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 smp_rmb(); 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 1239 spin_lock_bh(&hw->mbx_lock); 1240 1241 e1000_rlpml_set_vf(hw, max_frame_size); 1242 1243 spin_unlock_bh(&hw->mbx_lock); 1244 } 1245 1246 static int igbvf_vlan_rx_add_vid(struct net_device *netdev, 1247 __be16 proto, u16 vid) 1248 { 1249 struct igbvf_adapter *adapter = netdev_priv(netdev); 1250 struct e1000_hw *hw = &adapter->hw; 1251 1252 spin_lock_bh(&hw->mbx_lock); 1253 1254 if (hw->mac.ops.set_vfta(hw, vid, true)) { 1255 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid); 1256 spin_unlock_bh(&hw->mbx_lock); 1257 return -EINVAL; 1258 } 1259 1260 spin_unlock_bh(&hw->mbx_lock); 1261 1262 set_bit(vid, adapter->active_vlans); 1263 return 0; 1264 } 1265 1266 static int igbvf_vlan_rx_kill_vid(struct net_device *netdev, 1267 __be16 proto, u16 vid) 1268 { 1269 struct igbvf_adapter *adapter = netdev_priv(netdev); 1270 struct e1000_hw *hw = &adapter->hw; 1271 1272 spin_lock_bh(&hw->mbx_lock); 1273 1274 if (hw->mac.ops.set_vfta(hw, vid, false)) { 1275 dev_err(&adapter->pdev->dev, 1276 "Failed to remove vlan id %d\n", vid); 1277 spin_unlock_bh(&hw->mbx_lock); 1278 return -EINVAL; 1279 } 1280 1281 spin_unlock_bh(&hw->mbx_lock); 1282 1283 clear_bit(vid, adapter->active_vlans); 1284 return 0; 1285 } 1286 1287 static void igbvf_restore_vlan(struct igbvf_adapter *adapter) 1288 { 1289 u16 vid; 1290 1291 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID) 1292 igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid); 1293 } 1294 1295 /** 1296 * igbvf_configure_tx - Configure Transmit Unit after Reset 1297 * @adapter: board private structure 1298 * 1299 * Configure the Tx unit of the MAC after a reset. 1300 **/ 1301 static void igbvf_configure_tx(struct igbvf_adapter *adapter) 1302 { 1303 struct e1000_hw *hw = &adapter->hw; 1304 struct igbvf_ring *tx_ring = adapter->tx_ring; 1305 u64 tdba; 1306 u32 txdctl, dca_txctrl; 1307 1308 /* disable transmits */ 1309 txdctl = er32(TXDCTL(0)); 1310 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE); 1311 e1e_flush(); 1312 msleep(10); 1313 1314 /* Setup the HW Tx Head and Tail descriptor pointers */ 1315 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc)); 1316 tdba = tx_ring->dma; 1317 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32))); 1318 ew32(TDBAH(0), (tdba >> 32)); 1319 ew32(TDH(0), 0); 1320 ew32(TDT(0), 0); 1321 tx_ring->head = E1000_TDH(0); 1322 tx_ring->tail = E1000_TDT(0); 1323 1324 /* Turn off Relaxed Ordering on head write-backs. The writebacks 1325 * MUST be delivered in order or it will completely screw up 1326 * our bookkeeping. 1327 */ 1328 dca_txctrl = er32(DCA_TXCTRL(0)); 1329 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN; 1330 ew32(DCA_TXCTRL(0), dca_txctrl); 1331 1332 /* enable transmits */ 1333 txdctl |= E1000_TXDCTL_QUEUE_ENABLE; 1334 ew32(TXDCTL(0), txdctl); 1335 1336 /* Setup Transmit Descriptor Settings for eop descriptor */ 1337 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS; 1338 1339 /* enable Report Status bit */ 1340 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS; 1341 } 1342 1343 /** 1344 * igbvf_setup_srrctl - configure the receive control registers 1345 * @adapter: Board private structure 1346 **/ 1347 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter) 1348 { 1349 struct e1000_hw *hw = &adapter->hw; 1350 u32 srrctl = 0; 1351 1352 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK | 1353 E1000_SRRCTL_BSIZEHDR_MASK | 1354 E1000_SRRCTL_BSIZEPKT_MASK); 1355 1356 /* Enable queue drop to avoid head of line blocking */ 1357 srrctl |= E1000_SRRCTL_DROP_EN; 1358 1359 /* Setup buffer sizes */ 1360 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >> 1361 E1000_SRRCTL_BSIZEPKT_SHIFT; 1362 1363 if (adapter->rx_buffer_len < 2048) { 1364 adapter->rx_ps_hdr_size = 0; 1365 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF; 1366 } else { 1367 adapter->rx_ps_hdr_size = 128; 1368 srrctl |= adapter->rx_ps_hdr_size << 1369 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT; 1370 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS; 1371 } 1372 1373 ew32(SRRCTL(0), srrctl); 1374 } 1375 1376 /** 1377 * igbvf_configure_rx - Configure Receive Unit after Reset 1378 * @adapter: board private structure 1379 * 1380 * Configure the Rx unit of the MAC after a reset. 1381 **/ 1382 static void igbvf_configure_rx(struct igbvf_adapter *adapter) 1383 { 1384 struct e1000_hw *hw = &adapter->hw; 1385 struct igbvf_ring *rx_ring = adapter->rx_ring; 1386 u64 rdba; 1387 u32 rxdctl; 1388 1389 /* disable receives */ 1390 rxdctl = er32(RXDCTL(0)); 1391 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE); 1392 e1e_flush(); 1393 msleep(10); 1394 1395 /* Setup the HW Rx Head and Tail Descriptor Pointers and 1396 * the Base and Length of the Rx Descriptor Ring 1397 */ 1398 rdba = rx_ring->dma; 1399 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32))); 1400 ew32(RDBAH(0), (rdba >> 32)); 1401 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc)); 1402 rx_ring->head = E1000_RDH(0); 1403 rx_ring->tail = E1000_RDT(0); 1404 ew32(RDH(0), 0); 1405 ew32(RDT(0), 0); 1406 1407 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE; 1408 rxdctl &= 0xFFF00000; 1409 rxdctl |= IGBVF_RX_PTHRESH; 1410 rxdctl |= IGBVF_RX_HTHRESH << 8; 1411 rxdctl |= IGBVF_RX_WTHRESH << 16; 1412 1413 igbvf_set_rlpml(adapter); 1414 1415 /* enable receives */ 1416 ew32(RXDCTL(0), rxdctl); 1417 } 1418 1419 /** 1420 * igbvf_set_multi - Multicast and Promiscuous mode set 1421 * @netdev: network interface device structure 1422 * 1423 * The set_multi entry point is called whenever the multicast address 1424 * list or the network interface flags are updated. This routine is 1425 * responsible for configuring the hardware for proper multicast, 1426 * promiscuous mode, and all-multi behavior. 1427 **/ 1428 static void igbvf_set_multi(struct net_device *netdev) 1429 { 1430 struct igbvf_adapter *adapter = netdev_priv(netdev); 1431 struct e1000_hw *hw = &adapter->hw; 1432 struct netdev_hw_addr *ha; 1433 u8 *mta_list = NULL; 1434 int i; 1435 1436 if (!netdev_mc_empty(netdev)) { 1437 mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN, 1438 GFP_ATOMIC); 1439 if (!mta_list) 1440 return; 1441 } 1442 1443 /* prepare a packed array of only addresses. */ 1444 i = 0; 1445 netdev_for_each_mc_addr(ha, netdev) 1446 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN); 1447 1448 spin_lock_bh(&hw->mbx_lock); 1449 1450 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0); 1451 1452 spin_unlock_bh(&hw->mbx_lock); 1453 kfree(mta_list); 1454 } 1455 1456 /** 1457 * igbvf_set_uni - Configure unicast MAC filters 1458 * @netdev: network interface device structure 1459 * 1460 * This routine is responsible for configuring the hardware for proper 1461 * unicast filters. 