1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright(c) 2007 - 2018 Intel Corporation. */ 3 4 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 5 6 #include <linux/module.h> 7 #include <linux/types.h> 8 #include <linux/init.h> 9 #include <linux/bitops.h> 10 #include <linux/vmalloc.h> 11 #include <linux/pagemap.h> 12 #include <linux/netdevice.h> 13 #include <linux/ipv6.h> 14 #include <linux/slab.h> 15 #include <net/checksum.h> 16 #include <net/ip6_checksum.h> 17 #include <net/pkt_sched.h> 18 #include <net/pkt_cls.h> 19 #include <linux/net_tstamp.h> 20 #include <linux/mii.h> 21 #include <linux/ethtool.h> 22 #include <linux/if.h> 23 #include <linux/if_vlan.h> 24 #include <linux/pci.h> 25 #include <linux/delay.h> 26 #include <linux/interrupt.h> 27 #include <linux/ip.h> 28 #include <linux/tcp.h> 29 #include <linux/sctp.h> 30 #include <linux/if_ether.h> 31 #include <linux/prefetch.h> 32 #include <linux/bpf.h> 33 #include <linux/bpf_trace.h> 34 #include <linux/pm_runtime.h> 35 #include <linux/etherdevice.h> 36 #ifdef CONFIG_IGB_DCA 37 #include <linux/dca.h> 38 #endif 39 #include <linux/i2c.h> 40 #include "igb.h" 41 42 enum queue_mode { 43 QUEUE_MODE_STRICT_PRIORITY, 44 QUEUE_MODE_STREAM_RESERVATION, 45 }; 46 47 enum tx_queue_prio { 48 TX_QUEUE_PRIO_HIGH, 49 TX_QUEUE_PRIO_LOW, 50 }; 51 52 char igb_driver_name[] = "igb"; 53 static const char igb_driver_string[] = 54 "Intel(R) Gigabit Ethernet Network Driver"; 55 static const char igb_copyright[] = 56 "Copyright (c) 2007-2014 Intel Corporation."; 57 58 static const struct e1000_info *igb_info_tbl[] = { 59 [board_82575] = &e1000_82575_info, 60 }; 61 62 static const struct pci_device_id igb_pci_tbl[] = { 63 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) }, 64 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) }, 65 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) }, 66 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 }, 67 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 }, 68 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 }, 69 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 }, 70 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 }, 71 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 }, 72 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 }, 73 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 }, 74 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 }, 75 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 }, 76 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 }, 77 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 }, 78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 }, 79 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 }, 80 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 }, 81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 }, 82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 }, 83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 }, 84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 }, 85 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 }, 86 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 }, 87 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 }, 88 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 }, 89 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 }, 90 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 }, 91 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 }, 92 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 }, 93 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 }, 94 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 }, 95 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 }, 96 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 }, 97 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 }, 98 /* required last entry */ 99 {0, } 100 }; 101 102 MODULE_DEVICE_TABLE(pci, igb_pci_tbl); 103 104 static int igb_setup_all_tx_resources(struct igb_adapter *); 105 static int igb_setup_all_rx_resources(struct igb_adapter *); 106 static void igb_free_all_tx_resources(struct igb_adapter *); 107 static void igb_free_all_rx_resources(struct igb_adapter *); 108 static void igb_setup_mrqc(struct igb_adapter *); 109 static int igb_probe(struct pci_dev *, const struct pci_device_id *); 110 static void igb_remove(struct pci_dev *pdev); 111 static void igb_init_queue_configuration(struct igb_adapter *adapter); 112 static int igb_sw_init(struct igb_adapter *); 113 int igb_open(struct net_device *); 114 int igb_close(struct net_device *); 115 static void igb_configure(struct igb_adapter *); 116 static void igb_configure_tx(struct igb_adapter *); 117 static void igb_configure_rx(struct igb_adapter *); 118 static void igb_clean_all_tx_rings(struct igb_adapter *); 119 static void igb_clean_all_rx_rings(struct igb_adapter *); 120 static void igb_clean_tx_ring(struct igb_ring *); 121 static void igb_clean_rx_ring(struct igb_ring *); 122 static void igb_set_rx_mode(struct net_device *); 123 static void igb_update_phy_info(struct timer_list *); 124 static void igb_watchdog(struct timer_list *); 125 static void igb_watchdog_task(struct work_struct *); 126 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *); 127 static void igb_get_stats64(struct net_device *dev, 128 struct rtnl_link_stats64 *stats); 129 static int igb_change_mtu(struct net_device *, int); 130 static int igb_set_mac(struct net_device *, void *); 131 static void igb_set_uta(struct igb_adapter *adapter, bool set); 132 static irqreturn_t igb_intr(int irq, void *); 133 static irqreturn_t igb_intr_msi(int irq, void *); 134 static irqreturn_t igb_msix_other(int irq, void *); 135 static irqreturn_t igb_msix_ring(int irq, void *); 136 #ifdef CONFIG_IGB_DCA 137 static void igb_update_dca(struct igb_q_vector *); 138 static void igb_setup_dca(struct igb_adapter *); 139 #endif /* CONFIG_IGB_DCA */ 140 static int igb_poll(struct napi_struct *, int); 141 static bool igb_clean_tx_irq(struct igb_q_vector *, int); 142 static int igb_clean_rx_irq(struct igb_q_vector *, int); 143 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd); 144 static void igb_tx_timeout(struct net_device *, unsigned int txqueue); 145 static void igb_reset_task(struct work_struct *); 146 static void igb_vlan_mode(struct net_device *netdev, 147 netdev_features_t features); 148 static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16); 149 static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16); 150 static void igb_restore_vlan(struct igb_adapter *); 151 static void igb_rar_set_index(struct igb_adapter *, u32); 152 static void igb_ping_all_vfs(struct igb_adapter *); 153 static void igb_msg_task(struct igb_adapter *); 154 static void igb_vmm_control(struct igb_adapter *); 155 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *); 156 static void igb_flush_mac_table(struct igb_adapter *); 157 static int igb_available_rars(struct igb_adapter *, u8); 158 static void igb_set_default_mac_filter(struct igb_adapter *); 159 static int igb_uc_sync(struct net_device *, const unsigned char *); 160 static int igb_uc_unsync(struct net_device *, const unsigned char *); 161 static void igb_restore_vf_multicasts(struct igb_adapter *adapter); 162 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac); 163 static int igb_ndo_set_vf_vlan(struct net_device *netdev, 164 int vf, u16 vlan, u8 qos, __be16 vlan_proto); 165 static int igb_ndo_set_vf_bw(struct net_device *, int, int, int); 166 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf, 167 bool setting); 168 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf, 169 bool setting); 170 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf, 171 struct ifla_vf_info *ivi); 172 static void igb_check_vf_rate_limit(struct igb_adapter *); 173 static void igb_nfc_filter_exit(struct igb_adapter *adapter); 174 static void igb_nfc_filter_restore(struct igb_adapter *adapter); 175 176 #ifdef CONFIG_PCI_IOV 177 static int igb_vf_configure(struct igb_adapter *adapter, int vf); 178 static int igb_disable_sriov(struct pci_dev *dev, bool reinit); 179 #endif 180 181 static int igb_suspend(struct device *); 182 static int igb_resume(struct device *); 183 static int igb_runtime_suspend(struct device *dev); 184 static int igb_runtime_resume(struct device *dev); 185 static int igb_runtime_idle(struct device *dev); 186 #ifdef CONFIG_PM 187 static const struct dev_pm_ops igb_pm_ops = { 188 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume) 189 SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume, 190 igb_runtime_idle) 191 }; 192 #endif 193 static void igb_shutdown(struct pci_dev *); 194 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs); 195 #ifdef CONFIG_IGB_DCA 196 static int igb_notify_dca(struct notifier_block *, unsigned long, void *); 197 static struct notifier_block dca_notifier = { 198 .notifier_call = igb_notify_dca, 199 .next = NULL, 200 .priority = 0 201 }; 202 #endif 203 #ifdef CONFIG_PCI_IOV 204 static unsigned int max_vfs; 205 module_param(max_vfs, uint, 0); 206 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function"); 207 #endif /* CONFIG_PCI_IOV */ 208 209 static pci_ers_result_t igb_io_error_detected(struct pci_dev *, 210 pci_channel_state_t); 211 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *); 212 static void igb_io_resume(struct pci_dev *); 213 214 static const struct pci_error_handlers igb_err_handler = { 215 .error_detected = igb_io_error_detected, 216 .slot_reset = igb_io_slot_reset, 217 .resume = igb_io_resume, 218 }; 219 220 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba); 221 222 static struct pci_driver igb_driver = { 223 .name = igb_driver_name, 224 .id_table = igb_pci_tbl, 225 .probe = igb_probe, 226 .remove = igb_remove, 227 #ifdef CONFIG_PM 228 .driver.pm = &igb_pm_ops, 229 #endif 230 .shutdown = igb_shutdown, 231 .sriov_configure = igb_pci_sriov_configure, 232 .err_handler = &igb_err_handler 233 }; 234 235 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>"); 236 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver"); 237 MODULE_LICENSE("GPL v2"); 238 239 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK) 240 static int debug = -1; 241 module_param(debug, int, 0); 242 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)"); 243 244 struct igb_reg_info { 245 u32 ofs; 246 char *name; 247 }; 248 249 static const struct igb_reg_info igb_reg_info_tbl[] = { 250 251 /* General Registers */ 252 {E1000_CTRL, "CTRL"}, 253 {E1000_STATUS, "STATUS"}, 254 {E1000_CTRL_EXT, "CTRL_EXT"}, 255 256 /* Interrupt Registers */ 257 {E1000_ICR, "ICR"}, 258 259 /* RX Registers */ 260 {E1000_RCTL, "RCTL"}, 261 {E1000_RDLEN(0), "RDLEN"}, 262 {E1000_RDH(0), "RDH"}, 263 {E1000_RDT(0), "RDT"}, 264 {E1000_RXDCTL(0), "RXDCTL"}, 265 {E1000_RDBAL(0), "RDBAL"}, 266 {E1000_RDBAH(0), "RDBAH"}, 267 268 /* TX Registers */ 269 {E1000_TCTL, "TCTL"}, 270 {E1000_TDBAL(0), "TDBAL"}, 271 {E1000_TDBAH(0), "TDBAH"}, 272 {E1000_TDLEN(0), "TDLEN"}, 273 {E1000_TDH(0), "TDH"}, 274 {E1000_TDT(0), "TDT"}, 275 {E1000_TXDCTL(0), "TXDCTL"}, 276 {E1000_TDFH, "TDFH"}, 277 {E1000_TDFT, "TDFT"}, 278 {E1000_TDFHS, "TDFHS"}, 279 {E1000_TDFPC, "TDFPC"}, 280 281 /* List Terminator */ 282 {} 283 }; 284 285 /* igb_regdump - register printout routine */ 286 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo) 287 { 288 int n = 0; 289 char rname[16]; 290 u32 regs[8]; 291 292 switch (reginfo->ofs) { 293 case E1000_RDLEN(0): 294 for (n = 0; n < 4; n++) 295 regs[n] = rd32(E1000_RDLEN(n)); 296 break; 297 case E1000_RDH(0): 298 for (n = 0; n < 4; n++) 299 regs[n] = rd32(E1000_RDH(n)); 300 break; 301 case E1000_RDT(0): 302 for (n = 0; n < 4; n++) 303 regs[n] = rd32(E1000_RDT(n)); 304 break; 305 case E1000_RXDCTL(0): 306 for (n = 0; n < 4; n++) 307 regs[n] = rd32(E1000_RXDCTL(n)); 308 break; 309 case E1000_RDBAL(0): 310 for (n = 0; n < 4; n++) 311 regs[n] = rd32(E1000_RDBAL(n)); 312 break; 313 case E1000_RDBAH(0): 314 for (n = 0; n < 4; n++) 315 regs[n] = rd32(E1000_RDBAH(n)); 316 break; 317 case E1000_TDBAL(0): 318 for (n = 0; n < 4; n++) 319 regs[n] = rd32(E1000_TDBAL(n)); 320 break; 321 case E1000_TDBAH(0): 322 for (n = 0; n < 4; n++) 323 regs[n] = rd32(E1000_TDBAH(n)); 324 break; 325 case E1000_TDLEN(0): 326 for (n = 0; n < 4; n++) 327 regs[n] = rd32(E1000_TDLEN(n)); 328 break; 329 case E1000_TDH(0): 330 for (n = 0; n < 4; n++) 331 regs[n] = rd32(E1000_TDH(n)); 332 break; 333 case E1000_TDT(0): 334 for (n = 0; n < 4; n++) 335 regs[n] = rd32(E1000_TDT(n)); 336 break; 337 case E1000_TXDCTL(0): 338 for (n = 0; n < 4; n++) 339 regs[n] = rd32(E1000_TXDCTL(n)); 340 break; 341 default: 342 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs)); 343 return; 344 } 345 346 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]"); 347 pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1], 348 regs[2], regs[3]); 349 } 350 351 /* igb_dump - Print registers, Tx-rings and Rx-rings */ 352 static void igb_dump(struct igb_adapter *adapter) 353 { 354 struct net_device *netdev = adapter->netdev; 355 struct e1000_hw *hw = &adapter->hw; 356 struct igb_reg_info *reginfo; 357 struct igb_ring *tx_ring; 358 union e1000_adv_tx_desc *tx_desc; 359 struct my_u0 { __le64 a; __le64 b; } *u0; 360 struct igb_ring *rx_ring; 361 union e1000_adv_rx_desc *rx_desc; 362 u32 staterr; 363 u16 i, n; 364 365 if (!netif_msg_hw(adapter)) 366 return; 367 368 /* Print netdevice Info */ 369 if (netdev) { 370 dev_info(&adapter->pdev->dev, "Net device Info\n"); 371 pr_info("Device Name state trans_start\n"); 372 pr_info("%-15s %016lX %016lX\n", netdev->name, 373 netdev->state, dev_trans_start(netdev)); 374 } 375 376 /* Print Registers */ 377 dev_info(&adapter->pdev->dev, "Register Dump\n"); 378 pr_info(" Register Name Value\n"); 379 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl; 380 reginfo->name; reginfo++) { 381 igb_regdump(hw, reginfo); 382 } 383 384 /* Print TX Ring Summary */ 385 if (!netdev || !netif_running(netdev)) 386 goto exit; 387 388 dev_info(&adapter->pdev->dev, "TX Rings Summary\n"); 389 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n"); 390 for (n = 0; n < adapter->num_tx_queues; n++) { 391 struct igb_tx_buffer *buffer_info; 392 tx_ring = adapter->tx_ring[n]; 393 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean]; 394 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n", 395 n, tx_ring->next_to_use, tx_ring->next_to_clean, 396 (u64)dma_unmap_addr(buffer_info, dma), 397 dma_unmap_len(buffer_info, len), 398 buffer_info->next_to_watch, 399 (u64)buffer_info->time_stamp); 400 } 401 402 /* Print TX Rings */ 403 if (!netif_msg_tx_done(adapter)) 404 goto rx_ring_summary; 405 406 dev_info(&adapter->pdev->dev, "TX Rings Dump\n"); 407 408 /* Transmit Descriptor Formats 409 * 410 * Advanced Transmit Descriptor 411 * +--------------------------------------------------------------+ 412 * 0 | Buffer Address [63:0] | 413 * +--------------------------------------------------------------+ 414 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN | 415 * +--------------------------------------------------------------+ 416 * 63 46 45 40 39 38 36 35 32 31 24 15 0 417 */ 418 419 for (n = 0; n < adapter->num_tx_queues; n++) { 420 tx_ring = adapter->tx_ring[n]; 421 pr_info("------------------------------------\n"); 422 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index); 423 pr_info("------------------------------------\n"); 424 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] [bi->dma ] leng ntw timestamp bi->skb\n"); 425 426 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) { 427 const char *next_desc; 428 struct igb_tx_buffer *buffer_info; 429 tx_desc = IGB_TX_DESC(tx_ring, i); 430 buffer_info = &tx_ring->tx_buffer_info[i]; 431 u0 = (struct my_u0 *)tx_desc; 432 if (i == tx_ring->next_to_use && 433 i == tx_ring->next_to_clean) 434 next_desc = " NTC/U"; 435 else if (i == tx_ring->next_to_use) 436 next_desc = " NTU"; 437 else if (i == tx_ring->next_to_clean) 438 next_desc = " NTC"; 439 else 440 next_desc = ""; 441 442 pr_info("T [0x%03X] %016llX %016llX %016llX %04X %p %016llX %p%s\n", 443 i, le64_to_cpu(u0->a), 444 le64_to_cpu(u0->b), 445 (u64)dma_unmap_addr(buffer_info, dma), 446 dma_unmap_len(buffer_info, len), 447 buffer_info->next_to_watch, 448 (u64)buffer_info->time_stamp, 449 buffer_info->skb, next_desc); 450 451 if (netif_msg_pktdata(adapter) && buffer_info->skb) 452 print_hex_dump(KERN_INFO, "", 453 DUMP_PREFIX_ADDRESS, 454 16, 1, buffer_info->skb->data, 455 dma_unmap_len(buffer_info, len), 456 true); 457 } 458 } 459 460 /* Print RX Rings Summary */ 461 rx_ring_summary: 462 dev_info(&adapter->pdev->dev, "RX Rings Summary\n"); 463 pr_info("Queue [NTU] [NTC]\n"); 464 for (n = 0; n < adapter->num_rx_queues; n++) { 465 rx_ring = adapter->rx_ring[n]; 466 pr_info(" %5d %5X %5X\n", 467 n, rx_ring->next_to_use, rx_ring->next_to_clean); 468 } 469 470 /* Print RX Rings */ 471 if (!netif_msg_rx_status(adapter)) 472 goto exit; 473 474 dev_info(&adapter->pdev->dev, "RX Rings Dump\n"); 475 476 /* Advanced Receive Descriptor (Read) Format 477 * 63 1 0 478 * +-----------------------------------------------------+ 479 * 0 | Packet Buffer Address [63:1] |A0/NSE| 480 * +----------------------------------------------+------+ 481 * 8 | Header Buffer Address [63:1] | DD | 482 * +-----------------------------------------------------+ 483 * 484 * 485 * Advanced Receive Descriptor (Write-Back) Format 486 * 487 * 63 48 47 32 31 30 21 20 17 16 4 3 0 488 * +------------------------------------------------------+ 489 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS | 490 * | Checksum Ident | | | | Type | Type | 491 * +------------------------------------------------------+ 492 * 8 | VLAN Tag | Length | Extended Error | Extended Status | 493 * +------------------------------------------------------+ 494 * 63 48 47 32 31 20 19 0 495 */ 496 497 for (n = 0; n < adapter->num_rx_queues; n++) { 498 rx_ring = adapter->rx_ring[n]; 499 pr_info("------------------------------------\n"); 500 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index); 501 pr_info("------------------------------------\n"); 502 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] [bi->dma ] [bi->skb] <-- Adv Rx Read format\n"); 503 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n"); 504 505 for (i = 0; i < rx_ring->count; i++) { 506 const char *next_desc; 507 struct igb_rx_buffer *buffer_info; 508 buffer_info = &rx_ring->rx_buffer_info[i]; 509 rx_desc = IGB_RX_DESC(rx_ring, i); 510 u0 = (struct my_u0 *)rx_desc; 511 staterr = le32_to_cpu(rx_desc->wb.upper.status_error); 512 513 if (i == rx_ring->next_to_use) 514 next_desc = " NTU"; 515 else if (i == rx_ring->next_to_clean) 516 next_desc = " NTC"; 517 else 518 next_desc = ""; 519 520 if (staterr & E1000_RXD_STAT_DD) { 521 /* Descriptor Done */ 522 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n", 523 "RWB", i, 524 le64_to_cpu(u0->a), 525 le64_to_cpu(u0->b), 526 next_desc); 527 } else { 528 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n", 529 "R ", i, 530 le64_to_cpu(u0->a), 531 le64_to_cpu(u0->b), 532 (u64)buffer_info->dma, 533 next_desc); 534 535 if (netif_msg_pktdata(adapter) && 536 buffer_info->dma && buffer_info->page) { 537 print_hex_dump(KERN_INFO, "", 538 DUMP_PREFIX_ADDRESS, 539 16, 1, 540 page_address(buffer_info->page) + 541 buffer_info->page_offset, 542 igb_rx_bufsz(rx_ring), true); 543 } 544 } 545 } 546 } 547 548 exit: 549 return; 550 } 551 552 /** 553 * igb_get_i2c_data - Reads the I2C SDA data bit 554 * @data: opaque pointer to adapter struct 555 * 556 * Returns the I2C data bit value 557 **/ 558 static int igb_get_i2c_data(void *data) 559 { 560 struct igb_adapter *adapter = (struct igb_adapter *)data; 561 struct e1000_hw *hw = &adapter->hw; 562 s32 i2cctl = rd32(E1000_I2CPARAMS); 563 564 return !!(i2cctl & E1000_I2C_DATA_IN); 565 } 566 567 /** 568 * igb_set_i2c_data - Sets the I2C data bit 569 * @data: pointer to hardware structure 570 * @state: I2C data value (0 or 1) to set 571 * 572 * Sets the I2C data bit 573 **/ 574 static void igb_set_i2c_data(void *data, int state) 575 { 576 struct igb_adapter *adapter = (struct igb_adapter *)data; 577 struct e1000_hw *hw = &adapter->hw; 578 s32 i2cctl = rd32(E1000_I2CPARAMS); 579 580 if (state) { 581 i2cctl |= E1000_I2C_DATA_OUT | E1000_I2C_DATA_OE_N; 582 } else { 583 i2cctl &= ~E1000_I2C_DATA_OE_N; 584 i2cctl &= ~E1000_I2C_DATA_OUT; 585 } 586 587 wr32(E1000_I2CPARAMS, i2cctl); 588 wrfl(); 589 } 590 591 /** 592 * igb_set_i2c_clk - Sets the I2C SCL clock 593 * @data: pointer to hardware structure 594 * @state: state to set clock 595 * 596 * Sets the I2C clock line to state 597 **/ 598 static void igb_set_i2c_clk(void *data, int state) 599 { 600 struct igb_adapter *adapter = (struct igb_adapter *)data; 601 struct e1000_hw *hw = &adapter->hw; 602 s32 i2cctl = rd32(E1000_I2CPARAMS); 603 604 if (state) { 605 i2cctl |= E1000_I2C_CLK_OUT | E1000_I2C_CLK_OE_N; 606 } else { 607 i2cctl &= ~E1000_I2C_CLK_OUT; 608 i2cctl &= ~E1000_I2C_CLK_OE_N; 609 } 610 wr32(E1000_I2CPARAMS, i2cctl); 611 wrfl(); 612 } 613 614 /** 615 * igb_get_i2c_clk - Gets the I2C SCL clock state 616 * @data: pointer to hardware structure 617 * 618 * Gets the I2C clock state 619 **/ 620 static int igb_get_i2c_clk(void *data) 621 { 622 struct igb_adapter *adapter = (struct igb_adapter *)data; 623 struct e1000_hw *hw = &adapter->hw; 624 s32 i2cctl = rd32(E1000_I2CPARAMS); 625 626 return !!(i2cctl & E1000_I2C_CLK_IN); 627 } 628 629 static const struct i2c_algo_bit_data igb_i2c_algo = { 630 .setsda = igb_set_i2c_data, 631 .setscl = igb_set_i2c_clk, 632 .getsda = igb_get_i2c_data, 633 .getscl = igb_get_i2c_clk, 634 .udelay = 5, 635 .timeout = 20, 636 }; 637 638 /** 639 * igb_get_hw_dev - return device 640 * @hw: pointer to hardware structure 641 * 642 * used by hardware layer to print debugging information 643 **/ 644 struct net_device *igb_get_hw_dev(struct e1000_hw *hw) 645 { 646 struct igb_adapter *adapter = hw->back; 647 return adapter->netdev; 648 } 649 650 /** 651 * igb_init_module - Driver Registration Routine 652 * 653 * igb_init_module is the first routine called when the driver is 654 * loaded. All it does is register with the PCI subsystem. 