1 /******************************************************************************* 2 3 Intel(R) 82576 Virtual Function Linux driver 4 Copyright(c) 2009 - 2012 Intel Corporation. 5 6 This program is free software; you can redistribute it and/or modify it 7 under the terms and conditions of the GNU General Public License, 8 version 2, as published by the Free Software Foundation. 9 10 This program is distributed in the hope it will be useful, but WITHOUT 11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 13 more details. 14 15 You should have received a copy of the GNU General Public License along with 16 this program; if not, see <http://www.gnu.org/licenses/>. 17 18 The full GNU General Public License is included in this distribution in 19 the file called "COPYING". 20 21 Contact Information: 22 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net> 23 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497 24 25 *******************************************************************************/ 26 27 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 28 29 #include <linux/module.h> 30 #include <linux/types.h> 31 #include <linux/init.h> 32 #include <linux/pci.h> 33 #include <linux/vmalloc.h> 34 #include <linux/pagemap.h> 35 #include <linux/delay.h> 36 #include <linux/netdevice.h> 37 #include <linux/tcp.h> 38 #include <linux/ipv6.h> 39 #include <linux/slab.h> 40 #include <net/checksum.h> 41 #include <net/ip6_checksum.h> 42 #include <linux/mii.h> 43 #include <linux/ethtool.h> 44 #include <linux/if_vlan.h> 45 #include <linux/prefetch.h> 46 #include <linux/sctp.h> 47 48 #include "igbvf.h" 49 50 #define DRV_VERSION "2.4.0-k" 51 char igbvf_driver_name[] = "igbvf"; 52 const char igbvf_driver_version[] = DRV_VERSION; 53 static const char igbvf_driver_string[] = 54 "Intel(R) Gigabit Virtual Function Network Driver"; 55 static const char igbvf_copyright[] = 56 "Copyright (c) 2009 - 2012 Intel Corporation."; 57 58 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK) 59 static int debug = -1; 60 module_param(debug, int, 0); 61 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)"); 62 63 static int igbvf_poll(struct napi_struct *napi, int budget); 64 static void igbvf_reset(struct igbvf_adapter *); 65 static void igbvf_set_interrupt_capability(struct igbvf_adapter *); 66 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *); 67 68 static struct igbvf_info igbvf_vf_info = { 69 .mac = e1000_vfadapt, 70 .flags = 0, 71 .pba = 10, 72 .init_ops = e1000_init_function_pointers_vf, 73 }; 74 75 static struct igbvf_info igbvf_i350_vf_info = { 76 .mac = e1000_vfadapt_i350, 77 .flags = 0, 78 .pba = 10, 79 .init_ops = e1000_init_function_pointers_vf, 80 }; 81 82 static const struct igbvf_info *igbvf_info_tbl[] = { 83 [board_vf] = &igbvf_vf_info, 84 [board_i350_vf] = &igbvf_i350_vf_info, 85 }; 86 87 /** 88 * igbvf_desc_unused - calculate if we have unused descriptors 89 * @rx_ring: address of receive ring structure 90 **/ 91 static int igbvf_desc_unused(struct igbvf_ring *ring) 92 { 93 if (ring->next_to_clean > ring->next_to_use) 94 return ring->next_to_clean - ring->next_to_use - 1; 95 96 return ring->count + ring->next_to_clean - ring->next_to_use - 1; 97 } 98 99 /** 100 * igbvf_receive_skb - helper function to handle Rx indications 101 * @adapter: board private structure 102 * @status: descriptor status field as written by hardware 103 * @vlan: descriptor vlan field as written by hardware (no le/be conversion) 104 * @skb: pointer to sk_buff to be indicated to stack 105 **/ 106 static void igbvf_receive_skb(struct igbvf_adapter *adapter, 107 struct net_device *netdev, 108 struct sk_buff *skb, 109 u32 status, u16 vlan) 110 { 111 u16 vid; 112 113 if (status & E1000_RXD_STAT_VP) { 114 if ((adapter->flags & IGBVF_FLAG_RX_LB_VLAN_BSWAP) && 115 (status & E1000_RXDEXT_STATERR_LB)) 116 vid = be16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK; 117 else 118 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK; 119 if (test_bit(vid, adapter->active_vlans)) 120 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid); 121 } 122 123 napi_gro_receive(&adapter->rx_ring->napi, skb); 124 } 125 126 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter, 127 u32 status_err, struct sk_buff *skb) 128 { 129 skb_checksum_none_assert(skb); 130 131 /* Ignore Checksum bit is set or checksum is disabled through ethtool */ 132 if ((status_err & E1000_RXD_STAT_IXSM) || 133 (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED)) 134 return; 135 136 /* TCP/UDP checksum error bit is set */ 137 if (status_err & 138 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) { 139 /* let the stack verify checksum errors */ 140 adapter->hw_csum_err++; 141 return; 142 } 143 144 /* It must be a TCP or UDP packet with a valid checksum */ 145 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)) 146 skb->ip_summed = CHECKSUM_UNNECESSARY; 147 148 adapter->hw_csum_good++; 149 } 150 151 /** 152 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split 153 * @rx_ring: address of ring structure to repopulate 154 * @cleaned_count: number of buffers to repopulate 155 **/ 156 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring, 157 int cleaned_count) 158 { 159 struct igbvf_adapter *adapter = rx_ring->adapter; 160 struct net_device *netdev = adapter->netdev; 161 struct pci_dev *pdev = adapter->pdev; 162 union e1000_adv_rx_desc *rx_desc; 163 struct igbvf_buffer *buffer_info; 164 struct sk_buff *skb; 165 unsigned int i; 166 int bufsz; 167 168 i = rx_ring->next_to_use; 169 buffer_info = &rx_ring->buffer_info[i]; 170 171 if (adapter->rx_ps_hdr_size) 172 bufsz = adapter->rx_ps_hdr_size; 173 else 174 bufsz = adapter->rx_buffer_len; 175 176 while (cleaned_count--) { 177 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i); 178 179 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) { 180 if (!buffer_info->page) { 181 buffer_info->page = alloc_page(GFP_ATOMIC); 182 if (!buffer_info->page) { 183 adapter->alloc_rx_buff_failed++; 184 goto no_buffers; 185 } 186 buffer_info->page_offset = 0; 187 } else { 188 buffer_info->page_offset ^= PAGE_SIZE / 2; 189 } 190 buffer_info->page_dma = 191 dma_map_page(&pdev->dev, buffer_info->page, 192 buffer_info->page_offset, 193 PAGE_SIZE / 2, 194 DMA_FROM_DEVICE); 195 if (dma_mapping_error(&pdev->dev, 196 buffer_info->page_dma)) { 197 __free_page(buffer_info->page); 198 buffer_info->page = NULL; 199 dev_err(&pdev->dev, "RX DMA map failed\n"); 200 break; 201 } 202 } 203 204 if (!buffer_info->skb) { 205 skb = netdev_alloc_skb_ip_align(netdev, bufsz); 206 if (!skb) { 207 adapter->alloc_rx_buff_failed++; 208 goto no_buffers; 209 } 210 211 buffer_info->skb = skb; 212 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, 213 bufsz, 214 DMA_FROM_DEVICE); 215 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) { 216 dev_kfree_skb(buffer_info->skb); 217 buffer_info->skb = NULL; 218 dev_err(&pdev->dev, "RX DMA map failed\n"); 219 goto no_buffers; 220 } 221 } 222 /* Refresh the desc even if buffer_addrs didn't change because 223 * each write-back erases this info. 224 */ 225 if (adapter->rx_ps_hdr_size) { 226 rx_desc->read.pkt_addr = 227 cpu_to_le64(buffer_info->page_dma); 228 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma); 229 } else { 230 rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->dma); 231 rx_desc->read.hdr_addr = 0; 232 } 233 234 i++; 235 if (i == rx_ring->count) 236 i = 0; 237 buffer_info = &rx_ring->buffer_info[i]; 238 } 239 240 no_buffers: 241 if (rx_ring->next_to_use != i) { 242 rx_ring->next_to_use = i; 243 if (i == 0) 244 i = (rx_ring->count - 1); 245 else 246 i--; 247 248 /* Force memory writes to complete before letting h/w 249 * know there are new descriptors to fetch. (Only 250 * applicable for weak-ordered memory model archs, 251 * such as IA-64). 252 */ 253 wmb(); 254 writel(i, adapter->hw.hw_addr + rx_ring->tail); 255 } 256 } 257 258 /** 259 * igbvf_clean_rx_irq - Send received data up the network stack; legacy 260 * @adapter: board private structure 261 * 262 * the return value indicates whether actual cleaning was done, there 263 * is no guarantee that everything was cleaned 264 **/ 265 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter, 266 int *work_done, int work_to_do) 267 { 268 struct igbvf_ring *rx_ring = adapter->rx_ring; 269 struct net_device *netdev = adapter->netdev; 270 struct pci_dev *pdev = adapter->pdev; 271 union e1000_adv_rx_desc *rx_desc, *next_rxd; 272 struct igbvf_buffer *buffer_info, *next_buffer; 273 struct sk_buff *skb; 274 bool cleaned = false; 275 int cleaned_count = 0; 276 unsigned int total_bytes = 0, total_packets = 0; 277 unsigned int i; 278 u32 length, hlen, staterr; 279 280 i = rx_ring->next_to_clean; 281 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i); 282 staterr = le32_to_cpu(rx_desc->wb.upper.status_error); 283 284 while (staterr & E1000_RXD_STAT_DD) { 285 if (*work_done >= work_to_do) 286 break; 287 (*work_done)++; 288 rmb(); /* read descriptor and rx_buffer_info after status DD */ 289 290 buffer_info = &rx_ring->buffer_info[i]; 291 292 /* HW will not DMA in data larger than the given buffer, even 293 * if it parses the (NFS, of course) header to be larger. In 294 * that case, it fills the header buffer and spills the rest 295 * into the page. 296 */ 297 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) 298 & E1000_RXDADV_HDRBUFLEN_MASK) >> 299 E1000_RXDADV_HDRBUFLEN_SHIFT; 300 if (hlen > adapter->rx_ps_hdr_size) 301 hlen = adapter->rx_ps_hdr_size; 302 303 length = le16_to_cpu(rx_desc->wb.upper.length); 304 cleaned = true; 305 cleaned_count++; 306 307 skb = buffer_info->skb; 308 prefetch(skb->data - NET_IP_ALIGN); 309 buffer_info->skb = NULL; 310 if (!adapter->rx_ps_hdr_size) { 311 dma_unmap_single(&pdev->dev, buffer_info->dma, 312 adapter->rx_buffer_len, 313 DMA_FROM_DEVICE); 314 buffer_info->dma = 0; 315 skb_put(skb, length); 316 goto send_up; 317 } 318 319 if (!skb_shinfo(skb)->nr_frags) { 320 dma_unmap_single(&pdev->dev, buffer_info->dma, 321 adapter->rx_ps_hdr_size, 322 DMA_FROM_DEVICE); 323 buffer_info->dma = 0; 324 skb_put(skb, hlen); 325 } 326 327 if (length) { 328 dma_unmap_page(&pdev->dev, buffer_info->page_dma, 329 PAGE_SIZE / 2, 330 DMA_FROM_DEVICE); 331 buffer_info->page_dma = 0; 332 333 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags, 334 buffer_info->page, 335 buffer_info->page_offset, 336 length); 337 338 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) || 339 (page_count(buffer_info->page) != 1)) 340 buffer_info->page = NULL; 341 else 342 get_page(buffer_info->page); 343 344 skb->len += length; 345 skb->data_len += length; 346 skb->truesize += PAGE_SIZE / 2; 347 } 348 send_up: 349 i++; 350 if (i == rx_ring->count) 351 i = 0; 352 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i); 353 prefetch(next_rxd); 354 next_buffer = &rx_ring->buffer_info[i]; 355 356 if (!