */ 1716 if (test_bit(__IGBVF_TESTING, &adapter->state)) 1717 return -EBUSY; 1718 1719 /* allocate transmit descriptors */ 1720 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring); 1721 if (err) 1722 goto err_setup_tx; 1723 1724 /* allocate receive descriptors */ 1725 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring); 1726 if (err) 1727 goto err_setup_rx; 1728 1729 /* before we allocate an interrupt, we must be ready to handle it. 1730 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt 1731 * as soon as we call pci_request_irq, so we have to setup our 1732 * clean_rx handler before we do so. 1733 */ 1734 igbvf_configure(adapter); 1735 1736 err = igbvf_request_irq(adapter); 1737 if (err) 1738 goto err_req_irq; 1739 1740 /* From here on the code is the same as igbvf_up() */ 1741 clear_bit(__IGBVF_DOWN, &adapter->state); 1742 1743 napi_enable(&adapter->rx_ring->napi); 1744 1745 /* clear any pending interrupts */ 1746 er32(EICR); 1747 1748 igbvf_irq_enable(adapter); 1749 1750 /* start the watchdog */ 1751 hw->mac.get_link_status = 1; 1752 mod_timer(&adapter->watchdog_timer, jiffies + 1); 1753 1754 return 0; 1755 1756 err_req_irq: 1757 igbvf_free_rx_resources(adapter->rx_ring); 1758 err_setup_rx: 1759 igbvf_free_tx_resources(adapter->tx_ring); 1760 err_setup_tx: 1761 igbvf_reset(adapter); 1762 1763 return err; 1764 } 1765 1766 /** 1767 * igbvf_close - Disables a network interface 1768 * @netdev: network interface device structure 1769 * 1770 * Returns 0, this is not allowed to fail 1771 * 1772 * The close entry point is called when an interface is de-activated 1773 * by the OS. The hardware is still under the drivers control, but 1774 * needs to be disabled. A global MAC reset is issued to stop the 1775 * hardware, and all transmit and receive resources are freed. 1776 **/ 1777 static int igbvf_close(struct net_device *netdev) 1778 { 1779 struct igbvf_adapter *adapter = netdev_priv(netdev); 1780 1781 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state)); 1782 igbvf_down(adapter); 1783 1784 igbvf_free_irq(adapter); 1785 1786 igbvf_free_tx_resources(adapter->tx_ring); 1787 igbvf_free_rx_resources(adapter->rx_ring); 1788 1789 return 0; 1790 } 1791 1792 /** 1793 * igbvf_set_mac - Change the Ethernet Address of the NIC 1794 * @netdev: network interface device structure 1795 * @p: pointer to an address structure 1796 * 1797 * Returns 0 on success, negative on failure 1798 **/ 1799 static int igbvf_set_mac(struct net_device *netdev, void *p) 1800 { 1801 struct igbvf_adapter *adapter = netdev_priv(netdev); 1802 struct e1000_hw *hw = &adapter->hw; 1803 struct sockaddr *addr = p; 1804 1805 if (!is_valid_ether_addr(addr->sa_data)) 1806 return -EADDRNOTAVAIL; 1807 1808 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len); 1809 1810 spin_lock_bh(&hw->mbx_lock); 1811 1812 hw->mac.ops.rar_set(hw, hw->mac.addr, 0); 1813 1814 spin_unlock_bh(&hw->mbx_lock); 1815 1816 if (!ether_addr_equal(addr->sa_data, hw->mac.addr)) 1817 return -EADDRNOTAVAIL; 1818 1819 eth_hw_addr_set(netdev, addr->sa_data); 1820 1821 return 0; 1822 } 1823 1824 #define UPDATE_VF_COUNTER(reg, name) \ 1825 { \ 1826 u32 current_counter = er32(reg); \ 1827 if (current_counter < adapter->stats.last_##name) \ 1828 adapter->stats.name += 0x100000000LL; \ 1829 adapter->stats.last_##name = current_counter; \ 1830 adapter->stats.name &= 0xFFFFFFFF00000000LL; \ 1831 adapter->stats.name |= current_counter; \ 1832 } 1833 1834 /** 1835 * igbvf_update_stats - Update the board statistics counters 1836 * @adapter: board private structure 1837 **/ 1838 void igbvf_update_stats(struct igbvf_adapter *adapter) 1839 { 1840 struct e1000_hw *hw = &adapter->hw; 1841 struct pci_dev *pdev = adapter->pdev; 1842 1843 /* Prevent stats update while adapter is being reset, link is down 1844 * or if the pci connection is down. 1845 */ 1846 if (adapter->link_speed == 0) 1847 return; 1848 1849 if (test_bit(__IGBVF_RESETTING, &adapter->state)) 1850 return; 1851 1852 if (pci_channel_offline(pdev)) 1853 return; 1854 1855 UPDATE_VF_COUNTER(VFGPRC, gprc); 1856 UPDATE_VF_COUNTER(VFGORC, gorc); 1857 UPDATE_VF_COUNTER(VFGPTC, gptc); 1858 UPDATE_VF_COUNTER(VFGOTC, gotc); 1859 UPDATE_VF_COUNTER(VFMPRC, mprc); 1860 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc); 1861 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc); 1862 UPDATE_VF_COUNTER(VFGORLBC, gorlbc); 1863 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc); 1864 1865 /* Fill out the OS statistics structure */ 1866 adapter->netdev->stats.multicast = adapter->stats.mprc; 1867 } 1868 1869 static void igbvf_print_link_info(struct igbvf_adapter *adapter) 1870 { 1871 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n", 1872 adapter->link_speed, 1873 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half"); 1874 } 1875 1876 static bool igbvf_has_link(struct igbvf_adapter *adapter) 1877 { 1878 struct e1000_hw *hw = &adapter->hw; 1879 s32 ret_val = E1000_SUCCESS; 1880 bool link_active; 1881 1882 /* If interface is down, stay link down */ 1883 if (test_bit(__IGBVF_DOWN, &adapter->state)) 1884 return false; 1885 1886 spin_lock_bh(&hw->mbx_lock); 1887 1888 ret_val = hw->mac.ops.check_for_link(hw); 1889 1890 spin_unlock_bh(&hw->mbx_lock); 1891 1892 link_active = !hw->mac.get_link_status; 1893 1894 /* if check for link returns error we will need to reset */ 1895 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ))) 1896 schedule_work(&adapter->reset_task); 1897 1898 return link_active; 1899 } 1900 1901 /** 1902 * igbvf_watchdog - Timer Call-back 1903 * @t: timer list pointer containing private struct 1904 **/ 1905 static void igbvf_watchdog(struct timer_list *t) 1906 { 1907 struct igbvf_adapter *adapter = from_timer(adapter, t, watchdog_timer); 1908 1909 /* Do the rest outside of interrupt context */ 1910 schedule_work(&adapter->watchdog_task); 1911 } 1912 1913 static void igbvf_watchdog_task(struct work_struct *work) 1914 { 1915 struct igbvf_adapter *adapter = container_of(work, 1916 struct igbvf_adapter, 1917 watchdog_task); 1918 struct net_device *netdev = adapter->netdev; 1919 struct e1000_mac_info *mac = &adapter->hw.mac; 1920 struct igbvf_ring *tx_ring = adapter->tx_ring; 1921 struct e1000_hw *hw = &adapter->hw; 1922 u32 link; 1923 int tx_pending = 0; 1924 1925 link = igbvf_has_link(adapter); 1926 1927 if (link) { 1928 if (!netif_carrier_ok(netdev)) { 1929 mac->ops.get_link_up_info(&adapter->hw, 1930 &adapter->link_speed, 1931 &adapter->link_duplex); 1932 igbvf_print_link_info(adapter); 1933 1934 netif_carrier_on(netdev); 1935 netif_wake_queue(netdev); 1936 } 1937 } else { 1938 if (netif_carrier_ok(netdev)) { 1939 adapter->link_speed = 0; 1940 adapter->link_duplex = 0; 1941 dev_info(&adapter->pdev->dev, "Link is Down\n"); 1942 netif_carrier_off(netdev); 1943 netif_stop_queue(netdev); 1944 } 1945 } 1946 1947 if (netif_carrier_ok(netdev)) { 1948 igbvf_update_stats(adapter); 1949 } else { 1950 tx_pending = (igbvf_desc_unused(tx_ring) + 1 < 1951 tx_ring->count); 1952 if (tx_pending) { 1953 /* We've lost link, so the controller stops DMA, 1954 * but we've got queued Tx work that's never going 1955 * to get done, so reset controller to flush Tx. 1956 * (Do the reset outside of interrupt context). 1957 */ 1958 adapter->tx_timeout_count++; 1959 schedule_work(&adapter->reset_task); 1960 } 1961 } 1962 1963 /* Cause software interrupt to ensure Rx ring is cleaned */ 1964 ew32(EICS, adapter->rx_ring->eims_value); 1965 1966 /* Reset the timer */ 1967 if (!test_bit(__IGBVF_DOWN, &adapter->state)) 1968 mod_timer(&adapter->watchdog_timer, 1969 round_jiffies(jiffies + (2 * HZ))); 1970 } 1971 1972 #define IGBVF_TX_FLAGS_CSUM 0x00000001 1973 #define IGBVF_TX_FLAGS_VLAN 0x00000002 1974 #define IGBVF_TX_FLAGS_TSO 0x00000004 1975 #define IGBVF_TX_FLAGS_IPV4 0x00000008 1976 #define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000 1977 #define IGBVF_TX_FLAGS_VLAN_SHIFT 16 1978 1979 static void igbvf_tx_ctxtdesc(struct igbvf_ring *tx_ring, u32 vlan_macip_lens, 1980 u32 type_tucmd, u32 mss_l4len_idx) 1981 { 1982 struct e1000_adv_tx_context_desc *context_desc; 1983 struct igbvf_buffer *buffer_info; 1984 u16 i = tx_ring->next_to_use; 1985 1986 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i); 1987 buffer_info = &tx_ring->buffer_info[i]; 1988 1989 i++; 1990 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; 1991 1992 /* set bits to identify this as an advanced context descriptor */ 1993 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT; 1994 1995 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens); 1996 context_desc->seqnum_seed = 0; 1997 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd); 1998 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx); 1999 2000 buffer_info->time_stamp = jiffies; 2001 buffer_info->dma = 0; 2002 } 2003 2004 static int igbvf_tso(struct igbvf_ring *tx_ring, 2005 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len) 2006 { 2007 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx; 2008 union { 2009 struct iphdr *v4; 2010 struct ipv6hdr *v6; 2011 unsigned char *hdr; 2012 } ip; 2013 union { 2014 struct tcphdr *tcp; 2015 unsigned char *hdr; 2016 } l4; 2017 u32 paylen, l4_offset; 2018 int err; 2019 2020 if (skb->ip_summed != CHECKSUM_PARTIAL) 2021 return 0; 2022 2023 if (!skb_is_gso(skb)) 2024 return 0; 2025 2026 err = skb_cow_head(skb, 0); 2027 if (err < 0) 2028 return err; 2029 2030 ip.hdr = skb_network_header(skb); 2031 l4.hdr = skb_checksum_start(skb); 2032 2033 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */ 2034 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP; 2035 2036 /* initialize outer IP header fields */ 2037 if (ip.v4->version == 4) { 2038 unsigned char *csum_start = skb_checksum_start(skb); 2039 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4); 2040 2041 /* IP header will have to cancel out any data that 2042 * is not a part of the outer IP header 2043 */ 2044 ip.v4->check = csum_fold(csum_partial(trans_start, 2045 csum_start - trans_start, 2046 0)); 2047 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4; 2048 2049 ip.v4->tot_len = 0; 2050 } else { 2051 ip.v6->payload_len = 0; 2052 } 2053 2054 /* determine offset of inner transport header */ 2055 l4_offset = l4.hdr - skb->data; 2056 2057 /* compute length of segmentation header */ 2058 *hdr_len = (l4.tcp->doff * 4) + l4_offset; 2059 2060 /* remove payload length from inner checksum */ 2061 paylen = skb->len - l4_offset; 2062 csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen)); 2063 2064 /* MSS L4LEN IDX */ 2065 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT; 2066 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT; 2067 2068 /* VLAN MACLEN IPLEN */ 2069 vlan_macip_lens = l4.hdr - ip.hdr; 2070 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT; 2071 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK; 2072 2073 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx); 2074 2075 return 1; 2076 } 2077 2078 static bool igbvf_tx_csum(struct igbvf_ring *tx_ring, struct sk_buff *skb, 2079 u32 tx_flags, __be16 protocol) 2080 { 2081 u32 vlan_macip_lens = 0; 2082 u32 type_tucmd = 0; 2083 2084 if (skb->ip_summed != CHECKSUM_PARTIAL) { 2085 csum_failed: 2086 if (!(tx_flags & IGBVF_TX_FLAGS_VLAN)) 2087 return false; 2088 goto no_csum; 2089 } 2090 2091 switch (skb->csum_offset) { 2092 case offsetof(struct tcphdr, check): 2093 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP; 2094 fallthrough; 2095 case offsetof(struct udphdr, check): 2096 break; 2097 case offsetof(struct sctphdr, checksum): 2098 /* validate that this is actually an SCTP request */ 2099 if (skb_csum_is_sctp(skb)) { 2100 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP; 2101 break; 2102 } 2103 fallthrough; 2104 default: 2105 skb_checksum_help(skb); 2106 goto csum_failed; 2107 } 2108 2109 vlan_macip_lens = skb_checksum_start_offset(skb) - 2110 skb_network_offset(skb); 2111 no_csum: 2112 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT; 2113 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK; 2114 2115 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0); 2116 return true; 2117 } 2118 2119 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size) 2120 { 2121 struct igbvf_adapter *adapter = netdev_priv(netdev); 2122 2123 /* there is enough descriptors then we don't need to worry */ 2124 if (igbvf_desc_unused(adapter->tx_ring) >= size) 2125 return 0; 2126 2127 netif_stop_queue(netdev); 2128 2129 /* Herbert's original patch had: 2130 * smp_mb__after_netif_stop_queue(); 2131 * but since that doesn't exist yet, just open code it. 2132 */ 2133 smp_mb(); 2134 2135 /* We need to check again just in case room has been made available */ 2136 if (igbvf_desc_unused(adapter->tx_ring) < size) 2137 return -EBUSY; 2138 2139 netif_wake_queue(netdev); 2140 2141 ++adapter->restart_queue; 2142 return 0; 2143 } 2144 2145 #define IGBVF_MAX_TXD_PWR 16 2146 #define IGBVF_MAX_DATA_PER_TXD (1u << IGBVF_MAX_TXD_PWR) 2147 2148 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter, 2149 struct igbvf_ring *tx_ring, 2150 struct sk_buff *skb) 2151 { 2152 struct igbvf_buffer *buffer_info; 2153 struct pci_dev *pdev = adapter->pdev; 2154 unsigned int len = skb_headlen(skb); 2155 unsigned int count = 0, i; 2156 unsigned int f; 2157 2158 i = tx_ring->next_to_use; 2159 2160 buffer_info = &tx_ring->buffer_info[i]; 2161 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD); 2162 buffer_info->length = len; 2163 /* set time_stamp *before* dma to help avoid a possible race */ 2164 buffer_info->time_stamp = jiffies; 2165 buffer_info->mapped_as_page = false; 2166 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len, 2167 DMA_TO_DEVICE); 2168 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) 2169 goto dma_error; 2170 2171 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) { 2172 const skb_frag_t *frag; 2173 2174 count++; 2175 i++; 2176 if (i == tx_ring->count) 2177 i = 0; 2178 2179 frag = &skb_shinfo(skb)->frags[f]; 2180 len = skb_frag_size(frag); 2181 2182 buffer_info = &tx_ring->buffer_info[i]; 2183 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD); 2184 buffer_info->length = len; 2185 buffer_info->time_stamp = jiffies; 2186 buffer_info->mapped_as_page = true; 2187 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len, 2188 DMA_TO_DEVICE); 2189 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) 2190 goto dma_error; 2191 } 2192 2193 tx_ring->buffer_info[i].skb = skb; 2194 2195 return ++count; 2196 2197 dma_error: 2198 dev_err(&pdev->dev, "TX DMA map failed\n"); 2199 2200 /* clear timestamp and dma