1462 **/ 1463 static int igbvf_set_uni(struct net_device *netdev) 1464 { 1465 struct igbvf_adapter *adapter = netdev_priv(netdev); 1466 struct e1000_hw *hw = &adapter->hw; 1467 1468 if (netdev_uc_count(netdev) > IGBVF_MAX_MAC_FILTERS) { 1469 pr_err("Too many unicast filters - No Space\n"); 1470 return -ENOSPC; 1471 } 1472 1473 spin_lock_bh(&hw->mbx_lock); 1474 1475 /* Clear all unicast MAC filters */ 1476 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_CLR, NULL); 1477 1478 spin_unlock_bh(&hw->mbx_lock); 1479 1480 if (!netdev_uc_empty(netdev)) { 1481 struct netdev_hw_addr *ha; 1482 1483 /* Add MAC filters one by one */ 1484 netdev_for_each_uc_addr(ha, netdev) { 1485 spin_lock_bh(&hw->mbx_lock); 1486 1487 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_ADD, 1488 ha->addr); 1489 1490 spin_unlock_bh(&hw->mbx_lock); 1491 udelay(200); 1492 } 1493 } 1494 1495 return 0; 1496 } 1497 1498 static void igbvf_set_rx_mode(struct net_device *netdev) 1499 { 1500 igbvf_set_multi(netdev); 1501 igbvf_set_uni(netdev); 1502 } 1503 1504 /** 1505 * igbvf_configure - configure the hardware for Rx and Tx 1506 * @adapter: private board structure 1507 **/ 1508 static void igbvf_configure(struct igbvf_adapter *adapter) 1509 { 1510 igbvf_set_rx_mode(adapter->netdev); 1511 1512 igbvf_restore_vlan(adapter); 1513 1514 igbvf_configure_tx(adapter); 1515 igbvf_setup_srrctl(adapter); 1516 igbvf_configure_rx(adapter); 1517 igbvf_alloc_rx_buffers(adapter->rx_ring, 1518 igbvf_desc_unused(adapter->rx_ring)); 1519 } 1520 1521 /* igbvf_reset - bring the hardware into a known good state 1522 * @adapter: private board structure 1523 * 1524 * This function boots the hardware and enables some settings that 1525 * require a configuration cycle of the hardware - those cannot be 1526 * set/changed during runtime. After reset the device needs to be 1527 * properly configured for Rx, Tx etc. 1528 */ 1529 static void igbvf_reset(struct igbvf_adapter *adapter) 1530 { 1531 struct e1000_mac_info *mac = &adapter->hw.mac; 1532 struct net_device *netdev = adapter->netdev; 1533 struct e1000_hw *hw = &adapter->hw; 1534 1535 spin_lock_bh(&hw->mbx_lock); 1536 1537 /* Allow time for pending master requests to run */ 1538 if (mac->ops.reset_hw(hw)) 1539 dev_err(&adapter->pdev->dev, "PF still resetting\n"); 1540 1541 mac->ops.init_hw(hw); 1542 1543 spin_unlock_bh(&hw->mbx_lock); 1544 1545 if (is_valid_ether_addr(adapter->hw.mac.addr)) { 1546 memcpy(netdev->dev_addr, adapter->hw.mac.addr, 1547 netdev->addr_len); 1548 memcpy(netdev->perm_addr, adapter->hw.mac.addr, 1549 netdev->addr_len); 1550 } 1551 1552 adapter->last_reset = jiffies; 1553 } 1554 1555 int igbvf_up(struct igbvf_adapter *adapter) 1556 { 1557 struct e1000_hw *hw = &adapter->hw; 1558 1559 /* hardware has been reset, we need to reload some things */ 1560 igbvf_configure(adapter); 1561 1562 clear_bit(__IGBVF_DOWN, &adapter->state); 1563 1564 napi_enable(&adapter->rx_ring->napi); 1565 if (adapter->msix_entries) 1566 igbvf_configure_msix(adapter); 1567 1568 /* Clear any pending interrupts. */ 1569 er32(EICR); 1570 igbvf_irq_enable(adapter); 1571 1572 /* start the watchdog */ 1573 hw->mac.get_link_status = 1; 1574 mod_timer(&adapter->watchdog_timer, jiffies + 1); 1575 1576 return 0; 1577 } 1578 1579 void igbvf_down(struct igbvf_adapter *adapter) 1580 { 1581 struct net_device *netdev = adapter->netdev; 1582 struct e1000_hw *hw = &adapter->hw; 1583 u32 rxdctl, txdctl; 1584 1585 /* signal that we're down so the interrupt handler does not 1586 * reschedule our watchdog timer 1587 */ 1588 set_bit(__IGBVF_DOWN, &adapter->state); 1589 1590 /* disable receives in the hardware */ 1591 rxdctl = er32(RXDCTL(0)); 1592 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE); 1593 1594 netif_carrier_off(netdev); 1595 netif_stop_queue(netdev); 1596 1597 /* disable transmits in the hardware */ 1598 txdctl = er32(TXDCTL(0)); 1599 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE); 1600 1601 /* flush both disables and wait for them to finish */ 1602 e1e_flush(); 1603 msleep(10); 1604 1605 napi_disable(&adapter->rx_ring->napi); 1606 1607 igbvf_irq_disable(adapter); 1608 1609 del_timer_sync(&adapter->watchdog_timer); 1610 1611 /* record the stats before reset*/ 1612 igbvf_update_stats(adapter); 1613 1614 adapter->link_speed = 0; 1615 adapter->link_duplex = 0; 1616 1617 igbvf_reset(adapter); 1618 igbvf_clean_tx_ring(adapter->tx_ring); 1619 igbvf_clean_rx_ring(adapter->rx_ring); 1620 } 1621 1622 void igbvf_reinit_locked(struct igbvf_adapter *adapter) 1623 { 1624 might_sleep(); 1625 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state)) 1626 usleep_range(1000, 2000); 1627 igbvf_down(adapter); 1628 igbvf_up(adapter); 1629 clear_bit(__IGBVF_RESETTING, &adapter->state); 1630 } 1631 1632 /** 1633 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter) 1634 * @adapter: board private structure to initialize 1635 * 1636 * igbvf_sw_init initializes the Adapter private data structure. 1637 * Fields are initialized based on PCI device information and 1638 * OS network device settings (MTU size). 1639 **/ 1640 static int igbvf_sw_init(struct igbvf_adapter *adapter) 1641 { 1642 struct net_device *netdev = adapter->netdev; 1643 s32 rc; 1644 1645 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN; 1646 adapter->rx_ps_hdr_size = 0; 1647 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN; 1648 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN; 1649 1650 adapter->tx_int_delay = 8; 1651 adapter->tx_abs_int_delay = 32; 1652 adapter->rx_int_delay = 0; 1653 adapter->rx_abs_int_delay = 8; 1654 adapter->requested_itr = 3; 1655 adapter->current_itr = IGBVF_START_ITR; 1656 1657 /* Set various function pointers */ 1658 adapter->ei->init_ops(&adapter->hw); 1659 1660 rc = adapter->hw.mac.ops.init_params(&adapter->hw); 1661 if (rc) 1662 return rc; 1663 1664 rc = adapter->hw.mbx.ops.init_params(&adapter->hw); 1665 if (rc) 1666 return rc; 1667 1668 igbvf_set_interrupt_capability(adapter); 1669 1670 if (igbvf_alloc_queues(adapter)) 1671 return -ENOMEM; 1672 1673 spin_lock_init(&adapter->tx_queue_lock); 1674 1675 /* Explicitly disable IRQ since the NIC can be in any state. */ 1676 igbvf_irq_disable(adapter); 1677 1678 spin_lock_init(&adapter->stats_lock); 1679 spin_lock_init(&adapter->hw.mbx_lock); 1680 1681 set_bit(__IGBVF_DOWN, &adapter->state); 1682 return 0; 1683 } 1684 1685 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter) 1686 { 1687 struct e1000_hw *hw = &adapter->hw; 1688 1689 adapter->stats.last_gprc = er32(VFGPRC); 1690 adapter->stats.last_gorc = er32(VFGORC); 1691 adapter->stats.last_gptc = er32(VFGPTC); 1692 adapter->stats.last_gotc = er32(VFGOTC); 1693 