(staterr & E1000_RXD_STAT_EOP)) { 357 buffer_info->skb = next_buffer->skb; 358 buffer_info->dma = next_buffer->dma; 359 next_buffer->skb = skb; 360 next_buffer->dma = 0; 361 goto next_desc; 362 } 363 364 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) { 365 dev_kfree_skb_irq(skb); 366 goto next_desc; 367 } 368 369 total_bytes += skb->len; 370 total_packets++; 371 372 igbvf_rx_checksum_adv(adapter, staterr, skb); 373 374 skb->protocol = eth_type_trans(skb, netdev); 375 376 igbvf_receive_skb(adapter, netdev, skb, staterr, 377 rx_desc->wb.upper.vlan); 378 379 next_desc: 380 rx_desc->wb.upper.status_error = 0; 381 382 /* return some buffers to hardware, one at a time is too slow */ 383 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) { 384 igbvf_alloc_rx_buffers(rx_ring, cleaned_count); 385 cleaned_count = 0; 386 } 387 388 /* use prefetched values */ 389 rx_desc = next_rxd; 390 buffer_info = next_buffer; 391 392 staterr = le32_to_cpu(rx_desc->wb.upper.status_error); 393 } 394 395 rx_ring->next_to_clean = i; 396 cleaned_count = igbvf_desc_unused(rx_ring); 397 398 if (cleaned_count) 399 igbvf_alloc_rx_buffers(rx_ring, cleaned_count); 400 401 adapter->total_rx_packets += total_packets; 402 adapter->total_rx_bytes += total_bytes; 403 netdev->stats.rx_bytes += total_bytes; 404 netdev->stats.rx_packets += total_packets; 405 return cleaned; 406 } 407 408 static void igbvf_put_txbuf(struct igbvf_adapter *adapter, 409 struct igbvf_buffer *buffer_info) 410 { 411 if (buffer_info->dma) { 412 if (buffer_info->mapped_as_page) 413 dma_unmap_page(&adapter->pdev->dev, 414 buffer_info->dma, 415 buffer_info->length, 416 DMA_TO_DEVICE); 417 else 418 dma_unmap_single(&adapter->pdev->dev, 419 buffer_info->dma, 420 buffer_info->length, 421 DMA_TO_DEVICE); 422 buffer_info->dma = 0; 423 } 424 if (buffer_info->skb) { 425 dev_kfree_skb_any(buffer_info->skb); 426 buffer_info->skb = NULL; 427 } 428 buffer_info->time_stamp = 0; 429 } 430 431 /** 432 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors) 433 * @adapter: board private structure 434 * 435 * Return 0 on success, negative on failure 436 **/ 437 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter, 438 struct igbvf_ring *tx_ring) 439 { 440 struct pci_dev *pdev = adapter->pdev; 441 int size; 442 443 size = sizeof(struct igbvf_buffer) * tx_ring->count; 444 tx_ring->buffer_info = vzalloc(size); 445 if (!tx_ring->buffer_info) 446 goto err; 447 448 /* round up to nearest 4K */ 449 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc); 450 tx_ring->size = ALIGN(tx_ring->size, 4096); 451 452 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size, 453 &tx_ring->dma, GFP_KERNEL); 454 if (!tx_ring->desc) 455 goto err; 456 457 tx_ring->adapter = adapter; 458 tx_ring->next_to_use = 0; 459 tx_ring->next_to_clean = 0; 460 461 return 0; 462 err: 463 vfree(tx_ring->buffer_info); 464 dev_err(&adapter->pdev->dev, 465 "Unable to allocate memory for the transmit descriptor ring\n"); 466 return -ENOMEM; 467 } 468 469 /** 470 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors) 471 * @adapter: board private structure 472 * 473 * Returns 0 on success, negative on failure 474 **/ 475 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter, 476 struct igbvf_ring *rx_ring) 477 { 478 struct pci_dev *pdev = adapter->pdev; 479 int size, desc_len; 480 481 size = sizeof(struct igbvf_buffer) * rx_ring->count; 482 rx_ring->buffer_info = vzalloc(size); 483 if (!rx_ring->buffer_info) 484 goto err; 485 486 desc_len = sizeof(union e1000_adv_rx_desc); 487 488 /* Round up to nearest 4K */ 489 rx_ring->size = rx_ring->count * desc_len; 490 rx_ring->size = ALIGN(rx_ring->size, 4096); 491 492 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size, 493 &rx_ring->dma, GFP_KERNEL); 494 if (!rx_ring->desc) 495 goto err; 496 497 rx_ring->next_to_clean = 0; 498 rx_ring->next_to_use = 0; 499 500 rx_ring->adapter = adapter; 501 502 return 0; 503 504 err: 505 vfree(rx_ring->buffer_info); 506 rx_ring->buffer_info = NULL; 507 dev_err(&adapter->pdev->dev, 508 "Unable to allocate memory for the receive descriptor ring\n"); 509 return -ENOMEM; 510 } 511 512 /** 513 * igbvf_clean_tx_ring - Free Tx Buffers 514 * @tx_ring: ring to be cleaned 515 **/ 516 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring) 517 { 518 struct igbvf_adapter *adapter = tx_ring->adapter; 519 struct igbvf_buffer *buffer_info; 520 unsigned long size; 521 unsigned int i; 522 523 if (!tx_ring->buffer_info) 524 return; 525 526 /* Free all the Tx ring sk_buffs */ 527 for (i = 0; i < tx_ring->count; i++) { 528 buffer_info = &tx_ring->buffer_info[i]; 529 igbvf_put_txbuf(adapter, buffer_info); 530 } 531 532 size = sizeof(struct igbvf_buffer) * tx_ring->count; 533 memset(tx_ring->buffer_info, 0, size); 534 535 /* Zero out the descriptor ring */ 536 memset(tx_ring->desc, 0, tx_ring->size); 537 538 tx_ring->next_to_use = 0; 539 tx_ring->next_to_clean = 0; 540 541 writel(0, adapter->hw.hw_addr + tx_ring->head); 542 writel(0, adapter->hw.hw_addr + tx_ring->tail); 543 } 544 545 /** 546 * igbvf_free_tx_resources - Free Tx Resources per Queue 547 * @tx_ring: ring to free resources from 548 * 549 * Free all transmit software resources 550 **/ 551 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring) 552 { 553 struct pci_dev *pdev = tx_ring->adapter->pdev; 554 555 igbvf_clean_tx_ring(tx_ring); 556 557 vfree(tx_ring->buffer_info); 558 tx_ring->buffer_info = NULL; 559 560 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc, 561 tx_ring->dma); 562 563 tx_ring->desc = NULL; 564 } 565 566 /** 567 * igbvf_clean_rx_ring - Free Rx Buffers per Queue 568 * @adapter: board private structure 569 **/ 570 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring) 571 { 572 struct igbvf_adapter *adapter = rx_ring->adapter; 573 struct igbvf_buffer *buffer_info; 574 struct pci_dev *pdev = adapter->pdev; 575 unsigned long size; 576 unsigned int i; 577 578 if (!rx_ring->buffer_info) 579 return; 580 581 /* Free all the Rx ring sk_buffs */ 582 for (i = 0; i < rx_ring->count; i++) { 583 buffer_info = &rx_ring->buffer_info[i]; 584 if (buffer_info->dma) { 585 if (adapter->rx_ps_hdr_size) { 586 dma_unmap_single(&pdev->dev, buffer_info->dma, 587 adapter->rx_ps_hdr_size, 588 DMA_FROM_DEVICE); 589 } else { 590 dma_unmap_single(&pdev->dev, buffer_info->dma, 591 adapter->rx_buffer_len, 592 DMA_FROM_DEVICE); 593 } 594 buffer_info->dma = 0; 595 } 596 597 if (buffer_info->skb) { 598 dev_kfree_skb(buffer_info->skb); 599 buffer_info->skb = NULL; 600 } 601 602 if (buffer_info->page) { 603 if (buffer_info->page_dma) 604 dma_unmap_page(&pdev->dev, 605 buffer_info->page_dma, 606 PAGE_SIZE / 2, 607 DMA_FROM_DEVICE); 608 put_page(buffer_info->page); 609 buffer_info->page = NULL; 610 buffer_info->page_dma = 0; 611 buffer_info->page_offset = 0; 612 } 613 } 614 615 size = sizeof(struct igbvf_buffer) * rx_ring->count; 616 memset(rx_ring->buffer_info, 0, size); 617 618 /* Zero out the descriptor ring */ 619 memset(rx_ring->desc, 0, rx_ring->size); 620 621 rx_ring->next_to_clean = 0; 622 rx_ring->next_to_use = 0; 623 624 writel(0, adapter->hw.hw_addr + rx_ring->head); 625 writel(0, adapter->hw.hw_addr + rx_ring->tail); 626 } 627 628 /** 629 * igbvf_free_rx_resources - Free Rx Resources 630 * @rx_ring: ring to clean the resources from 631 * 632 * Free all receive software resources 633 **/ 634 635 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring) 636 { 637 struct pci_dev *pdev = rx_ring->adapter->pdev; 638 639 igbvf_clean_rx_ring(rx_ring); 640 641 vfree(rx_ring->buffer_info); 642 rx_ring->buffer_info = NULL; 643 644 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc, 645 rx_ring->dma); 646 rx_ring->desc = NULL; 647 } 648 649 /** 650 * igbvf_update_itr - update the dynamic ITR value based on statistics 651 * @adapter: pointer to adapter 652 * @itr_setting: current adapter->itr 653 * @packets: the number of packets during this measurement interval 654 * @bytes: the number of bytes during this measurement interval 655 * 656 * Stores a new ITR value based on packets and byte counts during the last 657 * interrupt. The advantage of per interrupt computation is faster updates 658 * and more accurate ITR for the current traffic pattern. Constants in this 659 * function were computed based on theoretical maximum wire speed and thresholds 660 * were set based on testing data as well as attempting to minimize response 661 * time while increasing bulk throughput. 662 **/ 663 static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter, 664 enum latency_range itr_setting, 665 int packets, int bytes) 666 { 667 enum latency_range retval = itr_setting; 668 669 if (packets == 0) 670 goto update_itr_done; 671 672 switch (itr_setting) { 673 case lowest_latency: 674 /* handle TSO and jumbo frames */ 675 if (bytes/packets > 8000) 676 retval = bulk_latency; 677 else if ((packets < 5) && (bytes > 512)) 678 retval = low_latency; 679 break; 680 case low_latency: /* 50 usec aka 20000 ints/s */ 681 if (bytes > 10000) { 682 /* this if handles the TSO accounting */ 683 if (bytes/packets > 8000) 684 retval = bulk_latency; 685 else if ((packets < 10) || ((bytes/packets) > 1200)) 686 retval = bulk_latency; 687 else if ((packets > 35)) 688 retval = lowest_latency; 689 } else if (bytes/packets > 2000) { 690 retval = bulk_latency; 691 } else if (packets <= 2 && bytes < 512) { 692 retval = lowest_latency; 693 } 694 break; 695 case bulk_latency: /* 250 usec aka 4000 ints/s */ 696 if (bytes > 25000) { 697 if (packets > 35) 698 retval = low_latency; 699 } else if (bytes < 6000) { 700 retval = low_latency; 701 } 702 break; 703 default: 704 break; 705 } 706 707 update_itr_done: 708 return retval; 709 } 710 711 static int igbvf_range_to_itr(enum latency_range current_range) 712 { 713 int new_itr; 714 715 switch (current_range) { 716 /* counts and packets in update_itr are dependent on these numbers */ 717 case lowest_latency: 718 new_itr = IGBVF_70K_ITR; 719 break; 720 case low_latency: 721 new_itr = IGBVF_20K_ITR; 722 break; 723 case bulk_latency: 724 new_itr = IGBVF_4K_ITR; 725 break; 726 default: 727 new_itr = IGBVF_START_ITR; 728 break; 729 } 730 return new_itr; 731 } 732 733 static void igbvf_set_itr(struct igbvf_adapter *adapter) 734 { 735 u32 new_itr; 736 737 adapter->tx_ring->itr_range = 738 igbvf_update_itr(adapter, 739 adapter->tx_ring->itr_val, 740 adapter->total_tx_packets, 741 adapter->total_tx_bytes); 742 743 /* conservative mode (itr 3) eliminates the lowest_latency setting */ 744 if (adapter->requested_itr == 3 && 745 adapter->tx_ring->itr_range == lowest_latency) 746 adapter->tx_ring->itr_range = low_latency; 747 748 new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range); 749 750 if (new_itr != adapter->tx_ring->itr_val) { 751 u32 current_itr = adapter->tx_ring->itr_val; 752 /* this attempts to bias the interrupt rate towards Bulk 753 * by adding intermediate steps when interrupt rate is 754 * increasing 755 */ 756 new_itr = new_itr > current_itr ? 757 min(current_itr + (new_itr >> 2), new_itr) : 758 new_itr; 759 adapter->tx_ring->itr_val = new_itr; 760 761 adapter->tx_ring->set_itr = 1; 762 } 763 764 adapter->rx_ring->itr_range = 765 igbvf_update_itr(adapter, adapter->rx_ring->itr_val, 766 adapter->total_rx_packets, 767 adapter->total_rx_bytes); 768 if (adapter->requested_itr == 3 && 769 adapter->rx_ring->itr_range == lowest_latency) 770 adapter->rx_ring->itr_range = low_latency; 771 772 new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range); 773 774 if (new_itr != adapter->rx_ring->itr_val) { 775 u32 current_itr = adapter->rx_ring->itr_val; 776 777 new_itr = new_itr > current_itr ? 