mappings for failed buffer_info mapping */ 2201 buffer_info->dma = 0; 2202 buffer_info->time_stamp = 0; 2203 buffer_info->length = 0; 2204 buffer_info->mapped_as_page = false; 2205 if (count) 2206 count--; 2207 2208 /* clear timestamp and dma mappings for remaining portion of packet */ 2209 while (count--) { 2210 if (i == 0) 2211 i += tx_ring->count; 2212 i--; 2213 buffer_info = &tx_ring->buffer_info[i]; 2214 igbvf_put_txbuf(adapter, buffer_info); 2215 } 2216 2217 return 0; 2218 } 2219 2220 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter, 2221 struct igbvf_ring *tx_ring, 2222 int tx_flags, int count, 2223 unsigned int first, u32 paylen, 2224 u8 hdr_len) 2225 { 2226 union e1000_adv_tx_desc *tx_desc = NULL; 2227 struct igbvf_buffer *buffer_info; 2228 u32 olinfo_status = 0, cmd_type_len; 2229 unsigned int i; 2230 2231 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS | 2232 E1000_ADVTXD_DCMD_DEXT); 2233 2234 if (tx_flags & IGBVF_TX_FLAGS_VLAN) 2235 cmd_type_len |= E1000_ADVTXD_DCMD_VLE; 2236 2237 if (tx_flags & IGBVF_TX_FLAGS_TSO) { 2238 cmd_type_len |= E1000_ADVTXD_DCMD_TSE; 2239 2240 /* insert tcp checksum */ 2241 olinfo_status |= E1000_TXD_POPTS_TXSM << 8; 2242 2243 /* insert ip checksum */ 2244 if (tx_flags & IGBVF_TX_FLAGS_IPV4) 2245 olinfo_status |= E1000_TXD_POPTS_IXSM << 8; 2246 2247 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) { 2248 olinfo_status |= E1000_TXD_POPTS_TXSM << 8; 2249 } 2250 2251 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT); 2252 2253 i = tx_ring->next_to_use; 2254 while (count--) { 2255 buffer_info = &tx_ring->buffer_info[i]; 2256 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i); 2257 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma); 2258 tx_desc->read.cmd_type_len = 2259 cpu_to_le32(cmd_type_len | buffer_info->length); 2260 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status); 2261 i++; 2262 if (i == tx_ring->count) 2263 i = 0; 2264 } 2265 2266 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd); 2267 /* Force memory writes to complete before letting h/w 2268 * know there are new descriptors to fetch. (Only 2269 * applicable for weak-ordered memory model archs, 2270 * such as IA-64). 2271 */ 2272 wmb(); 2273 2274 tx_ring->buffer_info[first].next_to_watch = tx_desc; 2275 tx_ring->next_to_use = i; 2276 writel(i, adapter->hw.hw_addr + tx_ring->tail); 2277 } 2278 2279 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb, 2280 struct net_device *netdev, 2281 struct igbvf_ring *tx_ring) 2282 { 2283 struct igbvf_adapter *adapter = netdev_priv(netdev); 2284 unsigned int first, tx_flags = 0; 2285 u8 hdr_len = 0; 2286 int count = 0; 2287 int tso = 0; 2288 __be16 protocol = vlan_get_protocol(skb); 2289 2290 if (test_bit(__IGBVF_DOWN, &adapter->state)) { 2291 dev_kfree_skb_any(skb); 2292 return NETDEV_TX_OK; 2293 } 2294 2295 if (skb->len <= 0) { 2296 dev_kfree_skb_any(skb); 2297 return NETDEV_TX_OK; 2298 } 2299 2300 /* need: count + 4 desc gap to keep tail from touching 2301 * + 2 desc gap to keep tail from touching head, 2302 * + 1 desc for skb->data, 2303 * + 1 desc for context descriptor, 2304 * head, otherwise try next time 2305 */ 2306 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) { 2307 /* this is a hard error */ 2308 return NETDEV_TX_BUSY; 2309 } 2310 2311 if (skb_vlan_tag_present(skb)) { 2312 tx_flags |= IGBVF_TX_FLAGS_VLAN; 2313 tx_flags |= (skb_vlan_tag_get(skb) << 2314 IGBVF_TX_FLAGS_VLAN_SHIFT); 2315 } 2316 2317 if (protocol == htons(ETH_P_IP)) 2318 tx_flags |= IGBVF_TX_FLAGS_IPV4; 2319 2320 first = tx_ring->next_to_use; 2321 2322 tso = igbvf_tso(tx_ring, skb, tx_flags, &hdr_len); 2323 if (unlikely(tso < 0)) { 2324 dev_kfree_skb_any(skb); 2325 return NETDEV_TX_OK; 2326 } 2327 2328 if (tso) 2329 tx_flags |= IGBVF_TX_FLAGS_TSO; 2330 else if (igbvf_tx_csum(tx_ring, skb, tx_flags, protocol) && 2331 (skb->ip_summed == CHECKSUM_PARTIAL)) 2332 tx_flags |= IGBVF_TX_FLAGS_CSUM; 2333 2334 /* count reflects descriptors mapped, if 0 then mapping error 2335 * has occurred and we need to rewind the descriptor queue 2336 */ 2337 count = igbvf_tx_map_adv(adapter, tx_ring, skb); 2338 2339 if (count) { 2340 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count, 2341 first, skb->len, hdr_len); 2342 /* Make sure there is space in the ring for the next send. */ 2343 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4); 2344 } else { 2345 dev_kfree_skb_any(skb); 2346 tx_ring->buffer_info[first].time_stamp = 0; 2347 tx_ring->next_to_use = first; 2348 } 2349 2350 return NETDEV_TX_OK; 2351 } 2352 2353 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb, 2354 struct net_device *netdev) 2355 { 2356 struct igbvf_adapter *adapter = netdev_priv(netdev); 2357 struct igbvf_ring *tx_ring; 2358 2359 if (test_bit(__IGBVF_DOWN, &adapter->state)) { 2360 dev_kfree_skb_any(skb); 2361 return NETDEV_TX_OK; 2362 } 