655 **/ 656 static int __init igb_init_module(void) 657 { 658 int ret; 659 660 pr_info("%s\n", igb_driver_string); 661 pr_info("%s\n", igb_copyright); 662 663 #ifdef CONFIG_IGB_DCA 664 dca_register_notify(&dca_notifier); 665 #endif 666 ret = pci_register_driver(&igb_driver); 667 return ret; 668 } 669 670 module_init(igb_init_module); 671 672 /** 673 * igb_exit_module - Driver Exit Cleanup Routine 674 * 675 * igb_exit_module is called just before the driver is removed 676 * from memory. 677 **/ 678 static void __exit igb_exit_module(void) 679 { 680 #ifdef CONFIG_IGB_DCA 681 dca_unregister_notify(&dca_notifier); 682 #endif 683 pci_unregister_driver(&igb_driver); 684 } 685 686 module_exit(igb_exit_module); 687 688 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1)) 689 /** 690 * igb_cache_ring_register - Descriptor ring to register mapping 691 * @adapter: board private structure to initialize 692 * 693 * Once we know the feature-set enabled for the device, we'll cache 694 * the register offset the descriptor ring is assigned to. 695 **/ 696 static void igb_cache_ring_register(struct igb_adapter *adapter) 697 { 698 int i = 0, j = 0; 699 u32 rbase_offset = adapter->vfs_allocated_count; 700 701 switch (adapter->hw.mac.type) { 702 case e1000_82576: 703 /* The queues are allocated for virtualization such that VF 0 704 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc. 705 * In order to avoid collision we start at the first free queue 706 * and continue consuming queues in the same sequence 707 */ 708 if (adapter->vfs_allocated_count) { 709 for (; i < adapter->rss_queues; i++) 710 adapter->rx_ring[i]->reg_idx = rbase_offset + 711 Q_IDX_82576(i); 712 } 713 fallthrough; 714 case e1000_82575: 715 case e1000_82580: 716 case e1000_i350: 717 case e1000_i354: 718 case e1000_i210: 719 case e1000_i211: 720 default: 721 for (; i < adapter->num_rx_queues; i++) 722 adapter->rx_ring[i]->reg_idx = rbase_offset + i; 723 for (; j < adapter->num_tx_queues; j++) 724 adapter->tx_ring[j]->reg_idx = rbase_offset + j; 725 break; 726 } 727 } 728 729 u32 igb_rd32(struct e1000_hw *hw, u32 reg) 730 { 731 struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw); 732 u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr); 733 u32 value = 0; 734 735 if (E1000_REMOVED(hw_addr)) 736 return ~value; 737 738 value = readl(&hw_addr[reg]); 739 740 /* reads should not return all F's */ 741 if (!(~value) && (!reg || !(~readl(hw_addr)))) { 742 struct net_device *netdev = igb->netdev; 743 hw->hw_addr = NULL; 744 netdev_err(netdev, "PCIe link lost\n"); 745 WARN(pci_device_is_present(igb->pdev), 746 "igb: Failed to read reg 0x%x!\n", reg); 747 } 748 749 return value; 750 } 751 752 /** 753 * igb_write_ivar - configure ivar for given MSI-X vector 754 * @hw: pointer to the HW structure 755 * @msix_vector: vector number we are allocating to a given ring 756 * @index: row index of IVAR register to write within IVAR table 757 * @offset: column offset of in IVAR, should be multiple of 8 758 * 759 * This function is intended to handle the writing of the IVAR register 760 * for adapters 82576 and newer. The IVAR table consists of 2 columns, 761 * each containing an cause allocation for an Rx and Tx ring, and a 762 * variable number of rows depending on the number of queues supported. 763 **/ 764 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector, 765 int index, int offset) 766 { 767 u32 ivar = array_rd32(E1000_IVAR0, index); 768 769 /* clear any bits that are currently set */ 770 ivar &= ~((u32)0xFF << offset); 771 772 /* write vector and valid bit */ 773 ivar |= (msix_vector | E1000_IVAR_VALID) << offset; 774 775 array_wr32(E1000_IVAR0, index, ivar); 776 } 777 778 #define IGB_N0_QUEUE -1 779 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector) 780 { 781 struct igb_adapter *adapter = q_vector->adapter; 782 struct e1000_hw *hw = &adapter->hw; 783 int rx_queue = IGB_N0_QUEUE; 784 int tx_queue = IGB_N0_QUEUE; 785 u32 msixbm = 0; 786 787 if (q_vector->rx.ring) 788 rx_queue = q_vector->rx.ring->reg_idx; 789 if (q_vector->tx.ring) 790 tx_queue = q_vector->tx.ring->reg_idx; 791 792 switch (hw->mac.type) { 793 case e1000_82575: 794 /* The 82575 assigns vectors using a bitmask, which matches the 795 * bitmask for the EICR/EIMS/EIMC registers. To assign one 796 * or more queues to a vector, we write the appropriate bits 797 * into the MSIXBM register for that vector. 798 */ 799 if (rx_queue > IGB_N0_QUEUE) 800 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue; 801 if (tx_queue > IGB_N0_QUEUE) 802 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue; 803 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0) 804 msixbm |= E1000_EIMS_OTHER; 805 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm); 806 q_vector->eims_value = msixbm; 807 break; 808 case e1000_82576: 809 /* 82576 uses a table that essentially consists of 2 columns 810 * with 8 rows. The ordering is column-major so we use the 811 * lower 3 bits as the row index, and the 4th bit as the 812 * column offset. 813 */ 814 if (rx_queue > IGB_N0_QUEUE) 815 igb_write_ivar(hw, msix_vector, 816 rx_queue & 0x7, 817 (rx_queue & 0x8) << 1); 818 if (tx_queue > IGB_N0_QUEUE) 819 igb_write_ivar(hw, msix_vector, 820 tx_queue & 0x7, 821 ((tx_queue & 0x8) << 1) + 8); 822 q_vector->eims_value = BIT(msix_vector); 823 break; 824 case e1000_82580: 825 case e1000_i350: 826 case e1000_i354: 827 case e1000_i210: 828 case e1000_i211: 829 /* On 82580 and newer adapters the scheme is similar to 82576 830 * however instead of ordering column-major we have things 831 * ordered row-major. So we traverse the table by using 832 * bit 0 as the column offset, and the remaining bits as the 833 * row index. 834 */ 835 if (rx_queue > IGB_N0_QUEUE) 836 igb_write_ivar(hw, msix_vector, 837 rx_queue >> 1, 838 (rx_queue & 0x1) << 4); 839 if (tx_queue > IGB_N0_QUEUE) 840 igb_write_ivar(hw, msix_vector, 841 tx_queue >> 1, 842 ((tx_queue & 0x1) << 4) + 8); 843 q_vector->eims_value = BIT(msix_vector); 844 break; 845 default: 846 BUG(); 847 break; 848 } 849 850 /* add q_vector eims value to global eims_enable_mask */ 851 adapter->eims_enable_mask |= q_vector->eims_value; 852 853 /* configure q_vector to set itr on first interrupt */ 854 q_vector->set_itr = 1; 855 } 856 857 /** 858 * igb_configure_msix - Configure MSI-X hardware 859 * @adapter: board private structure to initialize 860 * 861 * igb_configure_msix sets up the hardware to properly 862 * generate MSI-X interrupts. 863 **/ 864 static void igb_configure_msix(struct igb_adapter *adapter) 865 { 866 u32 tmp; 867 int i, vector = 0; 868 struct e1000_hw *hw = &adapter->hw; 869 870 adapter->eims_enable_mask = 0; 871 872 /* set vector for other causes, i.e. link changes */ 873 switch (hw->mac.type) { 874 case e1000_82575: 875 tmp = rd32(E1000_CTRL_EXT); 876 /* enable MSI-X PBA support*/ 877 tmp |= E1000_CTRL_EXT_PBA_CLR; 878 879 /* Auto-Mask interrupts upon ICR read. */ 880 tmp |= E1000_CTRL_EXT_EIAME; 881 tmp |= E1000_CTRL_EXT_IRCA; 882 883 wr32(E1000_CTRL_EXT, tmp); 884 885 /* enable msix_other interrupt */ 886 array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER); 887 adapter->eims_other = E1000_EIMS_OTHER; 888 889 break; 890 891 case e1000_82576: 892 case e1000_82580: 893 case e1000_i350: 894 case e1000_i354: 895 case e1000_i210: 896 case e1000_i211: 897 /* Turn on MSI-X capability first, or our settings 898 * won't stick. And it will take days to debug. 899 */ 900 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE | 901 E1000_GPIE_PBA | E1000_GPIE_EIAME | 902 E1000_GPIE_NSICR); 903 904 /* enable msix_other interrupt */ 905 adapter->eims_other = BIT(vector); 906 tmp = (vector++ | E1000_IVAR_VALID) << 8; 907 908 wr32(E1000_IVAR_MISC, tmp); 909 break; 910 default: 911 /* do nothing, since nothing else supports MSI-X */ 912 break; 913 } /* switch (hw->mac.type) */ 914 915 adapter->eims_enable_mask |= adapter->eims_other; 916 917 for (i = 0; i < adapter->num_q_vectors; i++) 918 igb_assign_vector(adapter->q_vector[i], vector++); 919 920 wrfl(); 921 } 922 923 /** 924 * igb_request_msix - Initialize MSI-X interrupts 925 * @adapter: board private structure to initialize 926 * 927 * igb_request_msix allocates MSI-X vectors and requests interrupts from the 928 * kernel. 929 **/ 930 static int igb_request_msix(struct igb_adapter *adapter) 931 { 932 unsigned int num_q_vectors = adapter->num_q_vectors; 933 struct net_device *netdev = adapter->netdev; 934 int i, err = 0, vector = 0, free_vector = 0; 935 936 err = request_irq(adapter->msix_entries[vector].vector, 937 igb_msix_other, 0, netdev->name, adapter); 938 if (err) 939 goto err_out; 940 941 if (num_q_vectors > MAX_Q_VECTORS) { 942 num_q_vectors = MAX_Q_VECTORS; 943 dev_warn(&adapter->pdev->dev, 944 "The number of queue vectors (%d) is higher than max allowed (%d)\n", 945 adapter->num_q_vectors, MAX_Q_VECTORS); 946 } 947 for (i = 0; i < num_q_vectors; i++) { 948 struct igb_q_vector *q_vector = adapter->q_vector[i]; 949 950 vector++; 951 952 q_vector->itr_register = adapter->io_addr + E1000_EITR(vector); 953 954 if (q_vector->rx.ring && q_vector->tx.ring) 955 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name, 956 q_vector->rx.ring->queue_index); 957 else if (q_vector->tx.ring) 958 sprintf(q_vector->name, "%s-tx-%u", netdev->name, 959 q_vector->tx.ring->queue_index); 960 else if (q_vector->rx.ring) 961 sprintf(q_vector->name, "%s-rx-%u", netdev->name, 962 q_vector->rx.ring->queue_index); 963 else 964 sprintf(q_vector->name, "%s-unused", netdev->name); 965 966 err = request_irq(adapter->msix_entries[vector].vector, 967 igb_msix_ring, 0, q_vector->name, 968 q_vector); 969 if (err) 970 goto err_free; 971 } 972 973 igb_configure_msix(adapter); 974 return 0; 975 976 err_free: 977 /* free already assigned IRQs */ 978 free_irq(adapter->msix_entries[free_vector++].vector, adapter); 979 980 vector--; 981 for (i = 0; i < vector; i++) { 982 free_irq(adapter->msix_entries[free_vector++].vector, 983 adapter->q_vector[i]); 984 } 985 err_out: 986 return err; 987 } 988 989 /** 990 * igb_free_q_vector - Free memory allocated for specific interrupt vector 991 * @adapter: board private structure to initialize 992 * @v_idx: Index of vector to be freed 993 * 994 * This function frees the memory allocated to the q_vector. 995 **/ 996 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx) 997 { 998 struct igb_q_vector *q_vector = adapter->q_vector[v_idx]; 999 1000 adapter->q_vector[v_idx] = NULL; 1001 1002 /* igb_get_stats64() might access the rings on this vector, 1003 * we must wait a grace period before freeing it. 1004 */ 1005 if (q_vector) 1006 kfree_rcu(q_vector, rcu); 1007 } 1008 1009 /** 1010 * igb_reset_q_vector - Reset config for interrupt vector 1011 * @adapter: board private structure to initialize 1012 * @v_idx: Index of vector to be reset 1013 * 1014 * If NAPI is enabled it will delete any references to the 1015 * NAPI struct. This is preparation for igb_free_q_vector. 1016 **/ 1017 static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx) 1018 { 1019 struct igb_q_vector *q_vector = adapter->q_vector[v_idx]; 1020 1021 /* Coming from igb_set_interrupt_capability, the vectors are not yet 1022 * allocated. So, q_vector is NULL so we should stop here. 1023 */ 1024 if (!q_vector) 1025 return; 1026 1027 if (q_vector->tx.ring) 1028 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL; 1029 1030 if (q_vector->rx.ring) 1031 adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL; 1032 1033 netif_napi_del(&q_vector->napi); 1034 1035 } 1036 1037 static void igb_reset_interrupt_capability(struct igb_adapter *adapter) 1038 { 1039 int v_idx = adapter->num_q_vectors; 1040 1041 if (adapter->flags & IGB_FLAG_HAS_MSIX) 1042 pci_disable_msix(adapter->pdev); 1043 else if (adapter->flags & IGB_FLAG_HAS_MSI) 1044 pci_disable_msi(adapter->pdev); 1045 1046 while (v_idx--) 1047 igb_reset_q_vector(adapter, v_idx); 1048 } 1049 1050 /** 1051 * igb_free_q_vectors - Free memory allocated for interrupt vectors 1052 * @adapter: board private structure to initialize 1053 * 1054 * This function frees the memory allocated to the q_vectors. In addition if 1055 * NAPI is enabled it will delete any references to the NAPI struct prior 1056 * to freeing the q_vector. 1057 **/ 1058 static void igb_free_q_vectors(struct igb_adapter *adapter) 1059 { 1060 int v_idx = adapter->num_q_vectors; 1061 1062 adapter->num_tx_queues = 0; 1063 adapter->num_rx_queues = 0; 1064 adapter->num_q_vectors = 0; 1065 1066 while (v_idx--) { 1067 igb_reset_q_vector(adapter, v_idx); 1068 igb_free_q_vector(adapter, v_idx); 1069 } 1070 } 1071 1072 /** 1073 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts 1074 * @adapter: board private structure to initialize 1075 * 1076 * This function resets the device so that it has 0 Rx queues, Tx queues, and 1077 * MSI-X interrupts allocated. 1078 */ 1079 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter) 1080 { 1081 igb_free_q_vectors(adapter); 1082 igb_reset_interrupt_capability(adapter); 1083 } 1084 1085 /** 1086 * igb_set_interrupt_capability - set MSI or MSI-X if supported 1087 * @adapter: board private structure to initialize 1088 * @msix: boolean value of MSIX capability 1089 * 1090 * Attempt to configure interrupts using the best available 1091 * capabilities of the hardware and kernel. 1092 **/ 1093 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix) 1094 { 1095 int err; 1096 int numvecs, i; 1097 1098 if (!msix) 1099 goto msi_only; 1100 adapter->flags |= IGB_FLAG_HAS_MSIX; 1101 1102 /* Number of supported queues. */ 1103 adapter->num_rx_queues = adapter->rss_queues; 1104 if (adapter->vfs_allocated_count) 1105 adapter->num_tx_queues = 1; 1106 else 1107 adapter->num_tx_queues = adapter->rss_queues; 1108 1109 /* start with one vector for every Rx queue */ 1110 numvecs = adapter->num_rx_queues; 1111 1112 /* if Tx handler is separate add 1 for every Tx queue */ 1113 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS)) 1114 numvecs += adapter->num_tx_queues; 1115 1116 /* store the number of vectors reserved for queues */ 1117 adapter->num_q_vectors = numvecs; 1118 1119 /* add 1 vector for link status interrupts */ 1120 numvecs++; 1121 for (i = 0; i < numvecs; i++) 1122 adapter->msix_entries[i].entry = i; 1123 1124 err = pci_enable_msix_range(adapter->pdev, 1125 adapter->msix_entries, 1126 numvecs, 1127 numvecs); 1128 if (err > 0) 1129 return; 1130 1131 igb_reset_interrupt_capability(adapter); 1132 1133 /* If we can't do MSI-X, try MSI */ 1134 msi_only: 1135 adapter->flags &= ~IGB_FLAG_HAS_MSIX; 1136 #ifdef CONFIG_PCI_IOV 1137 /* disable SR-IOV for non MSI-X configurations */ 1138 if (adapter->vf_data) { 1139 struct e1000_hw *hw = &adapter->hw; 1140 /* disable iov and allow time for transactions to clear */ 1141 pci_disable_sriov(adapter->pdev); 1142 msleep(500); 1143 1144 kfree(adapter->vf_mac_list); 1145 adapter->vf_mac_list = NULL; 1146 kfree(adapter->vf_data); 1147 adapter->vf_data = NULL; 1148 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ); 1149 wrfl(); 1150 msleep(100); 1151 dev_info(&adapter->pdev->dev, "IOV Disabled\n"); 1152 } 1153 #endif 1154 adapter->vfs_allocated_count = 0; 1155 adapter->rss_queues = 1; 1156 adapter->flags |= IGB_FLAG_QUEUE_PAIRS; 1157 adapter->num_rx_queues = 1; 1158 adapter->num_tx_queues = 1; 1159 adapter->num_q_vectors = 1; 1160 if (!pci_enable_msi(adapter->pdev)) 1161 adapter->flags |= IGB_FLAG_HAS_MSI; 1162 } 1163 1164 static void igb_add_ring(struct igb_ring *ring, 1165 struct igb_ring_container *head) 1166 { 1167 head->ring = ring; 1168 head->count++; 1169 } 1170 1171 /** 1172 * igb_alloc_q_vector - Allocate memory for a single interrupt vector 1173 * @adapter: board private structure to initialize 1174 * @v_count: q_vectors allocated on adapter, used for ring interleaving 1175 * @v_idx: index of vector in adapter struct 1176 * @txr_count: total number of Tx rings to allocate 1177 * @txr_idx: index of first Tx ring to allocate 1178 * @rxr_count: total number of Rx rings to allocate 1179 * @rxr_idx: index of first Rx ring to allocate 1180 * 1181 * We allocate one q_vector. If allocation fails we return -ENOMEM. 1182 **/ 1183 static int igb_alloc_q_vector(struct igb_adapter *adapter, 1184 int v_count, int v_idx, 1185 int txr_count, int txr_idx, 1186 int rxr_count, int rxr_idx) 1187 { 1188 struct igb_q_vector *q_vector; 1189 struct igb_ring *ring; 1190 int ring_count; 1191 size_t size; 1192 1193 /* igb only supports 1 Tx and/or 1 Rx queue per vector */ 1194 if (txr_count > 1 || rxr_count > 1) 1195 return -ENOMEM; 1196 1197 ring_count = txr_count + rxr_count; 1198 size = kmalloc_size_roundup(struct_size(q_vector, ring, ring_count)); 1199 1200 /* allocate q_vector and rings */ 1201 q_vector = adapter->q_vector[v_idx]; 1202 if (!q_vector) { 1203 q_vector = kzalloc(size, GFP_KERNEL); 1204 } else if (size > ksize(q_vector)) { 1205 struct igb_q_vector *new_q_vector; 1206 1207 new_q_vector = kzalloc(size, GFP_KERNEL); 1208 if (new_q_vector) 1209 kfree_rcu(q_vector, rcu); 1210 q_vector = new_q_vector; 1211 } else { 1212 memset(q_vector, 0, size); 1213 } 1214 if (!q_vector) 1215 return -ENOMEM; 1216 1217 /* initialize NAPI */ 1218 netif_napi_add(adapter->netdev, &q_vector->napi, igb_poll); 1219 1220 /* tie q_vector and adapter together */ 1221 adapter->q_vector[v_idx] = q_vector; 1222 q_vector->adapter = adapter; 1223 1224 /* initialize work limits */ 1225 q_vector->tx.work_limit = adapter->tx_work_limit; 1226 1227 /* initialize ITR configuration */ 1228 q_vector->itr_register = adapter->io_addr + E1000_EITR(0); 1229 q_vector->itr_val = IGB_START_ITR; 1230 1231 /* initialize pointer to rings */ 1232 ring = q_vector->ring; 1233 1234 /* intialize ITR */ 1235 if (rxr_count) { 1236 /* rx or rx/tx vector */ 1237 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3) 1238 q_vector->itr_val = adapter->rx_itr_setting; 1239 } else { 1240 /* tx only vector */ 1241 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3) 1242 q_vector->itr_val = adapter->tx_itr_setting; 1243 } 1244 1245 if (txr_count) { 1246 /* assign generic ring traits */ 1247 ring->dev = &adapter->pdev->dev; 1248 ring->netdev = adapter->netdev; 1249 1250 /* configure backlink on ring */ 1251 ring->q_vector = q_vector; 1252 1253 /* update q_vector Tx values */ 1254 igb_add_ring(ring, &q_vector->tx); 1255 1256 /* For 82575, context index must be unique per ring. */ 1257 if (adapter->hw.mac.type == e1000_82575) 1258 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags); 1259 1260 /* apply Tx specific ring traits */ 1261 ring->count = adapter->tx_ring_count; 1262 ring->queue_index = txr_idx; 1263 1264 ring->cbs_enable = false; 1265 ring->idleslope = 0; 1266 ring->sendslope = 0; 1267 ring->hicredit = 0; 1268 ring->locredit = 0; 1269 1270 u64_stats_init(&ring->tx_syncp); 1271 u64_stats_init(&ring->tx_syncp2); 1272 1273 /* assign ring to adapter */ 1274 adapter->tx_ring[txr_idx] = ring; 1275 1276 /* push pointer to next ring */ 1277 ring++; 1278 } 1279 1280 if (rxr_count) { 1281 /* assign generic ring traits */ 1282 ring->dev = &adapter->pdev->dev; 1283 ring->netdev = adapter->netdev; 1284 1285 /* configure backlink on ring */ 1286 ring->q_vector = q_vector; 1287 1288 /* update q_vector Rx values */ 1289 igb_add_ring(ring, &q_vector->rx); 1290 1291 /* set flag indicating ring supports SCTP checksum offload */ 1292 if (adapter->hw.mac.type >= e1000_82576) 1293 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags); 1294 1295 /* On i350, i354, i210, and i211, loopback VLAN packets 1296 * have the tag byte-swapped. 1297 */ 1298 if (adapter->hw.mac.type >= e1000_i350) 1299 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags); 1300 1301 /* apply Rx specific ring traits */ 1302 ring->count = adapter->rx_ring_count; 1303 ring->queue_index = rxr_idx; 1304 1305 u64_stats_init(&ring->rx_syncp); 1306 1307 /* assign ring to adapter */ 1308 adapter->rx_ring[rxr_idx] = ring; 1309 } 1310 1311 return 0; 1312 } 1313 1314 1315 /** 1316 * igb_alloc_q_vectors - Allocate memory for interrupt vectors 1317 * @adapter: board private structure to initialize 1318 * 1319 * We allocate one q_vector per queue interrupt. If allocation fails we 1320 * return -ENOMEM. 1321 **/ 1322 static int igb_alloc_q_vectors(struct igb_adapter *adapter) 1323 { 1324 int q_vectors = adapter->num_q_vectors; 1325 int rxr_remaining = adapter->num_rx_queues; 1326 int txr_remaining = adapter->num_tx_queues; 1327 int rxr_idx = 0, txr_idx = 0, v_idx = 0; 1328 int err; 1329 1330 if (q_vectors >= (rxr_remaining + txr_remaining)) { 1331 for (; rxr_remaining; v_idx++) { 1332 err = igb_alloc_q_vector(adapter, q_vectors, v_idx, 1333 0, 0, 1, rxr_idx); 1334 1335 if (err) 1336 goto err_out; 1337 1338 /* update counts and index */ 1339 rxr_remaining--; 1340 rxr_idx++; 1341 } 1342 } 1343 1344 for (; v_idx < q_vectors; v_idx++) { 1345 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx); 1346 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx); 1347 1348 err = igb_alloc_q_vector(adapter, q_vectors, v_idx, 1349 tqpv, txr_idx, rqpv, rxr_idx); 1350 1351 if (err) 1352 goto err_out; 1353 1354 /* update counts and index */ 1355 rxr_remaining -= rqpv; 1356 txr_remaining -= tqpv; 1357 rxr_idx++; 1358 txr_idx++; 1359 } 1360 1361 return 0; 1362 1363 err_out: 1364 adapter->num_tx_queues = 0; 1365 adapter->num_rx_queues = 0; 1366 adapter->num_q_vectors = 0; 1367 1368 while (v_idx--) 1369 igb_free_q_vector(adapter, v_idx); 1370 1371 return -ENOMEM; 1372 } 1373 1374 /** 1375 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors 1376 * @adapter: board private structure to initialize 1377 * @msix: boolean value of MSIX capability 1378 * 1379 * This function initializes the interrupts and allocates all of the queues. 