778 min(current_itr + (new_itr >> 2), new_itr) : 779 new_itr; 780 adapter->rx_ring->itr_val = new_itr; 781 782 adapter->rx_ring->set_itr = 1; 783 } 784 } 785 786 /** 787 * igbvf_clean_tx_irq - Reclaim resources after transmit completes 788 * @adapter: board private structure 789 * 790 * returns true if ring is completely cleaned 791 **/ 792 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring) 793 { 794 struct igbvf_adapter *adapter = tx_ring->adapter; 795 struct net_device *netdev = adapter->netdev; 796 struct igbvf_buffer *buffer_info; 797 struct sk_buff *skb; 798 union e1000_adv_tx_desc *tx_desc, *eop_desc; 799 unsigned int total_bytes = 0, total_packets = 0; 800 unsigned int i, count = 0; 801 bool cleaned = false; 802 803 i = tx_ring->next_to_clean; 804 buffer_info = &tx_ring->buffer_info[i]; 805 eop_desc = buffer_info->next_to_watch; 806 807 do { 808 /* if next_to_watch is not set then there is no work pending */ 809 if (!eop_desc) 810 break; 811 812 /* prevent any other reads prior to eop_desc */ 813 read_barrier_depends(); 814 815 /* if DD is not set pending work has not been completed */ 816 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD))) 817 break; 818 819 /* clear next_to_watch to prevent false hangs */ 820 buffer_info->next_to_watch = NULL; 821 822 for (cleaned = false; !cleaned; count++) { 823 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i); 824 cleaned = (tx_desc == eop_desc); 825 skb = buffer_info->skb; 826 827 if (skb) { 828 unsigned int segs, bytecount; 829 830 /* gso_segs is currently only valid for tcp */ 831 segs = skb_shinfo(skb)->gso_segs ?: 1; 832 /* multiply data chunks by size of headers */ 833 bytecount = ((segs - 1) * skb_headlen(skb)) + 834 skb->len; 835 total_packets += segs; 836 total_bytes += bytecount; 837 } 838 839 igbvf_put_txbuf(adapter, buffer_info); 840 tx_desc->wb.status = 0; 841 842 i++; 843 if (i == tx_ring->count) 844 i = 0; 845 846 buffer_info = &tx_ring->buffer_info[i]; 847 } 848 849 eop_desc = buffer_info->next_to_watch; 850 } while (count < tx_ring->count); 851 852 tx_ring->next_to_clean = i; 853 854 if (unlikely(count && netif_carrier_ok(netdev) && 855 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) { 856 /* Make sure that anybody stopping the queue after this 857 * sees the new next_to_clean. 858 */ 859 smp_mb(); 860 if (netif_queue_stopped(netdev) && 861 !(test_bit(__IGBVF_DOWN, &adapter->state))) { 862 netif_wake_queue(netdev); 863 ++adapter->restart_queue; 864 } 865 } 866 867 netdev->stats.tx_bytes += total_bytes; 868 netdev->stats.tx_packets += total_packets; 869 return count < tx_ring->count; 870 } 871 872 static irqreturn_t igbvf_msix_other(int irq, void *data) 873 { 874 struct net_device *netdev = data; 875 struct igbvf_adapter *adapter = netdev_priv(netdev); 876 struct e1000_hw *hw = &adapter->hw; 877 878 adapter->int_counter1++; 879 880 hw->mac.get_link_status = 1; 881 if (!test_bit(__IGBVF_DOWN, &adapter->state)) 882 mod_timer(&adapter->watchdog_timer, jiffies + 1); 883 884 ew32(EIMS, adapter->eims_other); 885 886 return IRQ_HANDLED; 887 } 888 889 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data) 890 { 891 struct net_device *netdev = data; 892 struct igbvf_adapter *adapter = netdev_priv(netdev); 893 struct e1000_hw *hw = &adapter->hw; 894 struct igbvf_ring *tx_ring = adapter->tx_ring; 895 896 if (tx_ring->set_itr) { 897 writel(tx_ring->itr_val, 898 adapter->hw.hw_addr + tx_ring->itr_register); 899 adapter->tx_ring->set_itr = 0; 900 } 901 902 adapter->total_tx_bytes = 0; 903 adapter->total_tx_packets = 0; 904 905 /* auto mask will automatically re-enable the interrupt when we write 906 * EICS 907 */ 908 if (!igbvf_clean_tx_irq(tx_ring)) 909 /* Ring was not completely cleaned, so fire another interrupt */ 910 ew32(EICS, tx_ring->eims_value); 911 else 912 ew32(EIMS, tx_ring->eims_value); 913 914 return IRQ_HANDLED; 915 } 916 917 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data) 918 { 919 struct net_device *netdev = data; 920 struct igbvf_adapter *adapter = netdev_priv(netdev); 921 922 adapter->int_counter0++; 923 924 /* Write the ITR value calculated at the end of the 925 * previous interrupt. 926 */ 927 if (adapter->rx_ring->set_itr) { 928 writel(adapter->rx_ring->itr_val, 929 adapter->hw.hw_addr + adapter->rx_ring->itr_register); 930 adapter->rx_ring->set_itr = 0; 931 } 932 933 if (napi_schedule_prep(&adapter->rx_ring->napi)) { 934 adapter->total_rx_bytes = 0; 935 adapter->total_rx_packets = 0; 936 __napi_schedule(&adapter->rx_ring->napi); 937 } 938 939 return IRQ_HANDLED; 940 } 941 942 #define IGBVF_NO_QUEUE -1 943 944 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue, 945 int tx_queue, int msix_vector) 946 { 947 struct e1000_hw *hw = &adapter->hw; 948 u32 ivar, index; 949 950 /* 82576 uses a table-based method for assigning vectors. 951 * Each queue has a single entry in the table to which we write 952 * a vector number along with a "valid" bit. Sadly, the layout 953 * of the table is somewhat counterintuitive. 954 */ 955 if (rx_queue > IGBVF_NO_QUEUE) { 956 index = (rx_queue >> 1); 957 ivar = array_er32(IVAR0, index); 958 if (rx_queue & 0x1) { 959 /* vector goes into third byte of register */ 960 ivar = ivar & 0xFF00FFFF; 961 ivar |= (msix_vector | E1000_IVAR_VALID) << 16; 962 } else { 963 /* vector goes into low byte of register */ 964 ivar = ivar & 0xFFFFFF00; 965 ivar |= msix_vector | E1000_IVAR_VALID; 966 } 967 adapter->rx_ring[rx_queue].eims_value = BIT(msix_vector); 968 array_ew32(IVAR0, index, ivar); 969 } 970 if (tx_queue > IGBVF_NO_QUEUE) { 971 index = (tx_queue >> 1); 972 ivar = array_er32(IVAR0, index); 973 if (tx_queue & 0x1) { 974 /* vector goes into high byte of register */ 975 ivar = ivar & 0x00FFFFFF; 976 ivar |= (msix_vector | E1000_IVAR_VALID) << 24; 977 } else { 978 /* vector goes into second byte of register */ 979 ivar = ivar & 0xFFFF00FF; 980 ivar |= (msix_vector | E1000_IVAR_VALID) << 8; 981 } 982 adapter->tx_ring[tx_queue].eims_value = BIT(msix_vector); 983 array_ew32(IVAR0, index, ivar); 984 } 985 } 986 987 /** 988 * igbvf_configure_msix - Configure MSI-X hardware 989 * @adapter: board private structure 990 * 991 * igbvf_configure_msix sets up the hardware to properly 992 * generate MSI-X interrupts. 993 **/ 994 static void igbvf_configure_msix(struct igbvf_adapter *adapter) 995 { 996 u32 tmp; 997 struct e1000_hw *hw = &adapter->hw; 998 struct igbvf_ring *tx_ring = adapter->tx_ring; 999 struct igbvf_ring *rx_ring = adapter->rx_ring; 1000 int vector = 0; 1001 1002 adapter->eims_enable_mask = 0; 1003 1004 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++); 1005 adapter->eims_enable_mask |= tx_ring->eims_value; 1006 writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register); 1007 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++); 1008 adapter->eims_enable_mask |= rx_ring->eims_value; 1009 writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register); 1010 1011 /* set vector for other causes, i.e. link changes */ 1012 1013 tmp = (vector++ | E1000_IVAR_VALID); 1014 1015 ew32(IVAR_MISC, tmp); 1016 1017 adapter->eims_enable_mask = GENMASK(vector - 1, 0); 1018 adapter->eims_other = BIT(vector - 1); 1019 e1e_flush(); 1020 } 1021 1022 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter) 1023 { 1024 if (adapter->msix_entries) { 1025 pci_disable_msix(adapter->pdev); 1026 kfree(adapter->msix_entries); 1027 adapter->msix_entries = NULL; 1028 } 1029 } 1030 1031 /** 1032 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported 1033 * @adapter: board private structure 1034 * 1035 * Attempt to configure interrupts using the best available 1036 * capabilities of the hardware and kernel. 1037 **/ 1038 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter) 1039 { 1040 int err = -ENOMEM; 1041 int i; 1042 1043 /* we allocate 3 vectors, 1 for Tx, 1 for Rx, one for PF messages */ 1044 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry), 1045 GFP_KERNEL); 1046 if (adapter->msix_entries) { 1047 for (i = 0; i < 3; i++) 1048 adapter->msix_entries[i].entry = i; 1049 1050 err = pci_enable_msix_range(adapter->pdev, 1051 adapter->msix_entries, 3, 3); 1052 } 1053 1054 if (err < 0) { 1055 /* MSI-X failed */ 1056 dev_err(&adapter->pdev->dev, 1057 "Failed to initialize MSI-X interrupts.\n"); 1058 igbvf_reset_interrupt_capability(adapter); 1059 } 1060 } 1061 1062 /** 1063 * igbvf_request_msix - Initialize MSI-X interrupts 1064 * @adapter: board private structure 1065 * 1066 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the 1067 * kernel. 1068 **/ 1069 static int igbvf_request_msix(struct igbvf_adapter *adapter) 1070 { 1071 struct net_device *netdev = adapter->netdev; 1072 int err = 0, vector = 0; 1073 1074 if (strlen(netdev->name) < (IFNAMSIZ - 5)) { 1075 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name); 1076 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name); 1077 } else { 1078 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ); 1079 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ); 1080 } 1081 1082 err = request_irq(adapter->msix_entries[vector].vector, 1083 igbvf_intr_msix_tx, 0, adapter->tx_ring->name, 1084 netdev); 1085 if (err) 1086 goto out; 1087 1088 adapter->tx_ring->itr_register = E1000_EITR(vector); 1089 adapter->tx_ring->itr_val = adapter->current_itr; 1090 vector++; 1091 1092 err = request_irq(adapter->msix_entries[vector].vector, 1093 igbvf_intr_msix_rx, 0, adapter->rx_ring->name, 1094 netdev); 1095 if (err) 1096 goto out; 1097 1098 adapter->rx_ring->itr_register = E1000_EITR(vector); 1099 adapter->rx_ring->itr_val = adapter->current_itr; 1100 vector++; 1101 1102 err = request_irq(adapter->msix_entries[vector].vector, 1103 igbvf_msix_other, 0, netdev->name, netdev); 1104 if (err) 1105 goto out; 1106 1107 igbvf_configure_msix(adapter); 1108 return 0; 1109 out: 1110 return err; 1111 } 1112 1113 /** 1114 * igbvf_alloc_queues - Allocate memory for all rings 1115 * @adapter: board private structure to initialize 1116 **/ 1117 static int igbvf_alloc_queues(struct igbvf_adapter *adapter) 1118 { 1119 struct net_device *netdev = adapter->netdev; 1120 1121 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL); 1122 if (!adapter->tx_ring) 1123 return -ENOMEM; 1124 1125 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL); 1126 if (!adapter->rx_ring) { 1127 kfree(adapter->tx_ring); 1128 return -ENOMEM; 1129 } 1130 1131 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64); 1132 1133 return 0; 1134 } 1135 1136 /** 1137 * igbvf_request_irq - initialize interrupts 1138 * @adapter: board private structure 1139 * 1140 * Attempts to configure interrupts using the best available 1141 * capabilities of the hardware and kernel. 