1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright(c) 1999 - 2024 Intel Corporation. */ 3 4 /****************************************************************************** 5 Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code 6 ******************************************************************************/ 7 8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 9 10 #include <linux/types.h> 11 #include <linux/bitops.h> 12 #include <linux/module.h> 13 #include <linux/pci.h> 14 #include <linux/netdevice.h> 15 #include <linux/vmalloc.h> 16 #include <linux/string.h> 17 #include <linux/in.h> 18 #include <linux/ip.h> 19 #include <linux/tcp.h> 20 #include <linux/sctp.h> 21 #include <linux/ipv6.h> 22 #include <linux/slab.h> 23 #include <net/checksum.h> 24 #include <net/ip6_checksum.h> 25 #include <linux/ethtool.h> 26 #include <linux/if.h> 27 #include <linux/if_vlan.h> 28 #include <linux/prefetch.h> 29 #include <net/mpls.h> 30 #include <linux/bpf.h> 31 #include <linux/bpf_trace.h> 32 #include <linux/atomic.h> 33 #include <net/xfrm.h> 34 35 #include "ixgbevf.h" 36 37 const char ixgbevf_driver_name[] = "ixgbevf"; 38 static const char ixgbevf_driver_string[] = 39 "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver"; 40 41 static char ixgbevf_copyright[] = 42 "Copyright (c) 2009 - 2024 Intel Corporation."; 43 44 static const struct ixgbevf_info *ixgbevf_info_tbl[] = { 45 [board_82599_vf] = &ixgbevf_82599_vf_info, 46 [board_82599_vf_hv] = &ixgbevf_82599_vf_hv_info, 47 [board_X540_vf] = &ixgbevf_X540_vf_info, 48 [board_X540_vf_hv] = &ixgbevf_X540_vf_hv_info, 49 [board_X550_vf] = &ixgbevf_X550_vf_info, 50 [board_X550_vf_hv] = &ixgbevf_X550_vf_hv_info, 51 [board_X550EM_x_vf] = &ixgbevf_X550EM_x_vf_info, 52 [board_X550EM_x_vf_hv] = &ixgbevf_X550EM_x_vf_hv_info, 53 [board_x550em_a_vf] = &ixgbevf_x550em_a_vf_info, 54 [board_e610_vf] = &ixgbevf_e610_vf_info, 55 [board_e610_vf_hv] = &ixgbevf_e610_vf_hv_info, 56 }; 57 58 /* ixgbevf_pci_tbl - PCI Device ID Table 59 * 60 * Wildcard entries (PCI_ANY_ID) should come last 61 * Last entry must be all 0s 62 * 63 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID, 64 * Class, Class Mask, private data (not used) } 65 */ 66 static const struct pci_device_id ixgbevf_pci_tbl[] = { 67 { 68 PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), 69 .driver_data = board_82599_vf, 70 }, { 71 PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), 72 .driver_data = board_82599_vf_hv, 73 }, { 74 PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), 75 .driver_data = board_X540_vf, 76 }, { 77 PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), 78 .driver_data = board_X540_vf_hv, 79 }, { 80 PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), 81 .driver_data = board_X550_vf, 82 }, { 83 PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), 84 .driver_data = board_X550_vf_hv, 85 }, { 86 PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), 87 .driver_data = board_X550EM_x_vf, 88 }, { 89 PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), 90 .driver_data = board_X550EM_x_vf_hv 91 }, { 92 PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), 93 .driver_data = board_x550em_a_vf, 94 }, { 95 PCI_VDEVICE_SUB(INTEL, IXGBE_DEV_ID_E610_VF, 96 PCI_ANY_ID, IXGBE_SUBDEV_ID_E610_VF_HV), 97 .driver_data = board_e610_vf_hv, 98 }, { 99 PCI_VDEVICE(INTEL, IXGBE_DEV_ID_E610_VF), 100 .driver_data = board_e610_vf, 101 }, 102 /* required last entry */ 103 { } 104 }; 105 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl); 106 107 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver"); 108 MODULE_LICENSE("GPL v2"); 109 110 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK) 111 static int debug = -1; 112 module_param(debug, int, 0); 113 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)"); 114 115 static struct workqueue_struct *ixgbevf_wq; 116 117 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter) 118 { 119 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) && 120 !test_bit(__IXGBEVF_REMOVING, &adapter->state) && 121 !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state)) 122 queue_work(ixgbevf_wq, &adapter->service_task); 123 } 124 125 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter) 126 { 127 BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state)); 128 129 /* flush memory to make sure state is correct before next watchdog */ 130 smp_mb__before_atomic(); 131 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state); 132 } 133 134 /* forward decls */ 135 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter); 136 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector); 137 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter); 138 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer); 139 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring, 140 struct ixgbevf_rx_buffer *old_buff); 141 142 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw) 143 { 144 struct ixgbevf_adapter *adapter = hw->back; 145 146 if (!hw->hw_addr) 147 return; 148 hw->hw_addr = NULL; 149 dev_err(&adapter->pdev->dev, "Adapter removed\n"); 150 if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state)) 151 ixgbevf_service_event_schedule(adapter); 152 } 153 154 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg) 155 { 156 u32 value; 157 158 /* The following check not only optimizes a bit by not 159 * performing a read on the status register when the 160 * register just read was a status register read that 161 * returned IXGBE_FAILED_READ_REG. It also blocks any 162 * potential recursion. 163 */ 164 if (reg == IXGBE_VFSTATUS) { 165 ixgbevf_remove_adapter(hw); 166 return; 167 } 168 value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS); 169 if (value == IXGBE_FAILED_READ_REG) 170 ixgbevf_remove_adapter(hw); 171 } 172 173 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg) 174 { 175 u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr); 176 u32 value; 177 178 if (IXGBE_REMOVED(reg_addr)) 179 return IXGBE_FAILED_READ_REG; 180 value = readl(reg_addr + reg); 181 if (unlikely(value == IXGBE_FAILED_READ_REG)) 182 ixgbevf_check_remove(hw, reg); 183 return value; 184 } 185 186 /** 187 * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors 188 * @adapter: pointer to adapter struct 189 * @direction: 0 for Rx, 1 for Tx, -1 for other causes 190 * @queue: queue to map the corresponding interrupt to 191 * @msix_vector: the vector to map to the corresponding queue 192 **/ 193 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction, 194 u8 queue, u8 msix_vector) 195 { 196 u32 ivar, index; 197 struct ixgbe_hw *hw = &adapter->hw; 198 199 if (direction == -1) { 200 /* other causes */ 201 msix_vector |= IXGBE_IVAR_ALLOC_VAL; 202 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC); 203 ivar &= ~0xFF; 204 ivar |= msix_vector; 205 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar); 206 } else { 207 /* Tx or Rx causes */ 208 msix_vector |= IXGBE_IVAR_ALLOC_VAL; 209 index = ((16 * (queue & 1)) + (8 * direction)); 210 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1)); 211 ivar &= ~(0xFF << index); 212 ivar |= (msix_vector << index); 213 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar); 214 } 215 } 216 217 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring) 218 { 219 return ring->stats.packets; 220 } 221 222 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring) 223 { 224 struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev); 225 struct ixgbe_hw *hw = &adapter->hw; 226 227 u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx)); 228 u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx)); 229 230 if (head != tail) 231 return (head < tail) ? 232 tail - head : (tail + ring->count - head); 233 234 return 0; 235 } 236 237 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring) 238 { 239 u32 tx_done = ixgbevf_get_tx_completed(tx_ring); 240 u32 tx_done_old = tx_ring->tx_stats.tx_done_old; 241 u32 tx_pending = ixgbevf_get_tx_pending(tx_ring); 242 243 clear_check_for_tx_hang(tx_ring); 244 245 /* Check for a hung queue, but be thorough. This verifies 246 * that a transmit has been completed since the previous 247 * check AND there is at least one packet pending. The 248 * ARMED bit is set to indicate a potential hang. 249 */ 250 if ((tx_done_old == tx_done) && tx_pending) { 251 /* make sure it is true for two checks in a row */ 252 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED, 253 &tx_ring->state); 254 } 255 /* reset the countdown */ 256 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state); 257 258 /* update completed stats and continue */ 259 tx_ring->tx_stats.tx_done_old = tx_done; 260 261 return false; 262 } 263 264 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter) 265 { 266 /* Do the reset outside of interrupt context */ 267 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) { 268 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state); 269 ixgbevf_service_event_schedule(adapter); 270 } 271 } 272 273 /** 274 * ixgbevf_tx_timeout - Respond to a Tx Hang 275 * @netdev: network interface device structure 276 * @txqueue: transmit queue hanging (unused) 277 **/ 278 static void ixgbevf_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue) 279 { 280 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 281 282 ixgbevf_tx_timeout_reset(adapter); 283 } 284 285 /** 286 * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes 287 * @q_vector: board private structure 288 * @tx_ring: tx ring to clean 289 * @napi_budget: Used to determine if we are in netpoll 290 **/ 291 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector, 292 struct ixgbevf_ring *tx_ring, int napi_budget) 293 { 294 struct ixgbevf_adapter *adapter = q_vector->adapter; 295 struct ixgbevf_tx_buffer *tx_buffer; 296 union ixgbe_adv_tx_desc *tx_desc; 297 unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0; 298 unsigned int budget = tx_ring->count / 2; 299 unsigned int i = tx_ring->next_to_clean; 300 301 if (test_bit(__IXGBEVF_DOWN, &adapter->state)) 302 return true; 303 304 tx_buffer = &tx_ring->tx_buffer_info[i]; 305 tx_desc = IXGBEVF_TX_DESC(tx_ring, i); 306 i -= tx_ring->count; 307 308 do { 309 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch; 310 311 /* if next_to_watch is not set then there is no work pending */ 312 if (!eop_desc) 313 break; 314 315 /* prevent any other reads prior to eop_desc */ 316 smp_rmb(); 317 318 /* if DD is not set pending work has not been completed */ 319 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD))) 320 break; 321 322 /* clear next_to_watch to prevent false hangs */ 323 tx_buffer->next_to_watch = NULL; 324 325 /* update the statistics for this packet */ 326 total_bytes += tx_buffer->bytecount; 327 total_packets += tx_buffer->gso_segs; 328 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC) 329 total_ipsec++; 330 331 /* free the skb */ 332 if (ring_is_xdp(tx_ring)) 333 page_frag_free(tx_buffer->data); 334 else 335 napi_consume_skb(tx_buffer->skb, napi_budget); 336 337 /* unmap skb header data */ 338 dma_unmap_single(tx_ring->dev, 339 dma_unmap_addr(tx_buffer, dma), 340 dma_unmap_len(tx_buffer, len), 341 DMA_TO_DEVICE); 342 343 /* clear tx_buffer data */ 344 dma_unmap_len_set(tx_buffer, len, 0); 345 346 /* unmap remaining buffers */ 347 while (tx_desc != eop_desc) { 348 tx_buffer++; 349 tx_desc++; 350 i++; 351 if (unlikely(!i)) { 352 i -= tx_ring->count; 353 tx_buffer = tx_ring->tx_buffer_info; 354 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0); 355 } 356 357 /* unmap any remaining paged data */ 358 if (dma_unmap_len(tx_buffer, len)) { 359 dma_unmap_page(tx_ring->dev, 360 dma_unmap_addr(tx_buffer, dma), 361 dma_unmap_len(tx_buffer, len), 362 DMA_TO_DEVICE); 363 dma_unmap_len_set(tx_buffer, len, 0); 364 } 365 } 366 367 /* move us one more past the eop_desc for start of next pkt */ 368 tx_buffer++; 369 tx_desc++; 370 i++; 371 if (unlikely(!i)) { 372 i -= tx_ring->count; 373 tx_buffer = tx_ring->tx_buffer_info; 374 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0); 375 } 376 377 /* issue prefetch for next Tx descriptor */ 378 prefetch(tx_desc); 379 380 /* update budget accounting */ 381 budget--; 382 } while (likely(budget)); 383 384 i += tx_ring->count; 385 tx_ring->next_to_clean = i; 386 u64_stats_update_begin(&tx_ring->syncp); 387 tx_ring->stats.bytes += total_bytes; 388 tx_ring->stats.packets += total_packets; 389 u64_stats_update_end(&tx_ring->syncp); 390 q_vector->tx.total_bytes += total_bytes; 391 q_vector->tx.total_packets += total_packets; 392 adapter->tx_ipsec += total_ipsec; 393 394 if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) { 395 struct ixgbe_hw *hw = &adapter->hw; 396 union ixgbe_adv_tx_desc *eop_desc; 397 398 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch; 399 400 pr_err("Detected Tx Unit Hang%s\n" 401 " Tx Queue <%d>\n" 402 " TDH, TDT <%x>, <%x>\n" 403 " next_to_use <%x>\n" 404 " next_to_clean <%x>\n" 405 "tx_buffer_info[next_to_clean]\n" 406 " next_to_watch <%p>\n" 407 " eop_desc->wb.status <%x>\n" 408 " time_stamp <%lx>\n" 409 " jiffies <%lx>\n", 410 ring_is_xdp(tx_ring) ? " XDP" : "", 411 tx_ring->queue_index, 412 IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)), 413 IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)), 414 tx_ring->next_to_use, i, 415 eop_desc, (eop_desc ? eop_desc->wb.status : 0), 416 tx_ring->tx_buffer_info[i].time_stamp, jiffies); 417 418 if (!ring_is_xdp(tx_ring)) 419 netif_stop_subqueue(tx_ring->netdev, 420 tx_ring->queue_index); 421 422 /* schedule immediate reset if we believe we hung */ 423 ixgbevf_tx_timeout_reset(adapter); 424 425 return true; 426 } 427 428 if (ring_is_xdp(tx_ring)) 429 return !!budget; 430 431 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2) 432 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) && 433 (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) { 434 /* Make sure that anybody stopping the queue after this 435 * sees the new next_to_clean. 436 */ 437 smp_mb(); 438 439 if (__netif_subqueue_stopped(tx_ring->netdev, 440 tx_ring->queue_index) && 441 !test_bit(__IXGBEVF_DOWN, &adapter->state)) { 442 netif_wake_subqueue(tx_ring->netdev, 443 tx_ring->queue_index); 444 ++tx_ring->tx_stats.restart_queue; 445 } 446 } 447 448 return !!budget; 449 } 450 451 /** 452 * ixgbevf_rx_skb - Helper function to determine proper Rx method 453 * @q_vector: structure containing interrupt and ring information 454 * @skb: packet to send up 455 **/ 456 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector, 457 struct sk_buff *skb) 458 { 459 napi_gro_receive(&q_vector->napi, skb); 460 } 461 462 #define IXGBE_RSS_L4_TYPES_MASK \ 463 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \ 464 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \ 465 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \ 466 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP)) 467 468 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring, 469 union ixgbe_adv_rx_desc *rx_desc, 470 struct sk_buff *skb) 471 { 472 u16 rss_type; 473 474 if (!(ring->netdev->features & NETIF_F_RXHASH)) 475 return; 476 477 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) & 478 IXGBE_RXDADV_RSSTYPE_MASK; 479 480 if (!rss_type) 481 return; 482 483 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss), 484 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ? 485 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3); 486 } 487 488 /** 489 * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum 490 * @ring: structure containig ring specific data 491 * @rx_desc: current Rx descriptor being processed 492 * @skb: skb currently being received and modified 493 **/ 494 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring, 495 union ixgbe_adv_rx_desc *rx_desc, 496 struct sk_buff *skb) 497 { 498 skb_checksum_none_assert(skb); 499 500 /* Rx csum disabled */ 501 if (!(ring->netdev->features & NETIF_F_RXCSUM)) 502 return; 503 504 /* if IP and error */ 505 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) && 506 ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) { 507 ring->rx_stats.csum_err++; 508 return; 509 } 510 511 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS)) 512 return; 513 514 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) { 515 ring->rx_stats.csum_err++; 516 return; 517 } 518 519 /* It must be a TCP or UDP packet with a valid checksum */ 520 skb->ip_summed = CHECKSUM_UNNECESSARY; 521 } 522 523 /** 524 * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor 525 * @rx_ring: rx descriptor ring packet is being transacted on 526 * @rx_desc: pointer to the EOP Rx descriptor 527 * @skb: pointer to current skb being populated 528 * 529 * This function checks the ring, descriptor, and packet information in 530 * order to populate the checksum, VLAN, protocol, and other fields within 531 * the skb. 532 **/ 533 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring, 534 union ixgbe_adv_rx_desc *rx_desc, 535 struct sk_buff *skb) 536 { 537 ixgbevf_rx_hash(rx_ring, rx_desc, skb); 538 ixgbevf_rx_checksum(rx_ring, rx_desc, skb); 539 540 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) { 541 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan); 542 unsigned long *active_vlans = netdev_priv(rx_ring->netdev); 543 544 if (test_bit(vid & VLAN_VID_MASK, active_vlans)) 545 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid); 546 } 547 548 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP)) 549 ixgbevf_ipsec_rx(rx_ring, rx_desc, skb); 550 551 skb->protocol = eth_type_trans(skb, rx_ring->netdev); 552 } 553 554 static 555 struct ixgbevf_rx_buffer *ixgbevf_get_rx_buffer(struct ixgbevf_ring *rx_ring, 556 const unsigned int size) 557 { 558 struct ixgbevf_rx_buffer *rx_buffer; 559 560 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean]; 561 prefetchw(rx_buffer->page); 562 563 /* we are reusing so sync this buffer for CPU use */ 564 dma_sync_single_range_for_cpu(rx_ring->dev, 565 rx_buffer->dma, 566 rx_buffer->page_offset, 567 size, 568 DMA_FROM_DEVICE); 569 570 rx_buffer->pagecnt_bias--; 571 572 return rx_buffer; 573 } 574 575 static void ixgbevf_put_rx_buffer(struct ixgbevf_ring *rx_ring, 576 struct ixgbevf_rx_buffer *rx_buffer, 577 struct sk_buff *skb) 578 { 579 if (ixgbevf_can_reuse_rx_page(rx_buffer)) { 580 /* hand second half of page back to the ring */ 581 ixgbevf_reuse_rx_page(rx_ring, rx_buffer); 582 } else { 583 if (IS_ERR(skb)) 584 /* We are not reusing the buffer so unmap it and free 585 * any references we are holding to it 586 */ 587 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma, 588 ixgbevf_rx_pg_size(rx_ring), 589 DMA_FROM_DEVICE, 590 IXGBEVF_RX_DMA_ATTR); 591 __page_frag_cache_drain(rx_buffer->page, 592 rx_buffer->pagecnt_bias); 593 } 594 595 /* clear contents of rx_buffer */ 596 rx_buffer->page = NULL; 597 } 598 599 /** 600 * ixgbevf_is_non_eop - process handling of non-EOP buffers 601 * @rx_ring: Rx ring being processed 602 * @rx_desc: Rx descriptor for current buffer 603 * 604 * This function updates next to clean. If the buffer is an EOP buffer 605 * this function exits returning false, otherwise it will place the 606 * sk_buff in the next buffer to be chained and return true indicating 607 * that this is in fact a non-EOP buffer. 