2363 2364 tx_ring = &adapter->tx_ring[0]; 2365 2366 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring); 2367 } 2368 2369 /** 2370 * igbvf_tx_timeout - Respond to a Tx Hang 2371 * @netdev: network interface device structure 2372 * @txqueue: queue timing out (unused) 2373 **/ 2374 static void igbvf_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue) 2375 { 2376 struct igbvf_adapter *adapter = netdev_priv(netdev); 2377 2378 /* Do the reset outside of interrupt context */ 2379 adapter->tx_timeout_count++; 2380 schedule_work(&adapter->reset_task); 2381 } 2382 2383 static void igbvf_reset_task(struct work_struct *work) 2384 { 2385 struct igbvf_adapter *adapter; 2386 2387 adapter = container_of(work, struct igbvf_adapter, reset_task); 2388 2389 igbvf_reinit_locked(adapter); 2390 } 2391 2392 /** 2393 * igbvf_change_mtu - Change the Maximum Transfer Unit 2394 * @netdev: network interface device structure 2395 * @new_mtu: new value for maximum frame size 2396 * 2397 * Returns 0 on success, negative on failure 2398 **/ 2399 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu) 2400 { 2401 struct igbvf_adapter *adapter = netdev_priv(netdev); 2402 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN; 2403 2404 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state)) 2405 usleep_range(1000, 2000); 2406 /* igbvf_down has a dependency on max_frame_size */ 2407 adapter->max_frame_size = max_frame; 2408 if (netif_running(netdev)) 2409 igbvf_down(adapter); 2410 2411 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN 2412 * means we reserve 2 more, this pushes us to allocate from the next 2413 * larger slab size. 2414 * i.e. RXBUFFER_2048 --> size-4096 slab 2415 * However with the new *_jumbo_rx* routines, jumbo receives will use 2416 * fragmented skbs 2417 */ 2418 2419 if (max_frame <= 1024) 2420 adapter->rx_buffer_len = 1024; 2421 else if (max_frame <= 2048) 2422 adapter->rx_buffer_len = 2048; 2423 else 2424 #if (PAGE_SIZE / 2) > 16384 2425 adapter->rx_buffer_len = 16384; 2426 #else 2427 adapter->rx_buffer_len = PAGE_SIZE / 2; 2428 #endif 2429 2430 /* adjust allocation if LPE protects us, and we aren't using SBP */ 2431 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) || 2432 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN)) 2433 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + 2434 ETH_FCS_LEN; 2435 2436 netdev_dbg(netdev, "changing MTU from %d to %d\n", 2437 netdev->mtu, new_mtu); 2438 netdev->mtu = new_mtu; 2439 2440 if (netif_running(netdev)) 2441 igbvf_up(adapter); 2442 else 2443 igbvf_reset(adapter); 2444 2445 clear_bit(__IGBVF_RESETTING, &adapter->state); 2446 2447 return 0; 2448 } 2449 2450 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd) 2451 { 2452 switch (cmd) { 2453 default: 2454 return -EOPNOTSUPP; 2455 } 2456 } 2457 2458 static int igbvf_suspend(struct device *dev_d) 2459 { 2460 struct net_device *netdev = dev_get_drvdata(dev_d); 2461 struct igbvf_adapter *adapter = netdev_priv(netdev); 2462 2463 netif_device_detach(netdev); 2464 2465 if (netif_running(netdev)) { 2466 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state)); 2467 igbvf_down(adapter); 2468 igbvf_free_irq(adapter); 2469 } 2470 2471 return 0; 2472 } 2473 2474 static int __maybe_unused igbvf_resume(struct device *dev_d) 2475 { 2476 struct pci_dev *pdev = to_pci_dev(dev_d); 2477 struct net_device *netdev = pci_get_drvdata(pdev); 2478 struct igbvf_adapter *adapter = netdev_priv(netdev); 2479 u32 err; 2480 2481 pci_set_master(pdev); 2482 2483 if (netif_running(netdev)) { 2484 err = igbvf_request_irq(adapter); 2485 if (err) 2486 return err; 2487 } 2488 2489 igbvf_reset(adapter); 2490 2491 if (netif_running(netdev)) 2492 igbvf_up(adapter); 2493 2494 netif_device_attach(netdev); 2495 2496 return 0; 2497 } 2498 2499 static void igbvf_shutdown(struct pci_dev *pdev) 2500 { 2501 igbvf_suspend(&pdev->dev); 2502 } 2503 2504 #ifdef CONFIG_NET_POLL_CONTROLLER 2505 /* Polling 'interrupt' - used by things like netconsole to send skbs 2506 * without having to re-enable interrupts. It's not called while 2507 * the interrupt routine is executing. 2508 */ 2509 static void igbvf_netpoll(struct net_device *netdev) 2510 { 2511 struct igbvf_adapter *adapter = netdev_priv(netdev); 2512 2513 disable_irq(adapter->pdev->irq); 2514 2515 igbvf_clean_tx_irq(adapter->tx_ring); 2516 2517 enable_irq(adapter->pdev->irq); 2518 } 2519 #endif 2520 2521 /** 2522 * igbvf_io_error_detected - called when PCI error is detected 2523 * @pdev: Pointer to PCI device 2524 * @state: The current pci connection state 2525 * 2526 * This function is called after a PCI bus error affecting 2527 * this device has been detected. 