1142 **/ 1143 static int igbvf_request_irq(struct igbvf_adapter *adapter) 1144 { 1145 int err = -1; 1146 1147 /* igbvf supports msi-x only */ 1148 if (adapter->msix_entries) 1149 err = igbvf_request_msix(adapter); 1150 1151 if (!err) 1152 return err; 1153 1154 dev_err(&adapter->pdev->dev, 1155 "Unable to allocate interrupt, Error: %d\n", err); 1156 1157 return err; 1158 } 1159 1160 static void igbvf_free_irq(struct igbvf_adapter *adapter) 1161 { 1162 struct net_device *netdev = adapter->netdev; 1163 int vector; 1164 1165 if (adapter->msix_entries) { 1166 for (vector = 0; vector < 3; vector++) 1167 free_irq(adapter->msix_entries[vector].vector, netdev); 1168 } 1169 } 1170 1171 /** 1172 * igbvf_irq_disable - Mask off interrupt generation on the NIC 1173 * @adapter: board private structure 1174 **/ 1175 static void igbvf_irq_disable(struct igbvf_adapter *adapter) 1176 { 1177 struct e1000_hw *hw = &adapter->hw; 1178 1179 ew32(EIMC, ~0); 1180 1181 if (adapter->msix_entries) 1182 ew32(EIAC, 0); 1183 } 1184 1185 /** 1186 * igbvf_irq_enable - Enable default interrupt generation settings 1187 * @adapter: board private structure 1188 **/ 1189 static void igbvf_irq_enable(struct igbvf_adapter *adapter) 1190 { 1191 struct e1000_hw *hw = &adapter->hw; 1192 1193 ew32(EIAC, adapter->eims_enable_mask); 1194 ew32(EIAM, adapter->eims_enable_mask); 1195 ew32(EIMS, adapter->eims_enable_mask); 1196 } 1197 1198 /** 1199 * igbvf_poll - NAPI Rx polling callback 1200 * @napi: struct associated with this polling callback 1201 * @budget: amount of packets driver is allowed to process this poll 1202 **/ 1203 static int igbvf_poll(struct napi_struct *napi, int budget) 1204 { 1205 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi); 1206 struct igbvf_adapter *adapter = rx_ring->adapter; 1207 struct e1000_hw *hw = &adapter->hw; 1208 int work_done = 0; 1209 1210 igbvf_clean_rx_irq(adapter, &work_done, budget); 1211 1212 /* If not enough Rx work done, exit the polling mode */ 1213 if (work_done < budget) { 1214 napi_complete_done(napi, work_done); 1215 1216 if (adapter->requested_itr & 3) 1217 igbvf_set_itr(adapter); 1218 1219 if (!test_bit(__IGBVF_DOWN, &adapter->state)) 1220 ew32(EIMS, adapter->rx_ring->eims_value); 1221 } 1222 1223 return work_done; 1224 } 1225 1226 /** 1227 * igbvf_set_rlpml - set receive large packet maximum length 1228 * @adapter: board private structure 1229 * 1230 * Configure the maximum size of packets that will be received 1231 */ 1232 static void igbvf_set_rlpml(struct igbvf_adapter *adapter) 1233 { 1234 int max_frame_size; 1235 struct e1000_hw *hw = &adapter->hw; 1236 1237 max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE; 1238 e1000_rlpml_set_vf(hw, max_frame_size); 1239 } 1240 1241 static int igbvf_vlan_rx_add_vid(struct net_device *netdev, 1242 __be16 proto, u16 vid) 1243 { 1244 struct igbvf_adapter *adapter = netdev_priv(netdev); 1245 struct e1000_hw *hw = &adapter->hw; 1246 1247 if (hw->mac.ops.set_vfta(hw, vid, true)) { 1248 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid); 1249 return -EINVAL; 1250 } 1251 set_bit(vid, adapter->active_vlans); 1252 return 0; 1253 } 1254 1255 static int igbvf_vlan_rx_kill_vid(struct net_device *netdev, 1256 __be16 proto, u16 vid) 1257 { 1258 struct igbvf_adapter *adapter = netdev_priv(netdev); 1259 struct e1000_hw *hw = &adapter->hw; 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 return -EINVAL; 1265 } 1266 clear_bit(vid, adapter->active_vlans); 1267 return 0; 1268 } 1269 1270 static void igbvf_restore_vlan(struct igbvf_adapter *adapter) 1271 { 1272 u16 vid; 1273 1274 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID) 1275 igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid); 1276 } 1277 1278 /** 1279 * igbvf_configure_tx - Configure Transmit Unit after Reset 1280 * @adapter: board private structure 1281 * 1282 * Configure the Tx unit of the MAC after a reset. 1283 **/ 1284 static void igbvf_configure_tx(struct igbvf_adapter *adapter) 1285 { 1286 struct e1000_hw *hw = &adapter->hw; 1287 struct igbvf_ring *tx_ring = adapter->tx_ring; 1288 u64 tdba; 1289 u32 txdctl, dca_txctrl; 1290 1291 /* disable transmits */ 1292 txdctl = er32(TXDCTL(0)); 1293 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE); 1294 e1e_flush(); 1295 msleep(10); 1296 1297 /* Setup the HW Tx Head and Tail descriptor pointers */ 1298 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc)); 1299 tdba = tx_ring->dma; 1300 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32))); 1301 ew32(TDBAH(0), (tdba >> 32)); 1302 ew32(TDH(0), 0); 1303 ew32(TDT(0), 0); 1304 tx_ring->head = E1000_TDH(0); 1305 tx_ring->tail = E1000_TDT(0); 1306 1307 /* Turn off Relaxed Ordering on head write-backs. The writebacks 1308 * MUST be delivered in order or it will completely screw up 1309 * our bookkeeping. 1310 */ 1311 dca_txctrl = er32(DCA_TXCTRL(0)); 1312 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN; 1313 ew32(DCA_TXCTRL(0), dca_txctrl); 1314 1315 /* enable transmits */ 1316 txdctl |= E1000_TXDCTL_QUEUE_ENABLE; 1317 ew32(TXDCTL(0), txdctl); 1318 1319 /* Setup Transmit Descriptor Settings for eop descriptor */ 1320 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS; 1321 1322 /* enable Report Status bit */ 1323 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS; 1324 } 1325 1326 /** 1327 * igbvf_setup_srrctl - configure the receive control registers 1328 * @adapter: Board private structure 1329 **/ 1330 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter) 1331 { 1332 struct e1000_hw *hw = &adapter->hw; 1333 u32 srrctl = 0; 1334 1335 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK | 1336 E1000_SRRCTL_BSIZEHDR_MASK | 1337 E1000_SRRCTL_BSIZEPKT_MASK); 1338 1339 /* Enable queue drop to avoid head of line blocking */ 1340 srrctl |= E1000_SRRCTL_DROP_EN; 1341 1342 /* Setup buffer sizes */ 1343 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >> 1344 E1000_SRRCTL_BSIZEPKT_SHIFT; 1345 1346 if (adapter->rx_buffer_len < 2048) { 1347 adapter->rx_ps_hdr_size = 0; 1348 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF; 1349 } else { 1350 adapter->rx_ps_hdr_size = 128; 1351 srrctl |= adapter->rx_ps_hdr_size << 1352 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT; 1353 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS; 1354 } 1355 1356 ew32(SRRCTL(0), srrctl); 1357 } 1358 1359 /** 1360 * igbvf_configure_rx - Configure Receive Unit after Reset 1361 * @adapter: board private structure 1362 * 1363 * Configure the Rx unit of the MAC after a reset. 1364 **/ 1365 static void igbvf_configure_rx(struct igbvf_adapter *adapter) 1366 { 1367 struct e1000_hw *hw = &adapter->hw; 1368 struct igbvf_ring *rx_ring = adapter->rx_ring; 1369 u64 rdba; 1370 u32 rxdctl; 1371 1372 /* disable receives */ 1373 rxdctl = er32(RXDCTL(0)); 1374 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE); 1375 e1e_flush(); 1376 msleep(10); 1377 1378 /* Setup the HW Rx Head and Tail Descriptor Pointers and 1379 * the Base and Length of the Rx Descriptor Ring 1380 */ 1381 rdba = rx_ring->dma; 1382 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32))); 1383 ew32(RDBAH(0), (rdba >> 32)); 1384 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc)); 1385 rx_ring->head = E1000_RDH(0); 1386 rx_ring->tail = E1000_RDT(0); 1387 ew32(RDH(0), 0); 1388 ew32(RDT(0), 0); 1389 1390 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE; 1391 rxdctl &= 0xFFF00000; 1392 rxdctl |= IGBVF_RX_PTHRESH; 1393 rxdctl |= IGBVF_RX_HTHRESH << 8; 1394 rxdctl |= IGBVF_RX_WTHRESH << 16; 1395 1396 igbvf_set_rlpml(adapter); 1397 1398 /* enable receives */ 1399 ew32(RXDCTL(0), rxdctl); 1400 } 1401 1402 /** 1403 * igbvf_set_multi - Multicast and Promiscuous mode set 1404 * @netdev: network interface device structure 1405 * 1406 * The set_multi entry point is called whenever the multicast address 1407 * list or the network interface flags are updated. This routine is 1408 * responsible for configuring the hardware for proper multicast, 1409 * promiscuous mode, and all-multi behavior. 1410 **/ 1411 static void igbvf_set_multi(struct net_device *netdev) 1412 { 1413 struct igbvf_adapter *adapter = netdev_priv(netdev); 1414 struct e1000_hw *hw = &adapter->hw; 1415 struct netdev_hw_addr *ha; 1416 u8 *mta_list = NULL; 1417 int i; 1418 1419 if (!netdev_mc_empty(netdev)) { 1420 mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN, 1421 GFP_ATOMIC); 1422 if (!mta_list) 1423 return; 1424 } 1425 1426 /* prepare a packed array of only addresses. */ 1427 i = 0; 1428 netdev_for_each_mc_addr(ha, netdev) 1429 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN); 1430 1431 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0); 1432 kfree(mta_list); 1433 } 1434 1435 /** 1436 * igbvf_set_uni - Configure unicast MAC filters 1437 * @netdev: network interface device structure 1438 * 1439 * This routine is responsible for configuring the hardware for proper 1440 * unicast filters. 1441 **/ 1442 static int igbvf_set_uni(struct net_device *netdev) 1443 { 1444 struct igbvf_adapter *adapter = netdev_priv(netdev); 1445 struct e1000_hw *hw = &adapter->hw; 1446 1447 if (netdev_uc_count(netdev) > IGBVF_MAX_MAC_FILTERS) { 1448 pr_err("Too many unicast filters - No Space\n"); 1449 return -ENOSPC; 1450 } 1451 1452 /* Clear all unicast MAC filters */ 1453 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_CLR, NULL); 1454 1455 if (!netdev_uc_empty(netdev)) { 1456 struct netdev_hw_addr *ha; 1457 1458 /* Add MAC filters one by one */ 1459 netdev_for_each_uc_addr(ha, netdev) { 1460 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_ADD, 1461 ha->addr); 1462 udelay(200); 1463 } 1464 } 1465 1466 return 0; 1467 } 1468 1469 static void igbvf_set_rx_mode(struct net_device *netdev) 1470 { 1471 igbvf_set_multi(netdev); 1472 igbvf_set_uni(netdev); 1473 } 1474 1475 /** 1476 * igbvf_configure - configure the hardware for Rx and Tx 1477 * @adapter: private board structure 1478 **/ 1479 static void igbvf_configure(struct igbvf_adapter *adapter) 1480 { 1481 igbvf_set_rx_mode(adapter->netdev); 1482 1483 igbvf_restore_vlan(adapter); 1484 1485 igbvf_configure_tx(adapter); 1486 igbvf_setup_srrctl(adapter); 1487 igbvf_configure_rx(adapter); 1488 igbvf_alloc_rx_buffers(adapter->rx_ring, 1489 igbvf_desc_unused(adapter->rx_ring)); 1490 } 1491 1492 /* igbvf_reset - bring the hardware into a known good state 1493 * @adapter: private board structure 1494 * 1495 * This function boots the hardware and enables some settings that 1496 * require a configuration cycle of the hardware - those cannot be 1497 * set/changed during runtime. After reset the device needs to be 1498 * properly configured for Rx, Tx etc. 1499 */ 1500 static void igbvf_reset(struct igbvf_adapter *adapter) 1501 { 1502 struct e1000_mac_info *mac = &adapter->hw.mac; 1503 struct net_device *netdev = adapter->netdev; 1504 struct e1000_hw *hw = &adapter->hw; 1505 1506 /* Allow time for pending master requests to run */ 1507 if (mac->ops.reset_hw(hw)) 1508 dev_err(&adapter->pdev->dev, "PF still resetting\n"); 1509 1510 mac->ops.init_hw(hw); 1511 1512 if (is_valid_ether_addr(adapter->hw.mac.addr)) { 1513 memcpy(netdev->dev_addr, adapter->hw.mac.addr, 1514 netdev->addr_len); 1515 memcpy(netdev->perm_addr, adapter->hw.mac.addr, 