608 **/ 609 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring, 610 union ixgbe_adv_rx_desc *rx_desc) 611 { 612 u32 ntc = rx_ring->next_to_clean + 1; 613 614 /* fetch, update, and store next to clean */ 615 ntc = (ntc < rx_ring->count) ? ntc : 0; 616 rx_ring->next_to_clean = ntc; 617 618 prefetch(IXGBEVF_RX_DESC(rx_ring, ntc)); 619 620 if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))) 621 return false; 622 623 return true; 624 } 625 626 static inline unsigned int ixgbevf_rx_offset(struct ixgbevf_ring *rx_ring) 627 { 628 return ring_uses_build_skb(rx_ring) ? IXGBEVF_SKB_PAD : 0; 629 } 630 631 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring, 632 struct ixgbevf_rx_buffer *bi) 633 { 634 struct page *page = bi->page; 635 dma_addr_t dma; 636 637 /* since we are recycling buffers we should seldom need to alloc */ 638 if (likely(page)) 639 return true; 640 641 /* alloc new page for storage */ 642 page = dev_alloc_pages(ixgbevf_rx_pg_order(rx_ring)); 643 if (unlikely(!page)) { 644 rx_ring->rx_stats.alloc_rx_page_failed++; 645 return false; 646 } 647 648 /* map page for use */ 649 dma = dma_map_page_attrs(rx_ring->dev, page, 0, 650 ixgbevf_rx_pg_size(rx_ring), 651 DMA_FROM_DEVICE, IXGBEVF_RX_DMA_ATTR); 652 653 /* if mapping failed free memory back to system since 654 * there isn't much point in holding memory we can't use 655 */ 656 if (dma_mapping_error(rx_ring->dev, dma)) { 657 __free_pages(page, ixgbevf_rx_pg_order(rx_ring)); 658 659 rx_ring->rx_stats.alloc_rx_page_failed++; 660 return false; 661 } 662 663 bi->dma = dma; 664 bi->page = page; 665 bi->page_offset = ixgbevf_rx_offset(rx_ring); 666 bi->pagecnt_bias = 1; 667 rx_ring->rx_stats.alloc_rx_page++; 668 669 return true; 670 } 671 672 /** 673 * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split 674 * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on 675 * @cleaned_count: number of buffers to replace 676 **/ 677 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring, 678 u16 cleaned_count) 679 { 680 union ixgbe_adv_rx_desc *rx_desc; 681 struct ixgbevf_rx_buffer *bi; 682 unsigned int i = rx_ring->next_to_use; 683 684 /* nothing to do or no valid netdev defined */ 685 if (!cleaned_count || !rx_ring->netdev) 686 return; 687 688 rx_desc = IXGBEVF_RX_DESC(rx_ring, i); 689 bi = &rx_ring->rx_buffer_info[i]; 690 i -= rx_ring->count; 691 692 do { 693 if (!ixgbevf_alloc_mapped_page(rx_ring, bi)) 694 break; 695 696 /* sync the buffer for use by the device */ 697 dma_sync_single_range_for_device(rx_ring->dev, bi->dma, 698 bi->page_offset, 699 ixgbevf_rx_bufsz(rx_ring), 700 DMA_FROM_DEVICE); 701 702 /* Refresh the desc even if pkt_addr didn't change 703 * because each write-back erases this info. 704 */ 705 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset); 706 707 rx_desc++; 708 bi++; 709 i++; 710 if (unlikely(!i)) { 711 rx_desc = IXGBEVF_RX_DESC(rx_ring, 0); 712 bi = rx_ring->rx_buffer_info; 713 i -= rx_ring->count; 714 } 715 716 /* clear the length for the next_to_use descriptor */ 717 rx_desc->wb.upper.length = 0; 718 719 cleaned_count--; 720 } while (cleaned_count); 721 722 i += rx_ring->count; 723 724 if (rx_ring->next_to_use != i) { 725 /* record the next descriptor to use */ 726 rx_ring->next_to_use = i; 727 728 /* update next to alloc since we have filled the ring */ 729 rx_ring->next_to_alloc = i; 730 731 /* Force memory writes to complete before letting h/w 732 * know there are new descriptors to fetch. (Only 733 * applicable for weak-ordered memory model archs, 734 * such as IA-64). 735 */ 736 wmb(); 737 ixgbevf_write_tail(rx_ring, i); 738 } 739 } 740 741 /** 742 * ixgbevf_cleanup_headers - Correct corrupted or empty headers 743 * @rx_ring: rx descriptor ring packet is being transacted on 744 * @rx_desc: pointer to the EOP Rx descriptor 745 * @skb: pointer to current skb being fixed 746 * 747 * Check for corrupted packet headers caused by senders on the local L2 748 * embedded NIC switch not setting up their Tx Descriptors right. These 749 * should be very rare. 750 * 751 * Also address the case where we are pulling data in on pages only 752 * and as such no data is present in the skb header. 753 * 754 * In addition if skb is not at least 60 bytes we need to pad it so that 755 * it is large enough to qualify as a valid Ethernet frame. 756 * 757 * Returns true if an error was encountered and skb was freed. 758 **/ 759 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring, 760 union ixgbe_adv_rx_desc *rx_desc, 761 struct sk_buff *skb) 762 { 763 /* verify that the packet does not have any known errors */ 764 if (unlikely(ixgbevf_test_staterr(rx_desc, 765 IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) { 766 struct net_device *netdev = rx_ring->netdev; 767 768 if (!(netdev->features & NETIF_F_RXALL)) { 769 dev_kfree_skb_any(skb); 770 return true; 771 } 772 } 773 774 /* if eth_skb_pad returns an error the skb was freed */ 775 if (eth_skb_pad(skb)) 776 return true; 777 778 return false; 779 } 780 781 /** 782 * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring 783 * @rx_ring: rx descriptor ring to store buffers on 784 * @old_buff: donor buffer to have page reused 785 * 786 * Synchronizes page for reuse by the adapter 787 **/ 788 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring, 789 struct ixgbevf_rx_buffer *old_buff) 790 { 791 struct ixgbevf_rx_buffer *new_buff; 792 u16 nta = rx_ring->next_to_alloc; 793 794 new_buff = &rx_ring->rx_buffer_info[nta]; 795 796 /* update, and store next to alloc */ 797 nta++; 798 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0; 799 800 /* transfer page from old buffer to new buffer */ 801 new_buff->page = old_buff->page; 802 new_buff->dma = old_buff->dma; 803 new_buff->page_offset = old_buff->page_offset; 804 new_buff->pagecnt_bias = old_buff->pagecnt_bias; 805 } 806 807 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer) 808 { 809 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias; 810 struct page *page = rx_buffer->page; 811 812 /* avoid re-using remote and pfmemalloc pages */ 813 if (!dev_page_is_reusable(page)) 814 return false; 815 816 #if (PAGE_SIZE < 8192) 817 /* if we are only owner of page we can reuse it */ 818 if (unlikely((page_ref_count(page) - pagecnt_bias) > 1)) 819 return false; 820 #else 821 #define IXGBEVF_LAST_OFFSET \ 822 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBEVF_RXBUFFER_2048) 823 824 if (rx_buffer->page_offset > IXGBEVF_LAST_OFFSET) 825 return false; 826 827 #endif 828 829 /* If we have drained the page fragment pool we need to update 830 * the pagecnt_bias and page count so that we fully restock the 831 * number of references the driver holds. 832 */ 833 if (unlikely(!pagecnt_bias)) { 834 page_ref_add(page, USHRT_MAX); 835 rx_buffer->pagecnt_bias = USHRT_MAX; 836 } 837 838 return true; 839 } 840 841 /** 842 * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff 843 * @rx_ring: rx descriptor ring to transact packets on 844 * @rx_buffer: buffer containing page to add 845 * @skb: sk_buff to place the data into 846 * @size: size of buffer to be added 847 * 848 * This function will add the data contained in rx_buffer->page to the skb. 849 **/ 850 static void ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring, 851 struct ixgbevf_rx_buffer *rx_buffer, 852 struct sk_buff *skb, 853 unsigned int size) 854 { 855 #if (PAGE_SIZE < 8192) 856 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2; 857 #else 858 unsigned int truesize = ring_uses_build_skb(rx_ring) ? 859 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) : 860 SKB_DATA_ALIGN(size); 861 #endif 862 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page, 863 rx_buffer->page_offset, size, truesize); 864 #if (PAGE_SIZE < 8192) 865 rx_buffer->page_offset ^= truesize; 866 #else 867 rx_buffer->page_offset += truesize; 868 #endif 869 } 870 871 static 872 struct sk_buff *ixgbevf_construct_skb(struct ixgbevf_ring *rx_ring, 873 struct ixgbevf_rx_buffer *rx_buffer, 874 struct xdp_buff *xdp, 875 union ixgbe_adv_rx_desc *rx_desc) 876 { 877 unsigned int size = xdp->data_end - xdp->data; 878 #if (PAGE_SIZE < 8192) 879 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2; 880 #else 881 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end - 882 xdp->data_hard_start); 883 #endif 884 unsigned int headlen; 885 struct sk_buff *skb; 886 887 /* prefetch first cache line of first page */ 888 net_prefetch(xdp->data); 889 890 /* Note, we get here by enabling legacy-rx via: 891 * 892 * ethtool --set-priv-flags <dev> legacy-rx on 893 * 894 * In this mode, we currently get 0 extra XDP headroom as 895 * opposed to having legacy-rx off, where we process XDP 896 * packets going to stack via ixgbevf_build_skb(). 897 * 898 * For ixgbevf_construct_skb() mode it means that the 899 * xdp->data_meta will always point to xdp->data, since 900 * the helper cannot expand the head. Should this ever 901 * changed in future for legacy-rx mode on, then lets also 902 * add xdp->data_meta handling here. 903 */ 904 905 /* allocate a skb to store the frags */ 906 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBEVF_RX_HDR_SIZE); 907 if (unlikely(!skb)) 908 return NULL; 909 910 /* Determine available headroom for copy */ 911 headlen = size; 912 if (headlen > IXGBEVF_RX_HDR_SIZE) 913 headlen = eth_get_headlen(skb->dev, xdp->data, 914 IXGBEVF_RX_HDR_SIZE); 915 916 /* align pull length to size of long to optimize memcpy performance */ 917 memcpy(__skb_put(skb, headlen), xdp->data, 918 ALIGN(headlen, sizeof(long))); 919 920 /* update all of the pointers */ 921 size -= headlen; 922 if (size) { 923 skb_add_rx_frag(skb, 0, rx_buffer->page, 924 (xdp->data + headlen) - 925 page_address(rx_buffer->page), 926 size, truesize); 927 #if (PAGE_SIZE < 8192) 928 rx_buffer->page_offset ^= truesize; 929 #else 930 rx_buffer->page_offset += truesize; 931 #endif 932 } else { 933 rx_buffer->pagecnt_bias++; 934 } 935 936 return skb; 937 } 938 939 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter, 940 u32 qmask) 941 { 942 struct ixgbe_hw *hw = &adapter->hw; 943 944 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask); 945 } 946 947 static struct sk_buff *ixgbevf_build_skb(struct ixgbevf_ring *rx_ring, 948 struct ixgbevf_rx_buffer *rx_buffer, 949 struct xdp_buff *xdp, 950 union ixgbe_adv_rx_desc *rx_desc) 951 { 952 unsigned int metasize = xdp->data - xdp->data_meta; 953 #if (PAGE_SIZE < 8192) 954 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2; 955 #else 956 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) + 957 SKB_DATA_ALIGN(xdp->data_end - 958 xdp->data_hard_start); 959 #endif 960 struct sk_buff *skb; 961 962 /* Prefetch first cache line of first page. If xdp->data_meta 963 * is unused, this points to xdp->data, otherwise, we likely 964 * have a consumer accessing first few bytes of meta data, 965 * and then actual data. 966 */ 967 net_prefetch(xdp->data_meta); 968 969 /* build an skb around the page buffer */ 970 skb = napi_build_skb(xdp->data_hard_start, truesize); 971 if (unlikely(!skb)) 972 return NULL; 973 974 /* update pointers within the skb to store the data */ 975 skb_reserve(skb, xdp->data - xdp->data_hard_start); 976 __skb_put(skb, xdp->data_end - xdp->data); 977 if (metasize) 978 skb_metadata_set(skb, metasize); 979 980 /* update buffer offset */ 981 #if (PAGE_SIZE < 8192) 982 rx_buffer->page_offset ^= truesize; 983 #else 984 rx_buffer->page_offset += truesize; 985 #endif 986 987 return skb; 988 } 989 990 #define IXGBEVF_XDP_PASS 0 991 #define IXGBEVF_XDP_CONSUMED 1 992 #define IXGBEVF_XDP_TX 2 993 994 static int ixgbevf_xmit_xdp_ring(struct ixgbevf_ring *ring, 995 struct xdp_buff *xdp) 996 { 997 struct ixgbevf_tx_buffer *tx_buffer; 998 union ixgbe_adv_tx_desc *tx_desc; 999 u32 len, cmd_type; 1000 dma_addr_t dma; 1001 u16 i; 1002 1003 len = xdp->data_end - xdp->data; 1004 1005 if (unlikely(!ixgbevf_desc_unused(ring))) 1006 return IXGBEVF_XDP_CONSUMED; 1007 1008 dma = dma_map_single(ring->dev, xdp->data, len, DMA_TO_DEVICE); 1009 if (dma_mapping_error(ring->dev, dma)) 1010 return IXGBEVF_XDP_CONSUMED; 1011 1012 /* record the location of the first descriptor for this packet */ 1013 i = ring->next_to_use; 1014 tx_buffer = &ring->tx_buffer_info[i]; 1015 1016 dma_unmap_len_set(tx_buffer, len, len); 1017 dma_unmap_addr_set(tx_buffer, dma, dma); 1018 tx_buffer->data = xdp->data; 1019 tx_buffer->bytecount = len; 1020 tx_buffer->gso_segs = 1; 1021 tx_buffer->protocol = 0; 1022 1023 /* Populate minimal context descriptor that will provide for the 1024 * fact that we are expected to process Ethernet frames. 1025 */ 1026 if (!test_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state)) { 1027 struct ixgbe_adv_tx_context_desc *context_desc; 1028 1029 set_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state); 1030 1031 context_desc = IXGBEVF_TX_CTXTDESC(ring, 0); 1032 context_desc->vlan_macip_lens = 1033 cpu_to_le32(ETH_HLEN << IXGBE_ADVTXD_MACLEN_SHIFT); 1034 context_desc->fceof_saidx = 0; 1035 context_desc->type_tucmd_mlhl = 1036 cpu_to_le32(IXGBE_TXD_CMD_DEXT | 1037 IXGBE_ADVTXD_DTYP_CTXT); 1038 context_desc->mss_l4len_idx = 0; 1039 1040 i = 1; 1041 } 1042 1043 /* put descriptor type bits */ 1044 cmd_type = IXGBE_ADVTXD_DTYP_DATA | 1045 IXGBE_ADVTXD_DCMD_DEXT | 1046 IXGBE_ADVTXD_DCMD_IFCS; 1047 cmd_type |= len | IXGBE_TXD_CMD; 1048 1049 tx_desc = IXGBEVF_TX_DESC(ring, i); 1050 tx_desc->read.buffer_addr = cpu_to_le64(dma); 1051 1052 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type); 1053 tx_desc->read.olinfo_status = 1054 cpu_to_le32((len << IXGBE_ADVTXD_PAYLEN_SHIFT) | 1055 IXGBE_ADVTXD_CC); 1056 1057 /* Avoid any potential race with cleanup */ 1058 smp_wmb(); 1059 1060 /* set next_to_watch value indicating a packet is present */ 1061 i++; 1062 if (i == ring->count) 1063 i = 0; 1064 1065 tx_buffer->next_to_watch = tx_desc; 1066 ring->next_to_use = i; 1067 1068 return IXGBEVF_XDP_TX; 1069 } 1070 1071 static int ixgbevf_run_xdp(struct ixgbevf_adapter *adapter, 1072 struct ixgbevf_ring *rx_ring, 1073 struct xdp_buff *xdp) 1074 { 1075 int result = IXGBEVF_XDP_PASS; 1076 struct ixgbevf_ring *xdp_ring; 1077 struct bpf_prog *xdp_prog; 1078 u32 act; 1079 1080 xdp_prog = READ_ONCE(rx_ring->xdp_prog); 1081 1082 if (!xdp_prog) 1083 goto xdp_out; 1084 1085 act = bpf_prog_run_xdp(xdp_prog, xdp); 1086 switch (act) { 1087 case XDP_PASS: 1088 break; 1089 case XDP_TX: 1090 xdp_ring = adapter->xdp_ring[rx_ring->queue_index]; 1091 result = ixgbevf_xmit_xdp_ring(xdp_ring, xdp); 1092 if (result == IXGBEVF_XDP_CONSUMED) 1093 goto out_failure; 1094 break; 1095 default: 1096 bpf_warn_invalid_xdp_action(rx_ring->netdev, xdp_prog, act); 1097 fallthrough; 1098 case XDP_ABORTED: 1099 out_failure: 1100 trace_xdp_exception(rx_ring->netdev, xdp_prog, act); 1101 fallthrough; /* handle aborts by dropping packet */ 1102 case XDP_DROP: 1103 result = IXGBEVF_XDP_CONSUMED; 1104 break; 1105 } 1106 xdp_out: 1107 return result; 1108 } 1109 1110 static unsigned int ixgbevf_rx_frame_truesize(struct ixgbevf_ring *rx_ring, 1111 unsigned int size) 1112 { 1113 unsigned int truesize; 1114 1115 #if (PAGE_SIZE < 8192) 1116 truesize = ixgbevf_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */ 1117 #else 1118 truesize = ring_uses_build_skb(rx_ring) ? 1119 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) + 1120 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) : 1121 SKB_DATA_ALIGN(size); 1122 #endif 1123 return truesize; 1124 } 1125 1126 static void ixgbevf_rx_buffer_flip(struct ixgbevf_ring *rx_ring, 1127 struct ixgbevf_rx_buffer *rx_buffer, 1128 unsigned int size) 1129 { 1130 unsigned int truesize = ixgbevf_rx_frame_truesize(rx_ring, size); 1131 1132 #if (PAGE_SIZE < 8192) 1133 rx_buffer->page_offset ^= truesize; 1134 #else 1135 rx_buffer->page_offset += truesize; 1136 #endif 1137 } 1138 1139 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector, 1140 struct ixgbevf_ring *rx_ring, 1141 int budget) 1142 { 1143 unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0; 1144 struct ixgbevf_adapter *adapter = q_vector->adapter; 1145 u16 cleaned_count = ixgbevf_desc_unused(rx_ring); 1146 struct sk_buff *skb = rx_ring->skb; 1147 bool xdp_xmit = false; 1148 struct xdp_buff xdp; 1149 int xdp_res = 0; 1150 1151 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */ 1152 #if (PAGE_SIZE < 8192) 1153 frame_sz = ixgbevf_rx_frame_truesize(rx_ring, 0); 1154 #endif 1155 xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq); 1156 1157 while (likely(total_rx_packets < budget)) { 1158 struct ixgbevf_rx_buffer *rx_buffer; 1159 union ixgbe_adv_rx_desc *rx_desc; 1160 unsigned int size; 1161 1162 /* return some buffers to hardware, one at a time is too slow */ 1163 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) { 1164 ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count); 1165 cleaned_count = 0; 1166 } 1167 1168 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean); 1169 size = le16_to_cpu(rx_desc->wb.upper.length); 1170 if (!size) 1171 break; 1172 1173 /* This memory barrier is needed to keep us from reading 1174 * any other fields out of the rx_desc until we know the 1175 * RXD_STAT_DD bit is set 1176 */ 1177 rmb(); 1178 1179 rx_buffer = ixgbevf_get_rx_buffer(rx_ring, size); 1180 1181 /* retrieve a buffer from the ring */ 1182 if (!skb) { 1183 unsigned int offset = ixgbevf_rx_offset(rx_ring); 1184 unsigned char *hard_start; 1185 1186 hard_start = page_address(rx_buffer->page) + 1187 rx_buffer->page_offset - offset; 1188 xdp_prepare_buff(&xdp, hard_start, offset, size, true); 1189 #if (PAGE_SIZE > 4096) 1190 /* At larger PAGE_SIZE, frame_sz depend on len size */ 1191 xdp.frame_sz = ixgbevf_rx_frame_truesize(rx_ring, size); 1192 #endif 1193 xdp_res = ixgbevf_run_xdp(adapter, rx_ring, &xdp); 1194 } 1195 1196 if (xdp_res) { 1197 if (xdp_res == IXGBEVF_XDP_TX) { 1198 xdp_xmit = true; 1199 ixgbevf_rx_buffer_flip(rx_ring, rx_buffer, 1200 size); 1201 } else { 1202 rx_buffer->pagecnt_bias++; 1203 } 1204 total_rx_packets++; 1205 total_rx_bytes += size; 1206 } else if (skb) { 1207 ixgbevf_add_rx_frag(rx_ring, rx_buffer, skb, size); 1208 } else if (ring_uses_build_skb(rx_ring)) { 1209 skb = ixgbevf_build_skb(rx_ring, rx_buffer, 1210 &xdp, rx_desc); 1211 } else { 1212 skb = ixgbevf_construct_skb(rx_ring, rx_buffer, 1213 &xdp, rx_desc); 1214 } 1215 1216 /* exit if we failed to retrieve a buffer */ 1217 if (!xdp_res && !skb) { 1218 rx_ring->rx_stats.alloc_rx_buff_failed++; 1219 rx_buffer->pagecnt_bias++; 1220 break; 1221 } 1222 1223 ixgbevf_put_rx_buffer(rx_ring, rx_buffer, skb); 1224 cleaned_count++; 1225 1226 /* fetch next buffer in frame if non-eop */ 1227 if (ixgbevf_is_non_eop(rx_ring, rx_desc)) 1228 continue; 1229 1230 /* verify the packet layout is correct */ 1231 if (xdp_res || ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) { 1232 skb = NULL; 1233 continue; 1234 } 1235 1236 /* probably a little skewed due to removing CRC */ 1237 total_rx_bytes += skb->len; 1238 1239 /* Workaround hardware that can't do proper VEPA multicast 1240 * source pruning. 