1380 **/ 1381 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix) 1382 { 1383 struct pci_dev *pdev = adapter->pdev; 1384 int err; 1385 1386 igb_set_interrupt_capability(adapter, msix); 1387 1388 err = igb_alloc_q_vectors(adapter); 1389 if (err) { 1390 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n"); 1391 goto err_alloc_q_vectors; 1392 } 1393 1394 igb_cache_ring_register(adapter); 1395 1396 return 0; 1397 1398 err_alloc_q_vectors: 1399 igb_reset_interrupt_capability(adapter); 1400 return err; 1401 } 1402 1403 /** 1404 * igb_request_irq - initialize interrupts 1405 * @adapter: board private structure to initialize 1406 * 1407 * Attempts to configure interrupts using the best available 1408 * capabilities of the hardware and kernel. 1409 **/ 1410 static int igb_request_irq(struct igb_adapter *adapter) 1411 { 1412 struct net_device *netdev = adapter->netdev; 1413 struct pci_dev *pdev = adapter->pdev; 1414 int err = 0; 1415 1416 if (adapter->flags & IGB_FLAG_HAS_MSIX) { 1417 err = igb_request_msix(adapter); 1418 if (!err) 1419 goto request_done; 1420 /* fall back to MSI */ 1421 igb_free_all_tx_resources(adapter); 1422 igb_free_all_rx_resources(adapter); 1423 1424 igb_clear_interrupt_scheme(adapter); 1425 err = igb_init_interrupt_scheme(adapter, false); 1426 if (err) 1427 goto request_done; 1428 1429 igb_setup_all_tx_resources(adapter); 1430 igb_setup_all_rx_resources(adapter); 1431 igb_configure(adapter); 1432 } 1433 1434 igb_assign_vector(adapter->q_vector[0], 0); 1435 1436 if (adapter->flags & IGB_FLAG_HAS_MSI) { 1437 err = request_irq(pdev->irq, igb_intr_msi, 0, 1438 netdev->name, adapter); 1439 if (!err) 1440 goto request_done; 1441 1442 /* fall back to legacy interrupts */ 1443 igb_reset_interrupt_capability(adapter); 1444 adapter->flags &= ~IGB_FLAG_HAS_MSI; 1445 } 1446 1447 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED, 1448 netdev->name, adapter); 1449 1450 if (err) 1451 dev_err(&pdev->dev, "Error %d getting interrupt\n", 1452 err); 1453 1454 request_done: 1455 return err; 1456 } 1457 1458 static void igb_free_irq(struct igb_adapter *adapter) 1459 { 1460 if (adapter->flags & IGB_FLAG_HAS_MSIX) { 1461 int vector = 0, i; 1462 1463 free_irq(adapter->msix_entries[vector++].vector, adapter); 1464 1465 for (i = 0; i < adapter->num_q_vectors; i++) 1466 free_irq(adapter->msix_entries[vector++].vector, 1467 adapter->q_vector[i]); 1468 } else { 1469 free_irq(adapter->pdev->irq, adapter); 1470 } 1471 } 1472 1473 /** 1474 * igb_irq_disable - Mask off interrupt generation on the NIC 1475 * @adapter: board private structure 1476 **/ 1477 static void igb_irq_disable(struct igb_adapter *adapter) 1478 { 1479 struct e1000_hw *hw = &adapter->hw; 1480 1481 /* we need to be careful when disabling interrupts. The VFs are also 1482 * mapped into these registers and so clearing the bits can cause 1483 * issues on the VF drivers so we only need to clear what we set 1484 */ 1485 if (adapter->flags & IGB_FLAG_HAS_MSIX) { 1486 u32 regval = rd32(E1000_EIAM); 1487 1488 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask); 1489 wr32(E1000_EIMC, adapter->eims_enable_mask); 1490 regval = rd32(E1000_EIAC); 1491 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask); 1492 } 1493 1494 wr32(E1000_IAM, 0); 1495 wr32(E1000_IMC, ~0); 1496 wrfl(); 1497 if (adapter->flags & IGB_FLAG_HAS_MSIX) { 1498 int i; 1499 1500 for (i = 0; i < adapter->num_q_vectors; i++) 1501 synchronize_irq(adapter->msix_entries[i].vector); 1502 } else { 1503 synchronize_irq(adapter->pdev->irq); 1504 } 1505 } 1506 1507 /** 1508 * igb_irq_enable - Enable default interrupt generation settings 1509 * @adapter: board private structure 1510 **/ 1511 static void igb_irq_enable(struct igb_adapter *adapter) 1512 { 1513 struct e1000_hw *hw = &adapter->hw; 1514 1515 if (adapter->flags & IGB_FLAG_HAS_MSIX) { 1516 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA; 1517 u32 regval = rd32(E1000_EIAC); 1518 1519 wr32(E1000_EIAC, regval | adapter->eims_enable_mask); 1520 regval = rd32(E1000_EIAM); 1521 wr32(E1000_EIAM, regval | adapter->eims_enable_mask); 1522 wr32(E1000_EIMS, adapter->eims_enable_mask); 1523 if (adapter->vfs_allocated_count) { 1524 wr32(E1000_MBVFIMR, 0xFF); 1525 ims |= E1000_IMS_VMMB; 1526 } 1527 wr32(E1000_IMS, ims); 1528 } else { 1529 wr32(E1000_IMS, IMS_ENABLE_MASK | 1530 E1000_IMS_DRSTA); 1531 wr32(E1000_IAM, IMS_ENABLE_MASK | 1532 E1000_IMS_DRSTA); 1533 } 1534 } 1535 1536 static void igb_update_mng_vlan(struct igb_adapter *adapter) 1537 { 1538 struct e1000_hw *hw = &adapter->hw; 1539 u16 pf_id = adapter->vfs_allocated_count; 1540 u16 vid = adapter->hw.mng_cookie.vlan_id; 1541 u16 old_vid = adapter->mng_vlan_id; 1542 1543 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) { 1544 /* add VID to filter table */ 1545 igb_vfta_set(hw, vid, pf_id, true, true); 1546 adapter->mng_vlan_id = vid; 1547 } else { 1548 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE; 1549 } 1550 1551 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) && 1552 (vid != old_vid) && 1553 !test_bit(old_vid, adapter->active_vlans)) { 1554 /* remove VID from filter table */ 1555 igb_vfta_set(hw, vid, pf_id, false, true); 1556 } 1557 } 1558 1559 /** 1560 * igb_release_hw_control - release control of the h/w to f/w 1561 * @adapter: address of board private structure 1562 * 1563 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit. 1564 * For ASF and Pass Through versions of f/w this means that the 1565 * driver is no longer loaded. 1566 **/ 1567 static void igb_release_hw_control(struct igb_adapter *adapter) 1568 { 1569 struct e1000_hw *hw = &adapter->hw; 1570 u32 ctrl_ext; 1571 1572 /* Let firmware take over control of h/w */ 1573 ctrl_ext = rd32(E1000_CTRL_EXT); 1574 wr32(E1000_CTRL_EXT, 1575 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD); 1576 } 1577 1578 /** 1579 * igb_get_hw_control - get control of the h/w from f/w 1580 * @adapter: address of board private structure 1581 * 1582 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit. 1583 * For ASF and Pass Through versions of f/w this means that 1584 * the driver is loaded. 1585 **/ 1586 static void igb_get_hw_control(struct igb_adapter *adapter) 1587 { 1588 struct e1000_hw *hw = &adapter->hw; 1589 u32 ctrl_ext; 1590 1591 /* Let firmware know the driver has taken over */ 1592 ctrl_ext = rd32(E1000_CTRL_EXT); 1593 wr32(E1000_CTRL_EXT, 1594 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD); 1595 } 1596 1597 static void enable_fqtss(struct igb_adapter *adapter, bool enable) 1598 { 1599 struct net_device *netdev = adapter->netdev; 1600 struct e1000_hw *hw = &adapter->hw; 1601 1602 WARN_ON(hw->mac.type != e1000_i210); 1603 1604 if (enable) 1605 adapter->flags |= IGB_FLAG_FQTSS; 1606 else 1607 adapter->flags &= ~IGB_FLAG_FQTSS; 1608 1609 if (netif_running(netdev)) 1610 schedule_work(&adapter->reset_task); 1611 } 1612 1613 static bool is_fqtss_enabled(struct igb_adapter *adapter) 1614 { 1615 return (adapter->flags & IGB_FLAG_FQTSS) ? true : false; 1616 } 1617 1618 static void set_tx_desc_fetch_prio(struct e1000_hw *hw, int queue, 1619 enum tx_queue_prio prio) 1620 { 1621 u32 val; 1622 1623 WARN_ON(hw->mac.type != e1000_i210); 1624 WARN_ON(queue < 0 || queue > 4); 1625 1626 val = rd32(E1000_I210_TXDCTL(queue)); 1627 1628 if (prio == TX_QUEUE_PRIO_HIGH) 1629 val |= E1000_TXDCTL_PRIORITY; 1630 else 1631 val &= ~E1000_TXDCTL_PRIORITY; 1632 1633 wr32(E1000_I210_TXDCTL(queue), val); 1634 } 1635 1636 static void set_queue_mode(struct e1000_hw *hw, int queue, enum queue_mode mode) 1637 { 1638 u32 val; 1639 1640 WARN_ON(hw->mac.type != e1000_i210); 1641 WARN_ON(queue < 0 || queue > 1); 1642 1643 val = rd32(E1000_I210_TQAVCC(queue)); 1644 1645 if (mode == QUEUE_MODE_STREAM_RESERVATION) 1646 val |= E1000_TQAVCC_QUEUEMODE; 1647 else 1648 val &= ~E1000_TQAVCC_QUEUEMODE; 1649 1650 wr32(E1000_I210_TQAVCC(queue), val); 1651 } 1652 1653 static bool is_any_cbs_enabled(struct igb_adapter *adapter) 1654 { 1655 int i; 1656 1657 for (i = 0; i < adapter->num_tx_queues; i++) { 1658 if (adapter->tx_ring[i]->cbs_enable) 1659 return true; 1660 } 1661 1662 return false; 1663 } 1664 1665 static bool is_any_txtime_enabled(struct igb_adapter *adapter) 1666 { 1667 int i; 1668 1669 for (i = 0; i < adapter->num_tx_queues; i++) { 1670 if (adapter->tx_ring[i]->launchtime_enable) 1671 return true; 1672 } 1673 1674 return false; 1675 } 1676 1677 /** 1678 * igb_config_tx_modes - Configure "Qav Tx mode" features on igb 1679 * @adapter: pointer to adapter struct 1680 * @queue: queue number 1681 * 1682 * Configure CBS and Launchtime for a given hardware queue. 1683 * Parameters are retrieved from the correct Tx ring, so 1684 * igb_save_cbs_params() and igb_save_txtime_params() should be used 1685 * for setting those correctly prior to this function being called. 1686 **/ 1687 static void igb_config_tx_modes(struct igb_adapter *adapter, int queue) 1688 { 1689 struct net_device *netdev = adapter->netdev; 1690 struct e1000_hw *hw = &adapter->hw; 1691 struct igb_ring *ring; 1692 u32 tqavcc, tqavctrl; 1693 u16 value; 1694 1695 WARN_ON(hw->mac.type != e1000_i210); 1696 WARN_ON(queue < 0 || queue > 1); 1697 ring = adapter->tx_ring[queue]; 1698 1699 /* If any of the Qav features is enabled, configure queues as SR and 1700 * with HIGH PRIO. If none is, then configure them with LOW PRIO and 1701 * as SP. 1702 */ 1703 if (ring->cbs_enable || ring->launchtime_enable) { 1704 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_HIGH); 1705 set_queue_mode(hw, queue, QUEUE_MODE_STREAM_RESERVATION); 1706 } else { 1707 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_LOW); 1708 set_queue_mode(hw, queue, QUEUE_MODE_STRICT_PRIORITY); 1709 } 1710 1711 /* If CBS is enabled, set DataTranARB and config its parameters. */ 1712 if (ring->cbs_enable || queue == 0) { 1713 /* i210 does not allow the queue 0 to be in the Strict 1714 * Priority mode while the Qav mode is enabled, so, 1715 * instead of disabling strict priority mode, we give 1716 * queue 0 the maximum of credits possible. 1717 * 1718 * See section 8.12.19 of the i210 datasheet, "Note: 1719 * Queue0 QueueMode must be set to 1b when 1720 * TransmitMode is set to Qav." 1721 */ 1722 if (queue == 0 && !ring->cbs_enable) { 1723 /* max "linkspeed" idleslope in kbps */ 1724 ring->idleslope = 1000000; 1725 ring->hicredit = ETH_FRAME_LEN; 1726 } 1727 1728 /* Always set data transfer arbitration to credit-based 1729 * shaper algorithm on TQAVCTRL if CBS is enabled for any of 1730 * the queues. 1731 */ 1732 tqavctrl = rd32(E1000_I210_TQAVCTRL); 1733 tqavctrl |= E1000_TQAVCTRL_DATATRANARB; 1734 wr32(E1000_I210_TQAVCTRL, tqavctrl); 1735 1736 /* According to i210 datasheet section 7.2.7.7, we should set 1737 * the 'idleSlope' field from TQAVCC register following the 1738 * equation: 1739 * 1740 * For 100 Mbps link speed: 1741 * 1742 * value = BW * 0x7735 * 0.2 (E1) 1743 * 1744 * For 1000Mbps link speed: 1745 * 1746 * value = BW * 0x7735 * 2 (E2) 1747 * 1748 * E1 and E2 can be merged into one equation as shown below. 1749 * Note that 'link-speed' is in Mbps. 1750 * 1751 * value = BW * 0x7735 * 2 * link-speed 1752 * -------------- (E3) 1753 * 1000 1754 * 1755 * 'BW' is the percentage bandwidth out of full link speed 1756 * which can be found with the following equation. Note that 1757 * idleSlope here is the parameter from this function which 1758 * is in kbps. 1759 * 1760 * BW = idleSlope 1761 * ----------------- (E4) 1762 * link-speed * 1000 1763 * 1764 * That said, we can come up with a generic equation to 1765 * calculate the value we should set it TQAVCC register by 1766 * replacing 'BW' in E3 by E4. The resulting equation is: 1767 * 1768 * value = idleSlope * 0x7735 * 2 * link-speed 1769 * ----------------- -------------- (E5) 1770 * link-speed * 1000 1000 1771 * 1772 * 'link-speed' is present in both sides of the fraction so 1773 * it is canceled out. The final equation is the following: 1774 * 1775 * value = idleSlope * 61034 1776 * ----------------- (E6) 1777 * 1000000 1778 * 1779 * NOTE: For i210, given the above, we can see that idleslope 1780 * is represented in 16.38431 kbps units by the value at 1781 * the TQAVCC register (1Gbps / 61034), which reduces 1782 * the granularity for idleslope increments. 1783 * For instance, if you want to configure a 2576kbps 1784 * idleslope, the value to be written on the register 1785 * would have to be 157.23. If rounded down, you end 1786 * up with less bandwidth available than originally 1787 * required (~2572 kbps). If rounded up, you end up 1788 * with a higher bandwidth (~2589 kbps). Below the 1789 * approach we take is to always round up the 1790 * calculated value, so the resulting bandwidth might 1791 * be slightly higher for some configurations. 1792 */ 1793 value = DIV_ROUND_UP_ULL(ring->idleslope * 61034ULL, 1000000); 1794 1795 tqavcc = rd32(E1000_I210_TQAVCC(queue)); 1796 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK; 1797 tqavcc |= value; 1798 wr32(E1000_I210_TQAVCC(queue), tqavcc); 1799 1800 wr32(E1000_I210_TQAVHC(queue), 1801 0x80000000 + ring->hicredit * 0x7735); 1802 } else { 1803 1804 /* Set idleSlope to zero. */ 1805 tqavcc = rd32(E1000_I210_TQAVCC(queue)); 1806 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK; 1807 wr32(E1000_I210_TQAVCC(queue), tqavcc); 1808 1809 /* Set hiCredit to zero. */ 1810 wr32(E1000_I210_TQAVHC(queue), 0); 1811 1812 /* If CBS is not enabled for any queues anymore, then return to 1813 * the default state of Data Transmission Arbitration on 1814 * TQAVCTRL. 1815 */ 1816 if (!is_any_cbs_enabled(adapter)) { 1817 tqavctrl = rd32(E1000_I210_TQAVCTRL); 1818 tqavctrl &= ~E1000_TQAVCTRL_DATATRANARB; 1819 wr32(E1000_I210_TQAVCTRL, tqavctrl); 1820 } 1821 } 1822 1823 /* If LaunchTime is enabled, set DataTranTIM. */ 1824 if (ring->launchtime_enable) { 1825 /* Always set DataTranTIM on TQAVCTRL if LaunchTime is enabled 1826 * for any of the SR queues, and configure fetchtime delta. 1827 * XXX NOTE: 1828 * - LaunchTime will be enabled for all SR queues. 1829 * - A fixed offset can be added relative to the launch 1830 * time of all packets if configured at reg LAUNCH_OS0. 1831 * We are keeping it as 0 for now (default value). 1832 */ 1833 tqavctrl = rd32(E1000_I210_TQAVCTRL); 1834 tqavctrl |= E1000_TQAVCTRL_DATATRANTIM | 1835 E1000_TQAVCTRL_FETCHTIME_DELTA; 1836 wr32(E1000_I210_TQAVCTRL, tqavctrl); 1837 } else { 1838 /* If Launchtime is not enabled for any SR queues anymore, 1839 * then clear DataTranTIM on TQAVCTRL and clear fetchtime delta, 1840 * effectively disabling Launchtime. 1841 */ 1842 if (!is_any_txtime_enabled(adapter)) { 1843 tqavctrl = rd32(E1000_I210_TQAVCTRL); 1844 tqavctrl &= ~E1000_TQAVCTRL_DATATRANTIM; 1845 tqavctrl &= ~E1000_TQAVCTRL_FETCHTIME_DELTA; 1846 wr32(E1000_I210_TQAVCTRL, tqavctrl); 1847 } 1848 } 1849 1850 /* XXX: In i210 controller the sendSlope and loCredit parameters from 1851 * CBS are not configurable by software so we don't do any 'controller 1852 * configuration' in respect to these parameters. 1853 */ 1854 1855 netdev_dbg(netdev, "Qav Tx mode: cbs %s, launchtime %s, queue %d idleslope %d sendslope %d hiCredit %d locredit %d\n", 1856 ring->cbs_enable ? "enabled" : "disabled", 1857 ring->launchtime_enable ? "enabled" : "disabled", 1858 queue, 1859 ring->idleslope, ring->sendslope, 1860 ring->hicredit, ring->locredit); 1861 } 1862 1863 static int igb_save_txtime_params(struct igb_adapter *adapter, int queue, 1864 bool enable) 1865 { 1866 struct igb_ring *ring; 1867 1868 if (queue < 0 || queue > adapter->num_tx_queues) 1869 return -EINVAL; 1870 1871 ring = adapter->tx_ring[queue]; 1872 ring->launchtime_enable = enable; 1873 1874 return 0; 1875 } 1876 1877 static int igb_save_cbs_params(struct igb_adapter *adapter, int queue, 1878 bool enable, int idleslope, int sendslope, 1879 int hicredit, int locredit) 1880 { 1881 struct igb_ring *ring; 1882 1883 if (queue < 0 || queue > adapter->num_tx_queues) 1884 return -EINVAL; 1885 1886 ring = adapter->tx_ring[queue]; 1887 1888 ring->cbs_enable = enable; 1889 ring->idleslope = idleslope; 1890 ring->sendslope = sendslope; 1891 ring->hicredit = hicredit; 1892 ring->locredit = locredit; 1893 1894 return 0; 1895 } 1896 1897 /** 1898 * igb_setup_tx_mode - Switch to/from Qav Tx mode when applicable 1899 * @adapter: pointer to adapter struct 1900 * 1901 * Configure TQAVCTRL register switching the controller's Tx mode 1902 * if FQTSS mode is enabled or disabled. Additionally, will issue 1903 * a call to igb_config_tx_modes() per queue so any previously saved 1904 * Tx parameters are applied. 1905 **/ 1906 static void igb_setup_tx_mode(struct igb_adapter *adapter) 1907 { 1908 struct net_device *netdev = adapter->netdev; 1909 struct e1000_hw *hw = &adapter->hw; 1910 u32 val; 1911 1912 /* Only i210 controller supports changing the transmission mode. */ 1913 if (hw->mac.type != e1000_i210) 1914 return; 1915 1916 if (is_fqtss_enabled(adapter)) { 1917 int i, max_queue; 1918 1919 /* Configure TQAVCTRL register: set transmit mode to 'Qav', 1920 * set data fetch arbitration to 'round robin', set SP_WAIT_SR 1921 * so SP queues wait for SR ones. 1922 */ 1923 val = rd32(E1000_I210_TQAVCTRL); 1924 val |= E1000_TQAVCTRL_XMIT_MODE | E1000_TQAVCTRL_SP_WAIT_SR; 1925 val &= ~E1000_TQAVCTRL_DATAFETCHARB; 1926 wr32(E1000_I210_TQAVCTRL, val); 1927 1928 /* Configure Tx and Rx packet buffers sizes as described in 1929 * i210 datasheet section 7.2.7.7. 1930 */ 1931 val = rd32(E1000_TXPBS); 1932 val &= ~I210_TXPBSIZE_MASK; 1933 val |= I210_TXPBSIZE_PB0_6KB | I210_TXPBSIZE_PB1_6KB | 1934 I210_TXPBSIZE_PB2_6KB | I210_TXPBSIZE_PB3_6KB; 1935 wr32(E1000_TXPBS, val); 1936 1937 val = rd32(E1000_RXPBS); 1938 val &= ~I210_RXPBSIZE_MASK; 1939 val |= I210_RXPBSIZE_PB_30KB; 1940 wr32(E1000_RXPBS, val); 1941 1942 /* Section 8.12.9 states that MAX_TPKT_SIZE from DTXMXPKTSZ 1943 * register should not exceed the buffer size programmed in 1944 * TXPBS. The smallest buffer size programmed in TXPBS is 4kB 1945 * so according to the datasheet we should set MAX_TPKT_SIZE to 1946 * 4kB / 64. 1947 * 1948 * However, when we do so, no frame from queue 2 and 3 are 1949 * transmitted. It seems the MAX_TPKT_SIZE should not be great 1950 * or _equal_ to the buffer size programmed in TXPBS. For this 1951 * reason, we set MAX_ TPKT_SIZE to (4kB - 1) / 64. 1952 */ 1953 val = (4096 - 1) / 64; 1954 wr32(E1000_I210_DTXMXPKTSZ, val); 1955 1956 /* Since FQTSS mode is enabled, apply any CBS configuration 1957 * previously set. If no previous CBS configuration has been 1958 * done, then the initial configuration is applied, which means 1959 * CBS is disabled. 1960 */ 1961 max_queue = (adapter->num_tx_queues < I210_SR_QUEUES_NUM) ? 1962 adapter->num_tx_queues : I210_SR_QUEUES_NUM; 1963 1964 for (i = 0; i < max_queue; i++) { 1965 igb_config_tx_modes(adapter, i); 1966 } 1967 } else { 1968 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT); 1969 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT); 1970 wr32(E1000_I210_DTXMXPKTSZ, I210_DTXMXPKTSZ_DEFAULT); 1971 1972 val = rd32(E1000_I210_TQAVCTRL); 1973 /* According to Section 8.12.21, the other flags we've set when 1974 * enabling FQTSS are not relevant when disabling FQTSS so we 1975 * don't set they here. 1976 */ 1977 val &= ~E1000_TQAVCTRL_XMIT_MODE; 1978 wr32(E1000_I210_TQAVCTRL, val); 1979 } 1980 1981 netdev_dbg(netdev, "FQTSS %s\n", (is_fqtss_enabled(adapter)) ? 1982 "enabled" : "disabled"); 1983 } 1984 1985 /** 1986 * igb_configure - configure the hardware for RX and TX 1987 * @adapter: private board structure 1988 **/ 1989 static void igb_configure(struct igb_adapter *adapter) 1990 { 1991 struct net_device *netdev = adapter->netdev; 1992 int i; 1993 1994 igb_get_hw_control(adapter); 1995 igb_set_rx_mode(netdev); 1996 igb_setup_tx_mode(adapter); 1997 1998 igb_restore_vlan(adapter); 1999 2000 igb_setup_tctl(adapter); 2001 igb_setup_mrqc(adapter); 2002 igb_setup_rctl(adapter); 2003 2004 igb_nfc_filter_restore(adapter); 2005 igb_configure_tx(adapter); 2006 igb_configure_rx(adapter); 2007 2008 igb_rx_fifo_flush_82575(&adapter->hw); 2009 2010 /* call igb_desc_unused which always leaves 2011 * at least 1 descriptor unused to make sure 2012 * next_to_use != next_to_clean 2013 */ 2014 for (i = 0; i < adapter->num_rx_queues; i++) { 2015 struct igb_ring *ring = adapter->rx_ring[i]; 2016 igb_alloc_rx_buffers(ring, igb_desc_unused(ring)); 2017 } 2018 } 2019 2020 /** 2021 * igb_power_up_link - Power up the phy/serdes link 2022 * @adapter: address of board private structure 2023 **/ 2024 void igb_power_up_link(struct igb_adapter *adapter) 2025 { 2026 igb_reset_phy(&adapter->hw); 2027 2028 if (adapter->hw.phy.media_type == e1000_media_type_copper) 2029 igb_power_up_phy_copper(&adapter->hw); 2030 else 2031 igb_power_up_serdes_link_82575(&adapter->hw); 2032 2033 igb_setup_link(&adapter->hw); 2034 } 2035 2036 /** 2037 * igb_power_down_link - Power down the phy/serdes link 2038 * @adapter: address of board private structure 2039 */ 2040 static void igb_power_down_link(struct igb_adapter *adapter) 2041 { 2042 if (adapter->hw.phy.media_type == e1000_media_type_copper) 2043 igb_power_down_phy_copper_82575(&adapter->hw); 2044 else 2045 igb_shutdown_serdes_link_82575(&adapter->hw); 2046 } 2047 2048 /** 2049 * igb_check_swap_media - Detect and switch function for Media Auto Sense 2050 * @adapter: address of the board private structure 2051 **/ 2052 static void igb_check_swap_media(struct igb_adapter *adapter) 2053 { 2054 struct e1000_hw *hw = &adapter->hw; 2055 u32 ctrl_ext, connsw; 2056 bool swap_now = false; 2057 2058 ctrl_ext = rd32(E1000_CTRL_EXT); 2059 connsw = rd32(E1000_CONNSW); 2060 2061 /* need to live swap if current media is copper and we have fiber/serdes 2062 * to go to. 2063 */ 2064 2065 if ((hw->phy.media_type == e1000_media_type_copper) && 2066 (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) { 2067 swap_now = true; 2068 } else if ((hw->phy.media_type != e1000_media_type_copper) && 2069 !(connsw & E1000_CONNSW_SERDESD)) { 2070 /* copper signal takes time to appear */ 2071 if (adapter->copper_tries < 4) { 2072 adapter->copper_tries++; 2073 connsw |= E1000_CONNSW_AUTOSENSE_CONF; 2074 wr32(E1000_CONNSW, connsw); 2075 return; 2076 } else { 2077 adapter->copper_tries = 0; 2078 if ((connsw & E1000_CONNSW_PHYSD) && 2079 (!