1516 netdev->addr_len); 1517 } 1518 1519 adapter->last_reset = jiffies; 1520 } 1521 1522 int igbvf_up(struct igbvf_adapter *adapter) 1523 { 1524 struct e1000_hw *hw = &adapter->hw; 1525 1526 /* hardware has been reset, we need to reload some things */ 1527 igbvf_configure(adapter); 1528 1529 clear_bit(__IGBVF_DOWN, &adapter->state); 1530 1531 napi_enable(&adapter->rx_ring->napi); 1532 if (adapter->msix_entries) 1533 igbvf_configure_msix(adapter); 1534 1535 /* Clear any pending interrupts. */ 1536 er32(EICR); 1537 igbvf_irq_enable(adapter); 1538 1539 /* start the watchdog */ 1540 hw->mac.get_link_status = 1; 1541 mod_timer(&adapter->watchdog_timer, jiffies + 1); 1542 1543 return 0; 1544 } 1545 1546 void igbvf_down(struct igbvf_adapter *adapter) 1547 { 1548 struct net_device *netdev = adapter->netdev; 1549 struct e1000_hw *hw = &adapter->hw; 1550 u32 rxdctl, txdctl; 1551 1552 /* signal that we're down so the interrupt handler does not 1553 * reschedule our watchdog timer 1554 */ 1555 set_bit(__IGBVF_DOWN, &adapter->state); 1556 1557 /* disable receives in the hardware */ 1558 rxdctl = er32(RXDCTL(0)); 1559 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE); 1560 1561 netif_carrier_off(netdev); 1562 netif_stop_queue(netdev); 1563 1564 /* disable transmits in the hardware */ 1565 txdctl = er32(TXDCTL(0)); 1566 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE); 1567 1568 /* flush both disables and wait for them to finish */ 1569 e1e_flush(); 1570 msleep(10); 1571 1572 napi_disable(&adapter->rx_ring->napi); 1573 1574 igbvf_irq_disable(adapter); 1575 1576 del_timer_sync(&adapter->watchdog_timer); 1577 1578 /* record the stats before reset*/ 1579 igbvf_update_stats(adapter); 1580 1581 adapter->link_speed = 0; 1582 adapter->link_duplex = 0; 1583 1584 igbvf_reset(adapter); 1585 igbvf_clean_tx_ring(adapter->tx_ring); 1586 igbvf_clean_rx_ring(adapter->rx_ring); 1587 } 1588 1589 void igbvf_reinit_locked(struct igbvf_adapter *adapter) 1590 { 1591 might_sleep(); 1592 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state)) 1593 usleep_range(1000, 2000); 1594 igbvf_down(adapter); 1595 igbvf_up(adapter); 1596 clear_bit(__IGBVF_RESETTING, &adapter->state); 1597 } 1598 1599 /** 1600 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter) 1601 * @adapter: board private structure to initialize 1602 * 1603 * igbvf_sw_init initializes the Adapter private data structure. 1604 * Fields are initialized based on PCI device information and 1605 * OS network device settings (MTU size). 1606 **/ 1607 static int igbvf_sw_init(struct igbvf_adapter *adapter) 1608 { 1609 struct net_device *netdev = adapter->netdev; 1610 s32 rc; 1611 1612 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN; 1613 adapter->rx_ps_hdr_size = 0; 1614 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN; 1615 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN; 1616 1617 adapter->tx_int_delay = 8; 1618 adapter->tx_abs_int_delay = 32; 1619 adapter->rx_int_delay = 0; 1620 adapter->rx_abs_int_delay = 8; 1621 adapter->requested_itr = 3; 1622 adapter->current_itr = IGBVF_START_ITR; 1623 1624 /* Set various function pointers */ 1625 adapter->ei->init_ops(&adapter->hw); 1626 1627 rc = adapter->hw.mac.ops.init_params(&adapter->hw); 1628 if (rc) 1629 return rc; 1630 1631 rc = adapter->hw.mbx.ops.init_params(&adapter->hw); 1632 if (rc) 1633 return rc; 1634 1635 igbvf_set_interrupt_capability(adapter); 1636 1637 if (igbvf_alloc_queues(adapter)) 1638 return -ENOMEM; 1639 1640 spin_lock_init(&adapter->tx_queue_lock); 1641 1642 /* Explicitly disable IRQ since the NIC can be in any state. */ 1643 igbvf_irq_disable(adapter); 1644 1645 spin_lock_init(&adapter->stats_lock); 1646 1647 set_bit(__IGBVF_DOWN, &adapter->state); 1648 return 0; 1649 } 1650 1651 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter) 1652 { 1653 struct e1000_hw *hw = &adapter->hw; 1654 1655 adapter->stats.last_gprc = er32(VFGPRC); 1656 adapter->stats.last_gorc = er32(VFGORC); 1657 adapter->stats.last_gptc = er32(VFGPTC); 1658 adapter->stats.last_gotc = er32(VFGOTC); 1659 adapter->stats.last_mprc = er32(VFMPRC); 1660 adapter->stats.last_gotlbc = er32(VFGOTLBC); 1661 adapter->stats.last_gptlbc = er32(VFGPTLBC); 1662 adapter->stats.last_gorlbc = er32(VFGORLBC); 1663 adapter->stats.last_gprlbc = er32(VFGPRLBC); 1664 1665 adapter->stats.base_gprc = er32(VFGPRC); 1666 adapter->stats.base_gorc = er32(VFGORC); 1667 adapter->stats.base_gptc = er32(VFGPTC); 1668 adapter->stats.base_gotc = er32(VFGOTC); 1669 adapter->stats.base_mprc = er32(VFMPRC); 1670 adapter->stats.base_gotlbc = er32(VFGOTLBC); 1671 adapter->stats.base_gptlbc = er32(VFGPTLBC); 1672 adapter->stats.base_gorlbc = er32(VFGORLBC); 1673 adapter->stats.base_gprlbc = er32(VFGPRLBC); 1674 } 1675 1676 /** 1677 * igbvf_open - Called when a network interface is made active 1678 * @netdev: network interface device structure 1679 * 1680 * Returns 0 on success, negative value on failure 1681 * 1682 * The open entry point is called when a network interface is made 1683 * active by the system (IFF_UP). At this point all resources needed 1684 * for transmit and receive operations are allocated, the interrupt 1685 * handler is registered with the OS, the watchdog timer is started, 1686 * and the stack is notified that the interface is ready. 1687 **/ 1688 static int igbvf_open(struct net_device *netdev) 1689 { 1690 struct igbvf_adapter *adapter = netdev_priv(netdev); 1691 struct e1000_hw *hw = &adapter->hw; 1692 int err; 1693 1694 /* disallow open during test */ 1695 if (test_bit(__IGBVF_TESTING, &adapter->state)) 1696 return -EBUSY; 1697 1698 /* allocate transmit descriptors */ 1699 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring); 1700 if (err) 1701 goto err_setup_tx; 1702 1703 /* allocate receive descriptors */ 1704 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring); 1705 if (err) 1706 goto err_setup_rx; 1707 1708 /* before we allocate an interrupt, we must be ready to handle it. 1709 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt 1710 * as soon as we call pci_request_irq, so we have to setup our 1711 * clean_rx handler before we do so. 1712 */ 1713 igbvf_configure(adapter); 1714 1715 err = igbvf_request_irq(adapter); 1716 if (err) 1717 goto err_req_irq; 1718 1719 /* From here on the code is the same as igbvf_up() */ 1720 clear_bit(__IGBVF_DOWN, &adapter->state); 1721 1722 napi_enable(&adapter->rx_ring->napi); 1723 1724 /* clear any pending interrupts */ 1725 er32(EICR); 1726 1727 igbvf_irq_enable(adapter); 1728 1729 /* start the watchdog */ 1730 hw->mac.get_link_status = 1; 1731 mod_timer(&adapter->watchdog_timer, jiffies + 1); 1732 1733 return 0; 1734 1735 err_req_irq: 1736 igbvf_free_rx_resources(adapter->rx_ring); 1737 err_setup_rx: 1738 igbvf_free_tx_resources(adapter->tx_ring); 1739 err_setup_tx: 1740 igbvf_reset(adapter); 1741 1742 return err; 1743 } 1744 1745 /** 1746 * igbvf_close - Disables a network interface 1747 * @netdev: network interface device structure 1748 * 1749 * Returns 0, this is not allowed to fail 1750 * 1751 * The close entry point is called when an interface is de-activated 1752 * by the OS. The hardware is still under the drivers control, but 1753 * needs to be disabled. A global MAC reset is issued to stop the 1754 * hardware, and all transmit and receive resources are freed. 1755 **/ 1756 static int igbvf_close(struct net_device *netdev) 1757 { 1758 struct igbvf_adapter *adapter = netdev_priv(netdev); 1759 1760 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state)); 1761 igbvf_down(adapter); 1762 1763 igbvf_free_irq(adapter); 1764 1765 igbvf_free_tx_resources(adapter->tx_ring); 1766 igbvf_free_rx_resources(adapter->rx_ring); 1767 1768 return 0; 1769 } 1770 1771 /** 1772 * igbvf_set_mac - Change the Ethernet Address of the NIC 1773 * @netdev: network interface device structure 1774 * @p: pointer to an address structure 1775 * 1776 * Returns 0 on success, negative on failure 1777 **/ 1778 static int igbvf_set_mac(struct net_device *netdev, void *p) 1779 { 1780 struct igbvf_adapter *adapter = netdev_priv(netdev); 1781 struct e1000_hw *hw = &adapter->hw; 1782 struct sockaddr *addr = p; 1783 1784 if (!is_valid_ether_addr(addr->sa_data)) 1785 return -EADDRNOTAVAIL; 1786 1787 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len); 1788 1789 hw->mac.ops.rar_set(hw, hw->mac.addr, 0); 1790 1791 if (!ether_addr_equal(addr->sa_data, hw->mac.addr)) 1792 return -EADDRNOTAVAIL; 1793 1794 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len); 1795 1796 return 0; 1797 } 1798 1799 #define UPDATE_VF_COUNTER(reg, name) \ 1800 { \ 1801 u32 current_counter = er32(reg); \ 1802 if (current_counter < adapter->stats.last_##name) \ 1803 adapter->stats.name += 0x100000000LL; \ 1804 adapter->stats.last_##name = current_counter; \ 1805 adapter->stats.name &= 0xFFFFFFFF00000000LL; \ 1806 adapter->stats.name |= current_counter; \ 1807 } 1808 1809 /** 1810 * igbvf_update_stats - Update the board statistics counters 1811 * @adapter: board private structure 1812 **/ 1813 void igbvf_update_stats(struct igbvf_adapter *adapter) 1814 { 1815 struct e1000_hw *hw = &adapter->hw; 1816 struct pci_dev *pdev = adapter->pdev; 1817 1818 /* Prevent stats update while adapter is being reset, link is down 1819 * or if the pci connection is down. 1820 */ 1821 if (adapter->link_speed == 0) 1822 return; 1823 1824 if (test_bit(__IGBVF_RESETTING, &adapter->state)) 1825 return; 1826 1827 if (pci_channel_offline(pdev)) 1828 return; 1829 1830 UPDATE_VF_COUNTER(VFGPRC, gprc); 1831 UPDATE_VF_COUNTER(VFGORC, gorc); 1832 UPDATE_VF_COUNTER(VFGPTC, gptc); 1833 UPDATE_VF_COUNTER(VFGOTC, gotc); 1834 UPDATE_VF_COUNTER(VFMPRC, mprc); 1835 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc); 1836 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc); 1837 UPDATE_VF_COUNTER(VFGORLBC, gorlbc); 1838 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc); 1839 1840 /* Fill out the OS statistics structure */ 1841 adapter->netdev->stats.multicast = adapter->stats.mprc; 1842 } 1843 1844 static void igbvf_print_link_info(struct igbvf_adapter *adapter) 1845 { 1846 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n", 1847 adapter->link_speed, 1848 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half"); 1849 } 1850 1851 static bool igbvf_has_link(struct igbvf_adapter *adapter) 1852 { 1853 struct e1000_hw *hw = &adapter->hw; 1854 s32 ret_val = E1000_SUCCESS; 1855 bool link_active; 1856 1857 /* If interface is down, stay link down */ 1858 if (test_bit(__IGBVF_DOWN, &adapter->state)) 1859 return false; 1860 1861 ret_val = hw->mac.ops.check_for_link(hw); 1862 link_active = !hw->mac.get_link_status; 1863 1864 /* if check for link returns error we will need to reset */ 1865 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ))) 1866 schedule_work(&adapter->reset_task); 1867 1868 return link_active; 1869 } 1870 1871 /** 1872 * igbvf_watchdog - Timer Call-back 1873 * @data: pointer to adapter cast into an unsigned long 1874 **/ 1875 static void igbvf_watchdog(unsigned long data) 1876 { 1877 struct igbvf_adapter *adapter = (struct igbvf_adapter *)data; 1878 1879 /* Do the rest outside of interrupt context */ 1880 schedule_work(&adapter->watchdog_task); 1881 } 1882 1883 static void igbvf_watchdog_task(struct work_struct *work) 1884 { 1885 struct igbvf_adapter *adapter = container_of(work, 1886 struct igbvf_adapter, 1887 watchdog_task); 1888 struct net_device *netdev = adapter->netdev; 1889 struct e1000_mac_info *mac = &adapter->hw.mac; 1890 struct igbvf_ring *tx_ring = adapter->tx_ring; 1891 struct e1000_hw *hw = &adapter->hw; 1892 u32 link; 1893 int tx_pending = 0; 1894 1895 link = igbvf_has_link(adapter); 1896 1897 if (link) { 1898 if (!netif_carrier_ok(netdev)) { 1899 mac->ops.get_link_up_info(&adapter->hw, 1900 &adapter->link_speed, 1901 &adapter->link_duplex); 1902 igbvf_print_link_info(adapter); 1903 1904 netif_carrier_on(netdev); 1905 netif_wake_queue(netdev); 1906 } 1907 } else { 1908 if (netif_carrier_ok(netdev)) { 1909 adapter->link_speed = 0; 1910 adapter->link_duplex = 0; 1911 dev_info(&adapter->pdev->dev, "Link is Down\n"); 1912 netif_carrier_off(netdev); 1913 netif_stop_queue(netdev); 1914 } 1915 } 1916 1917 if (netif_carrier_ok(netdev)) { 1918 igbvf_update_stats(adapter); 1919 } else { 1920 tx_pending = (igbvf_desc_unused(tx_ring) + 1 < 1921 tx_ring->count); 1922 if (tx_pending) { 1923 /* We've lost link, so the controller stops DMA, 1924 * but we've got queued Tx work that's never going 1925 * to get done, so reset controller to flush Tx. 1926 * (Do the reset outside of interrupt context). 