1241 */ 1242 if ((skb->pkt_type == PACKET_BROADCAST || 1243 skb->pkt_type == PACKET_MULTICAST) && 1244 ether_addr_equal(rx_ring->netdev->dev_addr, 1245 eth_hdr(skb)->h_source)) { 1246 dev_kfree_skb_irq(skb); 1247 skb = NULL; 1248 continue; 1249 } 1250 1251 /* populate checksum, VLAN, and protocol */ 1252 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb); 1253 1254 ixgbevf_rx_skb(q_vector, skb); 1255 1256 /* reset skb pointer */ 1257 skb = NULL; 1258 1259 /* update budget accounting */ 1260 total_rx_packets++; 1261 } 1262 1263 /* place incomplete frames back on ring for completion */ 1264 rx_ring->skb = skb; 1265 1266 if (xdp_xmit) { 1267 struct ixgbevf_ring *xdp_ring = 1268 adapter->xdp_ring[rx_ring->queue_index]; 1269 1270 /* Force memory writes to complete before letting h/w 1271 * know there are new descriptors to fetch. 1272 */ 1273 wmb(); 1274 ixgbevf_write_tail(xdp_ring, xdp_ring->next_to_use); 1275 } 1276 1277 u64_stats_update_begin(&rx_ring->syncp); 1278 rx_ring->stats.packets += total_rx_packets; 1279 rx_ring->stats.bytes += total_rx_bytes; 1280 u64_stats_update_end(&rx_ring->syncp); 1281 q_vector->rx.total_packets += total_rx_packets; 1282 q_vector->rx.total_bytes += total_rx_bytes; 1283 1284 return total_rx_packets; 1285 } 1286 1287 /** 1288 * ixgbevf_poll - NAPI polling calback 1289 * @napi: napi struct with our devices info in it 1290 * @budget: amount of work driver is allowed to do this pass, in packets 1291 * 1292 * This function will clean more than one or more rings associated with a 1293 * q_vector. 1294 **/ 1295 static int ixgbevf_poll(struct napi_struct *napi, int budget) 1296 { 1297 struct ixgbevf_q_vector *q_vector = 1298 container_of(napi, struct ixgbevf_q_vector, napi); 1299 struct ixgbevf_adapter *adapter = q_vector->adapter; 1300 struct ixgbevf_ring *ring; 1301 int per_ring_budget, work_done = 0; 1302 bool clean_complete = true; 1303 1304 ixgbevf_for_each_ring(ring, q_vector->tx) { 1305 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget)) 1306 clean_complete = false; 1307 } 1308 1309 if (budget <= 0) 1310 return budget; 1311 1312 /* attempt to distribute budget to each queue fairly, but don't allow 1313 * the budget to go below 1 because we'll exit polling 1314 */ 1315 if (q_vector->rx.count > 1) 1316 per_ring_budget = max(budget/q_vector->rx.count, 1); 1317 else 1318 per_ring_budget = budget; 1319 1320 ixgbevf_for_each_ring(ring, q_vector->rx) { 1321 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring, 1322 per_ring_budget); 1323 work_done += cleaned; 1324 if (cleaned >= per_ring_budget) 1325 clean_complete = false; 1326 } 1327 1328 /* If all work not completed, return budget and keep polling */ 1329 if (!clean_complete) 1330 return budget; 1331 1332 /* Exit the polling mode, but don't re-enable interrupts if stack might 1333 * poll us due to busy-polling 1334 */ 1335 if (likely(napi_complete_done(napi, work_done))) { 1336 if (adapter->rx_itr_setting == 1) 1337 ixgbevf_set_itr(q_vector); 1338 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) && 1339 !test_bit(__IXGBEVF_REMOVING, &adapter->state)) 1340 ixgbevf_irq_enable_queues(adapter, 1341 BIT(q_vector->v_idx)); 1342 } 1343 1344 return min(work_done, budget - 1); 1345 } 1346 1347 /** 1348 * ixgbevf_write_eitr - write VTEITR register in hardware specific way 1349 * @q_vector: structure containing interrupt and ring information 1350 **/ 1351 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector) 1352 { 1353 struct ixgbevf_adapter *adapter = q_vector->adapter; 1354 struct ixgbe_hw *hw = &adapter->hw; 1355 int v_idx = q_vector->v_idx; 1356 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR; 1357 1358 /* set the WDIS bit to not clear the timer bits and cause an 1359 * immediate assertion of the interrupt 1360 */ 1361 itr_reg |= IXGBE_EITR_CNT_WDIS; 1362 1363 IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg); 1364 } 1365 1366 /** 1367 * ixgbevf_configure_msix - Configure MSI-X hardware 1368 * @adapter: board private structure 1369 * 1370 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X 1371 * interrupts. 1372 **/ 1373 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter) 1374 { 1375 struct ixgbevf_q_vector *q_vector; 1376 int q_vectors, v_idx; 1377 1378 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS; 1379 adapter->eims_enable_mask = 0; 1380 1381 /* Populate the IVAR table and set the ITR values to the 1382 * corresponding register. 1383 */ 1384 for (v_idx = 0; v_idx < q_vectors; v_idx++) { 1385 struct ixgbevf_ring *ring; 1386 1387 q_vector = adapter->q_vector[v_idx]; 1388 1389 ixgbevf_for_each_ring(ring, q_vector->rx) 1390 ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx); 1391 1392 ixgbevf_for_each_ring(ring, q_vector->tx) 1393 ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx); 1394 1395 if (q_vector->tx.ring && !q_vector->rx.ring) { 1396 /* Tx only vector */ 1397 if (adapter->tx_itr_setting == 1) 1398 q_vector->itr = IXGBE_12K_ITR; 1399 else 1400 q_vector->itr = adapter->tx_itr_setting; 1401 } else { 1402 /* Rx or Rx/Tx vector */ 1403 if (adapter->rx_itr_setting == 1) 1404 q_vector->itr = IXGBE_20K_ITR; 1405 else 1406 q_vector->itr = adapter->rx_itr_setting; 1407 } 1408 1409 /* add q_vector eims value to global eims_enable_mask */ 1410 adapter->eims_enable_mask |= BIT(v_idx); 1411 1412 ixgbevf_write_eitr(q_vector); 1413 } 1414 1415 ixgbevf_set_ivar(adapter, -1, 1, v_idx); 1416 /* setup eims_other and add value to global eims_enable_mask */ 1417 adapter->eims_other = BIT(v_idx); 1418 adapter->eims_enable_mask |= adapter->eims_other; 1419 } 1420 1421 enum latency_range { 1422 lowest_latency = 0, 1423 low_latency = 1, 1424 bulk_latency = 2, 1425 latency_invalid = 255 1426 }; 1427 1428 /** 1429 * ixgbevf_update_itr - update the dynamic ITR value based on statistics 1430 * @q_vector: structure containing interrupt and ring information 1431 * @ring_container: structure containing ring performance data 1432 * 1433 * Stores a new ITR value based on packets and byte 1434 * counts during the last interrupt. The advantage of per interrupt 1435 * computation is faster updates and more accurate ITR for the current 1436 * traffic pattern. Constants in this function were computed 1437 * based on theoretical maximum wire speed and thresholds were set based 1438 * on testing data as well as attempting to minimize response time 1439 * while increasing bulk throughput. 1440 **/ 1441 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector, 1442 struct ixgbevf_ring_container *ring_container) 1443 { 1444 int bytes = ring_container->total_bytes; 1445 int packets = ring_container->total_packets; 1446 u32 timepassed_us; 1447 u64 bytes_perint; 1448 u8 itr_setting = ring_container->itr; 1449 1450 if (packets == 0) 1451 return; 1452 1453 /* simple throttle rate management 1454 * 0-20MB/s lowest (100000 ints/s) 1455 * 20-100MB/s low (20000 ints/s) 1456 * 100-1249MB/s bulk (12000 ints/s) 1457 */ 1458 /* what was last interrupt timeslice? */ 1459 timepassed_us = q_vector->itr >> 2; 1460 if (timepassed_us == 0) 1461 return; 1462 1463 bytes_perint = bytes / timepassed_us; /* bytes/usec */ 1464 1465 switch (itr_setting) { 1466 case lowest_latency: 1467 if (bytes_perint > 10) 1468 itr_setting = low_latency; 1469 break; 1470 case low_latency: 1471 if (bytes_perint > 20) 1472 itr_setting = bulk_latency; 1473 else if (bytes_perint <= 10) 1474 itr_setting = lowest_latency; 1475 break; 1476 case bulk_latency: 1477 if (bytes_perint <= 20) 1478 itr_setting = low_latency; 1479 break; 1480 } 1481 1482 /* clear work counters since we have the values we need */ 1483 ring_container->total_bytes = 0; 1484 ring_container->total_packets = 0; 1485 1486 /* write updated itr to ring container */ 1487 ring_container->itr = itr_setting; 1488 } 1489 1490 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector) 1491 { 1492 u32 new_itr = q_vector->itr; 1493 u8 current_itr; 1494 1495 ixgbevf_update_itr(q_vector, &q_vector->tx); 1496 ixgbevf_update_itr(q_vector, &q_vector->rx); 1497 1498 current_itr = max(q_vector->rx.itr, q_vector->tx.itr); 1499 1500 switch (current_itr) { 1501 /* counts and packets in update_itr are dependent on these numbers */ 1502 case lowest_latency: 1503 new_itr = IXGBE_100K_ITR; 1504 break; 1505 case low_latency: 1506 new_itr = IXGBE_20K_ITR; 1507 break; 1508 case bulk_latency: 1509 new_itr = IXGBE_12K_ITR; 1510 break; 1511 default: 1512 break; 1513 } 1514 1515 if (new_itr != q_vector->itr) { 1516 /* do an exponential smoothing */ 1517 new_itr = (10 * new_itr * q_vector->itr) / 1518 ((9 * new_itr) + q_vector->itr); 1519 1520 /* save the algorithm value here */ 1521 q_vector->itr = new_itr; 1522 1523 ixgbevf_write_eitr(q_vector); 1524 } 1525 } 1526 1527 static irqreturn_t ixgbevf_msix_other(int irq, void *data) 1528 { 1529 struct ixgbevf_adapter *adapter = data; 1530 struct ixgbe_hw *hw = &adapter->hw; 1531 1532 hw->mac.get_link_status = 1; 1533 1534 ixgbevf_service_event_schedule(adapter); 1535 1536 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other); 1537 1538 return IRQ_HANDLED; 1539 } 1540 1541 /** 1542 * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues) 1543 * @irq: unused 1544 * @data: pointer to our q_vector struct for this interrupt vector 1545 **/ 1546 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data) 1547 { 1548 struct ixgbevf_q_vector *q_vector = data; 1549 1550 /* EIAM disabled interrupts (on this vector) for us */ 1551 if (q_vector->rx.ring || q_vector->tx.ring) 1552 napi_schedule_irqoff(&q_vector->napi); 1553 1554 return IRQ_HANDLED; 1555 } 1556 1557 /** 1558 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts 1559 * @adapter: board private structure 1560 * 1561 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests 1562 * interrupts from the kernel. 1563 **/ 1564 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter) 1565 { 1566 struct net_device *netdev = adapter->netdev; 1567 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS; 1568 unsigned int ri = 0, ti = 0; 1569 int vector, err; 1570 1571 for (vector = 0; vector < q_vectors; vector++) { 1572 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector]; 1573 struct msix_entry *entry = &adapter->msix_entries[vector]; 1574 1575 if (q_vector->tx.ring && q_vector->rx.ring) { 1576 snprintf(q_vector->name, sizeof(q_vector->name), 1577 "%s-TxRx-%u", netdev->name, ri++); 1578 ti++; 1579 } else if (q_vector->rx.ring) { 1580 snprintf(q_vector->name, sizeof(q_vector->name), 1581 "%s-rx-%u", netdev->name, ri++); 1582 } else if (q_vector->tx.ring) { 1583 snprintf(q_vector->name, sizeof(q_vector->name), 1584 "%s-tx-%u", netdev->name, ti++); 1585 } else { 1586 /* skip this unused q_vector */ 1587 continue; 1588 } 1589 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0, 1590 q_vector->name, q_vector); 1591 if (err) { 1592 hw_dbg(&adapter->hw, 1593 "request_irq failed for MSIX interrupt Error: %d\n", 1594 err); 1595 goto free_queue_irqs; 1596 } 1597 } 1598 1599 err = request_irq(adapter->msix_entries[vector].vector, 1600 &ixgbevf_msix_other, 0, netdev->name, adapter); 1601 if (err) { 1602 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n", 1603 err); 1604 goto free_queue_irqs; 1605 } 1606 1607 return 0; 1608 1609 free_queue_irqs: 1610 while (vector) { 1611 vector--; 1612 free_irq(adapter->msix_entries[vector].vector, 1613 adapter->q_vector[vector]); 1614 } 1615 /* This failure is non-recoverable - it indicates the system is 1616 * out of MSIX vector resources and the VF driver cannot run 1617 * without them. Set the number of msix vectors to zero 1618 * indicating that not enough can be allocated. The error 1619 * will be returned to the user indicating device open failed. 1620 * Any further attempts to force the driver to open will also 1621 * fail. The only way to recover is to unload the driver and 1622 * reload it again. If the system has recovered some MSIX 1623 * vectors then it may succeed. 1624 */ 1625 adapter->num_msix_vectors = 0; 1626 return err; 1627 } 1628 1629 /** 1630 * ixgbevf_request_irq - initialize interrupts 1631 * @adapter: board private structure 1632 * 1633 * Attempts to configure interrupts using the best available 1634 * capabilities of the hardware and kernel. 1635 **/ 1636 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter) 1637 { 1638 int err = ixgbevf_request_msix_irqs(adapter); 1639 1640 if (err) 1641 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err); 1642 1643 return err; 1644 } 1645 1646 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter) 1647 { 1648 int i, q_vectors; 1649 1650 if (!adapter->msix_entries) 1651 return; 1652 1653 q_vectors = adapter->num_msix_vectors; 1654 i = q_vectors - 1; 1655 1656 free_irq(adapter->msix_entries[i].vector, adapter); 1657 i--; 1658 1659 for (; i >= 0; i--) { 1660 /* free only the irqs that were actually requested */ 1661 if (!adapter->q_vector[i]->rx.ring && 1662 !adapter->q_vector[i]->tx.ring) 1663 continue; 1664 1665 free_irq(adapter->msix_entries[i].vector, 1666 adapter->q_vector[i]); 1667 } 1668 } 1669 1670 /** 1671 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC 1672 * @adapter: board private structure 1673 **/ 1674 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter) 1675 { 1676 struct ixgbe_hw *hw = &adapter->hw; 1677 int i; 1678 1679 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0); 1680 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0); 1681 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0); 1682 1683 IXGBE_WRITE_FLUSH(hw); 1684 1685 for (i = 0; i < adapter->num_msix_vectors; i++) 1686 synchronize_irq(adapter->msix_entries[i].vector); 1687 } 1688 1689 /** 1690 * ixgbevf_irq_enable - Enable default interrupt generation settings 1691 * @adapter: board private structure 1692 **/ 1693 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter) 1694 { 1695 struct ixgbe_hw *hw = &adapter->hw; 1696 1697 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask); 1698 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask); 1699 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask); 1700 } 1701 1702 /** 1703 * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset 1704 * @adapter: board private structure 1705 * @ring: structure containing ring specific data 1706 * 1707 * Configure the Tx descriptor ring after a reset. 1708 **/ 1709 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter, 1710 struct ixgbevf_ring *ring) 1711 { 1712 struct ixgbe_hw *hw = &adapter->hw; 1713 u64 tdba = ring->dma; 1714 int wait_loop = 10; 1715 u32 txdctl = IXGBE_TXDCTL_ENABLE; 1716 u8 reg_idx = ring->reg_idx; 1717 1718 /* disable queue to avoid issues while updating state */ 1719 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH); 1720 IXGBE_WRITE_FLUSH(hw); 1721 1722 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32)); 1723 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32); 1724 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx), 1725 ring->count * sizeof(union ixgbe_adv_tx_desc)); 1726 1727 /* disable head writeback */ 1728 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0); 1729 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0); 1730 1731 /* enable relaxed ordering */ 1732 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx), 1733 (IXGBE_DCA_TXCTRL_DESC_RRO_EN | 1734 IXGBE_DCA_TXCTRL_DATA_RRO_EN)); 1735 1736 /* reset head and tail pointers */ 1737 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0); 1738 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0); 1739 ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx); 1740 1741 /* reset ntu and ntc to place SW in sync with hardwdare */ 1742 ring->next_to_clean = 0; 1743 ring->next_to_use = 0; 1744 1745 /* In order to avoid issues WTHRESH + PTHRESH should always be equal 1746 * to or less than the number of on chip descriptors, which is 1747 * currently 40. 1748 */ 1749 txdctl |= (8 << 16); /* WTHRESH = 8 */ 1750 1751 /* Setting PTHRESH to 32 both improves performance */ 1752 txdctl |= (1u << 8) | /* HTHRESH = 1 */ 1753 32; /* PTHRESH = 32 */ 1754 1755 /* reinitialize tx_buffer_info */ 1756 memset(ring->tx_buffer_info, 0, 1757 sizeof(struct ixgbevf_tx_buffer) * ring->count); 1758 1759 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state); 1760 clear_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state); 1761 1762 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl); 1763 1764 /* poll to verify queue is enabled */ 1765 do { 1766 usleep_range(1000, 2000); 1767 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx)); 1768 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE)); 1769 if (!wait_loop) 1770 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx); 1771 } 1772 1773 /** 1774 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset 1775 * @adapter: board private structure 1776 * 1777 * Configure the Tx unit of the MAC after a reset. 1778 **/ 1779 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter) 1780 { 1781 u32 i; 1782 1783 /* Setup the HW Tx Head and Tail descriptor pointers */ 1784 for (i = 0; i < adapter->num_tx_queues; i++) 1785 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]); 1786 for (i = 0; i < adapter->num_xdp_queues; i++) 1787 ixgbevf_configure_tx_ring(adapter, adapter->xdp_ring[i]); 1788 } 1789 1790 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2 1791 1792 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, 1793 struct ixgbevf_ring *ring, int index) 1794 { 1795 struct ixgbe_hw *hw = &adapter->hw; 1796 u32 srrctl; 1797 1798 srrctl = IXGBE_SRRCTL_DROP_EN; 1799 1800 srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT; 1801 if (ring_uses_large_buffer(ring)) 1802 srrctl |= IXGBEVF_RXBUFFER_3072 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT; 1803 else 1804 srrctl |= IXGBEVF_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT; 1805 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF; 1806 1807 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl); 1808 } 1809 1810 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter) 1811 { 1812 struct ixgbe_hw *hw = &adapter->hw; 1813 1814 /* PSRTYPE must be initialized in 82599 */ 1815 u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR | 1816 IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR | 1817 IXGBE_PSRTYPE_L2HDR; 1818 1819 if (adapter->num_rx_queues > 1) 1820 psrtype |= BIT(29); 1821 1822 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype); 1823 } 1824 1825 #define IXGBEVF_MAX_RX_DESC_POLL 10 1826 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter, 1827 struct ixgbevf_ring *ring) 1828 { 1829 struct ixgbe_hw *hw = &adapter->hw; 1830 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL; 1831 u32 rxdctl; 1832 u8 reg_idx = ring->reg_idx; 1833 1834 if (IXGBE_REMOVED(hw->hw_addr)) 1835 return; 1836 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx)); 1837 rxdctl &= ~IXGBE_RXDCTL_ENABLE; 1838 1839 /* write value back with RXDCTL.ENABLE bit cleared */ 1840 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl); 1841 1842 /* the hardware may take up to 100us to really disable the Rx queue */ 1843 do { 1844 udelay(10); 1845 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx)); 1846 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE)); 1847 1848 if (!wait_loop) 1849 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n", 1850 reg_idx); 1851 } 1852 1853 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter, 1854 struct ixgbevf_ring *ring) 1855 { 1856 struct ixgbe_hw *hw = &adapter->hw; 1857 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL; 1858 u32 rxdctl; 1859 u8 reg_idx = ring->reg_idx; 1860 1861 if (IXGBE_REMOVED(hw->hw_addr)) 1862 return; 1863 do { 1864 usleep_range(1000, 2000); 1865 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx)); 1866 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE)); 1867 1868 if (!wait_loop) 1869 pr_err("RXDCTL.ENABLE queue %d not set while polling\n", 1870 reg_idx); 1871 } 1872 1873 /** 1874 * ixgbevf_init_rss_key - Initialize adapter RSS key 1875 * @adapter: device handle 1876 * 1877 * Allocates and initializes the RSS key if it is not allocated. 