adapter->stats.last_mprc = er32(VFMPRC); 1694 adapter->stats.last_gotlbc = er32(VFGOTLBC); 1695 adapter->stats.last_gptlbc = er32(VFGPTLBC); 1696 adapter->stats.last_gorlbc = er32(VFGORLBC); 1697 adapter->stats.last_gprlbc = er32(VFGPRLBC); 1698 1699 adapter->stats.base_gprc = er32(VFGPRC); 1700 adapter->stats.base_gorc = er32(VFGORC); 1701 adapter->stats.base_gptc = er32(VFGPTC); 1702 adapter->stats.base_gotc = er32(VFGOTC); 1703 adapter->stats.base_mprc = er32(VFMPRC); 1704 adapter->stats.base_gotlbc = er32(VFGOTLBC); 1705 adapter->stats.base_gptlbc = er32(VFGPTLBC); 1706 adapter->stats.base_gorlbc = er32(VFGORLBC); 1707 adapter->stats.base_gprlbc = er32(VFGPRLBC); 1708 } 1709 1710 /** 1711 * igbvf_open - Called when a network interface is made active 1712 * @netdev: network interface device structure 1713 * 1714 * Returns 0 on success, negative value on failure 1715 * 1716 * The open entry point is called when a network interface is made 1717 * active by the system (IFF_UP). At this point all resources needed 1718 * for transmit and receive operations are allocated, the interrupt 1719 * handler is registered with the OS, the watchdog timer is started, 1720 * and the stack is notified that the interface is ready. 1721 **/ 1722 static int igbvf_open(struct net_device *netdev) 1723 { 1724 struct igbvf_adapter *adapter = netdev_priv(netdev); 1725 struct e1000_hw *hw = &adapter->hw; 1726 int err; 1727 1728 /* disallow open during test */ 1729 if (test_bit(__IGBVF_TESTING, &adapter->state)) 1730 return -EBUSY; 1731 1732 /* allocate transmit descriptors */ 1733 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring); 1734 if (err) 1735 goto err_setup_tx; 1736 1737 /* allocate receive descriptors */ 1738 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring); 1739 if (err) 1740 goto err_setup_rx; 1741 1742 /* before we allocate an interrupt, we must be ready to handle it. 1743 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt 1744 * as soon as we call pci_request_irq, so we have to setup our 1745 * clean_rx handler before we do so. 1746 */ 1747 igbvf_configure(adapter); 1748 1749 err = igbvf_request_irq(adapter); 1750 if (err) 1751 goto err_req_irq; 1752 1753 /* From here on the code is the same as igbvf_up() */ 1754 clear_bit(__IGBVF_DOWN, &adapter->state); 1755 1756 napi_enable(&adapter->rx_ring->napi); 1757 1758 /* clear any pending interrupts */ 1759 er32(EICR); 1760 1761 igbvf_irq_enable(adapter); 1762 1763 /* start the watchdog */ 1764 hw->mac.get_link_status = 1; 1765 mod_timer(&adapter->watchdog_timer, jiffies + 1); 1766 1767 return 0; 1768 1769 err_req_irq: 1770 igbvf_free_rx_resources(adapter->rx_ring); 1771 err_setup_rx: 1772 igbvf_free_tx_resources(adapter->tx_ring); 1773 err_setup_tx: 1774 igbvf_reset(adapter); 1775 1776 return err; 1777 } 1778 1779 /** 1780 * igbvf_close - Disables a network interface 1781 * @netdev: network interface device structure 1782 * 1783 * Returns 0, this is not allowed to fail 1784 * 1785 * The close entry point is called when an interface is de-activated 1786 * by the OS. The hardware is still under the drivers control, but 1787 * needs to be disabled. A global MAC reset is issued to stop the 1788 * hardware, and all transmit and receive resources are freed. 1789 **/ 1790 static int igbvf_close(struct net_device *netdev) 1791 { 1792 struct igbvf_adapter *adapter = netdev_priv(netdev); 1793 1794 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state)); 1795 igbvf_down(adapter); 1796 1797 igbvf_free_irq(adapter); 1798 1799 igbvf_free_tx_resources(adapter->tx_ring); 1800 igbvf_free_rx_resources(adapter->rx_ring); 1801 1802 return 0; 1803 } 1804 1805 /** 1806 * igbvf_set_mac - Change the Ethernet Address of the NIC 1807 * @netdev: network interface device structure 1808 * @p: pointer to an address structure 1809 * 1810 * Returns 0 on success, negative on failure 1811 **/ 1812 static int igbvf_set_mac(struct net_device *netdev, void *p) 1813 { 1814 struct igbvf_adapter *adapter = netdev_priv(netdev); 1815 struct e1000_hw *hw = &adapter->hw; 1816 struct sockaddr *addr = p; 1817 1818 if (!is_valid_ether_addr(addr->sa_data)) 1819 return -EADDRNOTAVAIL; 1820 1821 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len); 1822 1823 spin_lock_bh(&hw->mbx_lock); 1824 1825 hw->mac.ops.rar_set(hw, hw->mac.addr, 0); 1826 1827 spin_unlock_bh(&hw->mbx_lock); 1828 1829 if (!ether_addr_equal(addr->sa_data, hw->mac.addr)) 1830 return -EADDRNOTAVAIL; 1831 1832 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len); 1833 1834 return 0; 1835 } 1836 1837 #define UPDATE_VF_COUNTER(reg, name) \ 1838 { \ 1839 u32 current_counter = er32(reg); \ 1840 if (current_counter < adapter->stats.last_##name) \ 1841 adapter->stats.name += 0x100000000LL; \ 1842 adapter->stats.last_##name = current_counter; \ 1843 adapter->stats.name &= 0xFFFFFFFF00000000LL; \ 1844 adapter->stats.name |= current_counter; \ 1845 } 1846 1847 /** 1848 * igbvf_update_stats - Update the board statistics counters 1849 * @adapter: board private structure 1850 **/ 1851 void igbvf_update_stats(struct igbvf_adapter *adapter) 1852 { 1853 struct e1000_hw *hw = &adapter->hw; 1854 struct pci_dev *pdev = adapter->pdev; 1855 1856 /* Prevent stats update while adapter is being reset, link is down 1857 * or if the pci connection is down. 1858 */ 1859 if (adapter->link_speed == 0) 1860 return; 1861 1862 if (test_bit(__IGBVF_RESETTING, &adapter->state)) 1863 return; 1864 1865 if (pci_channel_offline(pdev)) 1866 return; 1867 1868 UPDATE_VF_COUNTER(VFGPRC, gprc); 1869 UPDATE_VF_COUNTER(VFGORC, gorc); 1870 UPDATE_VF_COUNTER(VFGPTC, gptc); 1871 UPDATE_VF_COUNTER(VFGOTC, gotc); 1872 UPDATE_VF_COUNTER(VFMPRC, mprc); 1873 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc); 1874 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc); 1875 UPDATE_VF_COUNTER(VFGORLBC, gorlbc); 1876 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc); 1877 1878 /* Fill out the OS statistics structure */ 1879 adapter->netdev->stats.multicast = adapter->stats.mprc; 1880 } 1881 1882 static void igbvf_print_link_info(struct igbvf_adapter *adapter) 1883 { 1884 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n", 1885 adapter->link_speed, 1886 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half"); 1887 } 1888 1889 static bool igbvf_has_link(struct igbvf_adapter *adapter) 1890 { 1891 struct e1000_hw *hw = &adapter->hw; 1892 s32 ret_val = E1000_SUCCESS; 1893 bool link_active; 1894 1895 /* If interface is down, stay link down */ 1896 if (test_bit(__IGBVF_DOWN, &adapter->state)) 1897 return false; 1898 1899 spin_lock_bh(&hw->mbx_lock); 1900 1901 ret_val = hw->mac.ops.check_for_link(hw); 1902 1903 spin_unlock_bh(&hw->mbx_lock); 1904 1905 link_active = !hw->mac.get_link_status; 1906 1907 /* if check for link returns error we will need to reset */ 1908 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ))) 1909 schedule_work(&adapter->reset_task); 1910 1911 return link_active; 1912 } 1913 1914 /** 1915 * igbvf_watchdog - Timer Call-back 1916 * @data: pointer to adapter cast into an unsigned long 1917 **/ 1918 static void igbvf_watchdog(struct timer_list *t) 1919 { 1920 struct igbvf_adapter *adapter = from_timer(adapter, t, watchdog_timer); 1921 1922 /* Do the rest outside of interrupt context */ 1923 schedule_work(&adapter->watchdog_task); 1924 } 1925 1926 static void igbvf_watchdog_task(struct work_struct *work) 1927 { 1928 struct igbvf_adapter *adapter = container_of(work, 1929 struct igbvf_adapter, 1930 watchdog_task); 1931 struct net_device *netdev = adapter->netdev; 1932 struct e1000_mac_info *mac = &adapter->hw.mac; 1933 struct igbvf_ring *tx_ring = adapter->tx_ring; 1934 struct e1000_hw *hw = &adapter->hw; 1935 u32 link; 1936 int tx_pending = 0; 1937 1938 link = igbvf_has_link(adapter); 1939 1940 if (link) { 1941 if (!netif_carrier_ok(netdev)) { 1942 mac->ops.get_link_up_info(&adapter->hw, 1943 &adapter->link_speed, 1944 &adapter->link_duplex); 1945 igbvf_print_link_info(adapter); 1946 1947 netif_carrier_on(netdev); 1948 netif_wake_queue(netdev); 1949 } 1950 } else { 1951 if (netif_carrier_ok(netdev)) { 1952 adapter->link_speed = 0; 1953 adapter->link_duplex = 0; 1954 dev_info(&adapter->pdev->dev, "Link is Down\n"); 1955 netif_carrier_off(netdev); 1956 netif_stop_queue(netdev); 1957 } 1958 } 1959 1960 if (netif_carrier_ok(netdev)) { 1961 igbvf_update_stats(adapter); 1962 } else { 1963 tx_pending = (igbvf_desc_unused(tx_ring) + 1 < 1964 tx_ring->count); 1965 if (tx_pending) { 1966 /* We've lost link, so the controller stops DMA, 1967 * but we've got queued Tx work that's never going 1968 * to get done, so reset controller to flush Tx. 1969 * (Do the reset outside of interrupt context). 1970 */ 1971 adapter->tx_timeout_count++; 1972 schedule_work(&adapter->reset_task); 1973 } 1974 } 1975 1976 /* Cause software interrupt to ensure Rx ring is cleaned */ 1977 ew32(EICS, adapter->rx_ring->eims_value); 1978 1979 /* Reset the timer */ 1980 if (!test_bit(__IGBVF_DOWN, &adapter->state)) 1981 mod_timer(&adapter->watchdog_timer, 1982 round_jiffies(jiffies + (2 * HZ))); 1983 } 1984 1985 #define IGBVF_TX_FLAGS_CSUM 0x00000001 1986 #define IGBVF_TX_FLAGS_VLAN 0x00000002 1987 #define IGBVF_TX_FLAGS_TSO 0x00000004 1988 #define IGBVF_TX_FLAGS_IPV4 0x00000008 1989 #define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000 1990 #define IGBVF_TX_FLAGS_VLAN_SHIFT 16 1991 1992 static void igbvf_tx_ctxtdesc(struct igbvf_ring *tx_ring, u32 vlan_macip_lens, 1993 u32 type_tucmd, u32 mss_l4len_idx) 1994 { 1995 struct e1000_adv_tx_context_desc *context_desc; 1996 struct igbvf_buffer *buffer_info; 1997 u16 i = tx_ring->next_to_use; 1998 1999 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i); 2000 buffer_info = &tx_ring->buffer_info[i]; 2001 2002 i++; 2003 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; 2004 2005 /* set bits to identify this as an advanced context descriptor */ 2006 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT; 2007 2008 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens); 2009 context_desc->seqnum_seed = 0; 2010 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd); 2011 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx); 2012 2013 buffer_info->time_stamp = jiffies; 2014 buffer_info->dma = 0; 2015 } 2016 2017 static int igbvf_tso(struct igbvf_ring *tx_ring, 2018 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len) 2019 { 2020 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx; 2021 union { 2022 struct iphdr *v4; 2023 struct ipv6hdr *v6; 2024 unsigned char *hdr; 2025 } ip; 2026 union { 2027 struct tcphdr *tcp; 2028 unsigned char *hdr; 2029 } l4; 2030 u32 paylen, l4_offset; 2031 int err; 2032 2033 if (skb->ip_summed != CHECKSUM_PARTIAL) 2034 return 0; 2035 2036 if (!skb_is_gso(skb)) 2037 return 0; 2038 2039 err = skb_cow_head(skb, 0); 2040 if (err < 0) 2041 return err; 2042 2043 ip.hdr = skb_network_header(skb); 2044 l4.hdr = skb_checksum_start(skb); 2045 2046 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */ 2047 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP; 2048 2049 /* initialize outer IP header fields */ 2050 if (ip.v4->version == 4) { 2051 unsigned char *csum_start = skb_checksum_start(skb); 2052 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4); 2053 2054 /* IP header will have to cancel out any data that 2055 * is not a part of the outer IP header 2056 */ 2057 ip.v4->check = csum_fold(csum_partial(trans_start, 2058 csum_start - trans_start, 2059 0)); 2060 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4; 2061 2062 ip.v4->tot_len = 0; 2063 } else { 2064 ip.v6->payload_len = 0; 2065 } 2066 2067 /* determine offset of inner transport header */ 2068 l4_offset = l4.hdr - skb->data; 2069 2070 /* compute length of segmentation header */ 2071 *hdr_len = (l4.tcp->doff * 4) + l4_offset; 2072 2073 /* remove payload length from inner checksum */ 2074 paylen = skb->len - l4_offset; 2075 csum_replace_by_diff(&l4.tcp->check, htonl(paylen)); 2076 2077 /* MSS L4LEN IDX */ 2078 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT; 2079 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT; 2080 2081 /* VLAN MACLEN IPLEN */ 2082 vlan_macip_lens = l4.hdr - ip.hdr; 2083 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT; 2084 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK; 2085 2086 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx); 2087 2088 return 1; 2089 } 2090 2091 static inline bool igbvf_ipv6_csum_is_sctp(struct sk_buff *skb) 2092 { 2093 unsigned int offset = 0; 2094 2095 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL); 2096 2097 return offset == skb_checksum_start_offset(skb); 2098 } 2099 2100 static bool igbvf_tx_csum(struct igbvf_ring *tx_ring, struct sk_buff *skb, 2101 u32 tx_flags, __be16 protocol) 2102 { 2103 u32 vlan_macip_lens = 0; 2104 u32 type_tucmd = 0; 2105 2106 if (skb->ip_summed != CHECKSUM_PARTIAL) { 2107 csum_failed: 2108 if (!