1927 */ 1928 adapter->tx_timeout_count++; 1929 schedule_work(&adapter->reset_task); 1930 } 1931 } 1932 1933 /* Cause software interrupt to ensure Rx ring is cleaned */ 1934 ew32(EICS, adapter->rx_ring->eims_value); 1935 1936 /* Reset the timer */ 1937 if (!test_bit(__IGBVF_DOWN, &adapter->state)) 1938 mod_timer(&adapter->watchdog_timer, 1939 round_jiffies(jiffies + (2 * HZ))); 1940 } 1941 1942 #define IGBVF_TX_FLAGS_CSUM 0x00000001 1943 #define IGBVF_TX_FLAGS_VLAN 0x00000002 1944 #define IGBVF_TX_FLAGS_TSO 0x00000004 1945 #define IGBVF_TX_FLAGS_IPV4 0x00000008 1946 #define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000 1947 #define IGBVF_TX_FLAGS_VLAN_SHIFT 16 1948 1949 static void igbvf_tx_ctxtdesc(struct igbvf_ring *tx_ring, u32 vlan_macip_lens, 1950 u32 type_tucmd, u32 mss_l4len_idx) 1951 { 1952 struct e1000_adv_tx_context_desc *context_desc; 1953 struct igbvf_buffer *buffer_info; 1954 u16 i = tx_ring->next_to_use; 1955 1956 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i); 1957 buffer_info = &tx_ring->buffer_info[i]; 1958 1959 i++; 1960 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; 1961 1962 /* set bits to identify this as an advanced context descriptor */ 1963 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT; 1964 1965 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens); 1966 context_desc->seqnum_seed = 0; 1967 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd); 1968 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx); 1969 1970 buffer_info->time_stamp = jiffies; 1971 buffer_info->dma = 0; 1972 } 1973 1974 static int igbvf_tso(struct igbvf_ring *tx_ring, 1975 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len) 1976 { 1977 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx; 1978 union { 1979 struct iphdr *v4; 1980 struct ipv6hdr *v6; 1981 unsigned char *hdr; 1982 } ip; 1983 union { 1984 struct tcphdr *tcp; 1985 unsigned char *hdr; 1986 } l4; 1987 u32 paylen, l4_offset; 1988 int err; 1989 1990 if (skb->ip_summed != CHECKSUM_PARTIAL) 1991 return 0; 1992 1993 if (!skb_is_gso(skb)) 1994 return 0; 1995 1996 err = skb_cow_head(skb, 0); 1997 if (err < 0) 1998 return err; 1999 2000 ip.hdr = skb_network_header(skb); 2001 l4.hdr = skb_checksum_start(skb); 2002 2003 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */ 2004 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP; 2005 2006 /* initialize outer IP header fields */ 2007 if (ip.v4->version == 4) { 2008 unsigned char *csum_start = skb_checksum_start(skb); 2009 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4); 2010 2011 /* IP header will have to cancel out any data that 2012 * is not a part of the outer IP header 2013 */ 2014 ip.v4->check = csum_fold(csum_partial(trans_start, 2015 csum_start - trans_start, 2016 0)); 2017 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4; 2018 2019 ip.v4->tot_len = 0; 2020 } else { 2021 ip.v6->payload_len = 0; 2022 } 2023 2024 /* determine offset of inner transport header */ 2025 l4_offset = l4.hdr - skb->data; 2026 2027 /* compute length of segmentation header */ 2028 *hdr_len = (l4.tcp->doff * 4) + l4_offset; 2029 2030 /* remove payload length from inner checksum */ 2031 paylen = skb->len - l4_offset; 2032 csum_replace_by_diff(&l4.tcp->check, htonl(paylen)); 2033 2034 /* MSS L4LEN IDX */ 2035 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT; 2036 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT; 2037 2038 /* VLAN MACLEN IPLEN */ 2039 vlan_macip_lens = l4.hdr - ip.hdr; 2040 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT; 2041 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK; 2042 2043 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx); 2044 2045 return 1; 2046 } 2047 2048 static inline bool igbvf_ipv6_csum_is_sctp(struct sk_buff *skb) 2049 { 2050 unsigned int offset = 0; 2051 2052 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL); 2053 2054 return offset == skb_checksum_start_offset(skb); 2055 } 2056 2057 static bool igbvf_tx_csum(struct igbvf_ring *tx_ring, struct sk_buff *skb, 2058 u32 tx_flags, __be16 protocol) 2059 { 2060 u32 vlan_macip_lens = 0; 2061 u32 type_tucmd = 0; 2062 2063 if (skb->ip_summed != CHECKSUM_PARTIAL) { 2064 csum_failed: 2065 if (!(tx_flags & IGBVF_TX_FLAGS_VLAN)) 2066 return false; 2067 goto no_csum; 2068 } 2069 2070 switch (skb->csum_offset) { 2071 case offsetof(struct tcphdr, check): 2072 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP; 2073 /* fall through */ 2074 case offsetof(struct udphdr, check): 2075 break; 2076 case offsetof(struct sctphdr, checksum): 2077 /* validate that this is actually an SCTP request */ 2078 if (((protocol == htons(ETH_P_IP)) && 2079 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) || 2080 ((protocol == htons(ETH_P_IPV6)) && 2081 igbvf_ipv6_csum_is_sctp(skb))) { 2082 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP; 2083 break; 2084 } 2085 default: 2086 skb_checksum_help(skb); 2087 goto csum_failed; 2088 } 2089 2090 vlan_macip_lens = skb_checksum_start_offset(skb) - 2091 skb_network_offset(skb); 2092 no_csum: 2093 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT; 2094 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK; 2095 2096 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0); 2097 return true; 2098 } 2099 2100 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size) 2101 { 2102 struct igbvf_adapter *adapter = netdev_priv(netdev); 2103 2104 /* there is enough descriptors then we don't need to worry */ 2105 if (igbvf_desc_unused(adapter->tx_ring) >= size) 2106 return 0; 2107 2108 netif_stop_queue(netdev); 2109 2110 /* Herbert's original patch had: 2111 * smp_mb__after_netif_stop_queue(); 2112 * but since that doesn't exist yet, just open code it. 2113 */ 2114 smp_mb(); 2115 2116 /* We need to check again just in case room has been made available */ 2117 if (igbvf_desc_unused(adapter->tx_ring) < size) 2118 return -EBUSY; 2119 2120 netif_wake_queue(netdev); 2121 2122 ++adapter->restart_queue; 2123 return 0; 2124 } 2125 2126 #define IGBVF_MAX_TXD_PWR 16 2127 #define IGBVF_MAX_DATA_PER_TXD (1u << IGBVF_MAX_TXD_PWR) 2128 2129 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter, 2130 struct igbvf_ring *tx_ring, 2131 struct sk_buff *skb) 2132 { 2133 struct igbvf_buffer *buffer_info; 2134 struct pci_dev *pdev = adapter->pdev; 2135 unsigned int len = skb_headlen(skb); 2136 unsigned int count = 0, i; 2137 unsigned int f; 2138 2139 i = tx_ring->next_to_use; 2140 2141 buffer_info = &tx_ring->buffer_info[i]; 2142 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD); 2143 buffer_info->length = len; 2144 /* set time_stamp *before* dma to help avoid a possible race */ 2145 buffer_info->time_stamp = jiffies; 2146 buffer_info->mapped_as_page = false; 2147 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len, 2148 DMA_TO_DEVICE); 2149 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) 2150 goto dma_error; 2151 2152 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) { 2153 const struct skb_frag_struct *frag; 2154 2155 count++; 2156 i++; 2157 if (i == tx_ring->count) 2158 i = 0; 2159 2160 frag = &skb_shinfo(skb)->frags[f]; 2161 len = skb_frag_size(frag); 2162 2163 buffer_info = &tx_ring->buffer_info[i]; 2164 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD); 2165 buffer_info->length = len; 2166 buffer_info->time_stamp = jiffies; 2167 buffer_info->mapped_as_page = true; 2168 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len, 2169 DMA_TO_DEVICE); 2170 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) 2171 goto dma_error; 2172 } 2173 2174 tx_ring->buffer_info[i].skb = skb; 2175 2176 return ++count; 2177 2178 dma_error: 2179 dev_err(&pdev->dev, "TX DMA map failed\n"); 2180 2181 /* clear timestamp and dma mappings for failed buffer_info mapping */ 2182 buffer_info->dma = 0; 2183 buffer_info->time_stamp = 0; 2184 buffer_info->length = 0; 2185 buffer_info->mapped_as_page = false; 2186 if (count) 2187 count--; 2188 2189 /* clear timestamp and dma mappings for remaining portion of packet */ 2190 while (count--) { 2191 if (i == 0) 2192 i += tx_ring->count; 2193 i--; 2194 buffer_info = &tx_ring->buffer_info[i]; 2195 igbvf_put_txbuf(adapter, buffer_info); 2196 } 2197 2198 return 0; 2199 } 2200 2201 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter, 2202 struct igbvf_ring *tx_ring, 2203 int tx_flags, int count, 2204 unsigned int first, u32 paylen, 2205 u8 hdr_len) 2206 { 2207 union e1000_adv_tx_desc *tx_desc = NULL; 2208 struct igbvf_buffer *buffer_info; 2209 u32 olinfo_status = 0, cmd_type_len; 2210 unsigned int i; 2211 2212 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS | 2213 E1000_ADVTXD_DCMD_DEXT); 2214 2215 if (tx_flags & IGBVF_TX_FLAGS_VLAN) 2216 cmd_type_len |= E1000_ADVTXD_DCMD_VLE; 2217 2218 if (tx_flags & IGBVF_TX_FLAGS_TSO) { 2219 cmd_type_len |= E1000_ADVTXD_DCMD_TSE; 2220 2221 /* insert tcp checksum */ 2222 olinfo_status |= E1000_TXD_POPTS_TXSM << 8; 2223 2224 /* insert ip checksum */ 2225 if (tx_flags & IGBVF_TX_FLAGS_IPV4) 2226 olinfo_status |= E1000_TXD_POPTS_IXSM << 8; 2227 2228 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) { 2229 olinfo_status |= E1000_TXD_POPTS_TXSM << 8; 2230 } 2231 2232 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT); 2233 2234 i = tx_ring->next_to_use; 2235 while (count--) { 2236 buffer_info = &tx_ring->buffer_info[i]; 2237 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i); 2238 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma); 2239 tx_desc->read.cmd_type_len = 2240 cpu_to_le32(cmd_type_len | buffer_info->length); 2241 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status); 2242 i++; 2243 if (i == tx_ring->count) 2244 i = 0; 2245 } 2246 2247 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd); 2248 /* Force memory writes to complete before letting h/w 2249 * know there are new descriptors to fetch. (Only 2250 * applicable for weak-ordered memory model archs, 2251 * such as IA-64). 