1878 **/ 1879 static inline int ixgbevf_init_rss_key(struct ixgbevf_adapter *adapter) 1880 { 1881 u32 *rss_key; 1882 1883 if (!adapter->rss_key) { 1884 rss_key = kzalloc(IXGBEVF_RSS_HASH_KEY_SIZE, GFP_KERNEL); 1885 if (unlikely(!rss_key)) 1886 return -ENOMEM; 1887 1888 netdev_rss_key_fill(rss_key, IXGBEVF_RSS_HASH_KEY_SIZE); 1889 adapter->rss_key = rss_key; 1890 } 1891 1892 return 0; 1893 } 1894 1895 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter) 1896 { 1897 struct ixgbe_hw *hw = &adapter->hw; 1898 u32 vfmrqc = 0, vfreta = 0; 1899 u16 rss_i = adapter->num_rx_queues; 1900 u8 i, j; 1901 1902 /* Fill out hash function seeds */ 1903 for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++) 1904 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), *(adapter->rss_key + i)); 1905 1906 for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) { 1907 if (j == rss_i) 1908 j = 0; 1909 1910 adapter->rss_indir_tbl[i] = j; 1911 1912 vfreta |= j << (i & 0x3) * 8; 1913 if ((i & 3) == 3) { 1914 IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta); 1915 vfreta = 0; 1916 } 1917 } 1918 1919 /* Perform hash on these packet types */ 1920 vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 | 1921 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP | 1922 IXGBE_VFMRQC_RSS_FIELD_IPV6 | 1923 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP; 1924 1925 vfmrqc |= IXGBE_VFMRQC_RSSEN; 1926 1927 IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc); 1928 } 1929 1930 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter, 1931 struct ixgbevf_ring *ring) 1932 { 1933 struct ixgbe_hw *hw = &adapter->hw; 1934 union ixgbe_adv_rx_desc *rx_desc; 1935 u64 rdba = ring->dma; 1936 u32 rxdctl; 1937 u8 reg_idx = ring->reg_idx; 1938 1939 /* disable queue to avoid issues while updating state */ 1940 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx)); 1941 ixgbevf_disable_rx_queue(adapter, ring); 1942 1943 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32)); 1944 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32); 1945 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx), 1946 ring->count * sizeof(union ixgbe_adv_rx_desc)); 1947 1948 #ifndef CONFIG_SPARC 1949 /* enable relaxed ordering */ 1950 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx), 1951 IXGBE_DCA_RXCTRL_DESC_RRO_EN); 1952 #else 1953 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx), 1954 IXGBE_DCA_RXCTRL_DESC_RRO_EN | 1955 IXGBE_DCA_RXCTRL_DATA_WRO_EN); 1956 #endif 1957 1958 /* reset head and tail pointers */ 1959 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0); 1960 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0); 1961 ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx); 1962 1963 /* initialize rx_buffer_info */ 1964 memset(ring->rx_buffer_info, 0, 1965 sizeof(struct ixgbevf_rx_buffer) * ring->count); 1966 1967 /* initialize Rx descriptor 0 */ 1968 rx_desc = IXGBEVF_RX_DESC(ring, 0); 1969 rx_desc->wb.upper.length = 0; 1970 1971 /* reset ntu and ntc to place SW in sync with hardwdare */ 1972 ring->next_to_clean = 0; 1973 ring->next_to_use = 0; 1974 ring->next_to_alloc = 0; 1975 1976 ixgbevf_configure_srrctl(adapter, ring, reg_idx); 1977 1978 /* RXDCTL.RLPML does not work on 82599 */ 1979 if (adapter->hw.mac.type != ixgbe_mac_82599_vf) { 1980 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK | 1981 IXGBE_RXDCTL_RLPML_EN); 1982 1983 #if (PAGE_SIZE < 8192) 1984 /* Limit the maximum frame size so we don't overrun the skb */ 1985 if (ring_uses_build_skb(ring) && 1986 !ring_uses_large_buffer(ring)) 1987 rxdctl |= IXGBEVF_MAX_FRAME_BUILD_SKB | 1988 IXGBE_RXDCTL_RLPML_EN; 1989 #endif 1990 } 1991 1992 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME; 1993 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl); 1994 1995 ixgbevf_rx_desc_queue_enable(adapter, ring); 1996 ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring)); 1997 } 1998 1999 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter, 2000 struct ixgbevf_ring *rx_ring) 2001 { 2002 struct net_device *netdev = adapter->netdev; 2003 unsigned int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN; 2004 2005 /* set build_skb and buffer size flags */ 2006 clear_ring_build_skb_enabled(rx_ring); 2007 clear_ring_uses_large_buffer(rx_ring); 2008 2009 if (adapter->flags & IXGBEVF_FLAGS_LEGACY_RX) 2010 return; 2011 2012 if (PAGE_SIZE < 8192) 2013 if (max_frame > IXGBEVF_MAX_FRAME_BUILD_SKB) 2014 set_ring_uses_large_buffer(rx_ring); 2015 2016 /* 82599 can't rely on RXDCTL.RLPML to restrict the size of the frame */ 2017 if (adapter->hw.mac.type == ixgbe_mac_82599_vf && !ring_uses_large_buffer(rx_ring)) 2018 return; 2019 2020 set_ring_build_skb_enabled(rx_ring); 2021 } 2022 2023 /** 2024 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset 2025 * @adapter: board private structure 2026 * 2027 * Configure the Rx unit of the MAC after a reset. 2028 **/ 2029 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter) 2030 { 2031 struct ixgbe_hw *hw = &adapter->hw; 2032 struct net_device *netdev = adapter->netdev; 2033 int i, ret; 2034 2035 ixgbevf_setup_psrtype(adapter); 2036 if (hw->mac.type >= ixgbe_mac_X550_vf) 2037 ixgbevf_setup_vfmrqc(adapter); 2038 2039 spin_lock_bh(&adapter->mbx_lock); 2040 /* notify the PF of our intent to use this size of frame */ 2041 ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN); 2042 spin_unlock_bh(&adapter->mbx_lock); 2043 if (ret) 2044 dev_err(&adapter->pdev->dev, 2045 "Failed to set MTU at %d\n", netdev->mtu); 2046 2047 /* Setup the HW Rx Head and Tail Descriptor Pointers and 2048 * the Base and Length of the Rx Descriptor Ring 2049 */ 2050 for (i = 0; i < adapter->num_rx_queues; i++) { 2051 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i]; 2052 2053 ixgbevf_set_rx_buffer_len(adapter, rx_ring); 2054 ixgbevf_configure_rx_ring(adapter, rx_ring); 2055 } 2056 } 2057 2058 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev, 2059 __be16 proto, u16 vid) 2060 { 2061 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 2062 struct ixgbe_hw *hw = &adapter->hw; 2063 int err; 2064 2065 spin_lock_bh(&adapter->mbx_lock); 2066 2067 /* add VID to filter table */ 2068 err = hw->mac.ops.set_vfta(hw, vid, 0, true); 2069 2070 spin_unlock_bh(&adapter->mbx_lock); 2071 2072 if (err) { 2073 netdev_err(netdev, "VF could not set VLAN %d\n", vid); 2074 2075 /* translate error return types so error makes sense */ 2076 if (err == IXGBE_ERR_MBX) 2077 return -EIO; 2078 2079 if (err == IXGBE_ERR_INVALID_ARGUMENT) 2080 return -EACCES; 2081 } 2082 2083 set_bit(vid, adapter->active_vlans); 2084 2085 return err; 2086 } 2087 2088 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev, 2089 __be16 proto, u16 vid) 2090 { 2091 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 2092 struct ixgbe_hw *hw = &adapter->hw; 2093 int err; 2094 2095 spin_lock_bh(&adapter->mbx_lock); 2096 2097 /* remove VID from filter table */ 2098 err = hw->mac.ops.set_vfta(hw, vid, 0, false); 2099 2100 spin_unlock_bh(&adapter->mbx_lock); 2101 2102 if (err) 2103 netdev_err(netdev, "Could not remove VLAN %d\n", vid); 2104 2105 clear_bit(vid, adapter->active_vlans); 2106 2107 return err; 2108 } 2109 2110 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter) 2111 { 2112 u16 vid; 2113 2114 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID) 2115 ixgbevf_vlan_rx_add_vid(adapter->netdev, 2116 htons(ETH_P_8021Q), vid); 2117 } 2118 2119 static int ixgbevf_write_uc_addr_list(struct net_device *netdev) 2120 { 2121 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 2122 struct ixgbe_hw *hw = &adapter->hw; 2123 int count = 0; 2124 2125 if (!netdev_uc_empty(netdev)) { 2126 struct netdev_hw_addr *ha; 2127 2128 netdev_for_each_uc_addr(ha, netdev) { 2129 hw->mac.ops.set_uc_addr(hw, ++count, ha->addr); 2130 udelay(200); 2131 } 2132 } else { 2133 /* If the list is empty then send message to PF driver to 2134 * clear all MAC VLANs on this VF. 2135 */ 2136 hw->mac.ops.set_uc_addr(hw, 0, NULL); 2137 } 2138 2139 return count; 2140 } 2141 2142 /** 2143 * ixgbevf_set_rx_mode - Multicast and unicast set 2144 * @netdev: network interface device structure 2145 * 2146 * The set_rx_method entry point is called whenever the multicast address 2147 * list, unicast address list or the network interface flags are updated. 2148 * This routine is responsible for configuring the hardware for proper 2149 * multicast mode and configuring requested unicast filters. 2150 **/ 2151 static void ixgbevf_set_rx_mode(struct net_device *netdev) 2152 { 2153 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 2154 struct ixgbe_hw *hw = &adapter->hw; 2155 unsigned int flags = netdev->flags; 2156 int xcast_mode; 2157 2158 /* request the most inclusive mode we need */ 2159 if (flags & IFF_PROMISC) 2160 xcast_mode = IXGBEVF_XCAST_MODE_PROMISC; 2161 else if (flags & IFF_ALLMULTI) 2162 xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI; 2163 else if (flags & (IFF_BROADCAST | IFF_MULTICAST)) 2164 xcast_mode = IXGBEVF_XCAST_MODE_MULTI; 2165 else 2166 xcast_mode = IXGBEVF_XCAST_MODE_NONE; 2167 2168 spin_lock_bh(&adapter->mbx_lock); 2169 2170 hw->mac.ops.update_xcast_mode(hw, xcast_mode); 2171 2172 /* reprogram multicast list */ 2173 hw->mac.ops.update_mc_addr_list(hw, netdev); 2174 2175 ixgbevf_write_uc_addr_list(netdev); 2176 2177 spin_unlock_bh(&adapter->mbx_lock); 2178 } 2179 2180 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter) 2181 { 2182 int q_idx; 2183 struct ixgbevf_q_vector *q_vector; 2184 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS; 2185 2186 for (q_idx = 0; q_idx < q_vectors; q_idx++) { 2187 q_vector = adapter->q_vector[q_idx]; 2188 napi_enable(&q_vector->napi); 2189 } 2190 } 2191 2192 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter) 2193 { 2194 int q_idx; 2195 struct ixgbevf_q_vector *q_vector; 2196 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS; 2197 2198 for (q_idx = 0; q_idx < q_vectors; q_idx++) { 2199 q_vector = adapter->q_vector[q_idx]; 2200 napi_disable(&q_vector->napi); 2201 } 2202 } 2203 2204 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter) 2205 { 2206 struct ixgbe_hw *hw = &adapter->hw; 2207 unsigned int def_q = 0; 2208 unsigned int num_tcs = 0; 2209 unsigned int num_rx_queues = adapter->num_rx_queues; 2210 unsigned int num_tx_queues = adapter->num_tx_queues; 2211 int err; 2212 2213 spin_lock_bh(&adapter->mbx_lock); 2214 2215 /* fetch queue configuration from the PF */ 2216 err = ixgbevf_get_queues(hw, &num_tcs, &def_q); 2217 2218 spin_unlock_bh(&adapter->mbx_lock); 2219 2220 if (err) 2221 return err; 2222 2223 if (num_tcs > 1) { 2224 /* we need only one Tx queue */ 2225 num_tx_queues = 1; 2226 2227 /* update default Tx ring register index */ 2228 adapter->tx_ring[0]->reg_idx = def_q; 2229 2230 /* we need as many queues as traffic classes */ 2231 num_rx_queues = num_tcs; 2232 } 2233 2234 /* if we have a bad config abort request queue reset */ 2235 if ((adapter->num_rx_queues != num_rx_queues) || 2236 (adapter->num_tx_queues != num_tx_queues)) { 2237 /* force mailbox timeout to prevent further messages */ 2238 hw->mbx.timeout = 0; 2239 2240 /* wait for watchdog to come around and bail us out */ 2241 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state); 2242 } 2243 2244 return 0; 2245 } 2246 2247 static void ixgbevf_configure(struct ixgbevf_adapter *adapter) 2248 { 2249 ixgbevf_configure_dcb(adapter); 2250 2251 ixgbevf_set_rx_mode(adapter->netdev); 2252 2253 ixgbevf_restore_vlan(adapter); 2254 ixgbevf_ipsec_restore(adapter); 2255 2256 ixgbevf_configure_tx(adapter); 2257 ixgbevf_configure_rx(adapter); 2258 } 2259 2260 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter) 2261 { 2262 /* Only save pre-reset stats if there are some */ 2263 if (adapter->stats.vfgprc || adapter->stats.vfgptc) { 2264 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc - 2265 adapter->stats.base_vfgprc; 2266 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc - 2267 adapter->stats.base_vfgptc; 2268 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc - 2269 adapter->stats.base_vfgorc; 2270 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc - 2271 adapter->stats.base_vfgotc; 2272 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc - 2273 adapter->stats.base_vfmprc; 2274 } 2275 } 2276 2277 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter) 2278 { 2279 struct ixgbe_hw *hw = &adapter->hw; 2280 2281 adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC); 2282 adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB); 2283 adapter->stats.last_vfgorc |= 2284 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32); 2285 adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC); 2286 adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB); 2287 adapter->stats.last_vfgotc |= 2288 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32); 2289 adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC); 2290 2291 adapter->stats.base_vfgprc = adapter->stats.last_vfgprc; 2292 adapter->stats.base_vfgorc = adapter->stats.last_vfgorc; 2293 adapter->stats.base_vfgptc = adapter->stats.last_vfgptc; 2294 adapter->stats.base_vfgotc = adapter->stats.last_vfgotc; 2295 adapter->stats.base_vfmprc = adapter->stats.last_vfmprc; 2296 } 2297 2298 /** 2299 * ixgbevf_set_features - Set features supported by PF 2300 * @adapter: pointer to the adapter struct 2301 * 2302 * Negotiate with PF supported features and then set pf_features accordingly. 2303 */ 2304 static void ixgbevf_set_features(struct ixgbevf_adapter *adapter) 2305 { 2306 u32 *pf_features = &adapter->pf_features; 2307 struct ixgbe_hw *hw = &adapter->hw; 2308 int err; 2309 2310 err = hw->mac.ops.negotiate_features(hw, pf_features); 2311 if (err && err != -EOPNOTSUPP) 2312 netdev_dbg(adapter->netdev, 2313 "PF feature negotiation failed.\n"); 2314 2315 /* Address also pre API 1.7 cases */ 2316 if (hw->api_version == ixgbe_mbox_api_14) 2317 *pf_features |= IXGBEVF_PF_SUP_IPSEC; 2318 else if (hw->api_version == ixgbe_mbox_api_15) 2319 *pf_features |= IXGBEVF_PF_SUP_ESX_MBX; 2320 } 2321 2322 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter) 2323 { 2324 struct ixgbe_hw *hw = &adapter->hw; 2325 static const int api[] = { 2326 ixgbe_mbox_api_17, 2327 ixgbe_mbox_api_16, 2328 ixgbe_mbox_api_15, 2329 ixgbe_mbox_api_14, 2330 ixgbe_mbox_api_13, 2331 ixgbe_mbox_api_12, 2332 ixgbe_mbox_api_11, 2333 ixgbe_mbox_api_10, 2334 ixgbe_mbox_api_unknown 2335 }; 2336 int err, idx = 0; 2337 2338 spin_lock_bh(&adapter->mbx_lock); 2339 2340 while (api[idx] != ixgbe_mbox_api_unknown) { 2341 err = hw->mac.ops.negotiate_api_version(hw, api[idx]); 2342 if (!err) 2343 break; 2344 idx++; 2345 } 2346 2347 ixgbevf_set_features(adapter); 2348 2349 if (adapter->pf_features & IXGBEVF_PF_SUP_ESX_MBX) { 2350 hw->mbx.ops.init_params(hw); 2351 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops, 2352 sizeof(struct ixgbe_mbx_operations)); 2353 } 2354 2355 spin_unlock_bh(&adapter->mbx_lock); 2356 } 2357 2358 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter) 2359 { 2360 struct net_device *netdev = adapter->netdev; 2361 struct pci_dev *pdev = adapter->pdev; 2362 struct ixgbe_hw *hw = &adapter->hw; 2363 bool state; 2364 2365 ixgbevf_configure_msix(adapter); 2366 2367 spin_lock_bh(&adapter->mbx_lock); 2368 2369 if (is_valid_ether_addr(hw->mac.addr)) 2370 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0); 2371 else 2372 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0); 2373 2374 spin_unlock_bh(&adapter->mbx_lock); 2375 2376 state = adapter->link_state; 2377 hw->mac.ops.get_link_state(hw, &adapter->link_state); 2378 if (state && state != adapter->link_state) 2379 dev_info(&pdev->dev, "VF is administratively disabled\n"); 2380 2381 smp_mb__before_atomic(); 2382 clear_bit(__IXGBEVF_DOWN, &adapter->state); 2383 ixgbevf_napi_enable_all(adapter); 2384 2385 /* clear any pending interrupts, may auto mask */ 2386 IXGBE_READ_REG(hw, IXGBE_VTEICR); 2387 ixgbevf_irq_enable(adapter); 2388 2389 /* enable transmits */ 2390 netif_tx_start_all_queues(netdev); 2391 2392 ixgbevf_save_reset_stats(adapter); 2393 ixgbevf_init_last_counter_stats(adapter); 2394 2395 hw->mac.get_link_status = 1; 2396 mod_timer(&adapter->service_timer, jiffies); 2397 } 2398 2399 void ixgbevf_up(struct ixgbevf_adapter *adapter) 2400 { 2401 ixgbevf_configure(adapter); 2402 2403 ixgbevf_up_complete(adapter); 2404 } 2405 2406 /** 2407 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue 2408 * @rx_ring: ring to free buffers from 2409 **/ 2410 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring) 2411 { 2412 u16 i = rx_ring->next_to_clean; 2413 2414 /* Free Rx ring sk_buff */ 2415 if (rx_ring->skb) { 2416 dev_kfree_skb(rx_ring->skb); 2417 rx_ring->skb = NULL; 2418 } 2419 2420 /* Free all the Rx ring pages */ 2421 while (i != rx_ring->next_to_alloc) { 2422 struct ixgbevf_rx_buffer *rx_buffer; 2423 2424 rx_buffer = &rx_ring->rx_buffer_info[i]; 2425 2426 /* Invalidate cache lines that may have been written to by 2427 * device so that we avoid corrupting memory. 