2528 */ 2529 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev, 2530 pci_channel_state_t state) 2531 { 2532 struct net_device *netdev = pci_get_drvdata(pdev); 2533 struct igbvf_adapter *adapter = netdev_priv(netdev); 2534 2535 netif_device_detach(netdev); 2536 2537 if (state == pci_channel_io_perm_failure) 2538 return PCI_ERS_RESULT_DISCONNECT; 2539 2540 if (netif_running(netdev)) 2541 igbvf_down(adapter); 2542 pci_disable_device(pdev); 2543 2544 /* Request a slot reset. */ 2545 return PCI_ERS_RESULT_NEED_RESET; 2546 } 2547 2548 /** 2549 * igbvf_io_slot_reset - called after the pci bus has been reset. 2550 * @pdev: Pointer to PCI device 2551 * 2552 * Restart the card from scratch, as if from a cold-boot. Implementation 2553 * resembles the first-half of the igbvf_resume routine. 2554 */ 2555 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev) 2556 { 2557 struct net_device *netdev = pci_get_drvdata(pdev); 2558 struct igbvf_adapter *adapter = netdev_priv(netdev); 2559 2560 if (pci_enable_device_mem(pdev)) { 2561 dev_err(&pdev->dev, 2562 "Cannot re-enable PCI device after reset.\n"); 2563 return PCI_ERS_RESULT_DISCONNECT; 2564 } 2565 pci_set_master(pdev); 2566 2567 igbvf_reset(adapter); 2568 2569 return PCI_ERS_RESULT_RECOVERED; 2570 } 2571 2572 /** 2573 * igbvf_io_resume - called when traffic can start flowing again. 2574 * @pdev: Pointer to PCI device 2575 * 2576 * This callback is called when the error recovery driver tells us that 2577 * its OK to resume normal operation. Implementation resembles the 2578 * second-half of the igbvf_resume routine. 2579 */ 2580 static void igbvf_io_resume(struct pci_dev *pdev) 2581 { 2582 struct net_device *netdev = pci_get_drvdata(pdev); 2583 struct igbvf_adapter *adapter = netdev_priv(netdev); 2584 2585 if (netif_running(netdev)) { 2586 if (igbvf_up(adapter)) { 2587 dev_err(&pdev->dev, 2588 "can't bring device back up after reset\n"); 2589 return; 2590 } 2591 } 2592 2593 netif_device_attach(netdev); 2594 } 2595 2596 /** 2597 * igbvf_io_prepare - prepare device driver for PCI reset 2598 * @pdev: PCI device information struct 2599 */ 2600 static void igbvf_io_prepare(struct pci_dev *pdev) 2601 { 2602 struct net_device *netdev = pci_get_drvdata(pdev); 2603 struct igbvf_adapter *adapter = netdev_priv(netdev); 2604 2605 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state)) 2606 usleep_range(1000, 2000); 2607 igbvf_down(adapter); 2608 } 2609 2610 /** 2611 * igbvf_io_reset_done - PCI reset done, device driver reset can begin 2612 * @pdev: PCI device information struct 2613 */ 2614 static void igbvf_io_reset_done(struct pci_dev *pdev) 2615 { 2616 struct net_device *netdev = pci_get_drvdata(pdev); 2617 struct igbvf_adapter *adapter = netdev_priv(netdev); 2618 2619 igbvf_up(adapter); 2620 clear_bit(__IGBVF_RESETTING, &adapter->state); 2621 } 2622 2623 static void igbvf_print_device_info(struct igbvf_adapter *adapter) 2624 { 2625 struct e1000_hw *hw = &adapter->hw; 2626 struct net_device *netdev = adapter->netdev; 2627 struct pci_dev *pdev = adapter->pdev; 2628 2629 if (hw->mac.type == e1000_vfadapt_i350) 2630 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n"); 2631 else 2632 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n"); 2633 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr); 2634 } 2635 2636 static int igbvf_set_features(struct net_device *netdev, 2637 netdev_features_t features) 2638 { 2639 struct igbvf_adapter *adapter = netdev_priv(netdev); 2640 2641 if (features & NETIF_F_RXCSUM) 2642 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED; 2643 else 2644 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED; 2645 2646 return 0; 2647 } 2648 2649 #define IGBVF_MAX_MAC_HDR_LEN 127 2650 #define IGBVF_MAX_NETWORK_HDR_LEN 511 2651 2652 static netdev_features_t 2653 igbvf_features_check(struct sk_buff *skb, struct net_device *dev, 2654 netdev_features_t features) 2655 { 2656 unsigned int network_hdr_len, mac_hdr_len; 2657 2658 /* Make certain the headers can be described by a context descriptor */ 2659 mac_hdr_len = skb_network_header(skb) - skb->data; 2660 if (unlikely(mac_hdr_len > IGBVF_MAX_MAC_HDR_LEN)) 2661 return features & ~(NETIF_F_HW_CSUM | 2662 NETIF_F_SCTP_CRC | 2663 NETIF_F_HW_VLAN_CTAG_TX | 2664 NETIF_F_TSO | 2665 NETIF_F_TSO6); 2666 2667 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb); 2668 if (unlikely(network_hdr_len > IGBVF_MAX_NETWORK_HDR_LEN)) 2669 return features & ~(NETIF_F_HW_CSUM | 2670 NETIF_F_SCTP_CRC | 2671 NETIF_F_TSO | 2672 NETIF_F_TSO6); 2673 2674 /* We can only support IPV4 TSO in tunnels if we can mangle the 2675 * inner IP ID field, so strip TSO if MANGLEID is not supported. 