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