(connsw & E1000_CONNSW_PHY_PDN))) { 2080 swap_now = true; 2081 connsw &= ~E1000_CONNSW_AUTOSENSE_CONF; 2082 wr32(E1000_CONNSW, connsw); 2083 } 2084 } 2085 } 2086 2087 if (!swap_now) 2088 return; 2089 2090 switch (hw->phy.media_type) { 2091 case e1000_media_type_copper: 2092 netdev_info(adapter->netdev, 2093 "MAS: changing media to fiber/serdes\n"); 2094 ctrl_ext |= 2095 E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES; 2096 adapter->flags |= IGB_FLAG_MEDIA_RESET; 2097 adapter->copper_tries = 0; 2098 break; 2099 case e1000_media_type_internal_serdes: 2100 case e1000_media_type_fiber: 2101 netdev_info(adapter->netdev, 2102 "MAS: changing media to copper\n"); 2103 ctrl_ext &= 2104 ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES; 2105 adapter->flags |= IGB_FLAG_MEDIA_RESET; 2106 break; 2107 default: 2108 /* shouldn't get here during regular operation */ 2109 netdev_err(adapter->netdev, 2110 "AMS: Invalid media type found, returning\n"); 2111 break; 2112 } 2113 wr32(E1000_CTRL_EXT, ctrl_ext); 2114 } 2115 2116 /** 2117 * igb_up - Open the interface and prepare it to handle traffic 2118 * @adapter: board private structure 2119 **/ 2120 int igb_up(struct igb_adapter *adapter) 2121 { 2122 struct e1000_hw *hw = &adapter->hw; 2123 int i; 2124 2125 /* hardware has been reset, we need to reload some things */ 2126 igb_configure(adapter); 2127 2128 clear_bit(__IGB_DOWN, &adapter->state); 2129 2130 for (i = 0; i < adapter->num_q_vectors; i++) 2131 napi_enable(&(adapter->q_vector[i]->napi)); 2132 2133 if (adapter->flags & IGB_FLAG_HAS_MSIX) 2134 igb_configure_msix(adapter); 2135 else 2136 igb_assign_vector(adapter->q_vector[0], 0); 2137 2138 /* Clear any pending interrupts. */ 2139 rd32(E1000_TSICR); 2140 rd32(E1000_ICR); 2141 igb_irq_enable(adapter); 2142 2143 /* notify VFs that reset has been completed */ 2144 if (adapter->vfs_allocated_count) { 2145 u32 reg_data = rd32(E1000_CTRL_EXT); 2146 2147 reg_data |= E1000_CTRL_EXT_PFRSTD; 2148 wr32(E1000_CTRL_EXT, reg_data); 2149 } 2150 2151 netif_tx_start_all_queues(adapter->netdev); 2152 2153 /* start the watchdog. */ 2154 hw->mac.get_link_status = 1; 2155 schedule_work(&adapter->watchdog_task); 2156 2157 if ((adapter->flags & IGB_FLAG_EEE) && 2158 (!hw->dev_spec._82575.eee_disable)) 2159 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T; 2160 2161 return 0; 2162 } 2163 2164 void igb_down(struct igb_adapter *adapter) 2165 { 2166 struct net_device *netdev = adapter->netdev; 2167 struct e1000_hw *hw = &adapter->hw; 2168 u32 tctl, rctl; 2169 int i; 2170 2171 /* signal that we're down so the interrupt handler does not 2172 * reschedule our watchdog timer 2173 */ 2174 set_bit(__IGB_DOWN, &adapter->state); 2175 2176 /* disable receives in the hardware */ 2177 rctl = rd32(E1000_RCTL); 2178 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN); 2179 /* flush and sleep below */ 2180 2181 igb_nfc_filter_exit(adapter); 2182 2183 netif_carrier_off(netdev); 2184 netif_tx_stop_all_queues(netdev); 2185 2186 /* disable transmits in the hardware */ 2187 tctl = rd32(E1000_TCTL); 2188 tctl &= ~E1000_TCTL_EN; 2189 wr32(E1000_TCTL, tctl); 2190 /* flush both disables and wait for them to finish */ 2191 wrfl(); 2192 usleep_range(10000, 11000); 2193 2194 igb_irq_disable(adapter); 2195 2196 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE; 2197 2198 for (i = 0; i < adapter->num_q_vectors; i++) { 2199 if (adapter->q_vector[i]) { 2200 napi_synchronize(&adapter->q_vector[i]->napi); 2201 napi_disable(&adapter->q_vector[i]->napi); 2202 } 2203 } 2204 2205 del_timer_sync(&adapter->watchdog_timer); 2206 del_timer_sync(&adapter->phy_info_timer); 2207 2208 /* record the stats before reset*/ 2209 spin_lock(&adapter->stats64_lock); 2210 igb_update_stats(adapter); 2211 spin_unlock(&adapter->stats64_lock); 2212 2213 adapter->link_speed = 0; 2214 adapter->link_duplex = 0; 2215 2216 if (!pci_channel_offline(adapter->pdev)) 2217 igb_reset(adapter); 2218 2219 /* clear VLAN promisc flag so VFTA will be updated if necessary */ 2220 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC; 2221 2222 igb_clean_all_tx_rings(adapter); 2223 igb_clean_all_rx_rings(adapter); 2224 #ifdef CONFIG_IGB_DCA 2225 2226 /* since we reset the hardware DCA settings were cleared */ 2227 igb_setup_dca(adapter); 2228 #endif 2229 } 2230 2231 void igb_reinit_locked(struct igb_adapter *adapter) 2232 { 2233 while (test_and_set_bit(__IGB_RESETTING, &adapter->state)) 2234 usleep_range(1000, 2000); 2235 igb_down(adapter); 2236 igb_up(adapter); 2237 clear_bit(__IGB_RESETTING, &adapter->state); 2238 } 2239 2240 /** igb_enable_mas - Media Autosense re-enable after swap 2241 * 2242 * @adapter: adapter struct 2243 **/ 2244 static void igb_enable_mas(struct igb_adapter *adapter) 2245 { 2246 struct e1000_hw *hw = &adapter->hw; 2247 u32 connsw = rd32(E1000_CONNSW); 2248 2249 /* configure for SerDes media detect */ 2250 if ((hw->phy.media_type == e1000_media_type_copper) && 2251 (!(connsw & E1000_CONNSW_SERDESD))) { 2252 connsw |= E1000_CONNSW_ENRGSRC; 2253 connsw |= E1000_CONNSW_AUTOSENSE_EN; 2254 wr32(E1000_CONNSW, connsw); 2255 wrfl(); 2256 } 2257 } 2258 2259 #ifdef CONFIG_IGB_HWMON 2260 /** 2261 * igb_set_i2c_bb - Init I2C interface 2262 * @hw: pointer to hardware structure 2263 **/ 2264 static void igb_set_i2c_bb(struct e1000_hw *hw) 2265 { 2266 u32 ctrl_ext; 2267 s32 i2cctl; 2268 2269 ctrl_ext = rd32(E1000_CTRL_EXT); 2270 ctrl_ext |= E1000_CTRL_I2C_ENA; 2271 wr32(E1000_CTRL_EXT, ctrl_ext); 2272 wrfl(); 2273 2274 i2cctl = rd32(E1000_I2CPARAMS); 2275 i2cctl |= E1000_I2CBB_EN 2276 | E1000_I2C_CLK_OE_N 2277 | E1000_I2C_DATA_OE_N; 2278 wr32(E1000_I2CPARAMS, i2cctl); 2279 wrfl(); 2280 } 2281 #endif 2282 2283 void igb_reset(struct igb_adapter *adapter) 2284 { 2285 struct pci_dev *pdev = adapter->pdev; 2286 struct e1000_hw *hw = &adapter->hw; 2287 struct e1000_mac_info *mac = &hw->mac; 2288 struct e1000_fc_info *fc = &hw->fc; 2289 u32 pba, hwm; 2290 2291 /* Repartition Pba for greater than 9k mtu 2292 * To take effect CTRL.RST is required. 2293 */ 2294 switch (mac->type) { 2295 case e1000_i350: 2296 case e1000_i354: 2297 case e1000_82580: 2298 pba = rd32(E1000_RXPBS); 2299 pba = igb_rxpbs_adjust_82580(pba); 2300 break; 2301 case e1000_82576: 2302 pba = rd32(E1000_RXPBS); 2303 pba &= E1000_RXPBS_SIZE_MASK_82576; 2304 break; 2305 case e1000_82575: 2306 case e1000_i210: 2307 case e1000_i211: 2308 default: 2309 pba = E1000_PBA_34K; 2310 break; 2311 } 2312 2313 if (mac->type == e1000_82575) { 2314 u32 min_rx_space, min_tx_space, needed_tx_space; 2315 2316 /* write Rx PBA so that hardware can report correct Tx PBA */ 2317 wr32(E1000_PBA, pba); 2318 2319 /* To maintain wire speed transmits, the Tx FIFO should be 2320 * large enough to accommodate two full transmit packets, 2321 * rounded up to the next 1KB and expressed in KB. Likewise, 2322 * the Rx FIFO should be large enough to accommodate at least 2323 * one full receive packet and is similarly rounded up and 2324 * expressed in KB. 2325 */ 2326 min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024); 2327 2328 /* The Tx FIFO also stores 16 bytes of information about the Tx 2329 * but don't include Ethernet FCS because hardware appends it. 2330 * We only need to round down to the nearest 512 byte block 2331 * count since the value we care about is 2 frames, not 1. 2332 */ 2333 min_tx_space = adapter->max_frame_size; 2334 min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN; 2335 min_tx_space = DIV_ROUND_UP(min_tx_space, 512); 2336 2337 /* upper 16 bits has Tx packet buffer allocation size in KB */ 2338 needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16); 2339 2340 /* If current Tx allocation is less than the min Tx FIFO size, 2341 * and the min Tx FIFO size is less than the current Rx FIFO 2342 * allocation, take space away from current Rx allocation. 2343 */ 2344 if (needed_tx_space < pba) { 2345 pba -= needed_tx_space; 2346 2347 /* if short on Rx space, Rx wins and must trump Tx 2348 * adjustment 2349 */ 2350 if (pba < min_rx_space) 2351 pba = min_rx_space; 2352 } 2353 2354 /* adjust PBA for jumbo frames */ 2355 wr32(E1000_PBA, pba); 2356 } 2357 2358 /* flow control settings 2359 * The high water mark must be low enough to fit one full frame 2360 * after transmitting the pause frame. As such we must have enough 2361 * space to allow for us to complete our current transmit and then 2362 * receive the frame that is in progress from the link partner. 2363 * Set it to: 2364 * - the full Rx FIFO size minus one full Tx plus one full Rx frame 2365 */ 2366 hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE); 2367 2368 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */ 2369 fc->low_water = fc->high_water - 16; 2370 fc->pause_time = 0xFFFF; 2371 fc->send_xon = 1; 2372 fc->current_mode = fc->requested_mode; 2373 2374 /* disable receive for all VFs and wait one second */ 2375 if (adapter->vfs_allocated_count) { 2376 int i; 2377 2378 for (i = 0 ; i < adapter->vfs_allocated_count; i++) 2379 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC; 2380 2381 /* ping all the active vfs to let them know we are going down */ 2382 igb_ping_all_vfs(adapter); 2383 2384 /* disable transmits and receives */ 2385 wr32(E1000_VFRE, 0); 2386 wr32(E1000_VFTE, 0); 2387 } 2388 2389 /* Allow time for pending master requests to run */ 2390 hw->mac.ops.reset_hw(hw); 2391 wr32(E1000_WUC, 0); 2392 2393 if (adapter->flags & IGB_FLAG_MEDIA_RESET) { 2394 /* need to resetup here after media swap */ 2395 adapter->ei.get_invariants(hw); 2396 adapter->flags &= ~IGB_FLAG_MEDIA_RESET; 2397 } 2398 if ((mac->type == e1000_82575 || mac->type == e1000_i350) && 2399 (adapter->flags & IGB_FLAG_MAS_ENABLE)) { 2400 igb_enable_mas(adapter); 2401 } 2402 if (hw->mac.ops.init_hw(hw)) 2403 dev_err(&pdev->dev, "Hardware Error\n"); 2404 2405 /* RAR registers were cleared during init_hw, clear mac table */ 2406 igb_flush_mac_table(adapter); 2407 __dev_uc_unsync(adapter->netdev, NULL); 2408 2409 /* Recover default RAR entry */ 2410 igb_set_default_mac_filter(adapter); 2411 2412 /* Flow control settings reset on hardware reset, so guarantee flow 2413 * control is off when forcing speed. 2414 */ 2415 if (!hw->mac.autoneg) 2416 igb_force_mac_fc(hw); 2417 2418 igb_init_dmac(adapter, pba); 2419 #ifdef CONFIG_IGB_HWMON 2420 /* Re-initialize the thermal sensor on i350 devices. */ 2421 if (!test_bit(__IGB_DOWN, &adapter->state)) { 2422 if (mac->type == e1000_i350 && hw->bus.func == 0) { 2423 /* If present, re-initialize the external thermal sensor 2424 * interface. 2425 */ 2426 if (adapter->ets) 2427 igb_set_i2c_bb(hw); 2428 mac->ops.init_thermal_sensor_thresh(hw); 2429 } 2430 } 2431 #endif 2432 /* Re-establish EEE setting */ 2433 if (hw->phy.media_type == e1000_media_type_copper) { 2434 switch (mac->type) { 2435 case e1000_i350: 2436 case e1000_i210: 2437 case e1000_i211: 2438 igb_set_eee_i350(hw, true, true); 2439 break; 2440 case e1000_i354: 2441 igb_set_eee_i354(hw, true, true); 2442 break; 2443 default: 2444 break; 2445 } 2446 } 2447 if (!netif_running(adapter->netdev)) 2448 igb_power_down_link(adapter); 2449 2450 igb_update_mng_vlan(adapter); 2451 2452 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */ 2453 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE); 2454 2455 /* Re-enable PTP, where applicable. */ 2456 if (adapter->ptp_flags & IGB_PTP_ENABLED) 2457 igb_ptp_reset(adapter); 2458 2459 igb_get_phy_info(hw); 2460 } 2461 2462 static netdev_features_t igb_fix_features(struct net_device *netdev, 2463 netdev_features_t features) 2464 { 2465 /* Since there is no support for separate Rx/Tx vlan accel 2466 * enable/disable make sure Tx flag is always in same state as Rx. 2467 */ 2468 if (features & NETIF_F_HW_VLAN_CTAG_RX) 2469 features |= NETIF_F_HW_VLAN_CTAG_TX; 2470 else 2471 features &= ~NETIF_F_HW_VLAN_CTAG_TX; 2472 2473 return features; 2474 } 2475 2476 static int igb_set_features(struct net_device *netdev, 2477 netdev_features_t features) 2478 { 2479 netdev_features_t changed = netdev->features ^ features; 2480 struct igb_adapter *adapter = netdev_priv(netdev); 2481 2482 if (changed & NETIF_F_HW_VLAN_CTAG_RX) 2483 igb_vlan_mode(netdev, features); 2484 2485 if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE))) 2486 return 0; 2487 2488 if (!(features & NETIF_F_NTUPLE)) { 2489 struct hlist_node *node2; 2490 struct igb_nfc_filter *rule; 2491 2492 spin_lock(&adapter->nfc_lock); 2493 hlist_for_each_entry_safe(rule, node2, 2494 &adapter->nfc_filter_list, nfc_node) { 2495 igb_erase_filter(adapter, rule); 2496 hlist_del(&rule->nfc_node); 2497 kfree(rule); 2498 } 2499 spin_unlock(&adapter->nfc_lock); 2500 adapter->nfc_filter_count = 0; 2501 } 2502 2503 netdev->features = features; 2504 2505 if (netif_running(netdev)) 2506 igb_reinit_locked(adapter); 2507 else 2508 igb_reset(adapter); 2509 2510 return 1; 2511 } 2512 2513 static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[], 2514 struct net_device *dev, 2515 const unsigned char *addr, u16 vid, 2516 u16 flags, 2517 struct netlink_ext_ack *extack) 2518 { 2519 /* guarantee we can provide a unique filter for the unicast address */ 2520 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) { 2521 struct igb_adapter *adapter = netdev_priv(dev); 2522 int vfn = adapter->vfs_allocated_count; 2523 2524 if (netdev_uc_count(dev) >= igb_available_rars(adapter, vfn)) 2525 return -ENOMEM; 2526 } 2527 2528 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags); 2529 } 2530 2531 #define IGB_MAX_MAC_HDR_LEN 127 2532 #define IGB_MAX_NETWORK_HDR_LEN 511 2533 2534 static netdev_features_t 2535 igb_features_check(struct sk_buff *skb, struct net_device *dev, 2536 netdev_features_t features) 2537 { 2538 unsigned int network_hdr_len, mac_hdr_len; 2539 2540 /* Make certain the headers can be described by a context descriptor */ 2541 mac_hdr_len = skb_network_header(skb) - skb->data; 2542 if (unlikely(mac_hdr_len > IGB_MAX_MAC_HDR_LEN)) 2543 return features & ~(NETIF_F_HW_CSUM | 2544 NETIF_F_SCTP_CRC | 2545 NETIF_F_GSO_UDP_L4 | 2546 NETIF_F_HW_VLAN_CTAG_TX | 2547 NETIF_F_TSO | 2548 NETIF_F_TSO6); 2549 2550 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb); 2551 if (unlikely(network_hdr_len > IGB_MAX_NETWORK_HDR_LEN)) 2552 return features & ~(NETIF_F_HW_CSUM | 2553 NETIF_F_SCTP_CRC | 2554 NETIF_F_GSO_UDP_L4 | 2555 NETIF_F_TSO | 2556 NETIF_F_TSO6); 2557 2558 /* We can only support IPV4 TSO in tunnels if we can mangle the 2559 * inner IP ID field, so strip TSO if MANGLEID is not supported. 2560 */ 2561 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) 2562 features &= ~NETIF_F_TSO; 2563 2564 return features; 2565 } 2566 2567 static void igb_offload_apply(struct igb_adapter *adapter, s32 queue) 2568 { 2569 if (!is_fqtss_enabled(adapter)) { 2570 enable_fqtss(adapter, true); 2571 return; 2572 } 2573 2574 igb_config_tx_modes(adapter, queue); 2575 2576 if (!is_any_cbs_enabled(adapter) && !is_any_txtime_enabled(adapter)) 2577 enable_fqtss(adapter, false); 2578 } 2579 2580 static int igb_offload_cbs(struct igb_adapter *adapter, 2581 struct tc_cbs_qopt_offload *qopt) 2582 { 2583 struct e1000_hw *hw = &adapter->hw; 2584 int err; 2585 2586 /* CBS offloading is only supported by i210 controller. */ 2587 if (hw->mac.type != e1000_i210) 2588 return -EOPNOTSUPP; 2589 2590 /* CBS offloading is only supported by queue 0 and queue 1. */ 2591 if (qopt->queue < 0 || qopt->queue > 1) 2592 return -EINVAL; 2593 2594 err = igb_save_cbs_params(adapter, qopt->queue, qopt->enable, 2595 qopt->idleslope, qopt->sendslope, 2596 qopt->hicredit, qopt->locredit); 2597 if (err) 2598 return err; 2599 2600 igb_offload_apply(adapter, qopt->queue); 2601 2602 return 0; 2603 } 2604 2605 #define ETHER_TYPE_FULL_MASK ((__force __be16)~0) 2606 #define VLAN_PRIO_FULL_MASK (0x07) 2607 2608 static int igb_parse_cls_flower(struct igb_adapter *adapter, 2609 struct flow_cls_offload *f, 2610 int traffic_class, 2611 struct igb_nfc_filter *input) 2612 { 2613 struct flow_rule *rule = flow_cls_offload_flow_rule(f); 2614 struct flow_dissector *dissector = rule->match.dissector; 2615 struct netlink_ext_ack *extack = f->common.extack; 2616 2617 if (dissector->used_keys & 2618 ~(BIT_ULL(FLOW_DISSECTOR_KEY_BASIC) | 2619 BIT_ULL(FLOW_DISSECTOR_KEY_CONTROL) | 2620 BIT_ULL(FLOW_DISSECTOR_KEY_ETH_ADDRS) | 2621 BIT_ULL(FLOW_DISSECTOR_KEY_VLAN))) { 2622 NL_SET_ERR_MSG_MOD(extack, 2623 "Unsupported key used, only BASIC, CONTROL, ETH_ADDRS and VLAN are supported"); 2624 return -EOPNOTSUPP; 2625 } 2626 2627 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) { 2628 struct flow_match_eth_addrs match; 2629 2630 flow_rule_match_eth_addrs(rule, &match); 2631 if (!is_zero_ether_addr(match.mask->dst)) { 2632 if (!is_broadcast_ether_addr(match.mask->dst)) { 2633 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for destination MAC address"); 2634 return -EINVAL; 2635 } 2636 2637 input->filter.match_flags |= 2638 IGB_FILTER_FLAG_DST_MAC_ADDR; 2639 ether_addr_copy(input->filter.dst_addr, match.key->dst); 2640 } 2641 2642 if (!is_zero_ether_addr(match.mask->src)) { 2643 if (!is_broadcast_ether_addr(match.mask->src)) { 2644 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for source MAC address"); 2645 return -EINVAL; 2646 } 2647 2648 input->filter.match_flags |= 2649 IGB_FILTER_FLAG_SRC_MAC_ADDR; 2650 ether_addr_copy(input->filter.src_addr, match.key->src); 2651 } 2652 } 2653 2654 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) { 2655 struct flow_match_basic match; 2656 2657 flow_rule_match_basic(rule, &match); 2658 if (match.mask->n_proto) { 2659 if (match.mask->n_proto != ETHER_TYPE_FULL_MASK) { 2660 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for EtherType filter"); 2661 return -EINVAL; 2662 } 2663 2664 input->filter.match_flags |= IGB_FILTER_FLAG_ETHER_TYPE; 2665 input->filter.etype = match.key->n_proto; 2666 } 2667 } 2668 2669 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) { 2670 struct flow_match_vlan match; 2671 2672 flow_rule_match_vlan(rule, &match); 2673 if (match.mask->vlan_priority) { 2674 if (match.mask->vlan_priority != VLAN_PRIO_FULL_MASK) { 2675 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for VLAN priority"); 2676 return -EINVAL; 2677 } 2678 2679 input->filter.match_flags |= IGB_FILTER_FLAG_VLAN_TCI; 2680 input->filter.vlan_tci = 2681 (__force __be16)match.key->vlan_priority; 2682 } 2683 } 2684 2685 input->action = traffic_class; 2686 input->cookie = f->cookie; 2687 2688 return 0; 2689 } 2690 2691 static int igb_configure_clsflower(struct igb_adapter *adapter, 2692 struct flow_cls_offload *cls_flower) 2693 { 2694 struct netlink_ext_ack *extack = cls_flower->common.extack; 2695 struct igb_nfc_filter *filter, *f; 2696 int err, tc; 2697 2698 tc = tc_classid_to_hwtc(adapter->netdev, cls_flower->classid); 2699 if (tc < 0) { 2700 NL_SET_ERR_MSG_MOD(extack, "Invalid traffic class"); 2701 return -EINVAL; 2702 } 2703 2704 filter = kzalloc(sizeof(*filter), GFP_KERNEL); 2705 if (!filter) 2706 return -ENOMEM; 2707 2708 err = igb_parse_cls_flower(adapter, cls_flower, tc, filter); 2709 if (err < 0) 2710 goto err_parse; 2711 2712 spin_lock(&adapter->nfc_lock); 2713 2714 hlist_for_each_entry(f, &adapter->nfc_filter_list, nfc_node) { 2715 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) { 2716 err = -EEXIST; 2717 NL_SET_ERR_MSG_MOD(extack, 2718 "This filter is already set in ethtool"); 2719 goto err_locked; 2720 } 2721 } 2722 2723 hlist_for_each_entry(f, &adapter->cls_flower_list, nfc_node) { 2724 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) { 2725 err = -EEXIST; 2726 NL_SET_ERR_MSG_MOD(extack, 2727 "This filter is already set in cls_flower"); 2728 goto err_locked; 2729 } 2730 } 2731 2732 err = igb_add_filter(adapter, filter); 2733 if (err < 0) { 2734 NL_SET_ERR_MSG_MOD(extack, "Could not add filter to the adapter"); 2735 goto err_locked; 2736 } 2737 2738 hlist_add_head(&filter->nfc_node, &adapter->cls_flower_list); 2739 2740 spin_unlock(&adapter->nfc_lock); 2741 2742 return 0; 2743 2744 err_locked: 2745 spin_unlock(&adapter->nfc_lock); 2746 2747 err_parse: 2748 kfree(filter); 2749 2750 return err; 2751 } 2752 2753 static int igb_delete_clsflower(struct igb_adapter *adapter, 2754 struct flow_cls_offload *cls_flower) 2755 { 2756 struct igb_nfc_filter *filter; 2757 int err; 2758 2759 spin_lock(&adapter->nfc_lock); 2760 2761 hlist_for_each_entry(filter, &adapter->cls_flower_list, nfc_node) 2762 if (filter->cookie == cls_flower->cookie) 2763 break; 2764 2765 if (!filter) { 2766 err = -ENOENT; 2767 goto out; 2768 } 2769 2770 err = igb_erase_filter(adapter, filter); 2771 if (err < 0) 2772 goto out; 2773 2774 hlist_del(&filter->nfc_node); 2775 kfree(filter); 2776 2777 out: 2778 spin_unlock(&adapter->nfc_lock); 2779 2780 return err; 2781 } 2782 2783 static int igb_setup_tc_cls_flower(struct igb_adapter *adapter, 2784 struct flow_cls_offload *cls_flower) 2785 { 2786 switch (cls_flower->command) { 2787 case FLOW_CLS_REPLACE: 2788 return igb_configure_clsflower(adapter, cls_flower); 2789 case FLOW_CLS_DESTROY: 2790 return igb_delete_clsflower(adapter, cls_flower); 2791 case FLOW_CLS_STATS: 2792 return -EOPNOTSUPP; 2793 default: 2794 return -EOPNOTSUPP; 2795 } 2796 } 2797 2798 static int igb_setup_tc_block_cb(enum tc_setup_type type, void *type_data, 2799 void *cb_priv) 2800 { 2801 struct igb_adapter *adapter = cb_priv; 2802 2803 if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data)) 2804 return -EOPNOTSUPP; 2805 2806 switch (type) { 2807 case TC_SETUP_CLSFLOWER: 2808 return igb_setup_tc_cls_flower(adapter, type_data); 2809 2810 default: 2811 return -EOPNOTSUPP; 2812 } 2813 } 2814 2815 static int igb_offload_txtime(struct igb_adapter *adapter, 2816 struct tc_etf_qopt_offload *qopt) 2817 { 2818 struct e1000_hw *hw = &adapter->hw; 2819 int err; 2820 2821 /* Launchtime offloading is only supported by i210 controller. */ 2822 if (hw->mac.type != e1000_i210) 2823 return -EOPNOTSUPP; 2824 2825 /* Launchtime offloading is only supported by queues 0 and 1. */ 2826 if (qopt->queue < 0 || qopt->queue > 1) 2827 return -EINVAL; 2828 2829 err = igb_save_txtime_params(adapter, qopt->queue, qopt->enable); 2830 if (err) 2831 return err; 2832 2833 igb_offload_apply(adapter, qopt->queue); 2834 2835 return 0; 2836 } 2837 2838 static int igb_tc_query_caps(struct igb_adapter *adapter, 2839 struct tc_query_caps_base *base) 2840 { 2841 switch (base->type) { 2842 case TC_SETUP_QDISC_TAPRIO: { 2843 struct tc_taprio_caps *caps = base->caps; 2844 2845 caps->broken_mqprio = true; 2846 2847 return 0; 2848 } 2849 default: 2850 return -EOPNOTSUPP; 2851 } 2852 } 2853 2854 static LIST_HEAD(igb_block_cb_list); 2855 2856 static int igb_setup_tc(struct net_device *dev, enum tc_setup_type type, 2857 void *type_data) 2858 { 2859 struct igb_adapter *adapter = netdev_priv(dev); 2860 2861 switch (type) { 2862 case TC_QUERY_CAPS: 2863 return igb_tc_query_caps(adapter, type_data); 2864 case TC_SETUP_QDISC_CBS: 2865 return igb_offload_cbs(adapter, type_data); 2866 case TC_SETUP_BLOCK: 2867 return flow_block_cb_setup_simple(type_data, 2868 &igb_block_cb_list, 2869 igb_setup_tc_block_cb, 2870 adapter, adapter, true); 2871 2872 case TC_SETUP_QDISC_ETF: 2873 return igb_offload_txtime(adapter, type_data); 2874 2875 default: 2876 return -EOPNOTSUPP; 2877 } 2878 } 2879 2880 static int igb_xdp_setup(struct net_device *dev, struct netdev_bpf *bpf) 2881 { 2882 int i, frame_size = dev->mtu + IGB_ETH_PKT_HDR_PAD; 2883 struct igb_adapter *adapter = netdev_priv(dev); 2884 struct bpf_prog *prog = bpf->prog, *old_prog; 2885 bool running = netif_running(dev); 2886 bool need_reset; 2887 2888 /* verify igb ring attributes are sufficient for XDP */ 2889 for (i = 0; i < adapter->num_rx_queues; i++) { 2890 struct igb_ring *ring = adapter->rx_ring[i]; 2891 2892 if (frame_size > igb_rx_bufsz(ring)) { 2893 NL_SET_ERR_MSG_MOD(bpf->extack, 2894 "The RX buffer size is too small for the frame size"); 2895 netdev_warn(dev, "XDP RX buffer size %d is too small for the frame size %d\n", 2896 igb_rx_bufsz(ring), frame_size); 2897 return -EINVAL; 2898 } 2899 } 2900 2901 old_prog = xchg(&adapter->xdp_prog, prog); 2902 need_reset = (!!prog != !!old_prog); 2903 2904 /* device is up and bpf is added/removed, must setup the RX queues */ 2905 if (need_reset && running) { 2906 igb_close(dev); 2907 } else { 2908 for (i = 0; i < adapter->num_rx_queues; i++) 2909 (void)xchg(&adapter->rx_ring[i]->xdp_prog, 2910 adapter->xdp_prog); 2911 } 2912 2913 if (old_prog) 2914 bpf_prog_put(old_prog); 2915 2916 /* bpf is just replaced, RXQ and MTU are already setup */ 2917 if (!need_reset) { 2918 return 0; 2919 } else { 2920 if (prog) 2921 xdp_features_set_redirect_target(dev, true); 2922 else 2923 xdp_features_clear_redirect_target(dev); 2924 } 2925 2926 if (running) 2927 igb_open(dev); 2928 2929 return 0; 2930 } 2931 2932 static int igb_xdp(struct net_device *dev, struct netdev_bpf *xdp) 2933 { 2934 switch (xdp->command) { 2935 case XDP_SETUP_PROG: 2936 return igb_xdp_setup(dev, xdp); 2937 default: 2938 return -EINVAL; 2939 } 2940 } 2941 2942 static void igb_xdp_ring_update_tail(struct igb_ring *ring) 2943 { 2944 /* Force memory writes to complete before letting h/w know there 2945 * are new descriptors to fetch. 