2428 */ 2429 dma_sync_single_range_for_cpu(rx_ring->dev, 2430 rx_buffer->dma, 2431 rx_buffer->page_offset, 2432 ixgbevf_rx_bufsz(rx_ring), 2433 DMA_FROM_DEVICE); 2434 2435 /* free resources associated with mapping */ 2436 dma_unmap_page_attrs(rx_ring->dev, 2437 rx_buffer->dma, 2438 ixgbevf_rx_pg_size(rx_ring), 2439 DMA_FROM_DEVICE, 2440 IXGBEVF_RX_DMA_ATTR); 2441 2442 __page_frag_cache_drain(rx_buffer->page, 2443 rx_buffer->pagecnt_bias); 2444 2445 i++; 2446 if (i == rx_ring->count) 2447 i = 0; 2448 } 2449 2450 rx_ring->next_to_alloc = 0; 2451 rx_ring->next_to_clean = 0; 2452 rx_ring->next_to_use = 0; 2453 } 2454 2455 /** 2456 * ixgbevf_clean_tx_ring - Free Tx Buffers 2457 * @tx_ring: ring to be cleaned 2458 **/ 2459 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring) 2460 { 2461 u16 i = tx_ring->next_to_clean; 2462 struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i]; 2463 2464 while (i != tx_ring->next_to_use) { 2465 union ixgbe_adv_tx_desc *eop_desc, *tx_desc; 2466 2467 /* Free all the Tx ring sk_buffs */ 2468 if (ring_is_xdp(tx_ring)) 2469 page_frag_free(tx_buffer->data); 2470 else 2471 dev_kfree_skb_any(tx_buffer->skb); 2472 2473 /* unmap skb header data */ 2474 dma_unmap_single(tx_ring->dev, 2475 dma_unmap_addr(tx_buffer, dma), 2476 dma_unmap_len(tx_buffer, len), 2477 DMA_TO_DEVICE); 2478 2479 /* check for eop_desc to determine the end of the packet */ 2480 eop_desc = tx_buffer->next_to_watch; 2481 tx_desc = IXGBEVF_TX_DESC(tx_ring, i); 2482 2483 /* unmap remaining buffers */ 2484 while (tx_desc != eop_desc) { 2485 tx_buffer++; 2486 tx_desc++; 2487 i++; 2488 if (unlikely(i == tx_ring->count)) { 2489 i = 0; 2490 tx_buffer = tx_ring->tx_buffer_info; 2491 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0); 2492 } 2493 2494 /* unmap any remaining paged data */ 2495 if (dma_unmap_len(tx_buffer, len)) 2496 dma_unmap_page(tx_ring->dev, 2497 dma_unmap_addr(tx_buffer, dma), 2498 dma_unmap_len(tx_buffer, len), 2499 DMA_TO_DEVICE); 2500 } 2501 2502 /* move us one more past the eop_desc for start of next pkt */ 2503 tx_buffer++; 2504 i++; 2505 if (unlikely(i == tx_ring->count)) { 2506 i = 0; 2507 tx_buffer = tx_ring->tx_buffer_info; 2508 } 2509 } 2510 2511 /* reset next_to_use and next_to_clean */ 2512 tx_ring->next_to_use = 0; 2513 tx_ring->next_to_clean = 0; 2514 2515 } 2516 2517 /** 2518 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues 2519 * @adapter: board private structure 2520 **/ 2521 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter) 2522 { 2523 int i; 2524 2525 for (i = 0; i < adapter->num_rx_queues; i++) 2526 ixgbevf_clean_rx_ring(adapter->rx_ring[i]); 2527 } 2528 2529 /** 2530 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues 2531 * @adapter: board private structure 2532 **/ 2533 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter) 2534 { 2535 int i; 2536 2537 for (i = 0; i < adapter->num_tx_queues; i++) 2538 ixgbevf_clean_tx_ring(adapter->tx_ring[i]); 2539 for (i = 0; i < adapter->num_xdp_queues; i++) 2540 ixgbevf_clean_tx_ring(adapter->xdp_ring[i]); 2541 } 2542 2543 void ixgbevf_down(struct ixgbevf_adapter *adapter) 2544 { 2545 struct net_device *netdev = adapter->netdev; 2546 struct ixgbe_hw *hw = &adapter->hw; 2547 int i; 2548 2549 /* signal that we are down to the interrupt handler */ 2550 if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state)) 2551 return; /* do nothing if already down */ 2552 2553 /* disable all enabled Rx queues */ 2554 for (i = 0; i < adapter->num_rx_queues; i++) 2555 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]); 2556 2557 usleep_range(10000, 20000); 2558 2559 netif_tx_stop_all_queues(netdev); 2560 2561 /* call carrier off first to avoid false dev_watchdog timeouts */ 2562 netif_carrier_off(netdev); 2563 netif_tx_disable(netdev); 2564 2565 ixgbevf_irq_disable(adapter); 2566 2567 ixgbevf_napi_disable_all(adapter); 2568 2569 timer_delete_sync(&adapter->service_timer); 2570 2571 /* disable transmits in the hardware now that interrupts are off */ 2572 for (i = 0; i < adapter->num_tx_queues; i++) { 2573 u8 reg_idx = adapter->tx_ring[i]->reg_idx; 2574 2575 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), 2576 IXGBE_TXDCTL_SWFLSH); 2577 } 2578 2579 for (i = 0; i < adapter->num_xdp_queues; i++) { 2580 u8 reg_idx = adapter->xdp_ring[i]->reg_idx; 2581 2582 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), 2583 IXGBE_TXDCTL_SWFLSH); 2584 } 2585 2586 if (!pci_channel_offline(adapter->pdev)) 2587 ixgbevf_reset(adapter); 2588 2589 ixgbevf_clean_all_tx_rings(adapter); 2590 ixgbevf_clean_all_rx_rings(adapter); 2591 } 2592 2593 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter) 2594 { 2595 while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state)) 2596 msleep(1); 2597 2598 ixgbevf_down(adapter); 2599 pci_set_master(adapter->pdev); 2600 ixgbevf_up(adapter); 2601 2602 clear_bit(__IXGBEVF_RESETTING, &adapter->state); 2603 } 2604 2605 void ixgbevf_reset(struct ixgbevf_adapter *adapter) 2606 { 2607 struct ixgbe_hw *hw = &adapter->hw; 2608 struct net_device *netdev = adapter->netdev; 2609 2610 if (hw->mac.ops.reset_hw(hw)) { 2611 hw_dbg(hw, "PF still resetting\n"); 2612 } else { 2613 hw->mac.ops.init_hw(hw); 2614 ixgbevf_negotiate_api(adapter); 2615 } 2616 2617 if (is_valid_ether_addr(adapter->hw.mac.addr)) { 2618 eth_hw_addr_set(netdev, adapter->hw.mac.addr); 2619 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr); 2620 } 2621 2622 adapter->last_reset = jiffies; 2623 } 2624 2625 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter, 2626 int vectors) 2627 { 2628 int vector_threshold; 2629 2630 /* We'll want at least 2 (vector_threshold): 2631 * 1) TxQ[0] + RxQ[0] handler 2632 * 2) Other (Link Status Change, etc.) 2633 */ 2634 vector_threshold = MIN_MSIX_COUNT; 2635 2636 /* The more we get, the more we will assign to Tx/Rx Cleanup 2637 * for the separate queues...where Rx Cleanup >= Tx Cleanup. 2638 * Right now, we simply care about how many we'll get; we'll 2639 * set them up later while requesting irq's. 2640 */ 2641 vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries, 2642 vector_threshold, vectors); 2643 2644 if (vectors < 0) { 2645 dev_err(&adapter->pdev->dev, 2646 "Unable to allocate MSI-X interrupts\n"); 2647 kfree(adapter->msix_entries); 2648 adapter->msix_entries = NULL; 2649 return vectors; 2650 } 2651 2652 /* Adjust for only the vectors we'll use, which is minimum 2653 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of 2654 * vectors we were allocated. 2655 */ 2656 adapter->num_msix_vectors = vectors; 2657 2658 return 0; 2659 } 2660 2661 /** 2662 * ixgbevf_set_num_queues - Allocate queues for device, feature dependent 2663 * @adapter: board private structure to initialize 2664 * 2665 * This is the top level queue allocation routine. The order here is very 2666 * important, starting with the "most" number of features turned on at once, 2667 * and ending with the smallest set of features. This way large combinations 2668 * can be allocated if they're turned on, and smaller combinations are the 2669 * fall through conditions. 2670 * 2671 **/ 2672 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter) 2673 { 2674 struct ixgbe_hw *hw = &adapter->hw; 2675 unsigned int def_q = 0; 2676 unsigned int num_tcs = 0; 2677 int err; 2678 2679 /* Start with base case */ 2680 adapter->num_rx_queues = 1; 2681 adapter->num_tx_queues = 1; 2682 adapter->num_xdp_queues = 0; 2683 2684 spin_lock_bh(&adapter->mbx_lock); 2685 2686 /* fetch queue configuration from the PF */ 2687 err = ixgbevf_get_queues(hw, &num_tcs, &def_q); 2688 2689 spin_unlock_bh(&adapter->mbx_lock); 2690 2691 if (err) 2692 return; 2693 2694 /* we need as many queues as traffic classes */ 2695 if (num_tcs > 1) { 2696 adapter->num_rx_queues = num_tcs; 2697 } else { 2698 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES); 2699 2700 switch (hw->api_version) { 2701 case ixgbe_mbox_api_11: 2702 case ixgbe_mbox_api_12: 2703 case ixgbe_mbox_api_13: 2704 case ixgbe_mbox_api_14: 2705 case ixgbe_mbox_api_15: 2706 case ixgbe_mbox_api_16: 2707 case ixgbe_mbox_api_17: 2708 if (adapter->xdp_prog && 2709 hw->mac.max_tx_queues == rss) 2710 rss = rss > 3 ? 2 : 1; 2711 2712 adapter->num_rx_queues = rss; 2713 adapter->num_tx_queues = rss; 2714 adapter->num_xdp_queues = adapter->xdp_prog ? rss : 0; 2715 break; 2716 default: 2717 break; 2718 } 2719 } 2720 } 2721 2722 /** 2723 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported 2724 * @adapter: board private structure to initialize 2725 * 2726 * Attempt to configure the interrupts using the best available 2727 * capabilities of the hardware and the kernel. 2728 **/ 2729 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter) 2730 { 2731 int vector, v_budget; 2732 2733 /* It's easy to be greedy for MSI-X vectors, but it really 2734 * doesn't do us much good if we have a lot more vectors 2735 * than CPU's. So let's be conservative and only ask for 2736 * (roughly) the same number of vectors as there are CPU's. 2737 * The default is to use pairs of vectors. 2738 */ 2739 v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues); 2740 v_budget = min_t(int, v_budget, num_online_cpus()); 2741 v_budget += NON_Q_VECTORS; 2742 2743 adapter->msix_entries = kzalloc_objs(struct msix_entry, v_budget); 2744 if (!adapter->msix_entries) 2745 return -ENOMEM; 2746 2747 for (vector = 0; vector < v_budget; vector++) 2748 adapter->msix_entries[vector].entry = vector; 2749 2750 /* A failure in MSI-X entry allocation isn't fatal, but the VF driver 2751 * does not support any other modes, so we will simply fail here. Note 2752 * that we clean up the msix_entries pointer else-where. 2753 */ 2754 return ixgbevf_acquire_msix_vectors(adapter, v_budget); 2755 } 2756 2757 static void ixgbevf_add_ring(struct ixgbevf_ring *ring, 2758 struct ixgbevf_ring_container *head) 2759 { 2760 ring->next = head->ring; 2761 head->ring = ring; 2762 head->count++; 2763 } 2764 2765 /** 2766 * ixgbevf_alloc_q_vector - Allocate memory for a single interrupt vector 2767 * @adapter: board private structure to initialize 2768 * @v_idx: index of vector in adapter struct 2769 * @txr_count: number of Tx rings for q vector 2770 * @txr_idx: index of first Tx ring to assign 2771 * @xdp_count: total number of XDP rings to allocate 2772 * @xdp_idx: index of first XDP ring to allocate 2773 * @rxr_count: number of Rx rings for q vector 2774 * @rxr_idx: index of first Rx ring to assign 2775 * 2776 * We allocate one q_vector. If allocation fails we return -ENOMEM. 2777 **/ 2778 static int ixgbevf_alloc_q_vector(struct ixgbevf_adapter *adapter, int v_idx, 2779 int txr_count, int txr_idx, 2780 int xdp_count, int xdp_idx, 2781 int rxr_count, int rxr_idx) 2782 { 2783 struct ixgbevf_q_vector *q_vector; 2784 int reg_idx = txr_idx + xdp_idx; 2785 struct ixgbevf_ring *ring; 2786 int ring_count, size; 2787 2788 ring_count = txr_count + xdp_count + rxr_count; 2789 size = sizeof(*q_vector) + (sizeof(*ring) * ring_count); 2790 2791 /* allocate q_vector and rings */ 2792 q_vector = kzalloc(size, GFP_KERNEL); 2793 if (!q_vector) 2794 return -ENOMEM; 2795 2796 /* initialize NAPI */ 2797 netif_napi_add(adapter->netdev, &q_vector->napi, ixgbevf_poll); 2798 2799 /* tie q_vector and adapter together */ 2800 adapter->q_vector[v_idx] = q_vector; 2801 q_vector->adapter = adapter; 2802 q_vector->v_idx = v_idx; 2803 2804 /* initialize pointer to rings */ 2805 ring = q_vector->ring; 2806 2807 while (txr_count) { 2808 /* assign generic ring traits */ 2809 ring->dev = &adapter->pdev->dev; 2810 ring->netdev = adapter->netdev; 2811 2812 /* configure backlink on ring */ 2813 ring->q_vector = q_vector; 2814 2815 /* update q_vector Tx values */ 2816 ixgbevf_add_ring(ring, &q_vector->tx); 2817 2818 /* apply Tx specific ring traits */ 2819 ring->count = adapter->tx_ring_count; 2820 ring->queue_index = txr_idx; 2821 ring->reg_idx = reg_idx; 2822 2823 /* assign ring to adapter */ 2824 adapter->tx_ring[txr_idx] = ring; 2825 2826 /* update count and index */ 2827 txr_count--; 2828 txr_idx++; 2829 reg_idx++; 2830 2831 /* push pointer to next ring */ 2832 ring++; 2833 } 2834 2835 while (xdp_count) { 2836 /* assign generic ring traits */ 2837 ring->dev = &adapter->pdev->dev; 2838 ring->netdev = adapter->netdev; 2839 2840 /* configure backlink on ring */ 2841 ring->q_vector = q_vector; 2842 2843 /* update q_vector Tx values */ 2844 ixgbevf_add_ring(ring, &q_vector->tx); 2845 2846 /* apply Tx specific ring traits */ 2847 ring->count = adapter->tx_ring_count; 2848 ring->queue_index = xdp_idx; 2849 ring->reg_idx = reg_idx; 2850 set_ring_xdp(ring); 2851 2852 /* assign ring to adapter */ 2853 adapter->xdp_ring[xdp_idx] = ring; 2854 2855 /* update count and index */ 2856 xdp_count--; 2857 xdp_idx++; 2858 reg_idx++; 2859 2860 /* push pointer to next ring */ 2861 ring++; 2862 } 2863 2864 while (rxr_count) { 2865 /* assign generic ring traits */ 2866 ring->dev = &adapter->pdev->dev; 2867 ring->netdev = adapter->netdev; 2868 2869 /* configure backlink on ring */ 2870 ring->q_vector = q_vector; 2871 2872 /* update q_vector Rx values */ 2873 ixgbevf_add_ring(ring, &q_vector->rx); 2874 2875 /* apply Rx specific ring traits */ 2876 ring->count = adapter->rx_ring_count; 2877 ring->queue_index = rxr_idx; 2878 ring->reg_idx = rxr_idx; 2879 2880 /* assign ring to adapter */ 2881 adapter->rx_ring[rxr_idx] = ring; 2882 2883 /* update count and index */ 2884 rxr_count--; 2885 rxr_idx++; 2886 2887 /* push pointer to next ring */ 2888 ring++; 2889 } 2890 2891 return 0; 2892 } 2893 2894 /** 2895 * ixgbevf_free_q_vector - Free memory allocated for specific interrupt vector 2896 * @adapter: board private structure to initialize 2897 * @v_idx: index of vector in adapter struct 2898 * 2899 * This function frees the memory allocated to the q_vector. In addition if 2900 * NAPI is enabled it will delete any references to the NAPI struct prior 2901 * to freeing the q_vector. 2902 **/ 2903 static void ixgbevf_free_q_vector(struct ixgbevf_adapter *adapter, int v_idx) 2904 { 2905 struct ixgbevf_q_vector *q_vector = adapter->q_vector[v_idx]; 2906 struct ixgbevf_ring *ring; 2907 2908 ixgbevf_for_each_ring(ring, q_vector->tx) { 2909 if (ring_is_xdp(ring)) 2910 adapter->xdp_ring[ring->queue_index] = NULL; 2911 else 2912 adapter->tx_ring[ring->queue_index] = NULL; 2913 } 2914 2915 ixgbevf_for_each_ring(ring, q_vector->rx) 2916 adapter->rx_ring[ring->queue_index] = NULL; 2917 2918 adapter->q_vector[v_idx] = NULL; 2919 netif_napi_del(&q_vector->napi); 2920 2921 /* ixgbevf_get_stats() might access the rings on this vector, 2922 * we must wait a grace period before freeing it. 2923 */ 2924 kfree_rcu(q_vector, rcu); 2925 } 2926 2927 /** 2928 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors 2929 * @adapter: board private structure to initialize 2930 * 2931 * We allocate one q_vector per queue interrupt. If allocation fails we 2932 * return -ENOMEM. 2933 **/ 2934 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter) 2935 { 2936 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS; 2937 int rxr_remaining = adapter->num_rx_queues; 2938 int txr_remaining = adapter->num_tx_queues; 2939 int xdp_remaining = adapter->num_xdp_queues; 2940 int rxr_idx = 0, txr_idx = 0, xdp_idx = 0, v_idx = 0; 2941 int err; 2942 2943 if (q_vectors >= (rxr_remaining + txr_remaining + xdp_remaining)) { 2944 for (; rxr_remaining; v_idx++, q_vectors--) { 2945 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors); 2946 2947 err = ixgbevf_alloc_q_vector(adapter, v_idx, 2948 0, 0, 0, 0, rqpv, rxr_idx); 2949 if (err) 2950 goto err_out; 2951 2952 /* update counts and index */ 2953 rxr_remaining -= rqpv; 2954 rxr_idx += rqpv; 2955 } 2956 } 2957 2958 for (; q_vectors; v_idx++, q_vectors--) { 2959 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors); 2960 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors); 2961 int xqpv = DIV_ROUND_UP(xdp_remaining, q_vectors); 2962 2963 err = ixgbevf_alloc_q_vector(adapter, v_idx, 2964 tqpv, txr_idx, 2965 xqpv, xdp_idx, 2966 rqpv, rxr_idx); 2967 2968 if (err) 2969 goto err_out; 2970 2971 /* update counts and index */ 2972 rxr_remaining -= rqpv; 2973 rxr_idx += rqpv; 2974 txr_remaining -= tqpv; 2975 txr_idx += tqpv; 2976 xdp_remaining -= xqpv; 2977 xdp_idx += xqpv; 2978 } 2979 2980 return 0; 2981 2982 err_out: 2983 while (v_idx) { 2984 v_idx--; 2985 ixgbevf_free_q_vector(adapter, v_idx); 2986 } 2987 2988 return -ENOMEM; 2989 } 2990 2991 /** 2992 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors 2993 * @adapter: board private structure to initialize 2994 * 2995 * This function frees the memory allocated to the q_vectors. In addition if 2996 * NAPI is enabled it will delete any references to the NAPI struct prior 2997 * to freeing the q_vector. 2998 **/ 2999 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter) 3000 { 3001 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS; 3002 3003 while (q_vectors) { 3004 q_vectors--; 3005 ixgbevf_free_q_vector(adapter, q_vectors); 3006 } 3007 } 3008 3009 /** 3010 * ixgbevf_reset_interrupt_capability - Reset MSIX setup 3011 * @adapter: board private structure 3012 * 3013 **/ 3014 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter) 3015 { 3016 if (!adapter->msix_entries) 3017 return; 3018 3019 pci_disable_msix(adapter->pdev); 3020 kfree(adapter->msix_entries); 3021 adapter->msix_entries = NULL; 3022 } 3023 3024 /** 3025 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init 3026 * @adapter: board private structure to initialize 3027 * 3028 **/ 3029 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter) 3030 { 3031 int err; 3032 3033 /* Number of supported queues */ 3034 ixgbevf_set_num_queues(adapter); 3035 3036 err = ixgbevf_set_interrupt_capability(adapter); 3037 if (err) { 3038 hw_dbg(&adapter->hw, 3039 "Unable to setup interrupt capabilities\n"); 3040 goto err_set_interrupt; 3041 } 3042 3043 err = ixgbevf_alloc_q_vectors(adapter); 3044 if (err) { 3045 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n"); 3046 goto err_alloc_q_vectors; 3047 } 3048 3049 hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u XDP Queue count %u\n", 3050 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled", 3051 adapter->num_rx_queues, adapter->num_tx_queues, 3052 adapter->num_xdp_queues); 3053 3054 set_bit(__IXGBEVF_DOWN, &adapter->state); 3055 3056 return 0; 3057 err_alloc_q_vectors: 3058 ixgbevf_reset_interrupt_capability(adapter); 3059 err_set_interrupt: 3060 return err; 3061 } 3062 3063 /** 3064 * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings 3065 * @adapter: board private structure to clear interrupt scheme on 3066 * 3067 * We go through and clear interrupt specific resources and reset the structure 3068 * to pre-load conditions 3069 **/ 3070 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter) 3071 { 3072 adapter->num_tx_queues = 0; 3073 adapter->num_xdp_queues = 0; 3074 adapter->num_rx_queues = 0; 3075 3076 ixgbevf_free_q_vectors(adapter); 3077 ixgbevf_reset_interrupt_capability(adapter); 3078 } 3079 3080 /** 3081 * ixgbevf_sw_init - Initialize general software structures 3082 * @adapter: board private structure to initialize 3083 * 3084 * ixgbevf_sw_init initializes the Adapter private data structure. 3085 * Fields are initialized based on PCI device information and 3086 * OS network device settings (MTU size). 