2676 */ 2677 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) 2678 features &= ~NETIF_F_TSO; 2679 2680 return features; 2681 } 2682 2683 static const struct net_device_ops igbvf_netdev_ops = { 2684 .ndo_open = igbvf_open, 2685 .ndo_stop = igbvf_close, 2686 .ndo_start_xmit = igbvf_xmit_frame, 2687 .ndo_set_rx_mode = igbvf_set_rx_mode, 2688 .ndo_set_mac_address = igbvf_set_mac, 2689 .ndo_change_mtu = igbvf_change_mtu, 2690 .ndo_eth_ioctl = igbvf_ioctl, 2691 .ndo_tx_timeout = igbvf_tx_timeout, 2692 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid, 2693 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid, 2694 #ifdef CONFIG_NET_POLL_CONTROLLER 2695 .ndo_poll_controller = igbvf_netpoll, 2696 #endif 2697 .ndo_set_features = igbvf_set_features, 2698 .ndo_features_check = igbvf_features_check, 2699 }; 2700 2701 /** 2702 * igbvf_probe - Device Initialization Routine 2703 * @pdev: PCI device information struct 2704 * @ent: entry in igbvf_pci_tbl 2705 * 2706 * Returns 0 on success, negative on failure 2707 * 2708 * igbvf_probe initializes an adapter identified by a pci_dev structure. 2709 * The OS initialization, configuring of the adapter private structure, 2710 * and a hardware reset occur. 2711 **/ 2712 static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 2713 { 2714 struct net_device *netdev; 2715 struct igbvf_adapter *adapter; 2716 struct e1000_hw *hw; 2717 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data]; 2718 static int cards_found; 2719 int err; 2720 2721 err = pci_enable_device_mem(pdev); 2722 if (err) 2723 return err; 2724 2725 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); 2726 if (err) { 2727 dev_err(&pdev->dev, 2728 "No usable DMA configuration, aborting\n"); 2729 goto err_dma; 2730 } 2731 2732 err = pci_request_regions(pdev, igbvf_driver_name); 2733 if (err) 2734 goto err_pci_reg; 2735 2736 pci_set_master(pdev); 2737 2738 err = -ENOMEM; 2739 netdev = alloc_etherdev(sizeof(struct igbvf_adapter)); 2740 if (!netdev) 2741 goto err_alloc_etherdev; 2742 2743 SET_NETDEV_DEV(netdev, &pdev->dev); 2744 2745 pci_set_drvdata(pdev, netdev); 2746 adapter = netdev_priv(netdev); 2747 hw = &adapter->hw; 2748 adapter->netdev = netdev; 2749 adapter->pdev = pdev; 2750 adapter->ei = ei; 2751 adapter->pba = ei->pba; 2752 adapter->flags = ei->flags; 2753 adapter->hw.back = adapter; 2754 adapter->hw.mac.type = ei->mac; 2755 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE); 2756 2757 /* PCI config space info */ 2758 2759 hw->vendor_id = pdev->vendor; 2760 hw->device_id = pdev->device; 2761 hw->subsystem_vendor_id = pdev->subsystem_vendor; 2762 hw->subsystem_device_id = pdev->subsystem_device; 2763 hw->revision_id = pdev->revision; 2764 2765 err = -EIO; 2766 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0), 2767 pci_resource_len(pdev, 0)); 2768 2769 if (!adapter->hw.hw_addr) 2770 goto err_ioremap; 2771 2772 if (ei->get_variants) { 2773 err = ei->get_variants(adapter); 2774 if (err) 2775 goto err_get_variants; 2776 } 2777 2778 /* setup adapter struct */ 2779 err = igbvf_sw_init(adapter); 2780 if (err) 2781 goto err_sw_init; 2782 2783 /* construct the net_device struct */ 2784 netdev->netdev_ops = &igbvf_netdev_ops; 2785 2786 igbvf_set_ethtool_ops(netdev); 2787 netdev->watchdog_timeo = 5 * HZ; 2788 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1); 2789 2790 adapter->bd_number = cards_found++; 2791 2792 netdev->hw_features = NETIF_F_SG | 2793 NETIF_F_TSO | 2794 NETIF_F_TSO6 | 2795 NETIF_F_RXCSUM | 2796 NETIF_F_HW_CSUM | 2797 NETIF_F_SCTP_CRC; 2798 2799 #define IGBVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \ 2800 NETIF_F_GSO_GRE_CSUM | \ 2801 NETIF_F_GSO_IPXIP4 | \ 2802 NETIF_F_GSO_IPXIP6 | \ 2803 NETIF_F_GSO_UDP_TUNNEL | \ 2804 NETIF_F_GSO_UDP_TUNNEL_CSUM) 2805 2806 netdev->gso_partial_features = IGBVF_GSO_PARTIAL_FEATURES; 2807 netdev->hw_features |= NETIF_F_GSO_PARTIAL | 2808 IGBVF_GSO_PARTIAL_FEATURES; 2809 2810 netdev->features = netdev->hw_features | NETIF_F_HIGHDMA; 2811 2812 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID; 2813 netdev->mpls_features |= NETIF_F_HW_CSUM; 2814 netdev->hw_enc_features |= netdev->vlan_features; 2815 2816 /* set this bit last since it cannot be part of vlan_features */ 2817 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER | 2818 NETIF_F_HW_VLAN_CTAG_RX | 2819 NETIF_F_HW_VLAN_CTAG_TX; 2820 2821 /* MTU range: 68 - 9216 */ 2822 netdev->min_mtu = ETH_MIN_MTU; 2823 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE; 2824 2825 spin_lock_bh(&hw->mbx_lock); 2826 2827 /*reset the controller to put the device in a known good state */ 2828 err = hw->mac.ops.reset_hw(hw); 2829 if (err) { 2830 dev_info(&pdev->dev, 2831 "PF still in reset state. Is the PF interface up?\n"); 2832 } else { 2833 err = hw->mac.ops.read_mac_addr(hw); 2834 if (err) 2835 dev_info(&pdev->dev, "Error reading MAC address.\n"); 2836 else if (is_zero_ether_addr(adapter->hw.mac.addr)) 2837 dev_info(&pdev->dev, 2838 "MAC address not assigned by administrator.\n"); 2839 eth_hw_addr_set(netdev, adapter->hw.mac.addr); 2840 } 2841 2842 spin_unlock_bh(&hw->mbx_lock); 2843 2844 if (!is_valid_ether_addr(netdev->dev_addr)) { 2845 dev_info(&pdev->dev, "Assigning random MAC address.