2946 */ 2947 wmb(); 2948 writel(ring->next_to_use, ring->tail); 2949 } 2950 2951 static struct igb_ring *igb_xdp_tx_queue_mapping(struct igb_adapter *adapter) 2952 { 2953 unsigned int r_idx = smp_processor_id(); 2954 2955 if (r_idx >= adapter->num_tx_queues) 2956 r_idx = r_idx % adapter->num_tx_queues; 2957 2958 return adapter->tx_ring[r_idx]; 2959 } 2960 2961 static int igb_xdp_xmit_back(struct igb_adapter *adapter, struct xdp_buff *xdp) 2962 { 2963 struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp); 2964 int cpu = smp_processor_id(); 2965 struct igb_ring *tx_ring; 2966 struct netdev_queue *nq; 2967 u32 ret; 2968 2969 if (unlikely(!xdpf)) 2970 return IGB_XDP_CONSUMED; 2971 2972 /* During program transitions its possible adapter->xdp_prog is assigned 2973 * but ring has not been configured yet. In this case simply abort xmit. 2974 */ 2975 tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL; 2976 if (unlikely(!tx_ring)) 2977 return IGB_XDP_CONSUMED; 2978 2979 nq = txring_txq(tx_ring); 2980 __netif_tx_lock(nq, cpu); 2981 /* Avoid transmit queue timeout since we share it with the slow path */ 2982 txq_trans_cond_update(nq); 2983 ret = igb_xmit_xdp_ring(adapter, tx_ring, xdpf); 2984 __netif_tx_unlock(nq); 2985 2986 return ret; 2987 } 2988 2989 static int igb_xdp_xmit(struct net_device *dev, int n, 2990 struct xdp_frame **frames, u32 flags) 2991 { 2992 struct igb_adapter *adapter = netdev_priv(dev); 2993 int cpu = smp_processor_id(); 2994 struct igb_ring *tx_ring; 2995 struct netdev_queue *nq; 2996 int nxmit = 0; 2997 int i; 2998 2999 if (unlikely(test_bit(__IGB_DOWN, &adapter->state))) 3000 return -ENETDOWN; 3001 3002 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) 3003 return -EINVAL; 3004 3005 /* During program transitions its possible adapter->xdp_prog is assigned 3006 * but ring has not been configured yet. In this case simply abort xmit. 3007 */ 3008 tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL; 3009 if (unlikely(!tx_ring)) 3010 return -ENXIO; 3011 3012 nq = txring_txq(tx_ring); 3013 __netif_tx_lock(nq, cpu); 3014 3015 /* Avoid transmit queue timeout since we share it with the slow path */ 3016 txq_trans_cond_update(nq); 3017 3018 for (i = 0; i < n; i++) { 3019 struct xdp_frame *xdpf = frames[i]; 3020 int err; 3021 3022 err = igb_xmit_xdp_ring(adapter, tx_ring, xdpf); 3023 if (err != IGB_XDP_TX) 3024 break; 3025 nxmit++; 3026 } 3027 3028 __netif_tx_unlock(nq); 3029 3030 if (unlikely(flags & XDP_XMIT_FLUSH)) 3031 igb_xdp_ring_update_tail(tx_ring); 3032 3033 return nxmit; 3034 } 3035 3036 static const struct net_device_ops igb_netdev_ops = { 3037 .ndo_open = igb_open, 3038 .ndo_stop = igb_close, 3039 .ndo_start_xmit = igb_xmit_frame, 3040 .ndo_get_stats64 = igb_get_stats64, 3041 .ndo_set_rx_mode = igb_set_rx_mode, 3042 .ndo_set_mac_address = igb_set_mac, 3043 .ndo_change_mtu = igb_change_mtu, 3044 .ndo_eth_ioctl = igb_ioctl, 3045 .ndo_tx_timeout = igb_tx_timeout, 3046 .ndo_validate_addr = eth_validate_addr, 3047 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid, 3048 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid, 3049 .ndo_set_vf_mac = igb_ndo_set_vf_mac, 3050 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan, 3051 .ndo_set_vf_rate = igb_ndo_set_vf_bw, 3052 .ndo_set_vf_spoofchk = igb_ndo_set_vf_spoofchk, 3053 .ndo_set_vf_trust = igb_ndo_set_vf_trust, 3054 .ndo_get_vf_config = igb_ndo_get_vf_config, 3055 .ndo_fix_features = igb_fix_features, 3056 .ndo_set_features = igb_set_features, 3057 .ndo_fdb_add = igb_ndo_fdb_add, 3058 .ndo_features_check = igb_features_check, 3059 .ndo_setup_tc = igb_setup_tc, 3060 .ndo_bpf = igb_xdp, 3061 .ndo_xdp_xmit = igb_xdp_xmit, 3062 }; 3063 3064 /** 3065 * igb_set_fw_version - Configure version string for ethtool 3066 * @adapter: adapter struct 3067 **/ 3068 void igb_set_fw_version(struct igb_adapter *adapter) 3069 { 3070 struct e1000_hw *hw = &adapter->hw; 3071 struct e1000_fw_version fw; 3072 3073 igb_get_fw_version(hw, &fw); 3074 3075 switch (hw->mac.type) { 3076 case e1000_i210: 3077 case e1000_i211: 3078 if (!(igb_get_flash_presence_i210(hw))) { 3079 snprintf(adapter->fw_version, 3080 sizeof(adapter->fw_version), 3081 "%2d.%2d-%d", 3082 fw.invm_major, fw.invm_minor, 3083 fw.invm_img_type); 3084 break; 3085 } 3086 fallthrough; 3087 default: 3088 /* if option is rom valid, display its version too */ 3089 if (fw.or_valid) { 3090 snprintf(adapter->fw_version, 3091 sizeof(adapter->fw_version), 3092 "%d.%d, 0x%08x, %d.%d.%d", 3093 fw.eep_major, fw.eep_minor, fw.etrack_id, 3094 fw.or_major, fw.or_build, fw.or_patch); 3095 /* no option rom */ 3096 } else if (fw.etrack_id != 0X0000) { 3097 snprintf(adapter->fw_version, 3098 sizeof(adapter->fw_version), 3099 "%d.%d, 0x%08x", 3100 fw.eep_major, fw.eep_minor, fw.etrack_id); 3101 } else { 3102 snprintf(adapter->fw_version, 3103 sizeof(adapter->fw_version), 3104 "%d.%d.%d", 3105 fw.eep_major, fw.eep_minor, fw.eep_build); 3106 } 3107 break; 3108 } 3109 } 3110 3111 /** 3112 * igb_init_mas - init Media Autosense feature if enabled in the NVM 3113 * 3114 * @adapter: adapter struct 3115 **/ 3116 static void igb_init_mas(struct igb_adapter *adapter) 3117 { 3118 struct e1000_hw *hw = &adapter->hw; 3119 u16 eeprom_data; 3120 3121 hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data); 3122 switch (hw->bus.func) { 3123 case E1000_FUNC_0: 3124 if (eeprom_data & IGB_MAS_ENABLE_0) { 3125 adapter->flags |= IGB_FLAG_MAS_ENABLE; 3126 netdev_info(adapter->netdev, 3127 "MAS: Enabling Media Autosense for port %d\n", 3128 hw->bus.func); 3129 } 3130 break; 3131 case E1000_FUNC_1: 3132 if (eeprom_data & IGB_MAS_ENABLE_1) { 3133 adapter->flags |= IGB_FLAG_MAS_ENABLE; 3134 netdev_info(adapter->netdev, 3135 "MAS: Enabling Media Autosense for port %d\n", 3136 hw->bus.func); 3137 } 3138 break; 3139 case E1000_FUNC_2: 3140 if (eeprom_data & IGB_MAS_ENABLE_2) { 3141 adapter->flags |= IGB_FLAG_MAS_ENABLE; 3142 netdev_info(adapter->netdev, 3143 "MAS: Enabling Media Autosense for port %d\n", 3144 hw->bus.func); 3145 } 3146 break; 3147 case E1000_FUNC_3: 3148 if (eeprom_data & IGB_MAS_ENABLE_3) { 3149 adapter->flags |= IGB_FLAG_MAS_ENABLE; 3150 netdev_info(adapter->netdev, 3151 "MAS: Enabling Media Autosense for port %d\n", 3152 hw->bus.func); 3153 } 3154 break; 3155 default: 3156 /* Shouldn't get here */ 3157 netdev_err(adapter->netdev, 3158 "MAS: Invalid port configuration, returning\n"); 3159 break; 3160 } 3161 } 3162 3163 /** 3164 * igb_init_i2c - Init I2C interface 3165 * @adapter: pointer to adapter structure 3166 **/ 3167 static s32 igb_init_i2c(struct igb_adapter *adapter) 3168 { 3169 s32 status = 0; 3170 3171 /* I2C interface supported on i350 devices */ 3172 if (adapter->hw.mac.type != e1000_i350) 3173 return 0; 3174 3175 /* Initialize the i2c bus which is controlled by the registers. 3176 * This bus will use the i2c_algo_bit structure that implements 3177 * the protocol through toggling of the 4 bits in the register. 3178 */ 3179 adapter->i2c_adap.owner = THIS_MODULE; 3180 adapter->i2c_algo = igb_i2c_algo; 3181 adapter->i2c_algo.data = adapter; 3182 adapter->i2c_adap.algo_data = &adapter->i2c_algo; 3183 adapter->i2c_adap.dev.parent = &adapter->pdev->dev; 3184 strscpy(adapter->i2c_adap.name, "igb BB", 3185 sizeof(adapter->i2c_adap.name)); 3186 status = i2c_bit_add_bus(&adapter->i2c_adap); 3187 return status; 3188 } 3189 3190 /** 3191 * igb_probe - Device Initialization Routine 3192 * @pdev: PCI device information struct 3193 * @ent: entry in igb_pci_tbl 3194 * 3195 * Returns 0 on success, negative on failure 3196 * 3197 * igb_probe initializes an adapter identified by a pci_dev structure. 3198 * The OS initialization, configuring of the adapter private structure, 3199 * and a hardware reset occur. 3200 **/ 3201 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 3202 { 3203 struct net_device *netdev; 3204 struct igb_adapter *adapter; 3205 struct e1000_hw *hw; 3206 u16 eeprom_data = 0; 3207 s32 ret_val; 3208 static int global_quad_port_a; /* global quad port a indication */ 3209 const struct e1000_info *ei = igb_info_tbl[ent->driver_data]; 3210 u8 part_str[E1000_PBANUM_LENGTH]; 3211 int err; 3212 3213 /* Catch broken hardware that put the wrong VF device ID in 3214 * the PCIe SR-IOV capability. 3215 */ 3216 if (pdev->is_virtfn) { 3217 WARN(1, KERN_ERR "%s (%x:%x) should not be a VF!\n", 3218 pci_name(pdev), pdev->vendor, pdev->device); 3219 return -EINVAL; 3220 } 3221 3222 err = pci_enable_device_mem(pdev); 3223 if (err) 3224 return err; 3225 3226 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); 3227 if (err) { 3228 dev_err(&pdev->dev, 3229 "No usable DMA configuration, aborting\n"); 3230 goto err_dma; 3231 } 3232 3233 err = pci_request_mem_regions(pdev, igb_driver_name); 3234 if (err) 3235 goto err_pci_reg; 3236 3237 pci_set_master(pdev); 3238 pci_save_state(pdev); 3239 3240 err = -ENOMEM; 3241 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter), 3242 IGB_MAX_TX_QUEUES); 3243 if (!netdev) 3244 goto err_alloc_etherdev; 3245 3246 SET_NETDEV_DEV(netdev, &pdev->dev); 3247 3248 pci_set_drvdata(pdev, netdev); 3249 adapter = netdev_priv(netdev); 3250 adapter->netdev = netdev; 3251 adapter->pdev = pdev; 3252 hw = &adapter->hw; 3253 hw->back = adapter; 3254 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE); 3255 3256 err = -EIO; 3257 adapter->io_addr = pci_iomap(pdev, 0, 0); 3258 if (!adapter->io_addr) 3259 goto err_ioremap; 3260 /* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */ 3261 hw->hw_addr = adapter->io_addr; 3262 3263 netdev->netdev_ops = &igb_netdev_ops; 3264 igb_set_ethtool_ops(netdev); 3265 netdev->watchdog_timeo = 5 * HZ; 3266 3267 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1); 3268 3269 netdev->mem_start = pci_resource_start(pdev, 0); 3270 netdev->mem_end = pci_resource_end(pdev, 0); 3271 3272 /* PCI config space info */ 3273 hw->vendor_id = pdev->vendor; 3274 hw->device_id = pdev->device; 3275 hw->revision_id = pdev->revision; 3276 hw->subsystem_vendor_id = pdev->subsystem_vendor; 3277 hw->subsystem_device_id = pdev->subsystem_device; 3278 3279 /* Copy the default MAC, PHY and NVM function pointers */ 3280 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops)); 3281 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops)); 3282 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops)); 3283 /* Initialize skew-specific constants */ 3284 err = ei->get_invariants(hw); 3285 if (err) 3286 goto err_sw_init; 3287 3288 /* setup the private structure */ 3289 err = igb_sw_init(adapter); 3290 if (err) 3291 goto err_sw_init; 3292 3293 igb_get_bus_info_pcie(hw); 3294 3295 hw->phy.autoneg_wait_to_complete = false; 3296 3297 /* Copper options */ 3298 if (hw->phy.media_type == e1000_media_type_copper) { 3299 hw->phy.mdix = AUTO_ALL_MODES; 3300 hw->phy.disable_polarity_correction = false; 3301 hw->phy.ms_type = e1000_ms_hw_default; 3302 } 3303 3304 if (igb_check_reset_block(hw)) 3305 dev_info(&pdev->dev, 3306 "PHY reset is blocked due to SOL/IDER session.\n"); 3307 3308 /* features is initialized to 0 in allocation, it might have bits 3309 * set by igb_sw_init so we should use an or instead of an 3310 * assignment. 3311 */ 3312 netdev->features |= NETIF_F_SG | 3313 NETIF_F_TSO | 3314 NETIF_F_TSO6 | 3315 NETIF_F_RXHASH | 3316 NETIF_F_RXCSUM | 3317 NETIF_F_HW_CSUM; 3318 3319 if (hw->mac.type >= e1000_82576) 3320 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4; 3321 3322 if (hw->mac.type >= e1000_i350) 3323 netdev->features |= NETIF_F_HW_TC; 3324 3325 #define IGB_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \ 3326 NETIF_F_GSO_GRE_CSUM | \ 3327 NETIF_F_GSO_IPXIP4 | \ 3328 NETIF_F_GSO_IPXIP6 | \ 3329 NETIF_F_GSO_UDP_TUNNEL | \ 3330 NETIF_F_GSO_UDP_TUNNEL_CSUM) 3331 3332 netdev->gso_partial_features = IGB_GSO_PARTIAL_FEATURES; 3333 netdev->features |= NETIF_F_GSO_PARTIAL | IGB_GSO_PARTIAL_FEATURES; 3334 3335 /* copy netdev features into list of user selectable features */ 3336 netdev->hw_features |= netdev->features | 3337 NETIF_F_HW_VLAN_CTAG_RX | 3338 NETIF_F_HW_VLAN_CTAG_TX | 3339 NETIF_F_RXALL; 3340 3341 if (hw->mac.type >= e1000_i350) 3342 netdev->hw_features |= NETIF_F_NTUPLE; 3343 3344 netdev->features |= NETIF_F_HIGHDMA; 3345 3346 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID; 3347 netdev->mpls_features |= NETIF_F_HW_CSUM; 3348 netdev->hw_enc_features |= netdev->vlan_features; 3349 3350 /* set this bit last since it cannot be part of vlan_features */ 3351 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER | 3352 NETIF_F_HW_VLAN_CTAG_RX | 3353 NETIF_F_HW_VLAN_CTAG_TX; 3354 3355 netdev->priv_flags |= IFF_SUPP_NOFCS; 3356 3357 netdev->priv_flags |= IFF_UNICAST_FLT; 3358 netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT; 3359 3360 /* MTU range: 68 - 9216 */ 3361 netdev->min_mtu = ETH_MIN_MTU; 3362 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE; 3363 3364 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw); 3365 3366 /* before reading the NVM, reset the controller to put the device in a 3367 * known good starting state 3368 */ 3369 hw->mac.ops.reset_hw(hw); 3370 3371 /* make sure the NVM is good , i211/i210 parts can have special NVM 3372 * that doesn't contain a checksum 3373 */ 3374 switch (hw->mac.type) { 3375 case e1000_i210: 3376 case e1000_i211: 3377 if (igb_get_flash_presence_i210(hw)) { 3378 if (hw->nvm.ops.validate(hw) < 0) { 3379 dev_err(&pdev->dev, 3380 "The NVM Checksum Is Not Valid\n"); 3381 err = -EIO; 3382 goto err_eeprom; 3383 } 3384 } 3385 break; 3386 default: 3387 if (hw->nvm.ops.validate(hw) < 0) { 3388 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n"); 3389 err = -EIO; 3390 goto err_eeprom; 3391 } 3392 break; 3393 } 3394 3395 if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) { 3396 /* copy the MAC address out of the NVM */ 3397 if (hw->mac.ops.read_mac_addr(hw)) 3398 dev_err(&pdev->dev, "NVM Read Error\n"); 3399 } 3400 3401 eth_hw_addr_set(netdev, hw->mac.addr); 3402 3403 if (!is_valid_ether_addr(netdev->dev_addr)) { 3404 dev_err(&pdev->dev, "Invalid MAC Address\n"); 3405 err = -EIO; 3406 goto err_eeprom; 3407 } 3408 3409 igb_set_default_mac_filter(adapter); 3410 3411 /* get firmware version for ethtool -i */ 3412 igb_set_fw_version(adapter); 3413 3414 /* configure RXPBSIZE and TXPBSIZE */ 3415 if (hw->mac.type == e1000_i210) { 3416 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT); 3417 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT); 3418 } 3419 3420 timer_setup(&adapter->watchdog_timer, igb_watchdog, 0); 3421 timer_setup(&adapter->phy_info_timer, igb_update_phy_info, 0); 3422 3423 INIT_WORK(&adapter->reset_task, igb_reset_task); 3424 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task); 3425 3426 /* Initialize link properties that are user-changeable */ 3427 adapter->fc_autoneg = true; 3428 hw->mac.autoneg = true; 3429 hw->phy.autoneg_advertised = 0x2f; 3430 3431 hw->fc.requested_mode = e1000_fc_default; 3432 hw->fc.current_mode = e1000_fc_default; 3433 3434 igb_validate_mdi_setting(hw); 3435 3436 /* By default, support wake on port A */ 3437 if (hw->bus.func == 0) 3438 adapter->flags |= IGB_FLAG_WOL_SUPPORTED; 3439 3440 /* Check the NVM for wake support on non-port A ports */ 3441 if (hw->mac.type >= e1000_82580) 3442 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A + 3443 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1, 3444 &eeprom_data); 3445 else if (hw->bus.func == 1) 3446 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data); 3447 3448 if (eeprom_data & IGB_EEPROM_APME) 3449 adapter->flags |= IGB_FLAG_WOL_SUPPORTED; 3450 3451 /* now that we have the eeprom settings, apply the special cases where 3452 * the eeprom may be wrong or the board simply won't support wake on 3453 * lan on a particular port 3454 */ 3455 switch (pdev->device) { 3456 case E1000_DEV_ID_82575GB_QUAD_COPPER: 3457 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED; 3458 break; 3459 case E1000_DEV_ID_82575EB_FIBER_SERDES: 3460 case E1000_DEV_ID_82576_FIBER: 3461 case E1000_DEV_ID_82576_SERDES: 3462 /* Wake events only supported on port A for dual fiber 3463 * regardless of eeprom setting 3464 */ 3465 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1) 3466 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED; 3467 break; 3468 case E1000_DEV_ID_82576_QUAD_COPPER: 3469 case E1000_DEV_ID_82576_QUAD_COPPER_ET2: 3470 /* if quad port adapter, disable WoL on all but port A */ 3471 if (global_quad_port_a != 0) 3472 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED; 3473 else 3474 adapter->flags |= IGB_FLAG_QUAD_PORT_A; 3475 /* Reset for multiple quad port adapters */ 3476 if (++global_quad_port_a == 4) 3477 global_quad_port_a = 0; 3478 break; 3479 default: 3480 /* If the device can't wake, don't set software support */ 3481 if (!device_can_wakeup(&adapter->pdev->dev)) 3482 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED; 3483 } 3484 3485 /* initialize the wol settings based on the eeprom settings */ 3486 if (adapter->flags & IGB_FLAG_WOL_SUPPORTED) 3487 adapter->wol |= E1000_WUFC_MAG; 3488 3489 /* Some vendors want WoL disabled by default, but still supported */ 3490 if ((hw->mac.type == e1000_i350) && 3491 (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) { 3492 adapter->flags |= IGB_FLAG_WOL_SUPPORTED; 3493 adapter->wol = 0; 3494 } 3495 3496 /* Some vendors want the ability to Use the EEPROM setting as 3497 * enable/disable only, and not for capability 3498 */ 3499 if (((hw->mac.type == e1000_i350) || 3500 (hw->mac.type == e1000_i354)) && 3501 (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) { 3502 adapter->flags |= IGB_FLAG_WOL_SUPPORTED; 3503 adapter->wol = 0; 3504 } 3505 if (hw->mac.type == e1000_i350) { 3506 if (((pdev->subsystem_device == 0x5001) || 3507 (pdev->subsystem_device == 0x5002)) && 3508 (hw->bus.func == 0)) { 3509 adapter->flags |= IGB_FLAG_WOL_SUPPORTED; 3510 adapter->wol = 0; 3511 } 3512 if (pdev->subsystem_device == 0x1F52) 3513 adapter->flags |= IGB_FLAG_WOL_SUPPORTED; 3514 } 3515 3516 device_set_wakeup_enable(&adapter->pdev->dev, 3517 adapter->flags & IGB_FLAG_WOL_SUPPORTED); 3518 3519 /* reset the hardware with the new settings */ 3520 igb_reset(adapter); 3521 3522 /* Init the I2C interface */ 3523 err = igb_init_i2c(adapter); 3524 if (err) { 3525 dev_err(&pdev->dev, "failed to init i2c interface\n"); 3526 goto err_eeprom; 3527 } 3528 3529 /* let the f/w know that the h/w is now under the control of the 3530 * driver. 3531 */ 3532 igb_get_hw_control(adapter); 3533 3534 strcpy(netdev->name, "eth%d"); 3535 err = register_netdev(netdev); 3536 if (err) 3537 goto err_register; 3538 3539 /* carrier off reporting is important to ethtool even BEFORE open */ 3540 netif_carrier_off(netdev); 3541 3542 #ifdef CONFIG_IGB_DCA 3543 if (dca_add_requester(&pdev->dev) == 0) { 3544 adapter->flags |= IGB_FLAG_DCA_ENABLED; 3545 dev_info(&pdev->dev, "DCA enabled\n"); 3546 igb_setup_dca(adapter); 3547 } 3548 3549 #endif 3550 #ifdef CONFIG_IGB_HWMON 3551 /* Initialize the thermal sensor on i350 devices. */ 3552 if (hw->mac.type == e1000_i350 && hw->bus.func == 0) { 3553 u16 ets_word; 3554 3555 /* Read the NVM to determine if this i350 device supports an 3556 * external thermal sensor. 