3087 **/ 3088 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter) 3089 { 3090 struct ixgbe_hw *hw = &adapter->hw; 3091 struct pci_dev *pdev = adapter->pdev; 3092 struct net_device *netdev = adapter->netdev; 3093 int err; 3094 3095 /* PCI config space info */ 3096 hw->vendor_id = pdev->vendor; 3097 hw->device_id = pdev->device; 3098 hw->revision_id = pdev->revision; 3099 hw->subsystem_vendor_id = pdev->subsystem_vendor; 3100 hw->subsystem_device_id = pdev->subsystem_device; 3101 3102 hw->mbx.ops.init_params(hw); 3103 3104 if (hw->mac.type >= ixgbe_mac_X550_vf) { 3105 err = ixgbevf_init_rss_key(adapter); 3106 if (err) 3107 goto out; 3108 } 3109 3110 /* assume legacy case in which PF would only give VF 2 queues */ 3111 hw->mac.max_tx_queues = 2; 3112 hw->mac.max_rx_queues = 2; 3113 3114 /* lock to protect mailbox accesses */ 3115 spin_lock_init(&adapter->mbx_lock); 3116 3117 err = hw->mac.ops.reset_hw(hw); 3118 if (err) { 3119 dev_info(&pdev->dev, 3120 "PF still in reset state. Is the PF interface up?\n"); 3121 } else { 3122 err = hw->mac.ops.init_hw(hw); 3123 if (err) { 3124 pr_err("init_shared_code failed: %d\n", err); 3125 goto out; 3126 } 3127 ixgbevf_negotiate_api(adapter); 3128 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr); 3129 if (err) 3130 dev_info(&pdev->dev, "Error reading MAC address\n"); 3131 else if (is_zero_ether_addr(adapter->hw.mac.addr)) 3132 dev_info(&pdev->dev, 3133 "MAC address not assigned by administrator.\n"); 3134 eth_hw_addr_set(netdev, hw->mac.addr); 3135 } 3136 3137 if (!is_valid_ether_addr(netdev->dev_addr)) { 3138 dev_info(&pdev->dev, "Assigning random MAC address\n"); 3139 eth_hw_addr_random(netdev); 3140 ether_addr_copy(hw->mac.addr, netdev->dev_addr); 3141 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr); 3142 } 3143 3144 /* Enable dynamic interrupt throttling rates */ 3145 adapter->rx_itr_setting = 1; 3146 adapter->tx_itr_setting = 1; 3147 3148 /* set default ring sizes */ 3149 adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD; 3150 adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD; 3151 3152 adapter->link_state = true; 3153 3154 set_bit(__IXGBEVF_DOWN, &adapter->state); 3155 return 0; 3156 3157 out: 3158 return err; 3159 } 3160 3161 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \ 3162 { \ 3163 u32 current_counter = IXGBE_READ_REG(hw, reg); \ 3164 if (current_counter < last_counter) \ 3165 counter += 0x100000000LL; \ 3166 last_counter = current_counter; \ 3167 counter &= 0xFFFFFFFF00000000LL; \ 3168 counter |= current_counter; \ 3169 } 3170 3171 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \ 3172 { \ 3173 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \ 3174 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \ 3175 u64 current_counter = (current_counter_msb << 32) | \ 3176 current_counter_lsb; \ 3177 if (current_counter < last_counter) \ 3178 counter += 0x1000000000LL; \ 3179 last_counter = current_counter; \ 3180 counter &= 0xFFFFFFF000000000LL; \ 3181 counter |= current_counter; \ 3182 } 3183 /** 3184 * ixgbevf_update_stats - Update the board statistics counters. 3185 * @adapter: board private structure 3186 **/ 3187 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter) 3188 { 3189 struct ixgbe_hw *hw = &adapter->hw; 3190 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0; 3191 u64 alloc_rx_page = 0, hw_csum_rx_error = 0; 3192 int i; 3193 3194 if (test_bit(__IXGBEVF_DOWN, &adapter->state) || 3195 test_bit(__IXGBEVF_RESETTING, &adapter->state)) 3196 return; 3197 3198 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc, 3199 adapter->stats.vfgprc); 3200 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc, 3201 adapter->stats.vfgptc); 3202 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB, 3203 adapter->stats.last_vfgorc, 3204 adapter->stats.vfgorc); 3205 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB, 3206 adapter->stats.last_vfgotc, 3207 adapter->stats.vfgotc); 3208 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc, 3209 adapter->stats.vfmprc); 3210 3211 for (i = 0; i < adapter->num_rx_queues; i++) { 3212 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i]; 3213 3214 hw_csum_rx_error += rx_ring->rx_stats.csum_err; 3215 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed; 3216 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed; 3217 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page; 3218 } 3219 3220 adapter->hw_csum_rx_error = hw_csum_rx_error; 3221 adapter->alloc_rx_page_failed = alloc_rx_page_failed; 3222 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed; 3223 adapter->alloc_rx_page = alloc_rx_page; 3224 } 3225 3226 /** 3227 * ixgbevf_service_timer - Timer Call-back 3228 * @t: pointer to timer_list struct 3229 **/ 3230 static void ixgbevf_service_timer(struct timer_list *t) 3231 { 3232 struct ixgbevf_adapter *adapter = timer_container_of(adapter, t, 3233 service_timer); 3234 3235 /* Reset the timer */ 3236 mod_timer(&adapter->service_timer, (HZ * 2) + jiffies); 3237 3238 ixgbevf_service_event_schedule(adapter); 3239 } 3240 3241 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter) 3242 { 3243 if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state)) 3244 return; 3245 3246 rtnl_lock(); 3247 /* If we're already down or resetting, just bail */ 3248 if (test_bit(__IXGBEVF_DOWN, &adapter->state) || 3249 test_bit(__IXGBEVF_REMOVING, &adapter->state) || 3250 test_bit(__IXGBEVF_RESETTING, &adapter->state)) { 3251 rtnl_unlock(); 3252 return; 3253 } 3254 3255 adapter->tx_timeout_count++; 3256 3257 ixgbevf_reinit_locked(adapter); 3258 rtnl_unlock(); 3259 } 3260 3261 /** 3262 * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts 3263 * @adapter: pointer to the device adapter structure 3264 * 3265 * This function serves two purposes. First it strobes the interrupt lines 3266 * in order to make certain interrupts are occurring. Secondly it sets the 3267 * bits needed to check for TX hangs. As a result we should immediately 3268 * determine if a hang has occurred. 3269 **/ 3270 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter) 3271 { 3272 struct ixgbe_hw *hw = &adapter->hw; 3273 u32 eics = 0; 3274 int i; 3275 3276 /* If we're down or resetting, just bail */ 3277 if (test_bit(__IXGBEVF_DOWN, &adapter->state) || 3278 test_bit(__IXGBEVF_RESETTING, &adapter->state)) 3279 return; 3280 3281 /* Force detection of hung controller */ 3282 if (netif_carrier_ok(adapter->netdev)) { 3283 for (i = 0; i < adapter->num_tx_queues; i++) 3284 set_check_for_tx_hang(adapter->tx_ring[i]); 3285 for (i = 0; i < adapter->num_xdp_queues; i++) 3286 set_check_for_tx_hang(adapter->xdp_ring[i]); 3287 } 3288 3289 /* get one bit for every active Tx/Rx interrupt vector */ 3290 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) { 3291 struct ixgbevf_q_vector *qv = adapter->q_vector[i]; 3292 3293 if (qv->rx.ring || qv->tx.ring) 3294 eics |= BIT(i); 3295 } 3296 3297 /* Cause software interrupt to ensure rings are cleaned */ 3298 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics); 3299 } 3300 3301 /** 3302 * ixgbevf_watchdog_update_link - update the link status 3303 * @adapter: pointer to the device adapter structure 3304 **/ 3305 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter) 3306 { 3307 struct ixgbe_hw *hw = &adapter->hw; 3308 u32 link_speed = adapter->link_speed; 3309 bool link_up = adapter->link_up; 3310 s32 err; 3311 3312 spin_lock_bh(&adapter->mbx_lock); 3313 3314 err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false); 3315 3316 spin_unlock_bh(&adapter->mbx_lock); 3317 3318 /* if check for link returns error we will need to reset */ 3319 if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) { 3320 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state); 3321 link_up = false; 3322 } 3323 3324 adapter->link_up = link_up; 3325 adapter->link_speed = link_speed; 3326 } 3327 3328 /** 3329 * ixgbevf_watchdog_link_is_up - update netif_carrier status and 3330 * print link up message 3331 * @adapter: pointer to the device adapter structure 3332 **/ 3333 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter) 3334 { 3335 struct net_device *netdev = adapter->netdev; 3336 3337 /* only continue if link was previously down */ 3338 if (netif_carrier_ok(netdev)) 3339 return; 3340 3341 dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n", 3342 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ? 3343 "10 Gbps" : 3344 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ? 3345 "1 Gbps" : 3346 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ? 3347 "100 Mbps" : 3348 "unknown speed"); 3349 3350 netif_carrier_on(netdev); 3351 } 3352 3353 /** 3354 * ixgbevf_watchdog_link_is_down - update netif_carrier status and 3355 * print link down message 3356 * @adapter: pointer to the adapter structure 3357 **/ 3358 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter) 3359 { 3360 struct net_device *netdev = adapter->netdev; 3361 3362 adapter->link_speed = 0; 3363 3364 /* only continue if link was up previously */ 3365 if (!netif_carrier_ok(netdev)) 3366 return; 3367 3368 dev_info(&adapter->pdev->dev, "NIC Link is Down\n"); 3369 3370 netif_carrier_off(netdev); 3371 } 3372 3373 /** 3374 * ixgbevf_watchdog_subtask - worker thread to bring link up 3375 * @adapter: board private structure 3376 **/ 3377 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter) 3378 { 3379 /* if interface is down do nothing */ 3380 if (test_bit(__IXGBEVF_DOWN, &adapter->state) || 3381 test_bit(__IXGBEVF_RESETTING, &adapter->state)) 3382 return; 3383 3384 ixgbevf_watchdog_update_link(adapter); 3385 3386 if (adapter->link_up && adapter->link_state) 3387 ixgbevf_watchdog_link_is_up(adapter); 3388 else 3389 ixgbevf_watchdog_link_is_down(adapter); 3390 3391 ixgbevf_update_stats(adapter); 3392 } 3393 3394 /** 3395 * ixgbevf_service_task - manages and runs subtasks 3396 * @work: pointer to work_struct containing our data 3397 **/ 3398 static void ixgbevf_service_task(struct work_struct *work) 3399 { 3400 struct ixgbevf_adapter *adapter = container_of(work, 3401 struct ixgbevf_adapter, 3402 service_task); 3403 struct ixgbe_hw *hw = &adapter->hw; 3404 3405 if (IXGBE_REMOVED(hw->hw_addr)) { 3406 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) { 3407 rtnl_lock(); 3408 ixgbevf_down(adapter); 3409 rtnl_unlock(); 3410 } 3411 return; 3412 } 3413 3414 ixgbevf_queue_reset_subtask(adapter); 3415 ixgbevf_reset_subtask(adapter); 3416 ixgbevf_watchdog_subtask(adapter); 3417 ixgbevf_check_hang_subtask(adapter); 3418 3419 ixgbevf_service_event_complete(adapter); 3420 } 3421 3422 /** 3423 * ixgbevf_free_tx_resources - Free Tx Resources per Queue 3424 * @tx_ring: Tx descriptor ring for a specific queue 3425 * 3426 * Free all transmit software resources 3427 **/ 3428 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring) 3429 { 3430 ixgbevf_clean_tx_ring(tx_ring); 3431 3432 vfree(tx_ring->tx_buffer_info); 3433 tx_ring->tx_buffer_info = NULL; 3434 3435 /* if not set, then don't free */ 3436 if (!tx_ring->desc) 3437 return; 3438 3439 dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc, 3440 tx_ring->dma); 3441 3442 tx_ring->desc = NULL; 3443 } 3444 3445 /** 3446 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues 3447 * @adapter: board private structure 3448 * 3449 * Free all transmit software resources 3450 **/ 3451 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter) 3452 { 3453 int i; 3454 3455 for (i = 0; i < adapter->num_tx_queues; i++) 3456 if (adapter->tx_ring[i]->desc) 3457 ixgbevf_free_tx_resources(adapter->tx_ring[i]); 3458 for (i = 0; i < adapter->num_xdp_queues; i++) 3459 if (adapter->xdp_ring[i]->desc) 3460 ixgbevf_free_tx_resources(adapter->xdp_ring[i]); 3461 } 3462 3463 /** 3464 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors) 3465 * @tx_ring: Tx descriptor ring (for a specific queue) to setup 3466 * 3467 * Return 0 on success, negative on failure 3468 **/ 3469 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring) 3470 { 3471 struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev); 3472 int size; 3473 3474 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count; 3475 tx_ring->tx_buffer_info = vmalloc(size); 3476 if (!tx_ring->tx_buffer_info) 3477 goto err; 3478 3479 u64_stats_init(&tx_ring->syncp); 3480 3481 /* round up to nearest 4K */ 3482 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc); 3483 tx_ring->size = ALIGN(tx_ring->size, 4096); 3484 3485 tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size, 3486 &tx_ring->dma, GFP_KERNEL); 3487 if (!tx_ring->desc) 3488 goto err; 3489 3490 return 0; 3491 3492 err: 3493 vfree(tx_ring->tx_buffer_info); 3494 tx_ring->tx_buffer_info = NULL; 3495 hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n"); 3496 return -ENOMEM; 3497 } 3498 3499 /** 3500 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources 3501 * @adapter: board private structure 3502 * 3503 * If this function returns with an error, then it's possible one or 3504 * more of the rings is populated (while the rest are not). It is the 3505 * callers duty to clean those orphaned rings. 3506 * 3507 * Return 0 on success, negative on failure 3508 **/ 3509 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter) 3510 { 3511 int i, j = 0, err = 0; 3512 3513 for (i = 0; i < adapter->num_tx_queues; i++) { 3514 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]); 3515 if (!err) 3516 continue; 3517 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i); 3518 goto err_setup_tx; 3519 } 3520 3521 for (j = 0; j < adapter->num_xdp_queues; j++) { 3522 err = ixgbevf_setup_tx_resources(adapter->xdp_ring[j]); 3523 if (!err) 3524 continue; 3525 hw_dbg(&adapter->hw, "Allocation for XDP Queue %u failed\n", j); 3526 goto err_setup_tx; 3527 } 3528 3529 return 0; 3530 err_setup_tx: 3531 /* rewind the index freeing the rings as we go */ 3532 while (j--) 3533 ixgbevf_free_tx_resources(adapter->xdp_ring[j]); 3534 while (i--) 3535 ixgbevf_free_tx_resources(adapter->tx_ring[i]); 3536 3537 return err; 3538 } 3539 3540 /** 3541 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors) 3542 * @adapter: board private structure 3543 * @rx_ring: Rx descriptor ring (for a specific queue) to setup 3544 * 3545 * Returns 0 on success, negative on failure 3546 **/ 3547 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter, 3548 struct ixgbevf_ring *rx_ring) 3549 { 3550 int size; 3551 3552 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count; 3553 rx_ring->rx_buffer_info = vmalloc(size); 3554 if (!rx_ring->rx_buffer_info) 3555 goto err; 3556 3557 u64_stats_init(&rx_ring->syncp); 3558 3559 /* Round up to nearest 4K */ 3560 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc); 3561 rx_ring->size = ALIGN(rx_ring->size, 4096); 3562 3563 rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size, 3564 &rx_ring->dma, GFP_KERNEL); 3565 3566 if (!rx_ring->desc) 3567 goto err; 3568 3569 /* XDP RX-queue info */ 3570 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev, 3571 rx_ring->queue_index, 0) < 0) 3572 goto err; 3573 3574 rx_ring->xdp_prog = adapter->xdp_prog; 3575 3576 return 0; 3577 err: 3578 vfree(rx_ring->rx_buffer_info); 3579 rx_ring->rx_buffer_info = NULL; 3580 dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n"); 3581 return -ENOMEM; 3582 } 3583 3584 /** 3585 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources 3586 * @adapter: board private structure 3587 * 3588 * If this function returns with an error, then it's possible one or 3589 * more of the rings is populated (while the rest are not). It is the 3590 * callers duty to clean those orphaned rings. 3591 * 3592 * Return 0 on success, negative on failure 3593 **/ 3594 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter) 3595 { 3596 int i, err = 0; 3597 3598 for (i = 0; i < adapter->num_rx_queues; i++) { 3599 err = ixgbevf_setup_rx_resources(adapter, adapter->rx_ring[i]); 3600 if (!err) 3601 continue; 3602 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i); 3603 goto err_setup_rx; 3604 } 3605 3606 return 0; 3607 err_setup_rx: 3608 /* rewind the index freeing the rings as we go */ 3609 while (i--) 3610 ixgbevf_free_rx_resources(adapter->rx_ring[i]); 3611 return err; 3612 } 3613 3614 /** 3615 * ixgbevf_free_rx_resources - Free Rx Resources 3616 * @rx_ring: ring to clean the resources from 3617 * 3618 * Free all receive software resources 3619 **/ 3620 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring) 3621 { 3622 ixgbevf_clean_rx_ring(rx_ring); 3623 3624 rx_ring->xdp_prog = NULL; 3625 xdp_rxq_info_unreg(&rx_ring->xdp_rxq); 3626 vfree(rx_ring->rx_buffer_info); 3627 rx_ring->rx_buffer_info = NULL; 3628 3629 dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc, 3630 rx_ring->dma); 3631 3632 rx_ring->desc = NULL; 3633 } 3634 3635 /** 3636 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues 3637 * @adapter: board private structure 3638 * 3639 * Free all receive software resources 3640 **/ 3641 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter) 3642 { 3643 int i; 3644 3645 for (i = 0; i < adapter->num_rx_queues; i++) 3646 if (adapter->rx_ring[i]->desc) 3647 ixgbevf_free_rx_resources(adapter->rx_ring[i]); 3648 } 3649 3650 /** 3651 * ixgbevf_open - Called when a network interface is made active 3652 * @netdev: network interface device structure 3653 * 3654 * Returns 0 on success, negative value on failure 3655 * 3656 * The open entry point is called when a network interface is made 3657 * active by the system (IFF_UP). At this point all resources needed 3658 * for transmit and receive operations are allocated, the interrupt 3659 * handler is registered with the OS, the watchdog timer is started, 3660 * and the stack is notified that the interface is ready. 3661 **/ 3662 int ixgbevf_open(struct net_device *netdev) 3663 { 3664 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 3665 struct ixgbe_hw *hw = &adapter->hw; 3666 int err; 3667 3668 /* A previous failure to open the device because of a lack of 3669 * available MSIX vector resources may have reset the number 3670 * of msix vectors variable to zero. The only way to recover 3671 * is to unload/reload the driver and hope that the system has 3672 * been able to recover some MSIX vector resources. 3673 */ 3674 if (!adapter->num_msix_vectors) 3675 return -ENOMEM; 3676 3677 if (hw->adapter_stopped) { 3678 ixgbevf_reset(adapter); 3679 /* if adapter is still stopped then PF isn't up and 3680 * the VF can't start. 3681 */ 3682 if (hw->adapter_stopped) { 3683 err = IXGBE_ERR_MBX; 3684 pr_err("Unable to start - perhaps the PF Driver isn't up yet\n"); 3685 goto err_setup_reset; 3686 } 3687 } 3688 3689 /* disallow open during test */ 3690 if (test_bit(__IXGBEVF_TESTING, &adapter->state)) 3691 return -EBUSY; 3692 3693 netif_carrier_off(netdev); 3694 3695 /* allocate transmit descriptors */ 3696 err = ixgbevf_setup_all_tx_resources(adapter); 3697 if (err) 3698 goto err_setup_tx; 3699 3700 /* allocate receive descriptors */ 3701 err = ixgbevf_setup_all_rx_resources(adapter); 3702 if (err) 3703 goto err_setup_rx; 3704 3705 ixgbevf_configure(adapter); 3706 3707 err = ixgbevf_request_irq(adapter); 3708 if (err) 3709 goto err_req_irq; 3710 3711 /* Notify the stack of the actual queue counts. */ 3712 err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues); 3713 if (err) 3714 goto err_set_queues; 3715 3716 err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues); 3717 if (err) 3718 goto err_set_queues; 3719 3720 ixgbevf_up_complete(adapter); 3721 3722 return 0; 3723 3724 err_set_queues: 3725 ixgbevf_free_irq(adapter); 3726 err_req_irq: 3727 ixgbevf_free_all_rx_resources(adapter); 3728 err_setup_rx: 3729 ixgbevf_free_all_tx_resources(adapter); 3730 err_setup_tx: 3731 ixgbevf_reset(adapter); 3732 err_setup_reset: 3733 3734 return err; 3735 } 3736 3737 /** 3738 * ixgbevf_close_suspend - actions necessary to both suspend and close flows 3739 * @adapter: the private adapter struct 3740 * 3741 * This function should contain the necessary work common to both suspending 3742 * and closing of the device. 