2252 */ 2253 wmb(); 2254 2255 tx_ring->buffer_info[first].next_to_watch = tx_desc; 2256 tx_ring->next_to_use = i; 2257 writel(i, adapter->hw.hw_addr + tx_ring->tail); 2258 /* we need this if more than one processor can write to our tail 2259 * at a time, it synchronizes IO on IA64/Altix systems 2260 */ 2261 mmiowb(); 2262 } 2263 2264 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb, 2265 struct net_device *netdev, 2266 struct igbvf_ring *tx_ring) 2267 { 2268 struct igbvf_adapter *adapter = netdev_priv(netdev); 2269 unsigned int first, tx_flags = 0; 2270 u8 hdr_len = 0; 2271 int count = 0; 2272 int tso = 0; 2273 __be16 protocol = vlan_get_protocol(skb); 2274 2275 if (test_bit(__IGBVF_DOWN, &adapter->state)) { 2276 dev_kfree_skb_any(skb); 2277 return NETDEV_TX_OK; 2278 } 2279 2280 if (skb->len <= 0) { 2281 dev_kfree_skb_any(skb); 2282 return NETDEV_TX_OK; 2283 } 2284 2285 /* need: count + 4 desc gap to keep tail from touching 2286 * + 2 desc gap to keep tail from touching head, 2287 * + 1 desc for skb->data, 2288 * + 1 desc for context descriptor, 2289 * head, otherwise try next time 2290 */ 2291 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) { 2292 /* this is a hard error */ 2293 return NETDEV_TX_BUSY; 2294 } 2295 2296 if (skb_vlan_tag_present(skb)) { 2297 tx_flags |= IGBVF_TX_FLAGS_VLAN; 2298 tx_flags |= (skb_vlan_tag_get(skb) << 2299 IGBVF_TX_FLAGS_VLAN_SHIFT); 2300 } 2301 2302 if (protocol == htons(ETH_P_IP)) 2303 tx_flags |= IGBVF_TX_FLAGS_IPV4; 2304 2305 first = tx_ring->next_to_use; 2306 2307 tso = igbvf_tso(tx_ring, skb, tx_flags, &hdr_len); 2308 if (unlikely(tso < 0)) { 2309 dev_kfree_skb_any(skb); 2310 return NETDEV_TX_OK; 2311 } 2312 2313 if (tso) 2314 tx_flags |= IGBVF_TX_FLAGS_TSO; 2315 else if (igbvf_tx_csum(tx_ring, skb, tx_flags, protocol) && 2316 (skb->ip_summed == CHECKSUM_PARTIAL)) 2317 tx_flags |= IGBVF_TX_FLAGS_CSUM; 2318 2319 /* count reflects descriptors mapped, if 0 then mapping error 2320 * has occurred and we need to rewind the descriptor queue 2321 */ 2322 count = igbvf_tx_map_adv(adapter, tx_ring, skb); 2323 2324 if (count) { 2325 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count, 2326 first, skb->len, hdr_len); 2327 /* Make sure there is space in the ring for the next send. */ 2328 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4); 2329 } else { 2330 dev_kfree_skb_any(skb); 2331 tx_ring->buffer_info[first].time_stamp = 0; 2332 tx_ring->next_to_use = first; 2333 } 2334 2335 return NETDEV_TX_OK; 2336 } 2337 2338 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb, 2339 struct net_device *netdev) 2340 { 2341 struct igbvf_adapter *adapter = netdev_priv(netdev); 2342 struct igbvf_ring *tx_ring; 2343 2344 if (test_bit(__IGBVF_DOWN, &adapter->state)) { 2345 dev_kfree_skb_any(skb); 2346 return NETDEV_TX_OK; 2347 } 2348 2349 tx_ring = &adapter->tx_ring[0]; 2350 2351 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring); 2352 } 2353 2354 /** 2355 * igbvf_tx_timeout - Respond to a Tx Hang 2356 * @netdev: network interface device structure 2357 **/ 2358 static void igbvf_tx_timeout(struct net_device *netdev) 2359 { 2360 struct igbvf_adapter *adapter = netdev_priv(netdev); 2361 2362 /* Do the reset outside of interrupt context */ 2363 adapter->tx_timeout_count++; 2364 schedule_work(&adapter->reset_task); 2365 } 2366 2367 static void igbvf_reset_task(struct work_struct *work) 2368 { 2369 struct igbvf_adapter *adapter; 2370 2371 adapter = container_of(work, struct igbvf_adapter, reset_task); 2372 2373 igbvf_reinit_locked(adapter); 2374 } 2375 2376 /** 2377 * igbvf_change_mtu - Change the Maximum Transfer Unit 2378 * @netdev: network interface device structure 2379 * @new_mtu: new value for maximum frame size 2380 * 2381 * Returns 0 on success, negative on failure 2382 **/ 2383 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu) 2384 { 2385 struct igbvf_adapter *adapter = netdev_priv(netdev); 2386 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN; 2387 2388 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state)) 2389 usleep_range(1000, 2000); 2390 /* igbvf_down has a dependency on max_frame_size */ 2391 adapter->max_frame_size = max_frame; 2392 if (netif_running(netdev)) 2393 igbvf_down(adapter); 2394 2395 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN 2396 * means we reserve 2 more, this pushes us to allocate from the next 2397 * larger slab size. 2398 * i.e. RXBUFFER_2048 --> size-4096 slab 2399 * However with the new *_jumbo_rx* routines, jumbo receives will use 2400 * fragmented skbs 2401 */ 2402 2403 if (max_frame <= 1024) 2404 adapter->rx_buffer_len = 1024; 2405 else if (max_frame <= 2048) 2406 adapter->rx_buffer_len = 2048; 2407 else 2408 #if (PAGE_SIZE / 2) > 16384 2409 adapter->rx_buffer_len = 16384; 2410 #else 2411 adapter->rx_buffer_len = PAGE_SIZE / 2; 2412 #endif 2413 2414 /* adjust allocation if LPE protects us, and we aren't using SBP */ 2415 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) || 2416 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN)) 2417 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + 2418 ETH_FCS_LEN; 2419 2420 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n", 2421 netdev->mtu, new_mtu); 2422 netdev->mtu = new_mtu; 2423 2424 if (netif_running(netdev)) 2425 igbvf_up(adapter); 2426 else 2427 igbvf_reset(adapter); 2428 2429 clear_bit(__IGBVF_RESETTING, &adapter->state); 2430 2431 return 0; 2432 } 2433 2434 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd) 2435 { 2436 switch (cmd) { 2437 default: 2438 return -EOPNOTSUPP; 2439 } 2440 } 2441 2442 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state) 2443 { 2444 struct net_device *netdev = pci_get_drvdata(pdev); 2445 struct igbvf_adapter *adapter = netdev_priv(netdev); 2446 #ifdef CONFIG_PM 2447 int retval = 0; 2448 #endif 2449 2450 netif_device_detach(netdev); 2451 2452 if (netif_running(netdev)) { 2453 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state)); 2454 igbvf_down(adapter); 2455 igbvf_free_irq(adapter); 2456 } 2457 2458 #ifdef CONFIG_PM 2459 retval = pci_save_state(pdev); 2460 if (retval) 2461 return retval; 2462 #endif 2463 2464 pci_disable_device(pdev); 2465 2466 return 0; 2467 } 2468 2469 #ifdef CONFIG_PM 2470 static int igbvf_resume(struct pci_dev *pdev) 2471 { 2472 struct net_device *netdev = pci_get_drvdata(pdev); 2473 struct igbvf_adapter *adapter = netdev_priv(netdev); 2474 u32 err; 2475 2476 pci_restore_state(pdev); 2477 err = pci_enable_device_mem(pdev); 2478 if (err) { 2479 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n"); 2480 return err; 2481 } 2482 2483 pci_set_master(pdev); 2484 2485 if (netif_running(netdev)) { 2486 err = igbvf_request_irq(adapter); 2487 if (err) 2488 return err; 2489 } 2490 2491 igbvf_reset(adapter); 2492 2493 if (netif_running(netdev)) 2494 igbvf_up(adapter); 2495 2496 netif_device_attach(netdev); 2497 2498 return 0; 2499 } 2500 #endif 2501 2502 static void igbvf_shutdown(struct pci_dev *pdev) 2503 { 2504 igbvf_suspend(pdev, PMSG_SUSPEND); 2505 } 2506 2507 #ifdef CONFIG_NET_POLL_CONTROLLER 2508 /* Polling 'interrupt' - used by things like netconsole to send skbs 2509 * without having to re-enable interrupts. It's not called while 2510 * the interrupt routine is executing. 2511 */ 2512 static void igbvf_netpoll(struct net_device *netdev) 2513 { 2514 struct igbvf_adapter *adapter = netdev_priv(netdev); 2515 2516 disable_irq(adapter->pdev->irq); 2517 2518 igbvf_clean_tx_irq(adapter->tx_ring); 2519 2520 enable_irq(adapter->pdev->irq); 2521 } 2522 #endif 2523 2524 /** 2525 * igbvf_io_error_detected - called when PCI error is detected 2526 * @pdev: Pointer to PCI device 2527 * @state: The current pci connection state 2528 * 2529 * This function is called after a PCI bus error affecting 2530 * this device has been detected. 2531 */ 2532 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev, 2533 pci_channel_state_t state) 2534 { 2535 struct net_device *netdev = pci_get_drvdata(pdev); 2536 struct igbvf_adapter *adapter = netdev_priv(netdev); 2537 2538 netif_device_detach(netdev); 2539 2540 if (state == pci_channel_io_perm_failure) 2541 return PCI_ERS_RESULT_DISCONNECT; 2542 2543 if (netif_running(netdev)) 2544 igbvf_down(adapter); 2545 pci_disable_device(pdev); 2546 2547 /* Request a slot slot reset. */ 2548 return PCI_ERS_RESULT_NEED_RESET; 2549 } 2550 2551 /** 2552 * igbvf_io_slot_reset - called after the pci bus has been reset. 2553 * @pdev: Pointer to PCI device 2554 * 2555 * Restart the card from scratch, as if from a cold-boot. Implementation 2556 * resembles the first-half of the igbvf_resume routine. 2557 */ 2558 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev) 2559 { 2560 struct net_device *netdev = pci_get_drvdata(pdev); 2561 struct igbvf_adapter *adapter = netdev_priv(netdev); 2562 2563 if (pci_enable_device_mem(pdev)) { 2564 dev_err(&pdev->dev, 2565 "Cannot re-enable PCI device after reset.\n"); 2566 return PCI_ERS_RESULT_DISCONNECT; 2567 } 2568 pci_set_master(pdev); 2569 2570 igbvf_reset(adapter); 2571 2572 return PCI_ERS_RESULT_RECOVERED; 2573 } 2574 2575 /** 2576 * igbvf_io_resume - called when traffic can start flowing again. 2577 * @pdev: Pointer to PCI device 2578 * 2579 * This callback is called when the error recovery driver tells us that 2580 * its OK to resume normal operation. Implementation resembles the 2581 * second-half of the igbvf_resume routine. 2582 */ 2583 static void igbvf_io_resume(struct pci_dev *pdev) 2584 { 2585 struct net_device *netdev = pci_get_drvdata(pdev); 2586 struct igbvf_adapter *adapter = netdev_priv(netdev); 2587 2588 if (netif_running(netdev)) { 2589 if (igbvf_up(adapter)) { 2590 dev_err(&pdev->dev, 2591 "can't bring device back up after reset\n"); 2592 return; 2593 } 2594 } 2595 2596 netif_device_attach(netdev); 2597 } 2598 2599 static void igbvf_print_device_info(struct igbvf_adapter *adapter) 2600 { 2601 struct e1000_hw *hw = &adapter->hw; 2602 struct net_device *netdev = adapter->netdev; 2603 struct pci_dev *pdev = adapter->pdev; 2604 2605 if (hw->mac.type == e1000_vfadapt_i350) 2606 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n"); 2607 else 2608 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n"); 2609 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr); 2610 } 2611 2612 static int igbvf_set_features(struct net_device *netdev, 2613 netdev_features_t features) 2614 { 2615 struct igbvf_adapter *adapter = netdev_priv(netdev); 2616 2617 if (features & NETIF_F_RXCSUM) 2618 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED; 2619 else 2620 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED; 2621 2622 return 0; 2623 } 2624 2625 #define IGBVF_MAX_MAC_HDR_LEN 127 2626 #define IGBVF_MAX_NETWORK_HDR_LEN 511 2627 2628 static netdev_features_t 2629 igbvf_features_check(struct sk_buff *skb, struct net_device *dev, 2630 netdev_features_t features) 2631 { 2632 unsigned int network_hdr_len, mac_hdr_len; 2633 2634 /* Make certain the headers can be described by a context descriptor */ 2635 mac_hdr_len = skb_network_header(skb) - skb->data; 2636 if (unlikely(mac_hdr_len > IGBVF_MAX_MAC_HDR_LEN)) 2637 return features & ~(NETIF_F_HW_CSUM | 2638 NETIF_F_SCTP_CRC | 2639 NETIF_F_HW_VLAN_CTAG_TX | 2640 NETIF_F_TSO | 2641 NETIF_F_TSO6); 2642 2643 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb); 2644 if (unlikely(network_hdr_len > IGBVF_MAX_NETWORK_HDR_LEN)) 2645 return features & ~(NETIF_F_HW_CSUM | 2646 NETIF_F_SCTP_CRC | 2647 NETIF_F_TSO | 2648 NETIF_F_TSO6); 2649 2650 /* We can only support IPV4 TSO in tunnels if we can mangle the 2651 * inner IP ID field, so strip TSO if MANGLEID is not supported. 