3557 */ 3558 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word); 3559 if (ets_word != 0x0000 && ets_word != 0xFFFF) 3560 adapter->ets = true; 3561 else 3562 adapter->ets = false; 3563 /* Only enable I2C bit banging if an external thermal 3564 * sensor is supported. 3565 */ 3566 if (adapter->ets) 3567 igb_set_i2c_bb(hw); 3568 hw->mac.ops.init_thermal_sensor_thresh(hw); 3569 if (igb_sysfs_init(adapter)) 3570 dev_err(&pdev->dev, 3571 "failed to allocate sysfs resources\n"); 3572 } else { 3573 adapter->ets = false; 3574 } 3575 #endif 3576 /* Check if Media Autosense is enabled */ 3577 adapter->ei = *ei; 3578 if (hw->dev_spec._82575.mas_capable) 3579 igb_init_mas(adapter); 3580 3581 /* do hw tstamp init after resetting */ 3582 igb_ptp_init(adapter); 3583 3584 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n"); 3585 /* print bus type/speed/width info, not applicable to i354 */ 3586 if (hw->mac.type != e1000_i354) { 3587 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n", 3588 netdev->name, 3589 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" : 3590 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" : 3591 "unknown"), 3592 ((hw->bus.width == e1000_bus_width_pcie_x4) ? 3593 "Width x4" : 3594 (hw->bus.width == e1000_bus_width_pcie_x2) ? 3595 "Width x2" : 3596 (hw->bus.width == e1000_bus_width_pcie_x1) ? 3597 "Width x1" : "unknown"), netdev->dev_addr); 3598 } 3599 3600 if ((hw->mac.type == e1000_82576 && 3601 rd32(E1000_EECD) & E1000_EECD_PRES) || 3602 (hw->mac.type >= e1000_i210 || 3603 igb_get_flash_presence_i210(hw))) { 3604 ret_val = igb_read_part_string(hw, part_str, 3605 E1000_PBANUM_LENGTH); 3606 } else { 3607 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND; 3608 } 3609 3610 if (ret_val) 3611 strcpy(part_str, "Unknown"); 3612 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str); 3613 dev_info(&pdev->dev, 3614 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n", 3615 (adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" : 3616 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy", 3617 adapter->num_rx_queues, adapter->num_tx_queues); 3618 if (hw->phy.media_type == e1000_media_type_copper) { 3619 switch (hw->mac.type) { 3620 case e1000_i350: 3621 case e1000_i210: 3622 case e1000_i211: 3623 /* Enable EEE for internal copper PHY devices */ 3624 err = igb_set_eee_i350(hw, true, true); 3625 if ((!err) && 3626 (!hw->dev_spec._82575.eee_disable)) { 3627 adapter->eee_advert = 3628 MDIO_EEE_100TX | MDIO_EEE_1000T; 3629 adapter->flags |= IGB_FLAG_EEE; 3630 } 3631 break; 3632 case e1000_i354: 3633 if ((rd32(E1000_CTRL_EXT) & 3634 E1000_CTRL_EXT_LINK_MODE_SGMII)) { 3635 err = igb_set_eee_i354(hw, true, true); 3636 if ((!err) && 3637 (!hw->dev_spec._82575.eee_disable)) { 3638 adapter->eee_advert = 3639 MDIO_EEE_100TX | MDIO_EEE_1000T; 3640 adapter->flags |= IGB_FLAG_EEE; 3641 } 3642 } 3643 break; 3644 default: 3645 break; 3646 } 3647 } 3648 3649 dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE); 3650 3651 pm_runtime_put_noidle(&pdev->dev); 3652 return 0; 3653 3654 err_register: 3655 igb_release_hw_control(adapter); 3656 memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap)); 3657 err_eeprom: 3658 if (!igb_check_reset_block(hw)) 3659 igb_reset_phy(hw); 3660 3661 if (hw->flash_address) 3662 iounmap(hw->flash_address); 3663 err_sw_init: 3664 kfree(adapter->mac_table); 3665 kfree(adapter->shadow_vfta); 3666 igb_clear_interrupt_scheme(adapter); 3667 #ifdef CONFIG_PCI_IOV 3668 igb_disable_sriov(pdev, false); 3669 #endif 3670 pci_iounmap(pdev, adapter->io_addr); 3671 err_ioremap: 3672 free_netdev(netdev); 3673 err_alloc_etherdev: 3674 pci_release_mem_regions(pdev); 3675 err_pci_reg: 3676 err_dma: 3677 pci_disable_device(pdev); 3678 return err; 3679 } 3680 3681 #ifdef CONFIG_PCI_IOV 3682 static int igb_sriov_reinit(struct pci_dev *dev) 3683 { 3684 struct net_device *netdev = pci_get_drvdata(dev); 3685 struct igb_adapter *adapter = netdev_priv(netdev); 3686 struct pci_dev *pdev = adapter->pdev; 3687 3688 rtnl_lock(); 3689 3690 if (netif_running(netdev)) 3691 igb_close(netdev); 3692 else 3693 igb_reset(adapter); 3694 3695 igb_clear_interrupt_scheme(adapter); 3696 3697 igb_init_queue_configuration(adapter); 3698 3699 if (igb_init_interrupt_scheme(adapter, true)) { 3700 rtnl_unlock(); 3701 dev_err(&pdev->dev, "Unable to allocate memory for queues\n"); 3702 return -ENOMEM; 3703 } 3704 3705 if (netif_running(netdev)) 3706 igb_open(netdev); 3707 3708 rtnl_unlock(); 3709 3710 return 0; 3711 } 3712 3713 static int igb_disable_sriov(struct pci_dev *pdev, bool reinit) 3714 { 3715 struct net_device *netdev = pci_get_drvdata(pdev); 3716 struct igb_adapter *adapter = netdev_priv(netdev); 3717 struct e1000_hw *hw = &adapter->hw; 3718 unsigned long flags; 3719 3720 /* reclaim resources allocated to VFs */ 3721 if (adapter->vf_data) { 3722 /* disable iov and allow time for transactions to clear */ 3723 if (pci_vfs_assigned(pdev)) { 3724 dev_warn(&pdev->dev, 3725 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n"); 3726 return -EPERM; 3727 } else { 3728 pci_disable_sriov(pdev); 3729 msleep(500); 3730 } 3731 spin_lock_irqsave(&adapter->vfs_lock, flags); 3732 kfree(adapter->vf_mac_list); 3733 adapter->vf_mac_list = NULL; 3734 kfree(adapter->vf_data); 3735 adapter->vf_data = NULL; 3736 adapter->vfs_allocated_count = 0; 3737 spin_unlock_irqrestore(&adapter->vfs_lock, flags); 3738 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ); 3739 wrfl(); 3740 msleep(100); 3741 dev_info(&pdev->dev, "IOV Disabled\n"); 3742 3743 /* Re-enable DMA Coalescing flag since IOV is turned off */ 3744 adapter->flags |= IGB_FLAG_DMAC; 3745 } 3746 3747 return reinit ? igb_sriov_reinit(pdev) : 0; 3748 } 3749 3750 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs, bool reinit) 3751 { 3752 struct net_device *netdev = pci_get_drvdata(pdev); 3753 struct igb_adapter *adapter = netdev_priv(netdev); 3754 int old_vfs = pci_num_vf(pdev); 3755 struct vf_mac_filter *mac_list; 3756 int err = 0; 3757 int num_vf_mac_filters, i; 3758 3759 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) { 3760 err = -EPERM; 3761 goto out; 3762 } 3763 if (!num_vfs) 3764 goto out; 3765 3766 if (old_vfs) { 3767 dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n", 3768 old_vfs, max_vfs); 3769 adapter->vfs_allocated_count = old_vfs; 3770 } else 3771 adapter->vfs_allocated_count = num_vfs; 3772 3773 adapter->vf_data = kcalloc(adapter->vfs_allocated_count, 3774 sizeof(struct vf_data_storage), GFP_KERNEL); 3775 3776 /* if allocation failed then we do not support SR-IOV */ 3777 if (!adapter->vf_data) { 3778 adapter->vfs_allocated_count = 0; 3779 err = -ENOMEM; 3780 goto out; 3781 } 3782 3783 /* Due to the limited number of RAR entries calculate potential 3784 * number of MAC filters available for the VFs. Reserve entries 3785 * for PF default MAC, PF MAC filters and at least one RAR entry 3786 * for each VF for VF MAC. 3787 */ 3788 num_vf_mac_filters = adapter->hw.mac.rar_entry_count - 3789 (1 + IGB_PF_MAC_FILTERS_RESERVED + 3790 adapter->vfs_allocated_count); 3791 3792 adapter->vf_mac_list = kcalloc(num_vf_mac_filters, 3793 sizeof(struct vf_mac_filter), 3794 GFP_KERNEL); 3795 3796 mac_list = adapter->vf_mac_list; 3797 INIT_LIST_HEAD(&adapter->vf_macs.l); 3798 3799 if (adapter->vf_mac_list) { 3800 /* Initialize list of VF MAC filters */ 3801 for (i = 0; i < num_vf_mac_filters; i++) { 3802 mac_list->vf = -1; 3803 mac_list->free = true; 3804 list_add(&mac_list->l, &adapter->vf_macs.l); 3805 mac_list++; 3806 } 3807 } else { 3808 /* If we could not allocate memory for the VF MAC filters 3809 * we can continue without this feature but warn user. 3810 */ 3811 dev_err(&pdev->dev, 3812 "Unable to allocate memory for VF MAC filter list\n"); 3813 } 3814 3815 dev_info(&pdev->dev, "%d VFs allocated\n", 3816 adapter->vfs_allocated_count); 3817 for (i = 0; i < adapter->vfs_allocated_count; i++) 3818 igb_vf_configure(adapter, i); 3819 3820 /* DMA Coalescing is not supported in IOV mode. */ 3821 adapter->flags &= ~IGB_FLAG_DMAC; 3822 3823 if (reinit) { 3824 err = igb_sriov_reinit(pdev); 3825 if (err) 3826 goto err_out; 3827 } 3828 3829 /* only call pci_enable_sriov() if no VFs are allocated already */ 3830 if (!old_vfs) 3831 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count); 3832 3833 goto out; 3834 3835 err_out: 3836 kfree(adapter->vf_mac_list); 3837 adapter->vf_mac_list = NULL; 3838 kfree(adapter->vf_data); 3839 adapter->vf_data = NULL; 3840 adapter->vfs_allocated_count = 0; 3841 out: 3842 return err; 3843 } 3844 3845 #endif 3846 /** 3847 * igb_remove_i2c - Cleanup I2C interface 3848 * @adapter: pointer to adapter structure 3849 **/ 3850 static void igb_remove_i2c(struct igb_adapter *adapter) 3851 { 3852 /* free the adapter bus structure */ 3853 i2c_del_adapter(&adapter->i2c_adap); 3854 } 3855 3856 /** 3857 * igb_remove - Device Removal Routine 3858 * @pdev: PCI device information struct 3859 * 3860 * igb_remove is called by the PCI subsystem to alert the driver 3861 * that it should release a PCI device. The could be caused by a 3862 * Hot-Plug event, or because the driver is going to be removed from 3863 * memory. 3864 **/ 3865 static void igb_remove(struct pci_dev *pdev) 3866 { 3867 struct net_device *netdev = pci_get_drvdata(pdev); 3868 struct igb_adapter *adapter = netdev_priv(netdev); 3869 struct e1000_hw *hw = &adapter->hw; 3870 3871 pm_runtime_get_noresume(&pdev->dev); 3872 #ifdef CONFIG_IGB_HWMON 3873 igb_sysfs_exit(adapter); 3874 #endif 3875 igb_remove_i2c(adapter); 3876 igb_ptp_stop(adapter); 3877 /* The watchdog timer may be rescheduled, so explicitly 3878 * disable watchdog from being rescheduled. 3879 */ 3880 set_bit(__IGB_DOWN, &adapter->state); 3881 del_timer_sync(&adapter->watchdog_timer); 3882 del_timer_sync(&adapter->phy_info_timer); 3883 3884 cancel_work_sync(&adapter->reset_task); 3885 cancel_work_sync(&adapter->watchdog_task); 3886 3887 #ifdef CONFIG_IGB_DCA 3888 if (adapter->flags & IGB_FLAG_DCA_ENABLED) { 3889 dev_info(&pdev->dev, "DCA disabled\n"); 3890 dca_remove_requester(&pdev->dev); 3891 adapter->flags &= ~IGB_FLAG_DCA_ENABLED; 3892 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE); 3893 } 3894 #endif 3895 3896 /* Release control of h/w to f/w. If f/w is AMT enabled, this 3897 * would have already happened in close and is redundant. 3898 */ 3899 igb_release_hw_control(adapter); 3900 3901 #ifdef CONFIG_PCI_IOV 3902 igb_disable_sriov(pdev, false); 3903 #endif 3904 3905 unregister_netdev(netdev); 3906 3907 igb_clear_interrupt_scheme(adapter); 3908 3909 pci_iounmap(pdev, adapter->io_addr); 3910 if (hw->flash_address) 3911 iounmap(hw->flash_address); 3912 pci_release_mem_regions(pdev); 3913 3914 kfree(adapter->mac_table); 3915 kfree(adapter->shadow_vfta); 3916 free_netdev(netdev); 3917 3918 pci_disable_device(pdev); 3919 } 3920 3921 /** 3922 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space 3923 * @adapter: board private structure to initialize 3924 * 3925 * This function initializes the vf specific data storage and then attempts to 3926 * allocate the VFs. The reason for ordering it this way is because it is much 3927 * mor expensive time wise to disable SR-IOV than it is to allocate and free 3928 * the memory for the VFs. 3929 **/ 3930 static void igb_probe_vfs(struct igb_adapter *adapter) 3931 { 3932 #ifdef CONFIG_PCI_IOV 3933 struct pci_dev *pdev = adapter->pdev; 3934 struct e1000_hw *hw = &adapter->hw; 3935 3936 /* Virtualization features not supported on i210 and 82580 family. */ 3937 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211) || 3938 (hw->mac.type == e1000_82580)) 3939 return; 3940 3941 /* Of the below we really only want the effect of getting 3942 * IGB_FLAG_HAS_MSIX set (if available), without which 3943 * igb_enable_sriov() has no effect. 3944 */ 3945 igb_set_interrupt_capability(adapter, true); 3946 igb_reset_interrupt_capability(adapter); 3947 3948 pci_sriov_set_totalvfs(pdev, 7); 3949 igb_enable_sriov(pdev, max_vfs, false); 3950 3951 #endif /* CONFIG_PCI_IOV */ 3952 } 3953 3954 unsigned int igb_get_max_rss_queues(struct igb_adapter *adapter) 3955 { 3956 struct e1000_hw *hw = &adapter->hw; 3957 unsigned int max_rss_queues; 3958 3959 /* Determine the maximum number of RSS queues supported. */ 3960 switch (hw->mac.type) { 3961 case e1000_i211: 3962 max_rss_queues = IGB_MAX_RX_QUEUES_I211; 3963 break; 3964 case e1000_82575: 3965 case e1000_i210: 3966 max_rss_queues = IGB_MAX_RX_QUEUES_82575; 3967 break; 3968 case e1000_i350: 3969 /* I350 cannot do RSS and SR-IOV at the same time */ 3970 if (!!adapter->vfs_allocated_count) { 3971 max_rss_queues = 1; 3972 break; 3973 } 3974 fallthrough; 3975 case e1000_82576: 3976 if (!!adapter->vfs_allocated_count) { 3977 max_rss_queues = 2; 3978 break; 3979 } 3980 fallthrough; 3981 case e1000_82580: 3982 case e1000_i354: 3983 default: 3984 max_rss_queues = IGB_MAX_RX_QUEUES; 3985 break; 3986 } 3987 3988 return max_rss_queues; 3989 } 3990 3991 static void igb_init_queue_configuration(struct igb_adapter *adapter) 3992 { 3993 u32 max_rss_queues; 3994 3995 max_rss_queues = igb_get_max_rss_queues(adapter); 3996 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus()); 3997 3998 igb_set_flag_queue_pairs(adapter, max_rss_queues); 3999 } 4000 4001 void igb_set_flag_queue_pairs(struct igb_adapter *adapter, 4002 const u32 max_rss_queues) 4003 { 4004 struct e1000_hw *hw = &adapter->hw; 4005 4006 /* Determine if we need to pair queues. */ 4007 switch (hw->mac.type) { 4008 case e1000_82575: 4009 case e1000_i211: 4010 /* Device supports enough interrupts without queue pairing. */ 4011 break; 4012 case e1000_82576: 4013 case e1000_82580: 4014 case e1000_i350: 4015 case e1000_i354: 4016 case e1000_i210: 4017 default: 4018 /* If rss_queues > half of max_rss_queues, pair the queues in 4019 * order to conserve interrupts due to limited supply. 4020 */ 4021 if (adapter->rss_queues > (max_rss_queues / 2)) 4022 adapter->flags |= IGB_FLAG_QUEUE_PAIRS; 4023 else 4024 adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS; 4025 break; 4026 } 4027 } 4028 4029 /** 4030 * igb_sw_init - Initialize general software structures (struct igb_adapter) 4031 * @adapter: board private structure to initialize 4032 * 4033 * igb_sw_init initializes the Adapter private data structure. 4034 * Fields are initialized based on PCI device information and 4035 * OS network device settings (MTU size). 4036 **/ 4037 static int igb_sw_init(struct igb_adapter *adapter) 4038 { 4039 struct e1000_hw *hw = &adapter->hw; 4040 struct net_device *netdev = adapter->netdev; 4041 struct pci_dev *pdev = adapter->pdev; 4042 4043 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word); 4044 4045 /* set default ring sizes */ 4046 adapter->tx_ring_count = IGB_DEFAULT_TXD; 4047 adapter->rx_ring_count = IGB_DEFAULT_RXD; 4048 4049 /* set default ITR values */ 4050 adapter->rx_itr_setting = IGB_DEFAULT_ITR; 4051 adapter->tx_itr_setting = IGB_DEFAULT_ITR; 4052 4053 /* set default work limits */ 4054 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK; 4055 4056 adapter->max_frame_size = netdev->mtu + IGB_ETH_PKT_HDR_PAD; 4057 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN; 4058 4059 spin_lock_init(&adapter->nfc_lock); 4060 spin_lock_init(&adapter->stats64_lock); 4061 4062 /* init spinlock to avoid concurrency of VF resources */ 4063 spin_lock_init(&adapter->vfs_lock); 4064 #ifdef CONFIG_PCI_IOV 4065 switch (hw->mac.type) { 4066 case e1000_82576: 4067 case e1000_i350: 4068 if (max_vfs > 7) { 4069 dev_warn(&pdev->dev, 4070 "Maximum of 7 VFs per PF, using max\n"); 4071 max_vfs = adapter->vfs_allocated_count = 7; 4072 } else 4073 adapter->vfs_allocated_count = max_vfs; 4074 if (adapter->vfs_allocated_count) 4075 dev_warn(&pdev->dev, 4076 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n"); 4077 break; 4078 default: 4079 break; 4080 } 4081 #endif /* CONFIG_PCI_IOV */ 4082 4083 /* Assume MSI-X interrupts, will be checked during IRQ allocation */ 4084 adapter->flags |= IGB_FLAG_HAS_MSIX; 4085 4086 adapter->mac_table = kcalloc(hw->mac.rar_entry_count, 4087 sizeof(struct igb_mac_addr), 4088 GFP_KERNEL); 4089 if (!adapter->mac_table) 4090 return -ENOMEM; 4091 4092 igb_probe_vfs(adapter); 4093 4094 igb_init_queue_configuration(adapter); 4095 4096 /* Setup and initialize a copy of the hw vlan table array */ 4097 adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32), 4098 GFP_KERNEL); 4099 if (!adapter->shadow_vfta) 4100 return -ENOMEM; 4101 4102 /* This call may decrease the number of queues */ 4103 if (igb_init_interrupt_scheme(adapter, true)) { 4104 dev_err(&pdev->dev, "Unable to allocate memory for queues\n"); 4105 return -ENOMEM; 4106 } 4107 4108 /* Explicitly disable IRQ since the NIC can be in any state. */ 4109 igb_irq_disable(adapter); 4110 4111 if (hw->mac.type >= e1000_i350) 4112 adapter->flags &= ~IGB_FLAG_DMAC; 4113 4114 set_bit(__IGB_DOWN, &adapter->state); 4115 return 0; 4116 } 4117 4118 /** 4119 * __igb_open - Called when a network interface is made active 4120 * @netdev: network interface device structure 4121 * @resuming: indicates whether we are in a resume call 4122 * 4123 * Returns 0 on success, negative value on failure 4124 * 4125 * The open entry point is called when a network interface is made 4126 * active by the system (IFF_UP). At this point all resources needed 4127 * for transmit and receive operations are allocated, the interrupt 4128 * handler is registered with the OS, the watchdog timer is started, 4129 * and the stack is notified that the interface is ready. 4130 **/ 4131 static int __igb_open(struct net_device *netdev, bool resuming) 4132 { 4133 struct igb_adapter *adapter = netdev_priv(netdev); 4134 struct e1000_hw *hw = &adapter->hw; 4135 struct pci_dev *pdev = adapter->pdev; 4136 int err; 4137 int i; 4138 4139 /* disallow open during test */ 4140 if (test_bit(__IGB_TESTING, &adapter->state)) { 4141 WARN_ON(resuming); 4142 return -EBUSY; 4143 } 4144 4145 if (!resuming) 4146 pm_runtime_get_sync(&pdev->dev); 4147 4148 netif_carrier_off(netdev); 4149 4150 /* allocate transmit descriptors */ 4151 err = igb_setup_all_tx_resources(adapter); 4152 if (err) 4153 goto err_setup_tx; 4154 4155 /* allocate receive descriptors */ 4156 err = igb_setup_all_rx_resources(adapter); 4157 if (err) 4158 goto err_setup_rx; 4159 4160 igb_power_up_link(adapter); 4161 4162 /* before we allocate an interrupt, we must be ready to handle it. 4163 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt 4164 * as soon as we call pci_request_irq, so we have to setup our 4165 * clean_rx handler before we do so. 4166 */ 4167 igb_configure(adapter); 4168 4169 err = igb_request_irq(adapter); 4170 if (err) 4171 goto err_req_irq; 4172 4173 /* Notify the stack of the actual queue counts. */ 4174 err = netif_set_real_num_tx_queues(adapter->netdev, 4175 adapter->num_tx_queues); 4176 if (err) 4177 goto err_set_queues; 4178 4179 err = netif_set_real_num_rx_queues(adapter->netdev, 4180 adapter->num_rx_queues); 4181 if (err) 4182 goto err_set_queues; 4183 4184 /* From here on the code is the same as igb_up() */ 4185 clear_bit(__IGB_DOWN, &adapter->state); 4186 4187 for (i = 0; i < adapter->num_q_vectors; i++) 4188 napi_enable(&(adapter->q_vector[i]->napi)); 4189 4190 /* Clear any pending interrupts. */ 4191 rd32(E1000_TSICR); 4192 rd32(E1000_ICR); 4193 4194 igb_irq_enable(adapter); 4195 4196 /* notify VFs that reset has been completed */ 4197 if (adapter->vfs_allocated_count) { 4198 u32 reg_data = rd32(E1000_CTRL_EXT); 4199 4200 reg_data |= E1000_CTRL_EXT_PFRSTD; 4201 wr32(E1000_CTRL_EXT, reg_data); 4202 } 4203 4204 netif_tx_start_all_queues(netdev); 4205 4206 if (!resuming) 4207 pm_runtime_put(&pdev->dev); 4208 4209 /* start the watchdog. */ 4210 hw->mac.get_link_status = 1; 4211 schedule_work(&adapter->watchdog_task); 4212 4213 return 0; 4214 4215 err_set_queues: 4216 igb_free_irq(adapter); 4217 err_req_irq: 4218 igb_release_hw_control(adapter); 4219 igb_power_down_link(adapter); 4220 igb_free_all_rx_resources(adapter); 4221 err_setup_rx: 4222 igb_free_all_tx_resources(adapter); 4223 err_setup_tx: 4224 igb_reset(adapter); 4225 if (!resuming) 4226 pm_runtime_put(&pdev->dev); 4227 4228 return err; 4229 } 4230 4231 int igb_open(struct net_device *netdev) 4232 { 4233 return __igb_open(netdev, false); 4234 } 4235 4236 /** 4237 * __igb_close - Disables a network interface 4238 * @netdev: network interface device structure 4239 * @suspending: indicates we are in a suspend call 4240 * 4241 * Returns 0, this is not allowed to fail 4242 * 4243 * The close entry point is called when an interface is de-activated 4244 * by the OS. The hardware is still under the driver's control, but 4245 * needs to be disabled. A global MAC reset is issued to stop the 4246 * hardware, and all transmit and receive resources are freed. 