3743 */ 3744 static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter) 3745 { 3746 ixgbevf_down(adapter); 3747 ixgbevf_free_irq(adapter); 3748 ixgbevf_free_all_tx_resources(adapter); 3749 ixgbevf_free_all_rx_resources(adapter); 3750 } 3751 3752 /** 3753 * ixgbevf_close - Disables a network interface 3754 * @netdev: network interface device structure 3755 * 3756 * Returns 0, this is not allowed to fail 3757 * 3758 * The close entry point is called when an interface is de-activated 3759 * by the OS. The hardware is still under the drivers control, but 3760 * needs to be disabled. A global MAC reset is issued to stop the 3761 * hardware, and all transmit and receive resources are freed. 3762 **/ 3763 int ixgbevf_close(struct net_device *netdev) 3764 { 3765 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 3766 3767 if (netif_device_present(netdev)) 3768 ixgbevf_close_suspend(adapter); 3769 3770 return 0; 3771 } 3772 3773 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter) 3774 { 3775 struct net_device *dev = adapter->netdev; 3776 3777 if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, 3778 &adapter->state)) 3779 return; 3780 3781 /* if interface is down do nothing */ 3782 if (test_bit(__IXGBEVF_DOWN, &adapter->state) || 3783 test_bit(__IXGBEVF_RESETTING, &adapter->state)) 3784 return; 3785 3786 /* Hardware has to reinitialize queues and interrupts to 3787 * match packet buffer alignment. Unfortunately, the 3788 * hardware is not flexible enough to do this dynamically. 3789 */ 3790 rtnl_lock(); 3791 3792 if (netif_running(dev)) 3793 ixgbevf_close(dev); 3794 3795 ixgbevf_clear_interrupt_scheme(adapter); 3796 ixgbevf_init_interrupt_scheme(adapter); 3797 3798 if (netif_running(dev)) 3799 ixgbevf_open(dev); 3800 3801 rtnl_unlock(); 3802 } 3803 3804 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring, 3805 u32 vlan_macip_lens, u32 fceof_saidx, 3806 u32 type_tucmd, u32 mss_l4len_idx) 3807 { 3808 struct ixgbe_adv_tx_context_desc *context_desc; 3809 u16 i = tx_ring->next_to_use; 3810 3811 context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i); 3812 3813 i++; 3814 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; 3815 3816 /* set bits to identify this as an advanced context descriptor */ 3817 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT; 3818 3819 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens); 3820 context_desc->fceof_saidx = cpu_to_le32(fceof_saidx); 3821 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd); 3822 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx); 3823 } 3824 3825 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring, 3826 struct ixgbevf_tx_buffer *first, 3827 u8 *hdr_len, 3828 struct ixgbevf_ipsec_tx_data *itd) 3829 { 3830 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx; 3831 struct sk_buff *skb = first->skb; 3832 union { 3833 struct iphdr *v4; 3834 struct ipv6hdr *v6; 3835 unsigned char *hdr; 3836 } ip; 3837 union { 3838 struct tcphdr *tcp; 3839 unsigned char *hdr; 3840 } l4; 3841 u32 paylen, l4_offset; 3842 u32 fceof_saidx = 0; 3843 int err; 3844 3845 if (skb->ip_summed != CHECKSUM_PARTIAL) 3846 return 0; 3847 3848 if (!skb_is_gso(skb)) 3849 return 0; 3850 3851 err = skb_cow_head(skb, 0); 3852 if (err < 0) 3853 return err; 3854 3855 if (eth_p_mpls(first->protocol)) 3856 ip.hdr = skb_inner_network_header(skb); 3857 else 3858 ip.hdr = skb_network_header(skb); 3859 l4.hdr = skb_checksum_start(skb); 3860 3861 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */ 3862 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP; 3863 3864 /* initialize outer IP header fields */ 3865 if (ip.v4->version == 4) { 3866 unsigned char *csum_start = skb_checksum_start(skb); 3867 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4); 3868 int len = csum_start - trans_start; 3869 3870 /* IP header will have to cancel out any data that 3871 * is not a part of the outer IP header, so set to 3872 * a reverse csum if needed, else init check to 0. 3873 */ 3874 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ? 3875 csum_fold(csum_partial(trans_start, 3876 len, 0)) : 0; 3877 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4; 3878 3879 ip.v4->tot_len = 0; 3880 first->tx_flags |= IXGBE_TX_FLAGS_TSO | 3881 IXGBE_TX_FLAGS_CSUM | 3882 IXGBE_TX_FLAGS_IPV4; 3883 } else { 3884 ip.v6->payload_len = 0; 3885 first->tx_flags |= IXGBE_TX_FLAGS_TSO | 3886 IXGBE_TX_FLAGS_CSUM; 3887 } 3888 3889 /* determine offset of inner transport header */ 3890 l4_offset = l4.hdr - skb->data; 3891 3892 /* compute length of segmentation header */ 3893 *hdr_len = (l4.tcp->doff * 4) + l4_offset; 3894 3895 /* remove payload length from inner checksum */ 3896 paylen = skb->len - l4_offset; 3897 csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen)); 3898 3899 /* update gso size and bytecount with header size */ 3900 first->gso_segs = skb_shinfo(skb)->gso_segs; 3901 first->bytecount += (first->gso_segs - 1) * *hdr_len; 3902 3903 /* mss_l4len_id: use 1 as index for TSO */ 3904 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT; 3905 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT; 3906 mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT); 3907 3908 fceof_saidx |= itd->pfsa; 3909 type_tucmd |= itd->flags | itd->trailer_len; 3910 3911 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */ 3912 vlan_macip_lens = l4.hdr - ip.hdr; 3913 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT; 3914 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK; 3915 3916 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 3917 mss_l4len_idx); 3918 3919 return 1; 3920 } 3921 3922 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring, 3923 struct ixgbevf_tx_buffer *first, 3924 struct ixgbevf_ipsec_tx_data *itd) 3925 { 3926 struct sk_buff *skb = first->skb; 3927 u32 vlan_macip_lens = 0; 3928 u32 fceof_saidx = 0; 3929 u32 type_tucmd = 0; 3930 3931 if (skb->ip_summed != CHECKSUM_PARTIAL) 3932 goto no_csum; 3933 3934 switch (skb->csum_offset) { 3935 case offsetof(struct tcphdr, check): 3936 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP; 3937 fallthrough; 3938 case offsetof(struct udphdr, check): 3939 break; 3940 case offsetof(struct sctphdr, checksum): 3941 /* validate that this is actually an SCTP request */ 3942 if (skb_csum_is_sctp(skb)) { 3943 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP; 3944 break; 3945 } 3946 fallthrough; 3947 default: 3948 skb_checksum_help(skb); 3949 goto no_csum; 3950 } 3951 3952 if (first->protocol == htons(ETH_P_IP)) 3953 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4; 3954 3955 /* update TX checksum flag */ 3956 first->tx_flags |= IXGBE_TX_FLAGS_CSUM; 3957 vlan_macip_lens = skb_checksum_start_offset(skb) - 3958 skb_network_offset(skb); 3959 no_csum: 3960 /* vlan_macip_lens: MACLEN, VLAN tag */ 3961 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT; 3962 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK; 3963 3964 fceof_saidx |= itd->pfsa; 3965 type_tucmd |= itd->flags | itd->trailer_len; 3966 3967 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, 3968 fceof_saidx, type_tucmd, 0); 3969 } 3970 3971 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags) 3972 { 3973 /* set type for advanced descriptor with frame checksum insertion */ 3974 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA | 3975 IXGBE_ADVTXD_DCMD_IFCS | 3976 IXGBE_ADVTXD_DCMD_DEXT); 3977 3978 /* set HW VLAN bit if VLAN is present */ 3979 if (tx_flags & IXGBE_TX_FLAGS_VLAN) 3980 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE); 3981 3982 /* set segmentation enable bits for TSO/FSO */ 3983 if (tx_flags & IXGBE_TX_FLAGS_TSO) 3984 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE); 3985 3986 return cmd_type; 3987 } 3988 3989 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc, 3990 u32 tx_flags, unsigned int paylen) 3991 { 3992 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT); 3993 3994 /* enable L4 checksum for TSO and TX checksum offload */ 3995 if (tx_flags & IXGBE_TX_FLAGS_CSUM) 3996 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM); 3997 3998 /* enble IPv4 checksum for TSO */ 3999 if (tx_flags & IXGBE_TX_FLAGS_IPV4) 4000 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM); 4001 4002 /* enable IPsec */ 4003 if (tx_flags & IXGBE_TX_FLAGS_IPSEC) 4004 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IPSEC); 4005 4006 /* use index 1 context for TSO/FSO/FCOE/IPSEC */ 4007 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_IPSEC)) 4008 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT); 4009 4010 /* Check Context must be set if Tx switch is enabled, which it 4011 * always is for case where virtual functions are running 4012 */ 4013 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC); 4014 4015 tx_desc->read.olinfo_status = olinfo_status; 4016 } 4017 4018 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring, 4019 struct ixgbevf_tx_buffer *first, 4020 const u8 hdr_len) 4021 { 4022 struct sk_buff *skb = first->skb; 4023 struct ixgbevf_tx_buffer *tx_buffer; 4024 union ixgbe_adv_tx_desc *tx_desc; 4025 skb_frag_t *frag; 4026 dma_addr_t dma; 4027 unsigned int data_len, size; 4028 u32 tx_flags = first->tx_flags; 4029 __le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags); 4030 u16 i = tx_ring->next_to_use; 4031 4032 tx_desc = IXGBEVF_TX_DESC(tx_ring, i); 4033 4034 ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len); 4035 4036 size = skb_headlen(skb); 4037 data_len = skb->data_len; 4038 4039 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE); 4040 4041 tx_buffer = first; 4042 4043 for (frag = &skb_shinfo(skb)->frags[0];; frag++) { 4044 if (dma_mapping_error(tx_ring->dev, dma)) 4045 goto dma_error; 4046 4047 /* record length, and DMA address */ 4048 dma_unmap_len_set(tx_buffer, len, size); 4049 dma_unmap_addr_set(tx_buffer, dma, dma); 4050 4051 tx_desc->read.buffer_addr = cpu_to_le64(dma); 4052 4053 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) { 4054 tx_desc->read.cmd_type_len = 4055 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD); 4056 4057 i++; 4058 tx_desc++; 4059 if (i == tx_ring->count) { 4060 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0); 4061 i = 0; 4062 } 4063 tx_desc->read.olinfo_status = 0; 4064 4065 dma += IXGBE_MAX_DATA_PER_TXD; 4066 size -= IXGBE_MAX_DATA_PER_TXD; 4067 4068 tx_desc->read.buffer_addr = cpu_to_le64(dma); 4069 } 4070 4071 if (likely(!data_len)) 4072 break; 4073 4074 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size); 4075 4076 i++; 4077 tx_desc++; 4078 if (i == tx_ring->count) { 4079 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0); 4080 i = 0; 4081 } 4082 tx_desc->read.olinfo_status = 0; 4083 4084 size = skb_frag_size(frag); 4085 data_len -= size; 4086 4087 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size, 4088 DMA_TO_DEVICE); 4089 4090 tx_buffer = &tx_ring->tx_buffer_info[i]; 4091 } 4092 4093 /* write last descriptor with RS and EOP bits */ 4094 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD); 4095 tx_desc->read.cmd_type_len = cmd_type; 4096 4097 /* set the timestamp */ 4098 first->time_stamp = jiffies; 4099 4100 skb_tx_timestamp(skb); 4101 4102 /* Force memory writes to complete before letting h/w know there 4103 * are new descriptors to fetch. (Only applicable for weak-ordered 4104 * memory model archs, such as IA-64). 4105 * 4106 * We also need this memory barrier (wmb) to make certain all of the 4107 * status bits have been updated before next_to_watch is written. 4108 */ 4109 wmb(); 4110 4111 /* set next_to_watch value indicating a packet is present */ 4112 first->next_to_watch = tx_desc; 4113 4114 i++; 4115 if (i == tx_ring->count) 4116 i = 0; 4117 4118 tx_ring->next_to_use = i; 4119 4120 /* notify HW of packet */ 4121 ixgbevf_write_tail(tx_ring, i); 4122 4123 return; 4124 dma_error: 4125 dev_err(tx_ring->dev, "TX DMA map failed\n"); 4126 tx_buffer = &tx_ring->tx_buffer_info[i]; 4127 4128 /* clear dma mappings for failed tx_buffer_info map */ 4129 while (tx_buffer != first) { 4130 if (dma_unmap_len(tx_buffer, len)) 4131 dma_unmap_page(tx_ring->dev, 4132 dma_unmap_addr(tx_buffer, dma), 4133 dma_unmap_len(tx_buffer, len), 4134 DMA_TO_DEVICE); 4135 dma_unmap_len_set(tx_buffer, len, 0); 4136 4137 if (i-- == 0) 4138 i += tx_ring->count; 4139 tx_buffer = &tx_ring->tx_buffer_info[i]; 4140 } 4141 4142 if (dma_unmap_len(tx_buffer, len)) 4143 dma_unmap_single(tx_ring->dev, 4144 dma_unmap_addr(tx_buffer, dma), 4145 dma_unmap_len(tx_buffer, len), 4146 DMA_TO_DEVICE); 4147 dma_unmap_len_set(tx_buffer, len, 0); 4148 4149 dev_kfree_skb_any(tx_buffer->skb); 4150 tx_buffer->skb = NULL; 4151 4152 tx_ring->next_to_use = i; 4153 } 4154 4155 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size) 4156 { 4157 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index); 4158 /* Herbert's original patch had: 4159 * smp_mb__after_netif_stop_queue(); 4160 * but since that doesn't exist yet, just open code it. 4161 */ 4162 smp_mb(); 4163 4164 /* We need to check again in a case another CPU has just 4165 * made room available. 4166 */ 4167 if (likely(ixgbevf_desc_unused(tx_ring) < size)) 4168 return -EBUSY; 4169 4170 /* A reprieve! - use start_queue because it doesn't call schedule */ 4171 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index); 4172 ++tx_ring->tx_stats.restart_queue; 4173 4174 return 0; 4175 } 4176 4177 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size) 4178 { 4179 if (likely(ixgbevf_desc_unused(tx_ring) >= size)) 4180 return 0; 4181 return __ixgbevf_maybe_stop_tx(tx_ring, size); 4182 } 4183 4184 static int ixgbevf_xmit_frame_ring(struct sk_buff *skb, 4185 struct ixgbevf_ring *tx_ring) 4186 { 4187 struct ixgbevf_tx_buffer *first; 4188 int tso; 4189 u32 tx_flags = 0; 4190 u16 count = TXD_USE_COUNT(skb_headlen(skb)); 4191 struct ixgbevf_ipsec_tx_data ipsec_tx = { 0 }; 4192 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD 4193 unsigned short f; 4194 #endif 4195 u8 hdr_len = 0; 4196 u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL); 4197 4198 if (!dst_mac || is_link_local_ether_addr(dst_mac)) { 4199 dev_kfree_skb_any(skb); 4200 return NETDEV_TX_OK; 4201 } 4202 4203 /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD, 4204 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD, 4205 * + 2 desc gap to keep tail from touching head, 4206 * + 1 desc for context descriptor, 4207 * otherwise try next time 4208 */ 4209 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD 4210 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) { 4211 skb_frag_t *frag = &skb_shinfo(skb)->frags[f]; 4212 4213 count += TXD_USE_COUNT(skb_frag_size(frag)); 4214 } 4215 #else 4216 count += skb_shinfo(skb)->nr_frags; 4217 #endif 4218 if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) { 4219 tx_ring->tx_stats.tx_busy++; 4220 return NETDEV_TX_BUSY; 4221 } 4222 4223 /* record the location of the first descriptor for this packet */ 4224 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use]; 4225 first->skb = skb; 4226 first->bytecount = skb->len; 4227 first->gso_segs = 1; 4228 4229 if (skb_vlan_tag_present(skb)) { 4230 tx_flags |= skb_vlan_tag_get(skb); 4231 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT; 4232 tx_flags |= IXGBE_TX_FLAGS_VLAN; 4233 } 4234 4235 /* record initial flags and protocol */ 4236 first->tx_flags = tx_flags; 4237 first->protocol = vlan_get_protocol(skb); 4238 4239 #ifdef CONFIG_IXGBEVF_IPSEC 4240 if (xfrm_offload(skb) && !ixgbevf_ipsec_tx(tx_ring, first, &ipsec_tx)) 4241 goto out_drop; 4242 #endif 4243 tso = ixgbevf_tso(tx_ring, first, &hdr_len, &ipsec_tx); 4244 if (tso < 0) 4245 goto out_drop; 4246 else if (!tso) 4247 ixgbevf_tx_csum(tx_ring, first, &ipsec_tx); 4248 4249 ixgbevf_tx_map(tx_ring, first, hdr_len); 4250 4251 ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED); 4252 4253 return NETDEV_TX_OK; 4254 4255 out_drop: 4256 dev_kfree_skb_any(first->skb); 4257 first->skb = NULL; 4258 4259 return NETDEV_TX_OK; 4260 } 4261 4262 static netdev_tx_t ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev) 4263 { 4264 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 4265 struct ixgbevf_ring *tx_ring; 4266 4267 if (skb->len <= 0) { 4268 dev_kfree_skb_any(skb); 4269 return NETDEV_TX_OK; 4270 } 4271 4272 /* The minimum packet size for olinfo paylen is 17 so pad the skb 4273 * in order to meet this minimum size requirement. 