(tx_flags & IGBVF_TX_FLAGS_VLAN)) 2109 return false; 2110 goto no_csum; 2111 } 2112 2113 switch (skb->csum_offset) { 2114 case offsetof(struct tcphdr, check): 2115 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP; 2116 /* fall through */ 2117 case offsetof(struct udphdr, check): 2118 break; 2119 case offsetof(struct sctphdr, checksum): 2120 /* validate that this is actually an SCTP request */ 2121 if (((protocol == htons(ETH_P_IP)) && 2122 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) || 2123 ((protocol == htons(ETH_P_IPV6)) && 2124 igbvf_ipv6_csum_is_sctp(skb))) { 2125 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP; 2126 break; 2127 } 2128 default: 2129 skb_checksum_help(skb); 2130 goto csum_failed; 2131 } 2132 2133 vlan_macip_lens = skb_checksum_start_offset(skb) - 2134 skb_network_offset(skb); 2135 no_csum: 2136 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT; 2137 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK; 2138 2139 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0); 2140 return true; 2141 } 2142 2143 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size) 2144 { 2145 struct igbvf_adapter *adapter = netdev_priv(netdev); 2146 2147 /* there is enough descriptors then we don't need to worry */ 2148 if (igbvf_desc_unused(adapter->tx_ring) >= size) 2149 return 0; 2150 2151 netif_stop_queue(netdev); 2152 2153 /* Herbert's original patch had: 2154 * smp_mb__after_netif_stop_queue(); 2155 * but since that doesn't exist yet, just open code it. 2156 */ 2157 smp_mb(); 2158 2159 /* We need to check again just in case room has been made available */ 2160 if (igbvf_desc_unused(adapter->tx_ring) < size) 2161 return -EBUSY; 2162 2163 netif_wake_queue(netdev); 2164 2165 ++adapter->restart_queue; 2166 return 0; 2167 } 2168 2169 #define IGBVF_MAX_TXD_PWR 16 2170 #define IGBVF_MAX_DATA_PER_TXD (1u << IGBVF_MAX_TXD_PWR) 2171 2172 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter, 2173 struct igbvf_ring *tx_ring, 2174 struct sk_buff *skb) 2175 { 2176 struct igbvf_buffer *buffer_info; 2177 struct pci_dev *pdev = adapter->pdev; 2178 unsigned int len = skb_headlen(skb); 2179 unsigned int count = 0, i; 2180 unsigned int f; 2181 2182 i = tx_ring->next_to_use; 2183 2184 buffer_info = &tx_ring->buffer_info[i]; 2185 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD); 2186 buffer_info->length = len; 2187 /* set time_stamp *before* dma to help avoid a possible race */ 2188 buffer_info->time_stamp = jiffies; 2189 buffer_info->mapped_as_page = false; 2190 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len, 2191 DMA_TO_DEVICE); 2192 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) 2193 goto dma_error; 2194 2195 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) { 2196 const struct skb_frag_struct *frag; 2197 2198 count++; 2199 i++; 2200 if (i == tx_ring->count) 2201 i = 0; 2202 2203 frag = &skb_shinfo(skb)->frags[f]; 2204 len = skb_frag_size(frag); 2205 2206 buffer_info = &tx_ring->buffer_info[i]; 2207 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD); 2208 buffer_info->length = len; 2209 buffer_info->time_stamp = jiffies; 2210 buffer_info->mapped_as_page = true; 2211 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len, 2212 DMA_TO_DEVICE); 2213 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) 2214 goto dma_error; 2215 } 2216 2217 tx_ring->buffer_info[i].skb = skb; 2218 2219 return ++count; 2220 2221 dma_error: 2222 dev_err(&pdev->dev, "TX DMA map failed\n"); 2223 2224 /* clear timestamp and dma mappings for failed buffer_info mapping */ 2225 buffer_info->dma = 0; 2226 buffer_info->time_stamp = 0; 2227 buffer_info->length = 0; 2228 buffer_info->mapped_as_page = false; 2229 if (count) 2230 count--; 2231 2232 /* clear timestamp and dma mappings for remaining portion of packet */ 2233 while (count--) { 2234 if (i == 0) 2235 i += tx_ring->count; 2236 i--; 2237 buffer_info = &tx_ring->buffer_info[i]; 2238 igbvf_put_txbuf(adapter, buffer_info); 2239 } 2240 2241 return 0; 2242 } 2243 2244 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter, 2245 struct igbvf_ring *tx_ring, 2246 int tx_flags, int count, 2247 unsigned int first, u32 paylen, 2248 u8 hdr_len) 2249 { 2250 union e1000_adv_tx_desc *tx_desc = NULL; 2251 struct igbvf_buffer *buffer_info; 2252 u32 olinfo_status = 0, cmd_type_len; 2253 unsigned int i; 2254 2255 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS | 2256 E1000_ADVTXD_DCMD_DEXT); 2257 2258 if (tx_flags & IGBVF_TX_FLAGS_VLAN) 2259 cmd_type_len |= E1000_ADVTXD_DCMD_VLE; 2260 2261 if (tx_flags & IGBVF_TX_FLAGS_TSO) { 2262 cmd_type_len |= E1000_ADVTXD_DCMD_TSE; 2263 2264 /* insert tcp checksum */ 2265 olinfo_status |= E1000_TXD_POPTS_TXSM << 8; 2266 2267 /* insert ip checksum */ 2268 if (tx_flags & IGBVF_TX_FLAGS_IPV4) 2269 olinfo_status |= E1000_TXD_POPTS_IXSM << 8; 2270 2271 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) { 2272 olinfo_status |= E1000_TXD_POPTS_TXSM << 8; 2273 } 2274 2275 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT); 2276 2277 i = tx_ring->next_to_use; 2278 while (count--) { 2279 buffer_info = &tx_ring->buffer_info[i]; 2280 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i); 2281 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma); 2282 tx_desc->read.cmd_type_len = 2283 cpu_to_le32(cmd_type_len | buffer_info->length); 2284 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status); 2285 i++; 2286 if (i == tx_ring->count) 2287 i = 0; 2288 } 2289 2290 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd); 2291 /* Force memory writes to complete before letting h/w 2292 * know there are new descriptors to fetch. (Only 2293 * applicable for weak-ordered memory model archs, 2294 * such as IA-64). 2295 */ 2296 wmb(); 2297 2298 tx_ring->buffer_info[first].next_to_watch = tx_desc; 2299 tx_ring->next_to_use = i; 2300 writel(i, adapter->hw.hw_addr + tx_ring->tail); 2301 /* we need this if more than one processor can write to our tail 2302 * at a time, it synchronizes IO on IA64/Altix systems 2303 */ 2304 mmiowb(); 2305 } 2306 2307 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb, 2308 struct net_device *netdev, 2309 struct igbvf_ring *tx_ring) 2310 { 2311 struct igbvf_adapter *adapter = netdev_priv(netdev); 2312 unsigned int first, tx_flags = 0; 2313 u8 hdr_len = 0; 2314 int count = 0; 2315 int tso = 0; 2316 __be16 protocol = vlan_get_protocol(skb); 2317 2318 if (test_bit(__IGBVF_DOWN, &adapter->state)) { 2319 dev_kfree_skb_any(skb); 2320 return NETDEV_TX_OK; 2321 } 2322 2323 if (skb->len <= 0) { 2324 dev_kfree_skb_any(skb); 2325 return NETDEV_TX_OK; 2326 } 2327 2328 /* need: count + 4 desc gap to keep tail from touching 2329 * + 2 desc gap to keep tail from touching head, 2330 * + 1 desc for skb->data, 2331 * + 1 desc for context descriptor, 2332 * head, otherwise try next time 2333 */ 2334 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) { 2335 /* this is a hard error */ 2336 return NETDEV_TX_BUSY; 2337 } 2338 2339 if (skb_vlan_tag_present(skb)) { 2340 tx_flags |= IGBVF_TX_FLAGS_VLAN; 2341 tx_flags |= (skb_vlan_tag_get(skb) << 2342 IGBVF_TX_FLAGS_VLAN_SHIFT); 2343 } 2344 2345 if (protocol == htons(ETH_P_IP)) 2346 tx_flags |= IGBVF_TX_FLAGS_IPV4; 2347 2348 first = tx_ring->next_to_use; 2349 2350 tso = igbvf_tso(tx_ring, skb, tx_flags, &hdr_len); 2351 if (unlikely(tso < 0)) { 2352 dev_kfree_skb_any(skb); 