4247 **/ 4248 static int __igb_close(struct net_device *netdev, bool suspending) 4249 { 4250 struct igb_adapter *adapter = netdev_priv(netdev); 4251 struct pci_dev *pdev = adapter->pdev; 4252 4253 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state)); 4254 4255 if (!suspending) 4256 pm_runtime_get_sync(&pdev->dev); 4257 4258 igb_down(adapter); 4259 igb_free_irq(adapter); 4260 4261 igb_free_all_tx_resources(adapter); 4262 igb_free_all_rx_resources(adapter); 4263 4264 if (!suspending) 4265 pm_runtime_put_sync(&pdev->dev); 4266 return 0; 4267 } 4268 4269 int igb_close(struct net_device *netdev) 4270 { 4271 if (netif_device_present(netdev) || netdev->dismantle) 4272 return __igb_close(netdev, false); 4273 return 0; 4274 } 4275 4276 /** 4277 * igb_setup_tx_resources - allocate Tx resources (Descriptors) 4278 * @tx_ring: tx descriptor ring (for a specific queue) to setup 4279 * 4280 * Return 0 on success, negative on failure 4281 **/ 4282 int igb_setup_tx_resources(struct igb_ring *tx_ring) 4283 { 4284 struct device *dev = tx_ring->dev; 4285 int size; 4286 4287 size = sizeof(struct igb_tx_buffer) * tx_ring->count; 4288 4289 tx_ring->tx_buffer_info = vmalloc(size); 4290 if (!tx_ring->tx_buffer_info) 4291 goto err; 4292 4293 /* round up to nearest 4K */ 4294 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc); 4295 tx_ring->size = ALIGN(tx_ring->size, 4096); 4296 4297 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size, 4298 &tx_ring->dma, GFP_KERNEL); 4299 if (!tx_ring->desc) 4300 goto err; 4301 4302 tx_ring->next_to_use = 0; 4303 tx_ring->next_to_clean = 0; 4304 4305 return 0; 4306 4307 err: 4308 vfree(tx_ring->tx_buffer_info); 4309 tx_ring->tx_buffer_info = NULL; 4310 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n"); 4311 return -ENOMEM; 4312 } 4313 4314 /** 4315 * igb_setup_all_tx_resources - wrapper to allocate Tx resources 4316 * (Descriptors) for all queues 4317 * @adapter: board private structure 4318 * 4319 * Return 0 on success, negative on failure 4320 **/ 4321 static int igb_setup_all_tx_resources(struct igb_adapter *adapter) 4322 { 4323 struct pci_dev *pdev = adapter->pdev; 4324 int i, err = 0; 4325 4326 for (i = 0; i < adapter->num_tx_queues; i++) { 4327 err = igb_setup_tx_resources(adapter->tx_ring[i]); 4328 if (err) { 4329 dev_err(&pdev->dev, 4330 "Allocation for Tx Queue %u failed\n", i); 4331 for (i--; i >= 0; i--) 4332 igb_free_tx_resources(adapter->tx_ring[i]); 4333 break; 4334 } 4335 } 4336 4337 return err; 4338 } 4339 4340 /** 4341 * igb_setup_tctl - configure the transmit control registers 4342 * @adapter: Board private structure 4343 **/ 4344 void igb_setup_tctl(struct igb_adapter *adapter) 4345 { 4346 struct e1000_hw *hw = &adapter->hw; 4347 u32 tctl; 4348 4349 /* disable queue 0 which is enabled by default on 82575 and 82576 */ 4350 wr32(E1000_TXDCTL(0), 0); 4351 4352 /* Program the Transmit Control Register */ 4353 tctl = rd32(E1000_TCTL); 4354 tctl &= ~E1000_TCTL_CT; 4355 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC | 4356 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT); 4357 4358 igb_config_collision_dist(hw); 4359 4360 /* Enable transmits */ 4361 tctl |= E1000_TCTL_EN; 4362 4363 wr32(E1000_TCTL, tctl); 4364 } 4365 4366 /** 4367 * igb_configure_tx_ring - Configure transmit ring after Reset 4368 * @adapter: board private structure 4369 * @ring: tx ring to configure 4370 * 4371 * Configure a transmit ring after a reset. 4372 **/ 4373 void igb_configure_tx_ring(struct igb_adapter *adapter, 4374 struct igb_ring *ring) 4375 { 4376 struct e1000_hw *hw = &adapter->hw; 4377 u32 txdctl = 0; 4378 u64 tdba = ring->dma; 4379 int reg_idx = ring->reg_idx; 4380 4381 wr32(E1000_TDLEN(reg_idx), 4382 ring->count * sizeof(union e1000_adv_tx_desc)); 4383 wr32(E1000_TDBAL(reg_idx), 4384 tdba & 0x00000000ffffffffULL); 4385 wr32(E1000_TDBAH(reg_idx), tdba >> 32); 4386 4387 ring->tail = adapter->io_addr + E1000_TDT(reg_idx); 4388 wr32(E1000_TDH(reg_idx), 0); 4389 writel(0, ring->tail); 4390 4391 txdctl |= IGB_TX_PTHRESH; 4392 txdctl |= IGB_TX_HTHRESH << 8; 4393 txdctl |= IGB_TX_WTHRESH << 16; 4394 4395 /* reinitialize tx_buffer_info */ 4396 memset(ring->tx_buffer_info, 0, 4397 sizeof(struct igb_tx_buffer) * ring->count); 4398 4399 txdctl |= E1000_TXDCTL_QUEUE_ENABLE; 4400 wr32(E1000_TXDCTL(reg_idx), txdctl); 4401 } 4402 4403 /** 4404 * igb_configure_tx - Configure transmit Unit after Reset 4405 * @adapter: board private structure 4406 * 4407 * Configure the Tx unit of the MAC after a reset. 4408 **/ 4409 static void igb_configure_tx(struct igb_adapter *adapter) 4410 { 4411 struct e1000_hw *hw = &adapter->hw; 4412 int i; 4413 4414 /* disable the queues */ 4415 for (i = 0; i < adapter->num_tx_queues; i++) 4416 wr32(E1000_TXDCTL(adapter->tx_ring[i]->reg_idx), 0); 4417 4418 wrfl(); 4419 usleep_range(10000, 20000); 4420 4421 for (i = 0; i < adapter->num_tx_queues; i++) 4422 igb_configure_tx_ring(adapter, adapter->tx_ring[i]); 4423 } 4424 4425 /** 4426 * igb_setup_rx_resources - allocate Rx resources (Descriptors) 4427 * @rx_ring: Rx descriptor ring (for a specific queue) to setup 4428 * 4429 * Returns 0 on success, negative on failure 4430 **/ 4431 int igb_setup_rx_resources(struct igb_ring *rx_ring) 4432 { 4433 struct igb_adapter *adapter = netdev_priv(rx_ring->netdev); 4434 struct device *dev = rx_ring->dev; 4435 int size, res; 4436 4437 /* XDP RX-queue info */ 4438 if (xdp_rxq_info_is_reg(&rx_ring->xdp_rxq)) 4439 xdp_rxq_info_unreg(&rx_ring->xdp_rxq); 4440 res = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev, 4441 rx_ring->queue_index, 0); 4442 if (res < 0) { 4443 dev_err(dev, "Failed to register xdp_rxq index %u\n", 4444 rx_ring->queue_index); 4445 return res; 4446 } 4447 4448 size = sizeof(struct igb_rx_buffer) * rx_ring->count; 4449 4450 rx_ring->rx_buffer_info = vmalloc(size); 4451 if (!rx_ring->rx_buffer_info) 4452 goto err; 4453 4454 /* Round up to nearest 4K */ 4455 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc); 4456 rx_ring->size = ALIGN(rx_ring->size, 4096); 4457 4458 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size, 4459 &rx_ring->dma, GFP_KERNEL); 4460 if (!rx_ring->desc) 4461 goto err; 4462 4463 rx_ring->next_to_alloc = 0; 4464 rx_ring->next_to_clean = 0; 4465 rx_ring->next_to_use = 0; 4466 4467 rx_ring->xdp_prog = adapter->xdp_prog; 4468 4469 return 0; 4470 4471 err: 4472 xdp_rxq_info_unreg(&rx_ring->xdp_rxq); 4473 vfree(rx_ring->rx_buffer_info); 4474 rx_ring->rx_buffer_info = NULL; 4475 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n"); 4476 return -ENOMEM; 4477 } 4478 4479 /** 4480 * igb_setup_all_rx_resources - wrapper to allocate Rx resources 4481 * (Descriptors) for all queues 4482 * @adapter: board private structure 4483 * 4484 * Return 0 on success, negative on failure 4485 **/ 4486 static int igb_setup_all_rx_resources(struct igb_adapter *adapter) 4487 { 4488 struct pci_dev *pdev = adapter->pdev; 4489 int i, err = 0; 4490 4491 for (i = 0; i < adapter->num_rx_queues; i++) { 4492 err = igb_setup_rx_resources(adapter->rx_ring[i]); 4493 if (err) { 4494 dev_err(&pdev->dev, 4495 "Allocation for Rx Queue %u failed\n", i); 4496 for (i--; i >= 0; i--) 4497 igb_free_rx_resources(adapter->rx_ring[i]); 4498 break; 4499 } 4500 } 4501 4502 return err; 4503 } 4504 4505 /** 4506 * igb_setup_mrqc - configure the multiple receive queue control registers 4507 * @adapter: Board private structure 4508 **/ 4509 static void igb_setup_mrqc(struct igb_adapter *adapter) 4510 { 4511 struct e1000_hw *hw = &adapter->hw; 4512 u32 mrqc, rxcsum; 4513 u32 j, num_rx_queues; 4514 u32 rss_key[10]; 4515 4516 netdev_rss_key_fill(rss_key, sizeof(rss_key)); 4517 for (j = 0; j < 10; j++) 4518 wr32(E1000_RSSRK(j), rss_key[j]); 4519 4520 num_rx_queues = adapter->rss_queues; 4521 4522 switch (hw->mac.type) { 4523 case e1000_82576: 4524 /* 82576 supports 2 RSS queues for SR-IOV */ 4525 if (adapter->vfs_allocated_count) 4526 num_rx_queues = 2; 4527 break; 4528 default: 4529 break; 4530 } 4531 4532 if (adapter->rss_indir_tbl_init != num_rx_queues) { 4533 for (j = 0; j < IGB_RETA_SIZE; j++) 4534 adapter->rss_indir_tbl[j] = 4535 (j * num_rx_queues) / IGB_RETA_SIZE; 4536 adapter->rss_indir_tbl_init = num_rx_queues; 4537 } 4538 igb_write_rss_indir_tbl(adapter); 4539 4540 /* Disable raw packet checksumming so that RSS hash is placed in 4541 * descriptor on writeback. No need to enable TCP/UDP/IP checksum 4542 * offloads as they are enabled by default 4543 */ 4544 rxcsum = rd32(E1000_RXCSUM); 4545 rxcsum |= E1000_RXCSUM_PCSD; 4546 4547 if (adapter->hw.mac.type >= e1000_82576) 4548 /* Enable Receive Checksum Offload for SCTP */ 4549 rxcsum |= E1000_RXCSUM_CRCOFL; 4550 4551 /* Don't need to set TUOFL or IPOFL, they default to 1 */ 4552 wr32(E1000_RXCSUM, rxcsum); 4553 4554 /* Generate RSS hash based on packet types, TCP/UDP 4555 * port numbers and/or IPv4/v6 src and dst addresses 4556 */ 4557 mrqc = E1000_MRQC_RSS_FIELD_IPV4 | 4558 E1000_MRQC_RSS_FIELD_IPV4_TCP | 4559 E1000_MRQC_RSS_FIELD_IPV6 | 4560 E1000_MRQC_RSS_FIELD_IPV6_TCP | 4561 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX; 4562 4563 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP) 4564 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP; 4565 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP) 4566 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP; 4567 4568 /* If VMDq is enabled then we set the appropriate mode for that, else 4569 * we default to RSS so that an RSS hash is calculated per packet even 4570 * if we are only using one queue 4571 */ 4572 if (adapter->vfs_allocated_count) { 4573 if (hw->mac.type > e1000_82575) { 4574 /* Set the default pool for the PF's first queue */ 4575 u32 vtctl = rd32(E1000_VT_CTL); 4576 4577 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK | 4578 E1000_VT_CTL_DISABLE_DEF_POOL); 4579 vtctl |= adapter->vfs_allocated_count << 4580 E1000_VT_CTL_DEFAULT_POOL_SHIFT; 4581 wr32(E1000_VT_CTL, vtctl); 4582 } 4583 if (adapter->rss_queues > 1) 4584 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ; 4585 else 4586 mrqc |= E1000_MRQC_ENABLE_VMDQ; 4587 } else { 4588 mrqc |= E1000_MRQC_ENABLE_RSS_MQ; 4589 } 4590 igb_vmm_control(adapter); 4591 4592 wr32(E1000_MRQC, mrqc); 4593 } 4594 4595 /** 4596 * igb_setup_rctl - configure the receive control registers 4597 * @adapter: Board private structure 4598 **/ 4599 void igb_setup_rctl(struct igb_adapter *adapter) 4600 { 4601 struct e1000_hw *hw = &adapter->hw; 4602 u32 rctl; 4603 4604 rctl = rd32(E1000_RCTL); 4605 4606 rctl &= ~(3 << E1000_RCTL_MO_SHIFT); 4607 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC); 4608 4609 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF | 4610 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT); 4611 4612 /* enable stripping of CRC. It's unlikely this will break BMC 4613 * redirection as it did with e1000. Newer features require 4614 * that the HW strips the CRC. 4615 */ 4616 rctl |= E1000_RCTL_SECRC; 4617 4618 /* disable store bad packets and clear size bits. */ 4619 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256); 4620 4621 /* enable LPE to allow for reception of jumbo frames */ 4622 rctl |= E1000_RCTL_LPE; 4623 4624 /* disable queue 0 to prevent tail write w/o re-config */ 4625 wr32(E1000_RXDCTL(0), 0); 4626 4627 /* Attention!!! For SR-IOV PF driver operations you must enable 4628 * queue drop for all VF and PF queues to prevent head of line blocking 4629 * if an un-trusted VF does not provide descriptors to hardware. 4630 */ 4631 if (adapter->vfs_allocated_count) { 4632 /* set all queue drop enable bits */ 4633 wr32(E1000_QDE, ALL_QUEUES); 4634 } 4635 4636 /* This is useful for sniffing bad packets. */ 4637 if (adapter->netdev->features & NETIF_F_RXALL) { 4638 /* UPE and MPE will be handled by normal PROMISC logic 4639 * in e1000e_set_rx_mode 4640 */ 4641 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */ 4642 E1000_RCTL_BAM | /* RX All Bcast Pkts */ 4643 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */ 4644 4645 rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */ 4646 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */ 4647 /* Do not mess with E1000_CTRL_VME, it affects transmit as well, 4648 * and that breaks VLANs. 4649 */ 4650 } 4651 4652 wr32(E1000_RCTL, rctl); 4653 } 4654 4655 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size, 4656 int vfn) 4657 { 4658 struct e1000_hw *hw = &adapter->hw; 4659 u32 vmolr; 4660 4661 if (size > MAX_JUMBO_FRAME_SIZE) 4662 size = MAX_JUMBO_FRAME_SIZE; 4663 4664 vmolr = rd32(E1000_VMOLR(vfn)); 4665 vmolr &= ~E1000_VMOLR_RLPML_MASK; 4666 vmolr |= size | E1000_VMOLR_LPE; 4667 wr32(E1000_VMOLR(vfn), vmolr); 4668 4669 return 0; 4670 } 4671 4672 static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter, 4673 int vfn, bool enable) 4674 { 4675 struct e1000_hw *hw = &adapter->hw; 4676 u32 val, reg; 4677 4678 if (hw->mac.type < e1000_82576) 4679 return; 4680 4681 if (hw->mac.type == e1000_i350) 4682 reg = E1000_DVMOLR(vfn); 4683 else 4684 reg = E1000_VMOLR(vfn); 4685 4686 val = rd32(reg); 4687 if (enable) 4688 val |= E1000_VMOLR_STRVLAN; 4689 else 4690 val &= ~(E1000_VMOLR_STRVLAN); 4691 wr32(reg, val); 4692 } 4693 4694 static inline void igb_set_vmolr(struct igb_adapter *adapter, 4695 int vfn, bool aupe) 4696 { 4697 struct e1000_hw *hw = &adapter->hw; 4698 u32 vmolr; 4699 4700 /* This register exists only on 82576 and newer so if we are older then 4701 * we should exit and do nothing 4702 */ 4703 if (hw->mac.type < e1000_82576) 4704 return; 4705 4706 vmolr = rd32(E1000_VMOLR(vfn)); 4707 if (aupe) 4708 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */ 4709 else 4710 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */ 4711 4712 /* clear all bits that might not be set */ 4713 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE); 4714 4715 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count) 4716 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */ 4717 /* for VMDq only allow the VFs and pool 0 to accept broadcast and 4718 * multicast packets 4719 */ 4720 if (vfn <= adapter->vfs_allocated_count) 4721 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */ 4722 4723 wr32(E1000_VMOLR(vfn), vmolr); 4724 } 4725 4726 /** 4727 * igb_setup_srrctl - configure the split and replication receive control 4728 * registers 4729 * @adapter: Board private structure 4730 * @ring: receive ring to be configured 4731 **/ 4732 void igb_setup_srrctl(struct igb_adapter *adapter, struct igb_ring *ring) 4733 { 4734 struct e1000_hw *hw = &adapter->hw; 4735 int reg_idx = ring->reg_idx; 4736 u32 srrctl = 0; 4737 4738 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT; 4739 if (ring_uses_large_buffer(ring)) 4740 srrctl |= IGB_RXBUFFER_3072 >> E1000_SRRCTL_BSIZEPKT_SHIFT; 4741 else 4742 srrctl |= IGB_RXBUFFER_2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT; 4743 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF; 4744 if (hw->mac.type >= e1000_82580) 4745 srrctl |= E1000_SRRCTL_TIMESTAMP; 4746 /* Only set Drop Enable if VFs allocated, or we are supporting multiple 4747 * queues and rx flow control is disabled 4748 */ 4749 if (adapter->vfs_allocated_count || 4750 (!(hw->fc.current_mode & e1000_fc_rx_pause) && 4751 adapter->num_rx_queues > 1)) 4752 srrctl |= E1000_SRRCTL_DROP_EN; 4753 4754 wr32(E1000_SRRCTL(reg_idx), srrctl); 4755 } 4756 4757 /** 4758 * igb_configure_rx_ring - Configure a receive ring after Reset 4759 * @adapter: board private structure 4760 * @ring: receive ring to be configured 4761 * 4762 * Configure the Rx unit of the MAC after a reset. 4763 **/ 4764 void igb_configure_rx_ring(struct igb_adapter *adapter, 4765 struct igb_ring *ring) 4766 { 4767 struct e1000_hw *hw = &adapter->hw; 4768 union e1000_adv_rx_desc *rx_desc; 4769 u64 rdba = ring->dma; 4770 int reg_idx = ring->reg_idx; 4771 u32 rxdctl = 0; 4772 4773 xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq); 4774 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq, 4775 MEM_TYPE_PAGE_SHARED, NULL)); 4776 4777 /* disable the queue */ 4778 wr32(E1000_RXDCTL(reg_idx), 0); 4779 4780 /* Set DMA base address registers */ 4781 wr32(E1000_RDBAL(reg_idx), 4782 rdba & 0x00000000ffffffffULL); 4783 wr32(E1000_RDBAH(reg_idx), rdba >> 32); 4784 wr32(E1000_RDLEN(reg_idx), 4785 ring->count * sizeof(union e1000_adv_rx_desc)); 4786 4787 /* initialize head and tail */ 4788 ring->tail = adapter->io_addr + E1000_RDT(reg_idx); 4789 wr32(E1000_RDH(reg_idx), 0); 4790 writel(0, ring->tail); 4791 4792 /* set descriptor configuration */ 4793 igb_setup_srrctl(adapter, ring); 4794 4795 /* set filtering for VMDQ pools */ 4796 igb_set_vmolr(adapter, reg_idx & 0x7, true); 4797 4798 rxdctl |= IGB_RX_PTHRESH; 4799 rxdctl |= IGB_RX_HTHRESH << 8; 4800 rxdctl |= IGB_RX_WTHRESH << 16; 4801 4802 /* initialize rx_buffer_info */ 4803 memset(ring->rx_buffer_info, 0, 4804 sizeof(struct igb_rx_buffer) * ring->count); 4805 4806 /* initialize Rx descriptor 0 */ 4807 rx_desc = IGB_RX_DESC(ring, 0); 4808 rx_desc->wb.upper.length = 0; 4809 4810 /* enable receive descriptor fetching */ 4811 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE; 4812 wr32(E1000_RXDCTL(reg_idx), rxdctl); 4813 } 4814 4815 static void igb_set_rx_buffer_len(struct igb_adapter *adapter, 4816 struct igb_ring *rx_ring) 4817 { 4818 #if (PAGE_SIZE < 8192) 4819 struct e1000_hw *hw = &adapter->hw; 4820 #endif 4821 4822 /* set build_skb and buffer size flags */ 4823 clear_ring_build_skb_enabled(rx_ring); 4824 clear_ring_uses_large_buffer(rx_ring); 4825 4826 if (adapter->flags & IGB_FLAG_RX_LEGACY) 4827 return; 4828 4829 set_ring_build_skb_enabled(rx_ring); 4830 4831 #if (PAGE_SIZE < 8192) 4832 if (adapter->max_frame_size > IGB_MAX_FRAME_BUILD_SKB || 4833 rd32(E1000_RCTL) & E1000_RCTL_SBP) 4834 set_ring_uses_large_buffer(rx_ring); 4835 #endif 4836 } 4837 4838 /** 4839 * igb_configure_rx - Configure receive Unit after Reset 4840 * @adapter: board private structure 4841 * 4842 * Configure the Rx unit of the MAC after a reset. 4843 **/ 4844 static void igb_configure_rx(struct igb_adapter *adapter) 4845 { 4846 int i; 4847 4848 /* set the correct pool for the PF default MAC address in entry 0 */ 4849 igb_set_default_mac_filter(adapter); 4850 4851 /* Setup the HW Rx Head and Tail Descriptor Pointers and 4852 * the Base and Length of the Rx Descriptor Ring 4853 */ 4854 for (i = 0; i < adapter->num_rx_queues; i++) { 4855 struct igb_ring *rx_ring = adapter->rx_ring[i]; 4856 4857 igb_set_rx_buffer_len(adapter, rx_ring); 4858 igb_configure_rx_ring(adapter, rx_ring); 4859 } 4860 } 4861 4862 /** 4863 * igb_free_tx_resources - Free Tx Resources per Queue 4864 * @tx_ring: Tx descriptor ring for a specific queue 4865 * 4866 * Free all transmit software resources 4867 **/ 4868 void igb_free_tx_resources(struct igb_ring *tx_ring) 4869 { 4870 igb_clean_tx_ring(tx_ring); 4871 4872 vfree(tx_ring->tx_buffer_info); 4873 tx_ring->tx_buffer_info = NULL; 4874 4875 /* if not set, then don't free */ 4876 if (!tx_ring->desc) 4877 return; 4878 4879 dma_free_coherent(tx_ring->dev, tx_ring->size, 4880 tx_ring->desc, tx_ring->dma); 4881 4882 tx_ring->desc = NULL; 4883 } 4884 4885 /** 4886 * igb_free_all_tx_resources - Free Tx Resources for All Queues 4887 * @adapter: board private structure 4888 * 4889 * Free all transmit software resources 4890 **/ 4891 static void igb_free_all_tx_resources(struct igb_adapter *adapter) 4892 { 4893 int i; 4894 4895 for (i = 0; i < adapter->num_tx_queues; i++) 4896 if (adapter->tx_ring[i]) 4897 igb_free_tx_resources(adapter->tx_ring[i]); 4898 } 4899 4900 /** 4901 * igb_clean_tx_ring - Free Tx Buffers 4902 * @tx_ring: ring to be cleaned 4903 **/ 4904 static void igb_clean_tx_ring(struct igb_ring *tx_ring) 4905 { 4906 u16 i = tx_ring->next_to_clean; 4907 struct igb_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i]; 4908 4909 while (i != tx_ring->next_to_use) { 4910 union e1000_adv_tx_desc *eop_desc, *tx_desc; 4911 4912 /* Free all the Tx ring sk_buffs or xdp frames */ 4913 if (tx_buffer->type == IGB_TYPE_SKB) 4914 dev_kfree_skb_any(tx_buffer->skb); 4915 else 4916 xdp_return_frame(tx_buffer->xdpf); 4917 4918 /* unmap skb header data */ 4919 dma_unmap_single(tx_ring->dev, 4920 dma_unmap_addr(tx_buffer, dma), 4921 dma_unmap_len(tx_buffer, len), 4922 DMA_TO_DEVICE); 4923 4924 /* check for eop_desc to determine the end of the packet */ 4925 eop_desc = tx_buffer->next_to_watch; 4926 tx_desc = IGB_TX_DESC(tx_ring, i); 4927 4928 /* unmap remaining buffers */ 4929 while (tx_desc != eop_desc) { 4930 tx_buffer++; 4931 tx_desc++; 4932 i++; 4933 if (unlikely(i == tx_ring->count)) { 4934 i = 0; 4935 tx_buffer = tx_ring->tx_buffer_info; 4936 tx_desc = IGB_TX_DESC(tx_ring, 0); 4937 } 4938 4939 /* unmap any remaining paged data */ 4940 if (dma_unmap_len(tx_buffer, len)) 4941 dma_unmap_page(tx_ring->dev, 4942 dma_unmap_addr(tx_buffer, dma), 4943 dma_unmap_len(tx_buffer, len), 4944 DMA_TO_DEVICE); 4945 } 4946 4947 tx_buffer->next_to_watch = NULL; 4948 4949 /* move us one more past the eop_desc for start of next pkt */ 4950 tx_buffer++; 4951 i++; 4952 if (unlikely(i == tx_ring->count)) { 4953 i = 0; 4954 tx_buffer = tx_ring->tx_buffer_info; 4955 } 4956 } 4957 4958 /* reset BQL for queue */ 4959 netdev_tx_reset_queue(txring_txq(tx_ring)); 4960 4961 /* reset next_to_use and next_to_clean */ 4962 tx_ring->next_to_use = 0; 4963 tx_ring->next_to_clean = 0; 4964 } 4965 4966 /** 4967 * igb_clean_all_tx_rings - Free Tx Buffers for all queues 4968 * @adapter: board private structure 4969 **/ 4970 static void igb_clean_all_tx_rings(struct igb_adapter *adapter) 4971 { 4972 int i; 4973 4974 for (i = 0; i < adapter->num_tx_queues; i++) 4975 if (adapter->tx_ring[i]) 4976 igb_clean_tx_ring(adapter->tx_ring[i]); 4977 } 4978 4979 /** 4980 * igb_free_rx_resources - Free Rx Resources 4981 * @rx_ring: ring to clean the resources from 4982 * 4983 * Free all receive software resources 4984 **/ 4985 void igb_free_rx_resources(struct igb_ring *rx_ring) 4986 { 4987 igb_clean_rx_ring(rx_ring); 4988 4989 rx_ring->xdp_prog = NULL; 4990 xdp_rxq_info_unreg(&rx_ring->xdp_rxq); 4991 vfree(rx_ring->rx_buffer_info); 4992 rx_ring->rx_buffer_info = NULL; 4993 4994 /* if not set, then don't free */ 4995 if (!rx_ring->desc) 4996 return; 4997 4998 dma_free_coherent(rx_ring->dev, rx_ring->size, 4999 rx_ring->desc, rx_ring->dma); 5000 5001 rx_ring->desc = NULL; 5002 } 5003 5004 /** 5005 * igb_free_all_rx_resources - Free Rx Resources for All Queues 5006 * @adapter: board private structure 5007 * 5008 * Free all receive software resources 5009 **/ 5010 static void igb_free_all_rx_resources(struct igb_adapter *adapter) 5011 { 5012 int i; 5013 5014 for (i = 0; i < adapter->num_rx_queues; i++) 5015 if (adapter->rx_ring[i]) 5016 igb_free_rx_resources(adapter->rx_ring[i]); 5017 } 5018 5019 /** 5020 * igb_clean_rx_ring - Free Rx Buffers per Queue 5021 * @rx_ring: ring to free buffers from 5022 **/ 5023 static void igb_clean_rx_ring(struct igb_ring *rx_ring) 5024 { 5025 u16 i = rx_ring->next_to_clean; 5026 5027 dev_kfree_skb(rx_ring->skb); 5028 rx_ring->skb = NULL; 5029 5030 /* Free all the Rx ring sk_buffs */ 5031 while (i != rx_ring->next_to_alloc) { 5032 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i]; 5033 5034 /* Invalidate cache lines that may have been written to by 5035 * device so that we avoid corrupting memory. 