4274 */ 4275 if (skb->len < 17) { 4276 if (skb_padto(skb, 17)) 4277 return NETDEV_TX_OK; 4278 skb->len = 17; 4279 } 4280 4281 tx_ring = adapter->tx_ring[skb->queue_mapping]; 4282 return ixgbevf_xmit_frame_ring(skb, tx_ring); 4283 } 4284 4285 /** 4286 * ixgbevf_set_mac - Change the Ethernet Address of the NIC 4287 * @netdev: network interface device structure 4288 * @p: pointer to an address structure 4289 * 4290 * Returns 0 on success, negative on failure 4291 **/ 4292 static int ixgbevf_set_mac(struct net_device *netdev, void *p) 4293 { 4294 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 4295 struct ixgbe_hw *hw = &adapter->hw; 4296 struct sockaddr *addr = p; 4297 int err; 4298 4299 if (!is_valid_ether_addr(addr->sa_data)) 4300 return -EADDRNOTAVAIL; 4301 4302 spin_lock_bh(&adapter->mbx_lock); 4303 4304 err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0); 4305 4306 spin_unlock_bh(&adapter->mbx_lock); 4307 4308 if (err) 4309 return -EPERM; 4310 4311 ether_addr_copy(hw->mac.addr, addr->sa_data); 4312 ether_addr_copy(hw->mac.perm_addr, addr->sa_data); 4313 eth_hw_addr_set(netdev, addr->sa_data); 4314 4315 return 0; 4316 } 4317 4318 /** 4319 * ixgbevf_change_mtu - Change the Maximum Transfer Unit 4320 * @netdev: network interface device structure 4321 * @new_mtu: new value for maximum frame size 4322 * 4323 * Returns 0 on success, negative on failure 4324 **/ 4325 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu) 4326 { 4327 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 4328 struct ixgbe_hw *hw = &adapter->hw; 4329 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN; 4330 int ret; 4331 4332 /* prevent MTU being changed to a size unsupported by XDP */ 4333 if (adapter->xdp_prog) { 4334 dev_warn(&adapter->pdev->dev, "MTU cannot be changed while XDP program is loaded\n"); 4335 return -EPERM; 4336 } 4337 4338 spin_lock_bh(&adapter->mbx_lock); 4339 /* notify the PF of our intent to use this size of frame */ 4340 ret = hw->mac.ops.set_rlpml(hw, max_frame); 4341 spin_unlock_bh(&adapter->mbx_lock); 4342 if (ret) 4343 return -EINVAL; 4344 4345 hw_dbg(hw, "changing MTU from %d to %d\n", 4346 netdev->mtu, new_mtu); 4347 4348 /* must set new MTU before calling down or up */ 4349 WRITE_ONCE(netdev->mtu, new_mtu); 4350 4351 if (netif_running(netdev)) 4352 ixgbevf_reinit_locked(adapter); 4353 4354 return 0; 4355 } 4356 4357 static int ixgbevf_suspend(struct device *dev_d) 4358 { 4359 struct net_device *netdev = dev_get_drvdata(dev_d); 4360 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 4361 4362 rtnl_lock(); 4363 netif_device_detach(netdev); 4364 4365 if (netif_running(netdev)) 4366 ixgbevf_close_suspend(adapter); 4367 4368 ixgbevf_clear_interrupt_scheme(adapter); 4369 rtnl_unlock(); 4370 4371 return 0; 4372 } 4373 4374 static int ixgbevf_resume(struct device *dev_d) 4375 { 4376 struct pci_dev *pdev = to_pci_dev(dev_d); 4377 struct net_device *netdev = pci_get_drvdata(pdev); 4378 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 4379 int err; 4380 4381 adapter->hw.hw_addr = adapter->io_addr; 4382 smp_mb__before_atomic(); 4383 clear_bit(__IXGBEVF_DISABLED, &adapter->state); 4384 pci_set_master(pdev); 4385 4386 ixgbevf_reset(adapter); 4387 4388 rtnl_lock(); 4389 err = ixgbevf_init_interrupt_scheme(adapter); 4390 if (!err && netif_running(netdev)) 4391 err = ixgbevf_open(netdev); 4392 rtnl_unlock(); 4393 if (err) 4394 return err; 4395 4396 netif_device_attach(netdev); 4397 4398 return err; 4399 } 4400 4401 static void ixgbevf_shutdown(struct pci_dev *pdev) 4402 { 4403 ixgbevf_suspend(&pdev->dev); 4404 } 4405 4406 static void ixgbevf_get_tx_ring_stats(struct rtnl_link_stats64 *stats, 4407 const struct ixgbevf_ring *ring) 4408 { 4409 u64 bytes, packets; 4410 unsigned int start; 4411 4412 if (ring) { 4413 do { 4414 start = u64_stats_fetch_begin(&ring->syncp); 4415 bytes = ring->stats.bytes; 4416 packets = ring->stats.packets; 4417 } while (u64_stats_fetch_retry(&ring->syncp, start)); 4418 stats->tx_bytes += bytes; 4419 stats->tx_packets += packets; 4420 } 4421 } 4422 4423 static void ixgbevf_get_stats(struct net_device *netdev, 4424 struct rtnl_link_stats64 *stats) 4425 { 4426 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 4427 unsigned int start; 4428 u64 bytes, packets; 4429 const struct ixgbevf_ring *ring; 4430 int i; 4431 4432 ixgbevf_update_stats(adapter); 4433 4434 stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc; 4435 4436 rcu_read_lock(); 4437 for (i = 0; i < adapter->num_rx_queues; i++) { 4438 ring = adapter->rx_ring[i]; 4439 do { 4440 start = u64_stats_fetch_begin(&ring->syncp); 4441 bytes = ring->stats.bytes; 4442 packets = ring->stats.packets; 4443 } while (u64_stats_fetch_retry(&ring->syncp, start)); 4444 stats->rx_bytes += bytes; 4445 stats->rx_packets += packets; 4446 } 4447 4448 for (i = 0; i < adapter->num_tx_queues; i++) { 4449 ring = adapter->tx_ring[i]; 4450 ixgbevf_get_tx_ring_stats(stats, ring); 4451 } 4452 4453 for (i = 0; i < adapter->num_xdp_queues; i++) { 4454 ring = adapter->xdp_ring[i]; 4455 ixgbevf_get_tx_ring_stats(stats, ring); 4456 } 4457 rcu_read_unlock(); 4458 } 4459 4460 #define IXGBEVF_MAX_MAC_HDR_LEN 127 4461 #define IXGBEVF_MAX_NETWORK_HDR_LEN 511 4462 4463 static netdev_features_t 4464 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev, 4465 netdev_features_t features) 4466 { 4467 unsigned int network_hdr_len, mac_hdr_len; 4468 4469 /* Make certain the headers can be described by a context descriptor */ 4470 mac_hdr_len = skb_network_offset(skb); 4471 if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN)) 4472 return features & ~(NETIF_F_HW_CSUM | 4473 NETIF_F_SCTP_CRC | 4474 NETIF_F_HW_VLAN_CTAG_TX | 4475 NETIF_F_TSO | 4476 NETIF_F_TSO6); 4477 4478 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb); 4479 if (unlikely(network_hdr_len > IXGBEVF_MAX_NETWORK_HDR_LEN)) 4480 return features & ~(NETIF_F_HW_CSUM | 4481 NETIF_F_SCTP_CRC | 4482 NETIF_F_TSO | 4483 NETIF_F_TSO6); 4484 4485 /* We can only support IPV4 TSO in tunnels if we can mangle the 4486 * inner IP ID field, so strip TSO if MANGLEID is not supported. 4487 */ 4488 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) 4489 features &= ~NETIF_F_TSO; 4490 4491 return features; 4492 } 4493 4494 static int ixgbevf_xdp_setup(struct net_device *dev, struct bpf_prog *prog) 4495 { 4496 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN; 4497 struct ixgbevf_adapter *adapter = netdev_priv(dev); 4498 struct bpf_prog *old_prog; 4499 4500 /* verify ixgbevf ring attributes are sufficient for XDP */ 4501 for (i = 0; i < adapter->num_rx_queues; i++) { 4502 struct ixgbevf_ring *ring = adapter->rx_ring[i]; 4503 4504 if (frame_size > ixgbevf_rx_bufsz(ring)) 4505 return -EINVAL; 4506 } 4507 4508 old_prog = xchg(&adapter->xdp_prog, prog); 4509 4510 /* If transitioning XDP modes reconfigure rings */ 4511 if (!!prog != !!old_prog) { 4512 /* Hardware has to reinitialize queues and interrupts to 4513 * match packet buffer alignment. Unfortunately, the 4514 * hardware is not flexible enough to do this dynamically. 4515 */ 4516 if (netif_running(dev)) 4517 ixgbevf_close(dev); 4518 4519 ixgbevf_clear_interrupt_scheme(adapter); 4520 ixgbevf_init_interrupt_scheme(adapter); 4521 4522 if (netif_running(dev)) 4523 ixgbevf_open(dev); 4524 } else { 4525 for (i = 0; i < adapter->num_rx_queues; i++) 4526 xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog); 4527 } 4528 4529 if (old_prog) 4530 bpf_prog_put(old_prog); 4531 4532 return 0; 4533 } 4534 4535 static int ixgbevf_xdp(struct net_device *dev, struct netdev_bpf *xdp) 4536 { 4537 switch (xdp->command) { 4538 case XDP_SETUP_PROG: 4539 return ixgbevf_xdp_setup(dev, xdp->prog); 4540 default: 4541 return -EINVAL; 4542 } 4543 } 4544 4545 static const struct net_device_ops ixgbevf_netdev_ops = { 4546 .ndo_open = ixgbevf_open, 4547 .ndo_stop = ixgbevf_close, 4548 .ndo_start_xmit = ixgbevf_xmit_frame, 4549 .ndo_set_rx_mode = ixgbevf_set_rx_mode, 4550 .ndo_get_stats64 = ixgbevf_get_stats, 4551 .ndo_validate_addr = eth_validate_addr, 4552 .ndo_set_mac_address = ixgbevf_set_mac, 4553 .ndo_change_mtu = ixgbevf_change_mtu, 4554 .ndo_tx_timeout = ixgbevf_tx_timeout, 4555 .ndo_vlan_rx_add_vid = ixgbevf_vlan_rx_add_vid, 4556 .ndo_vlan_rx_kill_vid = ixgbevf_vlan_rx_kill_vid, 4557 .ndo_features_check = ixgbevf_features_check, 4558 .ndo_bpf = ixgbevf_xdp, 4559 }; 4560 4561 static void ixgbevf_assign_netdev_ops(struct net_device *dev) 4562 { 4563 dev->netdev_ops = &ixgbevf_netdev_ops; 4564 ixgbevf_set_ethtool_ops(dev); 4565 dev->watchdog_timeo = 5 * HZ; 4566 } 4567 4568 /** 4569 * ixgbevf_probe - Device Initialization Routine 4570 * @pdev: PCI device information struct 4571 * @ent: entry in ixgbevf_pci_tbl 4572 * 4573 * Returns 0 on success, negative on failure 4574 * 4575 * ixgbevf_probe initializes an adapter identified by a pci_dev structure. 4576 * The OS initialization, configuring of the adapter private structure, 4577 * and a hardware reset occur. 4578 **/ 4579 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 4580 { 4581 struct net_device *netdev; 4582 struct ixgbevf_adapter *adapter = NULL; 4583 struct ixgbe_hw *hw = NULL; 4584 const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data]; 4585 bool disable_dev = false; 4586 int err; 4587 4588 err = pci_enable_device(pdev); 4589 if (err) 4590 return err; 4591 4592 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); 4593 if (err) { 4594 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n"); 4595 goto err_dma; 4596 } 4597 4598 err = pci_request_regions(pdev, ixgbevf_driver_name); 4599 if (err) { 4600 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err); 4601 goto err_pci_reg; 4602 } 4603 4604 pci_set_master(pdev); 4605 4606 netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter), 4607 MAX_TX_QUEUES); 4608 if (!netdev) { 4609 err = -ENOMEM; 4610 goto err_alloc_etherdev; 4611 } 4612 4613 SET_NETDEV_DEV(netdev, &pdev->dev); 4614 4615 adapter = netdev_priv(netdev); 4616 4617 adapter->netdev = netdev; 4618 adapter->pdev = pdev; 4619 hw = &adapter->hw; 4620 hw->back = adapter; 4621 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE); 4622 4623 /* call save state here in standalone driver because it relies on 4624 * adapter struct to exist, and needs to call netdev_priv 4625 */ 4626 pci_save_state(pdev); 4627 4628 hw->hw_addr = ioremap(pci_resource_start(pdev, 0), 4629 pci_resource_len(pdev, 0)); 4630 adapter->io_addr = hw->hw_addr; 4631 if (!hw->hw_addr) { 4632 err = -EIO; 4633 goto err_ioremap; 4634 } 4635 4636 ixgbevf_assign_netdev_ops(netdev); 4637 4638 /* Setup HW API */ 4639 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops)); 4640 hw->mac.type = ii->mac; 4641 4642 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops_legacy, 4643 sizeof(struct ixgbe_mbx_operations)); 4644 4645 /* setup the private structure */ 4646 err = ixgbevf_sw_init(adapter); 4647 if (err) 4648 goto err_sw_init; 4649 4650 /* The HW MAC address was set and/or determined in sw_init */ 4651 if (!is_valid_ether_addr(netdev->dev_addr)) { 4652 pr_err("invalid MAC address\n"); 4653 err = -EIO; 4654 goto err_sw_init; 4655 } 4656 4657 netdev->hw_features = NETIF_F_SG | 4658 NETIF_F_TSO | 4659 NETIF_F_TSO6 | 4660 NETIF_F_RXCSUM | 4661 NETIF_F_HW_CSUM | 4662 NETIF_F_SCTP_CRC; 4663 4664 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \ 4665 NETIF_F_GSO_GRE_CSUM | \ 4666 NETIF_F_GSO_IPXIP4 | \ 4667 NETIF_F_GSO_IPXIP6 | \ 4668 NETIF_F_GSO_UDP_TUNNEL | \ 4669 NETIF_F_GSO_UDP_TUNNEL_CSUM) 4670 4671 netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES; 4672 netdev->hw_features |= NETIF_F_GSO_PARTIAL | 4673 IXGBEVF_GSO_PARTIAL_FEATURES; 4674 4675 netdev->features = netdev->hw_features | NETIF_F_HIGHDMA; 4676 4677 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID; 4678 netdev->mpls_features |= NETIF_F_SG | 4679 NETIF_F_TSO | 4680 NETIF_F_TSO6 | 4681 NETIF_F_HW_CSUM; 4682 netdev->mpls_features |= IXGBEVF_GSO_PARTIAL_FEATURES; 4683 netdev->hw_enc_features |= netdev->vlan_features; 4684 4685 /* set this bit last since it cannot be part of vlan_features */ 4686 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER | 4687 NETIF_F_HW_VLAN_CTAG_RX | 4688 NETIF_F_HW_VLAN_CTAG_TX; 4689 4690 netdev->priv_flags |= IFF_UNICAST_FLT; 4691 netdev->xdp_features = NETDEV_XDP_ACT_BASIC; 4692 4693 /* MTU range: 68 - 1504 or 9710 */ 4694 netdev->min_mtu = ETH_MIN_MTU; 4695 switch (adapter->hw.api_version) { 4696 case ixgbe_mbox_api_11: 4697 case ixgbe_mbox_api_12: 4698 case ixgbe_mbox_api_13: 4699 case ixgbe_mbox_api_14: 4700 case ixgbe_mbox_api_15: 4701 case ixgbe_mbox_api_16: 4702 case ixgbe_mbox_api_17: 4703 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - 4704 (ETH_HLEN + ETH_FCS_LEN); 4705 break; 4706 default: 4707 if (adapter->hw.mac.type != ixgbe_mac_82599_vf) 4708 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - 4709 (ETH_HLEN + ETH_FCS_LEN); 4710 else 4711 netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN; 4712 break; 4713 } 4714 4715 if (IXGBE_REMOVED(hw->hw_addr)) { 4716 err = -EIO; 4717 goto err_sw_init; 4718 } 4719 4720 timer_setup(&adapter->service_timer, ixgbevf_service_timer, 0); 4721 4722 INIT_WORK(&adapter->service_task, ixgbevf_service_task); 4723 set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state); 4724 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state); 4725 4726 err = ixgbevf_init_interrupt_scheme(adapter); 4727 if (err) 4728 goto err_sw_init; 4729 4730 strcpy(netdev->name, "eth%d"); 4731 4732 err = register_netdev(netdev); 4733 if (err) 4734 goto err_register; 4735 4736 pci_set_drvdata(pdev, netdev); 4737 netif_carrier_off(netdev); 4738 ixgbevf_init_ipsec_offload(adapter); 4739 4740 ixgbevf_init_last_counter_stats(adapter); 4741 4742 /* print the VF info */ 4743 dev_info(&pdev->dev, "%pM\n", netdev->dev_addr); 4744 dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type); 4745 4746 switch (hw->mac.type) { 4747 case ixgbe_mac_X550_vf: 4748 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n"); 4749 break; 4750 case ixgbe_mac_X540_vf: 4751 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n"); 4752 break; 4753 case ixgbe_mac_e610_vf: 4754 dev_info(&pdev->dev, "Intel(R) E610 Virtual Function\n"); 4755 break; 4756 case ixgbe_mac_82599_vf: 4757 default: 4758 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n"); 4759 break; 4760 } 4761 4762 return 0; 4763 4764 err_register: 4765 ixgbevf_clear_interrupt_scheme(adapter); 4766 err_sw_init: 4767 ixgbevf_reset_interrupt_capability(adapter); 4768 iounmap(adapter->io_addr); 4769 kfree(adapter->rss_key); 4770 err_ioremap: 4771 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state); 4772 free_netdev(netdev); 4773 err_alloc_etherdev: 4774 pci_release_regions(pdev); 4775 err_pci_reg: 4776 err_dma: 4777 if (!adapter || disable_dev) 4778 pci_disable_device(pdev); 4779 return err; 4780 } 4781 4782 /** 4783 * ixgbevf_remove - Device Removal Routine 4784 * @pdev: PCI device information struct 4785 * 4786 * ixgbevf_remove is called by the PCI subsystem to alert the driver 4787 * that it should release a PCI device. The could be caused by a 4788 * Hot-Plug event, or because the driver is going to be removed from 4789 * memory. 4790 **/ 4791 static void ixgbevf_remove(struct pci_dev *pdev) 4792 { 4793 struct net_device *netdev = pci_get_drvdata(pdev); 4794 struct ixgbevf_adapter *adapter; 4795 bool disable_dev; 4796 4797 if (!netdev) 4798 return; 4799 4800 adapter = netdev_priv(netdev); 4801 4802 set_bit(__IXGBEVF_REMOVING, &adapter->state); 4803 cancel_work_sync(&adapter->service_task); 4804 4805 if (netdev->reg_state == NETREG_REGISTERED) 4806 unregister_netdev(netdev); 4807 4808 ixgbevf_stop_ipsec_offload(adapter); 4809 ixgbevf_clear_interrupt_scheme(adapter); 4810 ixgbevf_reset_interrupt_capability(adapter); 4811 4812 iounmap(adapter->io_addr); 4813 pci_release_regions(pdev); 4814 4815 hw_dbg(&adapter->hw, "Remove complete\n"); 4816 4817 kfree(adapter->rss_key); 4818 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state); 4819 free_netdev(netdev); 4820 4821 if (disable_dev) 4822 pci_disable_device(pdev); 4823 } 4824 4825 /** 4826 * ixgbevf_io_error_detected - called when PCI error is detected 4827 * @pdev: Pointer to PCI device 4828 * @state: The current pci connection state 4829 * 4830 * This function is called after a PCI bus error affecting 4831 * this device has been detected. 4832 **/ 4833 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev, 4834 pci_channel_state_t state) 4835 { 4836 struct net_device *netdev = pci_get_drvdata(pdev); 4837 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 4838 4839 if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state)) 4840 return PCI_ERS_RESULT_DISCONNECT; 4841 4842 rtnl_lock(); 4843 netif_device_detach(netdev); 4844 4845 if (netif_running(netdev)) 4846 ixgbevf_close_suspend(adapter); 4847 4848 if (state == pci_channel_io_perm_failure) { 4849 rtnl_unlock(); 4850 return PCI_ERS_RESULT_DISCONNECT; 4851 } 4852 4853 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state)) 4854 pci_disable_device(pdev); 4855 rtnl_unlock(); 4856 4857 /* Request a slot reset. */ 4858 return PCI_ERS_RESULT_NEED_RESET; 4859 } 4860 4861 /** 4862 * ixgbevf_io_slot_reset - called after the pci bus has been reset. 4863 * @pdev: Pointer to PCI device 4864 * 4865 * Restart the card from scratch, as if from a cold-boot. Implementation 4866 * resembles the first-half of the ixgbevf_resume routine. 4867 **/ 4868 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev) 4869 { 4870 struct net_device *netdev = pci_get_drvdata(pdev); 4871 struct ixgbevf_adapter *adapter = netdev_priv(netdev); 4872 4873 if (pci_enable_device_mem(pdev)) { 4874 dev_err(&pdev->dev, 4875 "Cannot re-enable PCI device after reset.\n"); 4876 return PCI_ERS_RESULT_DISCONNECT; 4877 } 4878 4879 adapter->hw.hw_addr = adapter->io_addr; 4880 smp_mb__before_atomic(); 4881 clear_bit(__IXGBEVF_DISABLED, &adapter->state); 4882 pci_set_master(pdev); 4883 4884 ixgbevf_reset(adapter); 4885 4886 return PCI_ERS_RESULT_RECOVERED; 4887 } 4888 4889 /** 4890 * ixgbevf_io_resume - called when traffic can start flowing again. 4891 * @pdev: Pointer to PCI device 4892 * 4893 * This callback is called when the error recovery driver tells us that 4894 * its OK to resume normal operation. Implementation resembles the 4895 * second-half of the ixgbevf_resume routine. 4896 **/ 4897 static void ixgbevf_io_resume(struct pci_dev *pdev) 4898 { 4899 struct net_device *netdev = pci_get_drvdata(pdev); 4900 4901 rtnl_lock(); 4902 if (netif_running(netdev)) 4903 ixgbevf_open(netdev); 4904 4905 netif_device_attach(netdev); 4906 rtnl_unlock(); 4907 } 4908 4909 /* PCI Error Recovery (ERS) */ 4910 static const struct pci_error_handlers ixgbevf_err_handler = { 4911 .error_detected = ixgbevf_io_error_detected, 4912 .slot_reset = ixgbevf_io_slot_reset, 4913 .resume = ixgbevf_io_resume, 4914 }; 4915 4916 static DEFINE_SIMPLE_DEV_PM_OPS(ixgbevf_pm_ops, ixgbevf_suspend, ixgbevf_resume); 4917 4918 static struct pci_driver ixgbevf_driver = { 4919 .name = ixgbevf_driver_name, 4920 .id_table = ixgbevf_pci_tbl, 4921 .probe = ixgbevf_probe, 4922 .remove = ixgbevf_remove, 4923 4924 /* Power Management Hooks */ 4925 .driver.pm = pm_sleep_ptr(&ixgbevf_pm_ops), 4926 4927 .shutdown = ixgbevf_shutdown, 4928 .err_handler = &ixgbevf_err_handler 4929 }; 4930 4931 /** 4932 * ixgbevf_init_module - Driver Registration Routine 4933 * 4934 * ixgbevf_init_module is the first routine called when the driver is 4935 * loaded. All it does is register with the PCI subsystem. 4936 **/ 4937 static int __init ixgbevf_init_module(void) 4938 { 4939 int err; 4940 4941 pr_info("%s\n", ixgbevf_driver_string); 4942 pr_info("%s\n", ixgbevf_copyright); 4943 ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name); 4944 if (!ixgbevf_wq) { 4945 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name); 4946 return -ENOMEM; 4947 } 4948 4949 err = pci_register_driver(&ixgbevf_driver); 4950 if (err) { 4951 destroy_workqueue(ixgbevf_wq); 4952 return err; 4953 } 4954 4955 return 0; 4956 } 4957 4958 module_init(ixgbevf_init_module); 4959 4960 /** 4961 * ixgbevf_exit_module - Driver Exit Cleanup Routine 4962 * 4963 * ixgbevf_exit_module is called just before the driver is removed 4964 * from memory. 4965 **/ 4966 static void __exit ixgbevf_exit_module(void) 4967 { 4968 pci_unregister_driver(&ixgbevf_driver); 4969 if (ixgbevf_wq) { 4970 destroy_workqueue(ixgbevf_wq); 4971 ixgbevf_wq = NULL; 4972 } 4973 } 4974 4975 #ifdef DEBUG 4976 /** 4977 * ixgbevf_get_hw_dev_name - return device name string 4978 * used by hardware layer to print debugging information 4979 * @hw: pointer to private hardware struct 4980 **/ 4981 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw) 4982 { 4983 struct ixgbevf_adapter *adapter = hw->back; 4984 4985 return adapter->netdev->name; 4986 } 4987 4988 #endif 4989 module_exit(ixgbevf_exit_module); 4990 4991 /* ixgbevf_main.c */ 4992