2353 return NETDEV_TX_OK; 2354 } 2355 2356 if (tso) 2357 tx_flags |= IGBVF_TX_FLAGS_TSO; 2358 else if (igbvf_tx_csum(tx_ring, skb, tx_flags, protocol) && 2359 (skb->ip_summed == CHECKSUM_PARTIAL)) 2360 tx_flags |= IGBVF_TX_FLAGS_CSUM; 2361 2362 /* count reflects descriptors mapped, if 0 then mapping error 2363 * has occurred and we need to rewind the descriptor queue 2364 */ 2365 count = igbvf_tx_map_adv(adapter, tx_ring, skb); 2366 2367 if (count) { 2368 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count, 2369 first, skb->len, hdr_len); 2370 /* Make sure there is space in the ring for the next send. */ 2371 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4); 2372 } else { 2373 dev_kfree_skb_any(skb); 2374 tx_ring->buffer_info[first].time_stamp = 0; 2375 tx_ring->next_to_use = first; 2376 } 2377 2378 return NETDEV_TX_OK; 2379 } 2380 2381 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb, 2382 struct net_device *netdev) 2383 { 2384 struct igbvf_adapter *adapter = netdev_priv(netdev); 2385 struct igbvf_ring *tx_ring; 2386 2387 if (test_bit(__IGBVF_DOWN, &adapter->state)) { 2388 dev_kfree_skb_any(skb); 2389 return NETDEV_TX_OK; 2390 } 2391 2392 tx_ring = &adapter->tx_ring[0]; 2393 2394 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring); 2395 } 2396 2397 /** 2398 * igbvf_tx_timeout - Respond to a Tx Hang 2399 * @netdev: network interface device structure 2400 **/ 2401 static void igbvf_tx_timeout(struct net_device *netdev) 2402 { 2403 struct igbvf_adapter *adapter = netdev_priv(netdev); 2404 2405 /* Do the reset outside of interrupt context */ 2406 adapter->tx_timeout_count++; 2407 schedule_work(&adapter->reset_task); 2408 } 2409 2410 static void igbvf_reset_task(struct work_struct *work) 2411 { 2412 struct igbvf_adapter *adapter; 2413 2414 adapter = container_of(work, struct igbvf_adapter, reset_task); 2415 2416 igbvf_reinit_locked(adapter); 2417 } 2418 2419 /** 2420 * igbvf_change_mtu - Change the Maximum Transfer Unit 2421 * @netdev: network interface device structure 2422 * @new_mtu: new value for maximum frame size 2423 * 2424 * Returns 0 on success, negative on failure 2425 **/ 2426 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu) 2427 { 2428 struct igbvf_adapter *adapter = netdev_priv(netdev); 2429 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN; 2430 2431 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state)) 2432 usleep_range(1000, 2000); 2433 /* igbvf_down has a dependency on max_frame_size */ 2434 adapter->max_frame_size = max_frame; 2435 if (netif_running(netdev)) 2436 igbvf_down(adapter); 2437 2438 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN 2439 * means we reserve 2 more, this pushes us to allocate from the next 2440 * larger slab size. 2441 * i.e. RXBUFFER_2048 --> size-4096 slab 2442 * However with the new *_jumbo_rx* routines, jumbo receives will use 2443 * fragmented skbs 2444 */ 2445 2446 if (max_frame <= 1024) 2447 adapter->rx_buffer_len = 1024; 2448 else if (max_frame <= 2048) 2449 adapter->rx_buffer_len = 2048; 2450 else 2451 #if (PAGE_SIZE / 2) > 16384 2452 adapter->rx_buffer_len = 16384; 2453 #else 2454 adapter->rx_buffer_len = PAGE_SIZE / 2; 2455 #endif 2456 2457 /* adjust allocation if LPE protects us, and we aren't using SBP */ 2458 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) || 2459 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN)) 2460 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + 2461 ETH_FCS_LEN; 2462 2463 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n", 2464 netdev->mtu, new_mtu); 2465 netdev->mtu = new_mtu; 2466 2467 if (netif_running(netdev)) 2468 igbvf_up(adapter); 2469 else 2470 igbvf_reset(adapter); 2471 2472 clear_bit(__IGBVF_RESETTING, &adapter->state); 2473 2474 return 0; 2475 } 2476 2477 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd) 2478 { 2479 switch (cmd) { 2480 default: 2481 return -EOPNOTSUPP; 2482 } 2483 } 2484 2485 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state) 2486 { 2487 struct net_device *netdev = pci_get_drvdata(pdev); 2488 struct igbvf_adapter *adapter = netdev_priv(netdev); 2489 #ifdef CONFIG_PM 2490 int retval = 0; 2491 #endif 2492 2493 netif_device_detach(netdev); 2494 2495 if (netif_running(netdev)) { 2496 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state)); 2497 igbvf_down(adapter); 2498 igbvf_free_irq(adapter); 2499 } 2500 2501 #ifdef CONFIG_PM 2502 retval = pci_save_state(pdev); 2503 if (retval) 2504 return retval; 2505 #endif 2506 2507 pci_disable_device(pdev); 2508 2509 return 0; 2510 } 2511 2512 #ifdef CONFIG_PM 2513 static int igbvf_resume(struct pci_dev *pdev) 2514 { 2515 struct net_device *netdev = pci_get_drvdata(pdev); 2516 struct igbvf_adapter *adapter = netdev_priv(netdev); 2517 u32 err; 2518 2519 pci_restore_state(pdev); 2520 err = pci_enable_device_mem(pdev); 2521 if (err) { 2522 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n"); 2523 return err; 2524 } 2525 2526 pci_set_master(pdev); 2527 2528 if (netif_running(netdev)) { 2529 err = igbvf_request_irq(adapter); 2530 if (err) 2531 return err; 2532 } 2533 2534 igbvf_reset(adapter); 2535 2536 if (netif_running(netdev)) 2537 igbvf_up(adapter); 2538 2539 netif_device_attach(netdev); 2540 2541 return 0; 2542 } 2543 #endif 2544 2545 static void igbvf_shutdown(struct pci_dev *pdev) 2546 { 2547 igbvf_suspend(pdev, PMSG_SUSPEND); 2548 } 2549 2550 #ifdef CONFIG_NET_POLL_CONTROLLER 2551 /* Polling 'interrupt' - used by things like netconsole to send skbs 2552 * without having to re-enable interrupts. It's not called while 2553 * the interrupt routine is executing. 2554 */ 2555 static void igbvf_netpoll(struct net_device *netdev) 2556 { 2557 struct igbvf_adapter *adapter = netdev_priv(netdev); 2558 2559 disable_irq(adapter->pdev->irq); 2560 2561 igbvf_clean_tx_irq(adapter->tx_ring); 2562 2563 enable_irq(adapter->pdev->irq); 2564 } 2565 #endif 2566 2567 /** 2568 * igbvf_io_error_detected - called when PCI error is detected 2569 * @pdev: Pointer to PCI device 2570 * @state: The current pci connection state 2571 * 2572 * This function is called after a PCI bus error affecting 2573 * this device has been detected. 2574 */ 2575 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev, 2576 pci_channel_state_t state) 2577 { 2578 struct net_device *netdev = pci_get_drvdata(pdev); 2579 struct igbvf_adapter *adapter = netdev_priv(netdev); 2580 2581 netif_device_detach(netdev); 2582 2583 if (state == pci_channel_io_perm_failure) 2584 return PCI_ERS_RESULT_DISCONNECT; 2585 2586 if (netif_running(netdev)) 2587 igbvf_down(adapter); 2588 pci_disable_device(pdev); 2589 2590 /* Request a slot slot reset. */ 2591 return PCI_ERS_RESULT_NEED_RESET; 2592 } 2593 2594 /** 2595 * igbvf_io_slot_reset - called after the pci bus has been reset. 