\n"); 2846 eth_hw_addr_random(netdev); 2847 memcpy(adapter->hw.mac.addr, netdev->dev_addr, 2848 netdev->addr_len); 2849 } 2850 2851 timer_setup(&adapter->watchdog_timer, igbvf_watchdog, 0); 2852 2853 INIT_WORK(&adapter->reset_task, igbvf_reset_task); 2854 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task); 2855 2856 /* ring size defaults */ 2857 adapter->rx_ring->count = 1024; 2858 adapter->tx_ring->count = 1024; 2859 2860 /* reset the hardware with the new settings */ 2861 igbvf_reset(adapter); 2862 2863 /* set hardware-specific flags */ 2864 if (adapter->hw.mac.type == e1000_vfadapt_i350) 2865 adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP; 2866 2867 strcpy(netdev->name, "eth%d"); 2868 err = register_netdev(netdev); 2869 if (err) 2870 goto err_hw_init; 2871 2872 /* tell the stack to leave us alone until igbvf_open() is called */ 2873 netif_carrier_off(netdev); 2874 netif_stop_queue(netdev); 2875 2876 igbvf_print_device_info(adapter); 2877 2878 igbvf_initialize_last_counter_stats(adapter); 2879 2880 return 0; 2881 2882 err_hw_init: 2883 netif_napi_del(&adapter->rx_ring->napi); 2884 kfree(adapter->tx_ring); 2885 kfree(adapter->rx_ring); 2886 err_sw_init: 2887 igbvf_reset_interrupt_capability(adapter); 2888 err_get_variants: 2889 iounmap(adapter->hw.hw_addr); 2890 err_ioremap: 2891 free_netdev(netdev); 2892 err_alloc_etherdev: 2893 pci_release_regions(pdev); 2894 err_pci_reg: 2895 err_dma: 2896 pci_disable_device(pdev); 2897 return err; 2898 } 2899 2900 /** 2901 * igbvf_remove - Device Removal Routine 2902 * @pdev: PCI device information struct 2903 * 2904 * igbvf_remove is called by the PCI subsystem to alert the driver 2905 * that it should release a PCI device. The could be caused by a 2906 * Hot-Plug event, or because the driver is going to be removed from 2907 * memory. 2908 **/ 2909 static void igbvf_remove(struct pci_dev *pdev) 2910 { 2911 struct net_device *netdev = pci_get_drvdata(pdev); 2912 struct igbvf_adapter *adapter = netdev_priv(netdev); 2913 struct e1000_hw *hw = &adapter->hw; 2914 2915 /* The watchdog timer may be rescheduled, so explicitly 2916 * disable it from being rescheduled. 2917 */ 2918 set_bit(__IGBVF_DOWN, &adapter->state); 2919 del_timer_sync(&adapter->watchdog_timer); 2920 2921 cancel_work_sync(&adapter->reset_task); 2922 cancel_work_sync(&adapter->watchdog_task); 2923 2924 unregister_netdev(netdev); 2925 2926 igbvf_reset_interrupt_capability(adapter); 2927 2928 /* it is important to delete the NAPI struct prior to freeing the 2929 * Rx ring so that you do not end up with null pointer refs 2930 */ 2931 netif_napi_del(&adapter->rx_ring->napi); 2932 kfree(adapter->tx_ring); 2933 kfree(adapter->rx_ring); 2934 2935 iounmap(hw->hw_addr); 2936 if (hw->flash_address) 2937 iounmap(hw->flash_address); 2938 pci_release_regions(pdev); 2939 2940 free_netdev(netdev); 2941 2942 pci_disable_device(pdev); 2943 } 2944 2945 /* PCI Error Recovery (ERS) */ 2946 static const struct pci_error_handlers igbvf_err_handler = { 2947 .error_detected = igbvf_io_error_detected, 2948 .slot_reset = igbvf_io_slot_reset, 2949 .resume = igbvf_io_resume, 2950 .reset_prepare = igbvf_io_prepare, 2951 .reset_done = igbvf_io_reset_done, 2952 }; 2953 2954 static const struct pci_device_id igbvf_pci_tbl[] = { 2955 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf }, 2956 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf }, 2957 { } /* terminate list */ 2958 }; 2959 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl); 2960 2961 static SIMPLE_DEV_PM_OPS(igbvf_pm_ops, igbvf_suspend, igbvf_resume); 2962 2963 /* PCI Device API Driver */ 2964 static struct pci_driver igbvf_driver = { 2965 .name = igbvf_driver_name, 2966 .id_table = igbvf_pci_tbl, 2967 .probe = igbvf_probe, 2968 .remove = igbvf_remove, 2969 .driver.pm = &igbvf_pm_ops, 2970 .shutdown = igbvf_shutdown, 2971 .err_handler = &igbvf_err_handler 2972 }; 2973 2974 /** 2975 * igbvf_init_module - Driver Registration Routine 2976 * 2977 * igbvf_init_module is the first routine called when the driver is 2978 * loaded. All it does is register with the PCI subsystem. 2979 **/ 2980 static int __init igbvf_init_module(void) 2981 { 2982 int ret; 2983 2984 pr_info("%s\n", igbvf_driver_string); 2985 pr_info("%s\n", igbvf_copyright); 2986 2987 ret = pci_register_driver(&igbvf_driver); 2988 2989 return ret; 2990 } 2991 module_init(igbvf_init_module); 2992 2993 /** 2994 * igbvf_exit_module - Driver Exit Cleanup Routine 2995 * 2996 * igbvf_exit_module is called just before the driver is removed 2997 * from memory. 2998 **/ 2999 static void __exit igbvf_exit_module(void) 3000 { 3001 pci_unregister_driver(&igbvf_driver); 3002 } 3003 module_exit(igbvf_exit_module); 3004 3005 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>"); 3006 MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver"); 3007 MODULE_LICENSE("GPL v2"); 3008 3009 /* netdev.c */ 3010