2652 */ 2653 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) 2654 features &= ~NETIF_F_TSO; 2655 2656 return features; 2657 } 2658 2659 static const struct net_device_ops igbvf_netdev_ops = { 2660 .ndo_open = igbvf_open, 2661 .ndo_stop = igbvf_close, 2662 .ndo_start_xmit = igbvf_xmit_frame, 2663 .ndo_set_rx_mode = igbvf_set_rx_mode, 2664 .ndo_set_mac_address = igbvf_set_mac, 2665 .ndo_change_mtu = igbvf_change_mtu, 2666 .ndo_do_ioctl = igbvf_ioctl, 2667 .ndo_tx_timeout = igbvf_tx_timeout, 2668 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid, 2669 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid, 2670 #ifdef CONFIG_NET_POLL_CONTROLLER 2671 .ndo_poll_controller = igbvf_netpoll, 2672 #endif 2673 .ndo_set_features = igbvf_set_features, 2674 .ndo_features_check = igbvf_features_check, 2675 }; 2676 2677 /** 2678 * igbvf_probe - Device Initialization Routine 2679 * @pdev: PCI device information struct 2680 * @ent: entry in igbvf_pci_tbl 2681 * 2682 * Returns 0 on success, negative on failure 2683 * 2684 * igbvf_probe initializes an adapter identified by a pci_dev structure. 2685 * The OS initialization, configuring of the adapter private structure, 2686 * and a hardware reset occur. 2687 **/ 2688 static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 2689 { 2690 struct net_device *netdev; 2691 struct igbvf_adapter *adapter; 2692 struct e1000_hw *hw; 2693 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data]; 2694 2695 static int cards_found; 2696 int err, pci_using_dac; 2697 2698 err = pci_enable_device_mem(pdev); 2699 if (err) 2700 return err; 2701 2702 pci_using_dac = 0; 2703 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); 2704 if (!err) { 2705 pci_using_dac = 1; 2706 } else { 2707 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); 2708 if (err) { 2709 dev_err(&pdev->dev, 2710 "No usable DMA configuration, aborting\n"); 2711 goto err_dma; 2712 } 2713 } 2714 2715 err = pci_request_regions(pdev, igbvf_driver_name); 2716 if (err) 2717 goto err_pci_reg; 2718 2719 pci_set_master(pdev); 2720 2721 err = -ENOMEM; 2722 netdev = alloc_etherdev(sizeof(struct igbvf_adapter)); 2723 if (!netdev) 2724 goto err_alloc_etherdev; 2725 2726 SET_NETDEV_DEV(netdev, &pdev->dev); 2727 2728 pci_set_drvdata(pdev, netdev); 2729 adapter = netdev_priv(netdev); 2730 hw = &adapter->hw; 2731 adapter->netdev = netdev; 2732 adapter->pdev = pdev; 2733 adapter->ei = ei; 2734 adapter->pba = ei->pba; 2735 adapter->flags = ei->flags; 2736 adapter->hw.back = adapter; 2737 adapter->hw.mac.type = ei->mac; 2738 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE); 2739 2740 /* PCI config space info */ 2741 2742 hw->vendor_id = pdev->vendor; 2743 hw->device_id = pdev->device; 2744 hw->subsystem_vendor_id = pdev->subsystem_vendor; 2745 hw->subsystem_device_id = pdev->subsystem_device; 2746 hw->revision_id = pdev->revision; 2747 2748 err = -EIO; 2749 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0), 2750 pci_resource_len(pdev, 0)); 2751 2752 if (!adapter->hw.hw_addr) 2753 goto err_ioremap; 2754 2755 if (ei->get_variants) { 2756 err = ei->get_variants(adapter); 2757 if (err) 2758 goto err_get_variants; 2759 } 2760 2761 /* setup adapter struct */ 2762 err = igbvf_sw_init(adapter); 2763 if (err) 2764 goto err_sw_init; 2765 2766 /* construct the net_device struct */ 2767 netdev->netdev_ops = &igbvf_netdev_ops; 2768 2769 igbvf_set_ethtool_ops(netdev); 2770 netdev->watchdog_timeo = 5 * HZ; 2771 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1); 2772 2773 adapter->bd_number = cards_found++; 2774 2775 netdev->hw_features = NETIF_F_SG | 2776 NETIF_F_TSO | 2777 NETIF_F_TSO6 | 2778 NETIF_F_RXCSUM | 2779 NETIF_F_HW_CSUM | 2780 NETIF_F_SCTP_CRC; 2781 2782 #define IGBVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \ 2783 NETIF_F_GSO_GRE_CSUM | \ 2784 NETIF_F_GSO_IPXIP4 | \ 2785 NETIF_F_GSO_IPXIP6 | \ 2786 NETIF_F_GSO_UDP_TUNNEL | \ 2787 NETIF_F_GSO_UDP_TUNNEL_CSUM) 2788 2789 netdev->gso_partial_features = IGBVF_GSO_PARTIAL_FEATURES; 2790 netdev->hw_features |= NETIF_F_GSO_PARTIAL | 2791 IGBVF_GSO_PARTIAL_FEATURES; 2792 2793 netdev->features = netdev->hw_features; 2794 2795 if (pci_using_dac) 2796 netdev->features |= NETIF_F_HIGHDMA; 2797 2798 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID; 2799 netdev->mpls_features |= NETIF_F_HW_CSUM; 2800 netdev->hw_enc_features |= netdev->vlan_features; 2801 2802 /* set this bit last since it cannot be part of vlan_features */ 2803 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER | 2804 NETIF_F_HW_VLAN_CTAG_RX | 2805 NETIF_F_HW_VLAN_CTAG_TX; 2806 2807 /* MTU range: 68 - 9216 */ 2808 netdev->min_mtu = ETH_MIN_MTU; 2809 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE; 2810 2811 /*reset the controller to put the device in a known good state */ 2812 err = hw->mac.ops.reset_hw(hw); 2813 if (err) { 2814 dev_info(&pdev->dev, 2815 "PF still in reset state. Is the PF interface up?\n"); 2816 } else { 2817 err = hw->mac.ops.read_mac_addr(hw); 2818 if (err) 2819 dev_info(&pdev->dev, "Error reading MAC address.\n"); 2820 else if (is_zero_ether_addr(adapter->hw.mac.addr)) 2821 dev_info(&pdev->dev, 2822 "MAC address not assigned by administrator.\n"); 2823 memcpy(netdev->dev_addr, adapter->hw.mac.addr, 2824 netdev->addr_len); 2825 } 2826 2827 if (!is_valid_ether_addr(netdev->dev_addr)) { 2828 dev_info(&pdev->dev, "Assigning random MAC address.\n"); 2829 eth_hw_addr_random(netdev); 2830 memcpy(adapter->hw.mac.addr, netdev->dev_addr, 2831 netdev->addr_len); 2832 } 2833 2834 setup_timer(&adapter->watchdog_timer, &igbvf_watchdog, 2835 (unsigned long)adapter); 2836 2837 INIT_WORK(&adapter->reset_task, igbvf_reset_task); 2838 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task); 2839 2840 /* ring size defaults */ 2841 adapter->rx_ring->count = 1024; 2842 adapter->tx_ring->count = 1024; 2843 2844 /* reset the hardware with the new settings */ 2845 igbvf_reset(adapter); 2846 2847 /* set hardware-specific flags */ 2848 if (adapter->hw.mac.type == e1000_vfadapt_i350) 2849 adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP; 2850 2851 strcpy(netdev->name, "eth%d"); 2852 err = register_netdev(netdev); 2853 if (err) 2854 goto err_hw_init; 2855 2856 /* tell the stack to leave us alone until igbvf_open() is called */ 2857 netif_carrier_off(netdev); 2858 netif_stop_queue(netdev); 2859 2860 igbvf_print_device_info(adapter); 2861 2862 igbvf_initialize_last_counter_stats(adapter); 2863 2864 return 0; 2865 2866 err_hw_init: 2867 kfree(adapter->tx_ring); 2868 kfree(adapter->rx_ring); 2869 err_sw_init: 2870 igbvf_reset_interrupt_capability(adapter); 2871 err_get_variants: 2872 iounmap(adapter->hw.hw_addr); 2873 err_ioremap: 2874 free_netdev(netdev); 2875 err_alloc_etherdev: 2876 pci_release_regions(pdev); 2877 err_pci_reg: 2878 err_dma: 2879 pci_disable_device(pdev); 2880 return err; 2881 } 2882 2883 /** 2884 * igbvf_remove - Device Removal Routine 2885 * @pdev: PCI device information struct 2886 * 2887 * igbvf_remove is called by the PCI subsystem to alert the driver 2888 * that it should release a PCI device. The could be caused by a 2889 * Hot-Plug event, or because the driver is going to be removed from 2890 * memory. 2891 **/ 2892 static void igbvf_remove(struct pci_dev *pdev) 2893 { 2894 struct net_device *netdev = pci_get_drvdata(pdev); 2895 struct igbvf_adapter *adapter = netdev_priv(netdev); 2896 struct e1000_hw *hw = &adapter->hw; 2897 2898 /* The watchdog timer may be rescheduled, so explicitly 2899 * disable it from being rescheduled. 2900 */ 2901 set_bit(__IGBVF_DOWN, &adapter->state); 2902 del_timer_sync(&adapter->watchdog_timer); 2903 2904 cancel_work_sync(&adapter->reset_task); 2905 cancel_work_sync(&adapter->watchdog_task); 2906 2907 unregister_netdev(netdev); 2908 2909 igbvf_reset_interrupt_capability(adapter); 2910 2911 /* it is important to delete the NAPI struct prior to freeing the 2912 * Rx ring so that you do not end up with null pointer refs 2913 */ 2914 netif_napi_del(&adapter->rx_ring->napi); 2915 kfree(adapter->tx_ring); 2916 kfree(adapter->rx_ring); 2917 2918 iounmap(hw->hw_addr); 2919 if (hw->flash_address) 2920 iounmap(hw->flash_address); 2921 pci_release_regions(pdev); 2922 2923 free_netdev(netdev); 2924 2925 pci_disable_device(pdev); 2926 } 2927 2928 /* PCI Error Recovery (ERS) */ 2929 static const struct pci_error_handlers igbvf_err_handler = { 2930 .error_detected = igbvf_io_error_detected, 2931 .slot_reset = igbvf_io_slot_reset, 2932 .resume = igbvf_io_resume, 2933 }; 2934 2935 static const struct pci_device_id igbvf_pci_tbl[] = { 2936 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf }, 2937 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf }, 2938 { } /* terminate list */ 2939 }; 2940 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl); 2941 2942 /* PCI Device API Driver */ 2943 static struct pci_driver igbvf_driver = { 2944 .name = igbvf_driver_name, 2945 .id_table = igbvf_pci_tbl, 2946 .probe = igbvf_probe, 2947 .remove = igbvf_remove, 2948 #ifdef CONFIG_PM 2949 /* Power Management Hooks */ 2950 .suspend = igbvf_suspend, 2951 .resume = igbvf_resume, 2952 #endif 2953 .shutdown = igbvf_shutdown, 2954 .err_handler = &igbvf_err_handler 2955 }; 2956 2957 /** 2958 * igbvf_init_module - Driver Registration Routine 2959 * 2960 * igbvf_init_module is the first routine called when the driver is 2961 * loaded. All it does is register with the PCI subsystem. 2962 **/ 2963 static int __init igbvf_init_module(void) 2964 { 2965 int ret; 2966 2967 pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version); 2968 pr_info("%s\n", igbvf_copyright); 2969 2970 ret = pci_register_driver(&igbvf_driver); 2971 2972 return ret; 2973 } 2974 module_init(igbvf_init_module); 2975 2976 /** 2977 * igbvf_exit_module - Driver Exit Cleanup Routine 2978 * 2979 * igbvf_exit_module is called just before the driver is removed 2980 * from memory. 2981 **/ 2982 static void __exit igbvf_exit_module(void) 2983 { 2984 pci_unregister_driver(&igbvf_driver); 2985 } 2986 module_exit(igbvf_exit_module); 2987 2988 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>"); 2989 MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver"); 2990 MODULE_LICENSE("GPL"); 2991 MODULE_VERSION(DRV_VERSION); 2992 2993 /* netdev.c */ 2994