2596 * @pdev: Pointer to PCI device 2597 * 2598 * Restart the card from scratch, as if from a cold-boot. Implementation 2599 * resembles the first-half of the igbvf_resume routine. 2600 */ 2601 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev) 2602 { 2603 struct net_device *netdev = pci_get_drvdata(pdev); 2604 struct igbvf_adapter *adapter = netdev_priv(netdev); 2605 2606 if (pci_enable_device_mem(pdev)) { 2607 dev_err(&pdev->dev, 2608 "Cannot re-enable PCI device after reset.\n"); 2609 return PCI_ERS_RESULT_DISCONNECT; 2610 } 2611 pci_set_master(pdev); 2612 2613 igbvf_reset(adapter); 2614 2615 return PCI_ERS_RESULT_RECOVERED; 2616 } 2617 2618 /** 2619 * igbvf_io_resume - called when traffic can start flowing again. 2620 * @pdev: Pointer to PCI device 2621 * 2622 * This callback is called when the error recovery driver tells us that 2623 * its OK to resume normal operation. Implementation resembles the 2624 * second-half of the igbvf_resume routine. 2625 */ 2626 static void igbvf_io_resume(struct pci_dev *pdev) 2627 { 2628 struct net_device *netdev = pci_get_drvdata(pdev); 2629 struct igbvf_adapter *adapter = netdev_priv(netdev); 2630 2631 if (netif_running(netdev)) { 2632 if (igbvf_up(adapter)) { 2633 dev_err(&pdev->dev, 2634 "can't bring device back up after reset\n"); 2635 return; 2636 } 2637 } 2638 2639 netif_device_attach(netdev); 2640 } 2641 2642 static void igbvf_print_device_info(struct igbvf_adapter *adapter) 2643 { 2644 struct e1000_hw *hw = &adapter->hw; 2645 struct net_device *netdev = adapter->netdev; 2646 struct pci_dev *pdev = adapter->pdev; 2647 2648 if (hw->mac.type == e1000_vfadapt_i350) 2649 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n"); 2650 else 2651 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n"); 2652 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr); 2653 } 2654 2655 static int igbvf_set_features(struct net_device *netdev, 2656 netdev_features_t features) 2657 { 2658 struct igbvf_adapter *adapter = netdev_priv(netdev); 2659 2660 if (features & NETIF_F_RXCSUM) 2661 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED; 2662 else 2663 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED; 2664 2665 return 0; 2666 } 2667 2668 #define IGBVF_MAX_MAC_HDR_LEN 127 2669 #define IGBVF_MAX_NETWORK_HDR_LEN 511 2670 2671 static netdev_features_t 2672 igbvf_features_check(struct sk_buff *skb, struct net_device *dev, 2673 netdev_features_t features) 2674 { 2675 unsigned int network_hdr_len, mac_hdr_len; 2676 2677 /* Make certain the headers can be described by a context descriptor */ 2678 mac_hdr_len = skb_network_header(skb) - skb->data; 2679 if (unlikely(mac_hdr_len > IGBVF_MAX_MAC_HDR_LEN)) 2680 return features & ~(NETIF_F_HW_CSUM | 2681 NETIF_F_SCTP_CRC | 2682 NETIF_F_HW_VLAN_CTAG_TX | 2683 NETIF_F_TSO | 2684 NETIF_F_TSO6); 2685 2686 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb); 2687 if (unlikely(network_hdr_len > IGBVF_MAX_NETWORK_HDR_LEN)) 2688 return features & ~(NETIF_F_HW_CSUM | 2689 NETIF_F_SCTP_CRC | 2690 NETIF_F_TSO | 2691 NETIF_F_TSO6); 2692 2693 /* We can only support IPV4 TSO in tunnels if we can mangle the 2694 * inner IP ID field, so strip TSO if MANGLEID is not supported. 2695 */ 2696 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) 2697 features &= ~NETIF_F_TSO; 2698 2699 return features; 2700 } 2701 2702 static const struct net_device_ops igbvf_netdev_ops = { 2703 .ndo_open = igbvf_open, 2704 .ndo_stop = igbvf_close, 2705 .ndo_start_xmit = igbvf_xmit_frame, 2706 .ndo_set_rx_mode = igbvf_set_rx_mode, 2707 .ndo_set_mac_address = igbvf_set_mac, 2708 .ndo_change_mtu = igbvf_change_mtu, 2709 .ndo_do_ioctl = igbvf_ioctl, 2710 .ndo_tx_timeout = igbvf_tx_timeout, 2711 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid, 2712 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid, 2713 #ifdef CONFIG_NET_POLL_CONTROLLER 2714 .ndo_poll_controller = igbvf_netpoll, 2715 #endif 2716 .ndo_set_features = igbvf_set_features, 2717 .ndo_features_check = igbvf_features_check, 2718 }; 2719 2720 /** 2721 * igbvf_probe - Device Initialization Routine 2722 * @pdev: PCI device information struct 2723 * @ent: entry in igbvf_pci_tbl 2724 * 2725 * Returns 0 on success, negative on failure 2726 * 2727 * igbvf_probe initializes an adapter identified by a pci_dev structure. 2728 * The OS initialization, configuring of the adapter private structure, 2729 * and a hardware reset occur. 2730 **/ 2731 static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 2732 { 2733 struct net_device *netdev; 2734 struct igbvf_adapter *adapter; 2735 struct e1000_hw *hw; 2736 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data]; 2737 2738 static int cards_found; 2739 int err, pci_using_dac; 2740 2741 err = pci_enable_device_mem(pdev); 2742 if (err) 2743 return err; 2744 2745 pci_using_dac = 0; 2746 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); 2747 if (!err) { 2748 pci_using_dac = 1; 2749 } else { 2750 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); 2751 if (err) { 2752 dev_err(&pdev->dev, 2753 "No usable DMA configuration, aborting\n"); 2754 goto err_dma; 2755 } 2756 } 2757 2758 err = pci_request_regions(pdev, igbvf_driver_name); 2759 if (err) 2760 goto err_pci_reg; 2761 2762 pci_set_master(pdev); 2763 2764 err = -ENOMEM; 2765 netdev = alloc_etherdev(sizeof(struct igbvf_adapter)); 2766 if (!netdev) 2767 goto err_alloc_etherdev; 2768 2769 SET_NETDEV_DEV(netdev, &pdev->dev); 2770 2771 pci_set_drvdata(pdev, netdev); 2772 adapter = netdev_priv(netdev); 2773 hw = &adapter->hw; 2774 adapter->netdev = netdev; 2775 adapter->pdev = pdev; 2776 adapter->ei = ei; 2777 adapter->pba = ei->pba; 2778 adapter->flags = ei->flags; 2779 adapter->hw.back = adapter; 2780 adapter->hw.mac.type = ei->mac; 2781 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE); 2782 2783 /* PCI config space info */ 2784 2785 hw->vendor_id = pdev->vendor; 2786 hw->device_id = pdev->device; 2787 hw->subsystem_vendor_id = pdev->subsystem_vendor; 2788 hw->subsystem_device_id = pdev->subsystem_device; 2789 hw->revision_id = pdev->revision; 2790 2791 err = -EIO; 2792 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0), 2793 pci_resource_len(pdev, 0)); 2794 2795 if (!adapter->hw.hw_addr) 2796 goto err_ioremap; 2797 2798 if (ei->get_variants) { 2799 err = ei->get_variants(adapter); 2800 if (err) 2801 goto err_get_variants; 2802 } 2803 2804 /* setup adapter struct */ 2805 err = igbvf_sw_init(adapter); 2806 if (err) 2807 goto err_sw_init; 2808 2809 /* construct the net_device struct */ 2810 netdev->netdev_ops = &igbvf_netdev_ops; 2811 2812 igbvf_set_ethtool_ops(netdev); 2813 netdev->watchdog_timeo = 5 * HZ; 2814 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1); 2815 2816 adapter->bd_number = cards_found++; 2817 2818 netdev->hw_features = NETIF_F_SG | 2819 NETIF_F_TSO | 2820 NETIF_F_TSO6 | 2821 NETIF_F_RXCSUM | 2822 NETIF_F_HW_CSUM | 2823 NETIF_F_SCTP_CRC; 2824 2825 #define IGBVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \ 2826 NETIF_F_GSO_GRE_CSUM | \ 2827 NETIF_F_GSO_IPXIP4 | \ 2828 NETIF_F_GSO_IPXIP6 | \ 2829 NETIF_F_GSO_UDP_TUNNEL | \ 2830 NETIF_F_GSO_UDP_TUNNEL_CSUM) 2831 2832 netdev->gso_partial_features = IGBVF_GSO_PARTIAL_FEATURES; 2833 netdev->hw_features |= NETIF_F_GSO_PARTIAL | 2834 IGBVF_GSO_PARTIAL_FEATURES; 2835 2836 netdev->features = netdev->hw_features; 2837 2838 if (pci_using_dac) 2839 netdev->features |= NETIF_F_HIGHDMA; 2840 2841 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID; 2842 netdev->mpls_features |= NETIF_F_HW_CSUM; 2843 netdev->hw_enc_features |= netdev->vlan_features; 2844 2845 /* set this bit last since it cannot be part of vlan_features */ 2846 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER | 2847 NETIF_F_HW_VLAN_CTAG_RX | 2848 NETIF_F_HW_VLAN_CTAG_TX; 2849 2850 /* MTU range: 68 - 9216 */ 2851 netdev->min_mtu = ETH_MIN_MTU; 2852 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE; 2853 2854 spin_lock_bh(&hw->mbx_lock); 2855 2856 /*reset the controller to put the device in a known good state */ 2857 err = hw->mac.ops.reset_hw(hw); 2858 if (err) { 2859 dev_info(&pdev->dev, 2860 "PF still in reset state. Is the PF interface up?\n"); 2861 } else { 2862 err = hw->mac.ops.read_mac_addr(hw); 2863 if (err) 2864 dev_info(&pdev->dev, "Error reading MAC address.\n"); 2865 else if (is_zero_ether_addr(adapter->hw.mac.addr)) 2866 dev_info(&pdev->dev, 2867 "MAC address not assigned by administrator.\n"); 2868 memcpy(netdev->dev_addr, adapter->hw.mac.addr, 2869 netdev->addr_len); 2870 } 2871 2872 spin_unlock_bh(&hw->mbx_lock); 2873 2874 if (!is_valid_ether_addr(netdev->dev_addr)) { 2875 dev_info(&pdev->dev, "Assigning random MAC address.\n"); 2876 eth_hw_addr_random(netdev); 2877 memcpy(adapter->hw.mac.addr, netdev->dev_addr, 2878 netdev->addr_len); 2879 } 2880 2881 timer_setup(&adapter->watchdog_timer, igbvf_watchdog, 0); 2882 2883 INIT_WORK(&adapter->reset_task, igbvf_reset_task); 2884 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task); 2885 2886 /* ring size defaults */ 2887 adapter->rx_ring->count = 1024; 2888 adapter->tx_ring->count = 1024; 2889 2890 /* reset the hardware with the new settings */ 2891 igbvf_reset(adapter); 2892 2893 /* set hardware-specific flags */ 2894 if (adapter->hw.mac.type == e1000_vfadapt_i350) 2895 adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP; 2896 2897 strcpy(netdev->name, "eth%d"); 2898 err = register_netdev(netdev); 2899 if (err) 2900 goto err_hw_init; 2901 2902 /* tell the stack to leave us alone until igbvf_open() is called */ 2903 netif_carrier_off(netdev); 2904 netif_stop_queue(netdev); 2905 2906 igbvf_print_device_info(adapter); 2907 2908 igbvf_initialize_last_counter_stats(adapter); 2909 2910 return 0; 2911 2912 err_hw_init: 2913 kfree(adapter->tx_ring); 2914 kfree(adapter->rx_ring); 2915 err_sw_init: 2916 igbvf_reset_interrupt_capability(adapter); 2917 err_get_variants: 2918 iounmap(adapter->hw.hw_addr); 2919 err_ioremap: 2920 free_netdev(netdev); 2921 err_alloc_etherdev: 2922 pci_release_regions(pdev); 2923 err_pci_reg: 2924 err_dma: 2925 pci_disable_device(pdev); 2926 return err; 2927 } 2928 2929 /** 2930 * igbvf_remove - Device Removal Routine 2931 * @pdev: PCI device information struct 2932 * 2933 * igbvf_remove is called by the PCI subsystem to alert the driver 2934 * that it should release a PCI device. The could be caused by a 2935 * Hot-Plug event, or because the driver is going to be removed from 2936 * memory. 2937 **/ 2938 static void igbvf_remove(struct pci_dev *pdev) 2939 { 2940 struct net_device *netdev = pci_get_drvdata(pdev); 2941 struct igbvf_adapter *adapter = netdev_priv(netdev); 2942 struct e1000_hw *hw = &adapter->hw; 2943 2944 /* The watchdog timer may be rescheduled, so explicitly 2945 * disable it from being rescheduled. 2946 */ 2947 set_bit(__IGBVF_DOWN, &adapter->state); 2948 del_timer_sync(&adapter->watchdog_timer); 2949 2950 cancel_work_sync(&adapter->reset_task); 2951 cancel_work_sync(&adapter->watchdog_task); 2952 2953 unregister_netdev(netdev); 2954 2955 igbvf_reset_interrupt_capability(adapter); 2956 2957 /* it is important to delete the NAPI struct prior to freeing the 2958 * Rx ring so that you do not end up with null pointer refs 2959 */ 2960 netif_napi_del(&adapter->rx_ring->napi); 2961 kfree(adapter->tx_ring); 2962 kfree(adapter->rx_ring); 2963 2964 iounmap(hw->hw_addr); 2965 if (hw->flash_address) 2966 iounmap(hw->flash_address); 2967 pci_release_regions(pdev); 2968 2969 free_netdev(netdev); 2970 2971 pci_disable_device(pdev); 2972 } 2973 2974 /* PCI Error Recovery (ERS) */ 2975 static const struct pci_error_handlers igbvf_err_handler = { 2976 .error_detected = igbvf_io_error_detected, 2977 .slot_reset = igbvf_io_slot_reset, 2978 .resume = igbvf_io_resume, 2979 }; 2980 2981 static const struct pci_device_id igbvf_pci_tbl[] = { 2982 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf }, 2983 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf }, 2984 { } /* terminate list */ 2985 }; 2986 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl); 2987 2988 /* PCI Device API Driver */ 2989 static struct pci_driver igbvf_driver = { 2990 .name = igbvf_driver_name, 2991 .id_table = igbvf_pci_tbl, 2992 .probe = igbvf_probe, 2993 .remove = igbvf_remove, 2994 #ifdef CONFIG_PM 2995 /* Power Management Hooks */ 2996 .suspend = igbvf_suspend, 2997 .resume = igbvf_resume, 2998 #endif 2999 .shutdown = igbvf_shutdown, 3000 .err_handler = &igbvf_err_handler 3001 }; 3002 3003 /** 3004 * igbvf_init_module - Driver Registration Routine 3005 * 3006 * igbvf_init_module is the first routine called when the driver is 3007 * loaded. All it does is register with the PCI subsystem. 3008 **/ 3009 static int __init igbvf_init_module(void) 3010 { 3011 int ret; 3012 3013 pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version); 3014 pr_info("%s\n", igbvf_copyright); 3015 3016 ret = pci_register_driver(&igbvf_driver); 3017 3018 return ret; 3019 } 3020 module_init(igbvf_init_module); 3021 3022 /** 3023 * igbvf_exit_module - Driver Exit Cleanup Routine 3024 * 3025 * igbvf_exit_module is called just before the driver is removed 3026 * from memory. 3027 **/ 3028 static void __exit igbvf_exit_module(void) 3029 { 3030 pci_unregister_driver(&igbvf_driver); 3031 } 3032 module_exit(igbvf_exit_module); 3033 3034 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>"); 3035 MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver"); 3036 MODULE_LICENSE("GPL"); 3037 MODULE_VERSION(DRV_VERSION); 3038 3039 /* netdev.c */ 3040