5036 */ 5037 dma_sync_single_range_for_cpu(rx_ring->dev, 5038 buffer_info->dma, 5039 buffer_info->page_offset, 5040 igb_rx_bufsz(rx_ring), 5041 DMA_FROM_DEVICE); 5042 5043 /* free resources associated with mapping */ 5044 dma_unmap_page_attrs(rx_ring->dev, 5045 buffer_info->dma, 5046 igb_rx_pg_size(rx_ring), 5047 DMA_FROM_DEVICE, 5048 IGB_RX_DMA_ATTR); 5049 __page_frag_cache_drain(buffer_info->page, 5050 buffer_info->pagecnt_bias); 5051 5052 i++; 5053 if (i == rx_ring->count) 5054 i = 0; 5055 } 5056 5057 rx_ring->next_to_alloc = 0; 5058 rx_ring->next_to_clean = 0; 5059 rx_ring->next_to_use = 0; 5060 } 5061 5062 /** 5063 * igb_clean_all_rx_rings - Free Rx Buffers for all queues 5064 * @adapter: board private structure 5065 **/ 5066 static void igb_clean_all_rx_rings(struct igb_adapter *adapter) 5067 { 5068 int i; 5069 5070 for (i = 0; i < adapter->num_rx_queues; i++) 5071 if (adapter->rx_ring[i]) 5072 igb_clean_rx_ring(adapter->rx_ring[i]); 5073 } 5074 5075 /** 5076 * igb_set_mac - Change the Ethernet Address of the NIC 5077 * @netdev: network interface device structure 5078 * @p: pointer to an address structure 5079 * 5080 * Returns 0 on success, negative on failure 5081 **/ 5082 static int igb_set_mac(struct net_device *netdev, void *p) 5083 { 5084 struct igb_adapter *adapter = netdev_priv(netdev); 5085 struct e1000_hw *hw = &adapter->hw; 5086 struct sockaddr *addr = p; 5087 5088 if (!is_valid_ether_addr(addr->sa_data)) 5089 return -EADDRNOTAVAIL; 5090 5091 eth_hw_addr_set(netdev, addr->sa_data); 5092 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len); 5093 5094 /* set the correct pool for the new PF MAC address in entry 0 */ 5095 igb_set_default_mac_filter(adapter); 5096 5097 return 0; 5098 } 5099 5100 /** 5101 * igb_write_mc_addr_list - write multicast addresses to MTA 5102 * @netdev: network interface device structure 5103 * 5104 * Writes multicast address list to the MTA hash table. 5105 * Returns: -ENOMEM on failure 5106 * 0 on no addresses written 5107 * X on writing X addresses to MTA 5108 **/ 5109 static int igb_write_mc_addr_list(struct net_device *netdev) 5110 { 5111 struct igb_adapter *adapter = netdev_priv(netdev); 5112 struct e1000_hw *hw = &adapter->hw; 5113 struct netdev_hw_addr *ha; 5114 u8 *mta_list; 5115 int i; 5116 5117 if (netdev_mc_empty(netdev)) { 5118 /* nothing to program, so clear mc list */ 5119 igb_update_mc_addr_list(hw, NULL, 0); 5120 igb_restore_vf_multicasts(adapter); 5121 return 0; 5122 } 5123 5124 mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC); 5125 if (!mta_list) 5126 return -ENOMEM; 5127 5128 /* The shared function expects a packed array of only addresses. */ 5129 i = 0; 5130 netdev_for_each_mc_addr(ha, netdev) 5131 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN); 5132 5133 igb_update_mc_addr_list(hw, mta_list, i); 5134 kfree(mta_list); 5135 5136 return netdev_mc_count(netdev); 5137 } 5138 5139 static int igb_vlan_promisc_enable(struct igb_adapter *adapter) 5140 { 5141 struct e1000_hw *hw = &adapter->hw; 5142 u32 i, pf_id; 5143 5144 switch (hw->mac.type) { 5145 case e1000_i210: 5146 case e1000_i211: 5147 case e1000_i350: 5148 /* VLAN filtering needed for VLAN prio filter */ 5149 if (adapter->netdev->features & NETIF_F_NTUPLE) 5150 break; 5151 fallthrough; 5152 case e1000_82576: 5153 case e1000_82580: 5154 case e1000_i354: 5155 /* VLAN filtering needed for pool filtering */ 5156 if (adapter->vfs_allocated_count) 5157 break; 5158 fallthrough; 5159 default: 5160 return 1; 5161 } 5162 5163 /* We are already in VLAN promisc, nothing to do */ 5164 if (adapter->flags & IGB_FLAG_VLAN_PROMISC) 5165 return 0; 5166 5167 if (!adapter->vfs_allocated_count) 5168 goto set_vfta; 5169 5170 /* Add PF to all active pools */ 5171 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT; 5172 5173 for (i = E1000_VLVF_ARRAY_SIZE; --i;) { 5174 u32 vlvf = rd32(E1000_VLVF(i)); 5175 5176 vlvf |= BIT(pf_id); 5177 wr32(E1000_VLVF(i), vlvf); 5178 } 5179 5180 set_vfta: 5181 /* Set all bits in the VLAN filter table array */ 5182 for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;) 5183 hw->mac.ops.write_vfta(hw, i, ~0U); 5184 5185 /* Set flag so we don't redo unnecessary work */ 5186 adapter->flags |= IGB_FLAG_VLAN_PROMISC; 5187 5188 return 0; 5189 } 5190 5191 #define VFTA_BLOCK_SIZE 8 5192 static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset) 5193 { 5194 struct e1000_hw *hw = &adapter->hw; 5195 u32 vfta[VFTA_BLOCK_SIZE] = { 0 }; 5196 u32 vid_start = vfta_offset * 32; 5197 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32); 5198 u32 i, vid, word, bits, pf_id; 5199 5200 /* guarantee that we don't scrub out management VLAN */ 5201 vid = adapter->mng_vlan_id; 5202 if (vid >= vid_start && vid < vid_end) 5203 vfta[(vid - vid_start) / 32] |= BIT(vid % 32); 5204 5205 if (!adapter->vfs_allocated_count) 5206 goto set_vfta; 5207 5208 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT; 5209 5210 for (i = E1000_VLVF_ARRAY_SIZE; --i;) { 5211 u32 vlvf = rd32(E1000_VLVF(i)); 5212 5213 /* pull VLAN ID from VLVF */ 5214 vid = vlvf & VLAN_VID_MASK; 5215 5216 /* only concern ourselves with a certain range */ 5217 if (vid < vid_start || vid >= vid_end) 5218 continue; 5219 5220 if (vlvf & E1000_VLVF_VLANID_ENABLE) { 5221 /* record VLAN ID in VFTA */ 5222 vfta[(vid - vid_start) / 32] |= BIT(vid % 32); 5223 5224 /* if PF is part of this then continue */ 5225 if (test_bit(vid, adapter->active_vlans)) 5226 continue; 5227 } 5228 5229 /* remove PF from the pool */ 5230 bits = ~BIT(pf_id); 5231 bits &= rd32(E1000_VLVF(i)); 5232 wr32(E1000_VLVF(i), bits); 5233 } 5234 5235 set_vfta: 5236 /* extract values from active_vlans and write back to VFTA */ 5237 for (i = VFTA_BLOCK_SIZE; i--;) { 5238 vid = (vfta_offset + i) * 32; 5239 word = vid / BITS_PER_LONG; 5240 bits = vid % BITS_PER_LONG; 5241 5242 vfta[i] |= adapter->active_vlans[word] >> bits; 5243 5244 hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]); 5245 } 5246 } 5247 5248 static void igb_vlan_promisc_disable(struct igb_adapter *adapter) 5249 { 5250 u32 i; 5251 5252 /* We are not in VLAN promisc, nothing to do */ 5253 if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC)) 5254 return; 5255 5256 /* Set flag so we don't redo unnecessary work */ 5257 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC; 5258 5259 for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE) 5260 igb_scrub_vfta(adapter, i); 5261 } 5262 5263 /** 5264 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set 5265 * @netdev: network interface device structure 5266 * 5267 * The set_rx_mode entry point is called whenever the unicast or multicast 5268 * address lists or the network interface flags are updated. This routine is 5269 * responsible for configuring the hardware for proper unicast, multicast, 5270 * promiscuous mode, and all-multi behavior. 5271 **/ 5272 static void igb_set_rx_mode(struct net_device *netdev) 5273 { 5274 struct igb_adapter *adapter = netdev_priv(netdev); 5275 struct e1000_hw *hw = &adapter->hw; 5276 unsigned int vfn = adapter->vfs_allocated_count; 5277 u32 rctl = 0, vmolr = 0, rlpml = MAX_JUMBO_FRAME_SIZE; 5278 int count; 5279 5280 /* Check for Promiscuous and All Multicast modes */ 5281 if (netdev->flags & IFF_PROMISC) { 5282 rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE; 5283 vmolr |= E1000_VMOLR_MPME; 5284 5285 /* enable use of UTA filter to force packets to default pool */ 5286 if (hw->mac.type == e1000_82576) 5287 vmolr |= E1000_VMOLR_ROPE; 5288 } else { 5289 if (netdev->flags & IFF_ALLMULTI) { 5290 rctl |= E1000_RCTL_MPE; 5291 vmolr |= E1000_VMOLR_MPME; 5292 } else { 5293 /* Write addresses to the MTA, if the attempt fails 5294 * then we should just turn on promiscuous mode so 5295 * that we can at least receive multicast traffic 5296 */ 5297 count = igb_write_mc_addr_list(netdev); 5298 if (count < 0) { 5299 rctl |= E1000_RCTL_MPE; 5300 vmolr |= E1000_VMOLR_MPME; 5301 } else if (count) { 5302 vmolr |= E1000_VMOLR_ROMPE; 5303 } 5304 } 5305 } 5306 5307 /* Write addresses to available RAR registers, if there is not 5308 * sufficient space to store all the addresses then enable 5309 * unicast promiscuous mode 5310 */ 5311 if (__dev_uc_sync(netdev, igb_uc_sync, igb_uc_unsync)) { 5312 rctl |= E1000_RCTL_UPE; 5313 vmolr |= E1000_VMOLR_ROPE; 5314 } 5315 5316 /* enable VLAN filtering by default */ 5317 rctl |= E1000_RCTL_VFE; 5318 5319 /* disable VLAN filtering for modes that require it */ 5320 if ((netdev->flags & IFF_PROMISC) || 5321 (netdev->features & NETIF_F_RXALL)) { 5322 /* if we fail to set all rules then just clear VFE */ 5323 if (igb_vlan_promisc_enable(adapter)) 5324 rctl &= ~E1000_RCTL_VFE; 5325 } else { 5326 igb_vlan_promisc_disable(adapter); 5327 } 5328 5329 /* update state of unicast, multicast, and VLAN filtering modes */ 5330 rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE | 5331 E1000_RCTL_VFE); 5332 wr32(E1000_RCTL, rctl); 5333 5334 #if (PAGE_SIZE < 8192) 5335 if (!adapter->vfs_allocated_count) { 5336 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB) 5337 rlpml = IGB_MAX_FRAME_BUILD_SKB; 5338 } 5339 #endif 5340 wr32(E1000_RLPML, rlpml); 5341 5342 /* In order to support SR-IOV and eventually VMDq it is necessary to set 5343 * the VMOLR to enable the appropriate modes. Without this workaround 5344 * we will have issues with VLAN tag stripping not being done for frames 5345 * that are only arriving because we are the default pool 5346 */ 5347 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350)) 5348 return; 5349 5350 /* set UTA to appropriate mode */ 5351 igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE)); 5352 5353 vmolr |= rd32(E1000_VMOLR(vfn)) & 5354 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE); 5355 5356 /* enable Rx jumbo frames, restrict as needed to support build_skb */ 5357 vmolr &= ~E1000_VMOLR_RLPML_MASK; 5358 #if (PAGE_SIZE < 8192) 5359 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB) 5360 vmolr |= IGB_MAX_FRAME_BUILD_SKB; 5361 else 5362 #endif 5363 vmolr |= MAX_JUMBO_FRAME_SIZE; 5364 vmolr |= E1000_VMOLR_LPE; 5365 5366 wr32(E1000_VMOLR(vfn), vmolr); 5367 5368 igb_restore_vf_multicasts(adapter); 5369 } 5370 5371 static void igb_check_wvbr(struct igb_adapter *adapter) 5372 { 5373 struct e1000_hw *hw = &adapter->hw; 5374 u32 wvbr = 0; 5375 5376 switch (hw->mac.type) { 5377 case e1000_82576: 5378 case e1000_i350: 5379 wvbr = rd32(E1000_WVBR); 5380 if (!wvbr) 5381 return; 5382 break; 5383 default: 5384 break; 5385 } 5386 5387 adapter->wvbr |= wvbr; 5388 } 5389 5390 #define IGB_STAGGERED_QUEUE_OFFSET 8 5391 5392 static void igb_spoof_check(struct igb_adapter *adapter) 5393 { 5394 int j; 5395 5396 if (!adapter->wvbr) 5397 return; 5398 5399 for (j = 0; j < adapter->vfs_allocated_count; j++) { 5400 if (adapter->wvbr & BIT(j) || 5401 adapter->wvbr & BIT(j + IGB_STAGGERED_QUEUE_OFFSET)) { 5402 dev_warn(&adapter->pdev->dev, 5403 "Spoof event(s) detected on VF %d\n", j); 5404 adapter->wvbr &= 5405 ~(BIT(j) | 5406 BIT(j + IGB_STAGGERED_QUEUE_OFFSET)); 5407 } 5408 } 5409 } 5410 5411 /* Need to wait a few seconds after link up to get diagnostic information from 5412 * the phy 5413 */ 5414 static void igb_update_phy_info(struct timer_list *t) 5415 { 5416 struct igb_adapter *adapter = from_timer(adapter, t, phy_info_timer); 5417 igb_get_phy_info(&adapter->hw); 5418 } 5419 5420 /** 5421 * igb_has_link - check shared code for link and determine up/down 5422 * @adapter: pointer to driver private info 5423 **/ 5424 bool igb_has_link(struct igb_adapter *adapter) 5425 { 5426 struct e1000_hw *hw = &adapter->hw; 5427 bool link_active = false; 5428 5429 /* get_link_status is set on LSC (link status) interrupt or 5430 * rx sequence error interrupt. get_link_status will stay 5431 * false until the e1000_check_for_link establishes link 5432 * for copper adapters ONLY 5433 */ 5434 switch (hw->phy.media_type) { 5435 case e1000_media_type_copper: 5436 if (!hw->mac.get_link_status) 5437 return true; 5438 fallthrough; 5439 case e1000_media_type_internal_serdes: 5440 hw->mac.ops.check_for_link(hw); 5441 link_active = !hw->mac.get_link_status; 5442 break; 5443 default: 5444 case e1000_media_type_unknown: 5445 break; 5446 } 5447 5448 if (((hw->mac.type == e1000_i210) || 5449 (hw->mac.type == e1000_i211)) && 5450 (hw->phy.id == I210_I_PHY_ID)) { 5451 if (!netif_carrier_ok(adapter->netdev)) { 5452 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE; 5453 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) { 5454 adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE; 5455 adapter->link_check_timeout = jiffies; 5456 } 5457 } 5458 5459 return link_active; 5460 } 5461 5462 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event) 5463 { 5464 bool ret = false; 5465 u32 ctrl_ext, thstat; 5466 5467 /* check for thermal sensor event on i350 copper only */ 5468 if (hw->mac.type == e1000_i350) { 5469 thstat = rd32(E1000_THSTAT); 5470 ctrl_ext = rd32(E1000_CTRL_EXT); 5471 5472 if ((hw->phy.media_type == e1000_media_type_copper) && 5473 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII)) 5474 ret = !!(thstat & event); 5475 } 5476 5477 return ret; 5478 } 5479 5480 /** 5481 * igb_check_lvmmc - check for malformed packets received 5482 * and indicated in LVMMC register 5483 * @adapter: pointer to adapter 5484 **/ 5485 static void igb_check_lvmmc(struct igb_adapter *adapter) 5486 { 5487 struct e1000_hw *hw = &adapter->hw; 5488 u32 lvmmc; 5489 5490 lvmmc = rd32(E1000_LVMMC); 5491 if (lvmmc) { 5492 if (unlikely(net_ratelimit())) { 5493 netdev_warn(adapter->netdev, 5494 "malformed Tx packet detected and dropped, LVMMC:0x%08x\n", 5495 lvmmc); 5496 } 5497 } 5498 } 5499 5500 /** 5501 * igb_watchdog - Timer Call-back 5502 * @t: pointer to timer_list containing our private info pointer 5503 **/ 5504 static void igb_watchdog(struct timer_list *t) 5505 { 5506 struct igb_adapter *adapter = from_timer(adapter, t, watchdog_timer); 5507 /* Do the rest outside of interrupt context */ 5508 schedule_work(&adapter->watchdog_task); 5509 } 5510 5511 static void igb_watchdog_task(struct work_struct *work) 5512 { 5513 struct igb_adapter *adapter = container_of(work, 5514 struct igb_adapter, 5515 watchdog_task); 5516 struct e1000_hw *hw = &adapter->hw; 5517 struct e1000_phy_info *phy = &hw->phy; 5518 struct net_device *netdev = adapter->netdev; 5519 u32 link; 5520 int i; 5521 u32 connsw; 5522 u16 phy_data, retry_count = 20; 5523 5524 link = igb_has_link(adapter); 5525 5526 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) { 5527 if (time_after(jiffies, (adapter->link_check_timeout + HZ))) 5528 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE; 5529 else 5530 link = false; 5531 } 5532 5533 /* Force link down if we have fiber to swap to */ 5534 if (adapter->flags & IGB_FLAG_MAS_ENABLE) { 5535 if (hw->phy.media_type == e1000_media_type_copper) { 5536 connsw = rd32(E1000_CONNSW); 5537 if (!(connsw & E1000_CONNSW_AUTOSENSE_EN)) 5538 link = 0; 5539 } 5540 } 5541 if (link) { 5542 /* Perform a reset if the media type changed. */ 5543 if (hw->dev_spec._82575.media_changed) { 5544 hw->dev_spec._82575.media_changed = false; 5545 adapter->flags |= IGB_FLAG_MEDIA_RESET; 5546 igb_reset(adapter); 5547 } 5548 /* Cancel scheduled suspend requests. */ 5549 pm_runtime_resume(netdev->dev.parent); 5550 5551 if (!netif_carrier_ok(netdev)) { 5552 u32 ctrl; 5553 5554 hw->mac.ops.get_speed_and_duplex(hw, 5555 &adapter->link_speed, 5556 &adapter->link_duplex); 5557 5558 ctrl = rd32(E1000_CTRL); 5559 /* Links status message must follow this format */ 5560 netdev_info(netdev, 5561 "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n", 5562 netdev->name, 5563 adapter->link_speed, 5564 adapter->link_duplex == FULL_DUPLEX ? 5565 "Full" : "Half", 5566 (ctrl & E1000_CTRL_TFCE) && 5567 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" : 5568 (ctrl & E1000_CTRL_RFCE) ? "RX" : 5569 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None"); 5570 5571 /* disable EEE if enabled */ 5572 if ((adapter->flags & IGB_FLAG_EEE) && 5573 (adapter->link_duplex == HALF_DUPLEX)) { 5574 dev_info(&adapter->pdev->dev, 5575 "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n"); 5576 adapter->hw.dev_spec._82575.eee_disable = true; 5577 adapter->flags &= ~IGB_FLAG_EEE; 5578 } 5579 5580 /* check if SmartSpeed worked */ 5581 igb_check_downshift(hw); 5582 if (phy->speed_downgraded) 5583 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n"); 5584 5585 /* check for thermal sensor event */ 5586 if (igb_thermal_sensor_event(hw, 5587 E1000_THSTAT_LINK_THROTTLE)) 5588 netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n"); 5589 5590 /* adjust timeout factor according to speed/duplex */ 5591 adapter->tx_timeout_factor = 1; 5592 switch (adapter->link_speed) { 5593 case SPEED_10: 5594 adapter->tx_timeout_factor = 14; 5595 break; 5596 case SPEED_100: 5597 /* maybe add some timeout factor ? */ 5598 break; 5599 } 5600 5601 if (adapter->link_speed != SPEED_1000 || 5602 !hw->phy.ops.read_reg) 5603 goto no_wait; 5604 5605 /* wait for Remote receiver status OK */ 5606 retry_read_status: 5607 if (!igb_read_phy_reg(hw, PHY_1000T_STATUS, 5608 &phy_data)) { 5609 if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) && 5610 retry_count) { 5611 msleep(100); 5612 retry_count--; 5613 goto retry_read_status; 5614 } else if (!retry_count) { 5615 dev_err(&adapter->pdev->dev, "exceed max 2 second\n"); 5616 } 5617 } else { 5618 dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n"); 5619 } 5620 no_wait: 5621 netif_carrier_on(netdev); 5622 5623 igb_ping_all_vfs(adapter); 5624 igb_check_vf_rate_limit(adapter); 5625 5626 /* link state has changed, schedule phy info update */ 5627 if (!test_bit(__IGB_DOWN, &adapter->state)) 5628 mod_timer(&adapter->phy_info_timer, 5629 round_jiffies(jiffies + 2 * HZ)); 5630 } 5631 } else { 5632 if (netif_carrier_ok(netdev)) { 5633 adapter->link_speed = 0; 5634 adapter->link_duplex = 0; 5635 5636 /* check for thermal sensor event */ 5637 if (igb_thermal_sensor_event(hw, 5638 E1000_THSTAT_PWR_DOWN)) { 5639 netdev_err(netdev, "The network adapter was stopped because it overheated\n"); 5640 } 5641 5642 /* Links status message must follow this format */ 5643 netdev_info(netdev, "igb: %s NIC Link is Down\n", 5644 netdev->name); 5645 netif_carrier_off(netdev); 5646 5647 igb_ping_all_vfs(adapter); 5648 5649 /* link state has changed, schedule phy info update */ 5650 if (!test_bit(__IGB_DOWN, &adapter->state)) 5651 mod_timer(&adapter->phy_info_timer, 5652 round_jiffies(jiffies + 2 * HZ)); 5653 5654 /* link is down, time to check for alternate media */ 5655 if (adapter->flags & IGB_FLAG_MAS_ENABLE) { 5656 igb_check_swap_media(adapter); 5657 if (adapter->flags & IGB_FLAG_MEDIA_RESET) { 5658 schedule_work(&adapter->reset_task); 5659 /* return immediately */ 5660 return; 5661 } 5662 } 5663 pm_schedule_suspend(netdev->dev.parent, 5664 MSEC_PER_SEC * 5); 5665 5666 /* also check for alternate media here */ 5667 } else if (!netif_carrier_ok(netdev) && 5668 (adapter->flags & IGB_FLAG_MAS_ENABLE)) { 5669 igb_check_swap_media(adapter); 5670 if (adapter->flags & IGB_FLAG_MEDIA_RESET) { 5671 schedule_work(&adapter->reset_task); 5672 /* return immediately */ 5673 return; 5674 } 5675 } 5676 } 5677 5678 spin_lock(&adapter->stats64_lock); 5679 igb_update_stats(adapter); 5680 spin_unlock(&adapter->stats64_lock); 5681 5682 for (i = 0; i < adapter->num_tx_queues; i++) { 5683 struct igb_ring *tx_ring = adapter->tx_ring[i]; 5684 if (!netif_carrier_ok(netdev)) { 5685 /* We've lost link, so the controller stops DMA, 5686 * but we've got queued Tx work that's never going 5687 * to get done, so reset controller to flush Tx. 5688 * (Do the reset outside of interrupt context). 5689 */ 5690 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) { 5691 adapter->tx_timeout_count++; 5692 schedule_work(&adapter->reset_task); 5693 /* return immediately since reset is imminent */ 5694 return; 5695 } 5696 } 5697 5698 /* Force detection of hung controller every watchdog period */ 5699 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags); 5700 } 5701 5702 /* Cause software interrupt to ensure Rx ring is cleaned */ 5703 if (adapter->flags & IGB_FLAG_HAS_MSIX) { 5704 u32 eics = 0; 5705 5706 for (i = 0; i < adapter->num_q_vectors; i++) 5707 eics |= adapter->q_vector[i]->eims_value; 5708 wr32(E1000_EICS, eics); 5709 } else { 5710 wr32(E1000_ICS, E1000_ICS_RXDMT0); 5711 } 5712 5713 igb_spoof_check(adapter); 5714 igb_ptp_rx_hang(adapter); 5715 igb_ptp_tx_hang(adapter); 5716 5717 /* Check LVMMC register on i350/i354 only */ 5718 if ((adapter->hw.mac.type == e1000_i350) || 5719 (adapter->hw.mac.type == e1000_i354)) 5720 igb_check_lvmmc(adapter); 5721 5722 /* Reset the timer */ 5723 if (!test_bit(__IGB_DOWN, &adapter->state)) { 5724 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) 5725 mod_timer(&adapter->watchdog_timer, 5726 round_jiffies(jiffies + HZ)); 5727 else 5728 mod_timer(&adapter->watchdog_timer, 5729 round_jiffies(jiffies + 2 * HZ)); 5730 } 5731 } 5732 5733 enum latency_range { 5734 lowest_latency = 0, 5735 low_latency = 1, 5736 bulk_latency = 2, 5737 latency_invalid = 255 5738 }; 5739 5740 /** 5741 * igb_update_ring_itr - update the dynamic ITR value based on packet size 5742 * @q_vector: pointer to q_vector 5743 * 5744 * Stores a new ITR value based on strictly on packet size. This 5745 * algorithm is less sophisticated than that used in igb_update_itr, 5746 * due to the difficulty of synchronizing statistics across multiple 5747 * receive rings. The divisors and thresholds used by this function 5748 * were determined based on theoretical maximum wire speed and testing 5749 * data, in order to minimize response time while increasing bulk 5750 * throughput. 5751 * This functionality is controlled by ethtool's coalescing settings. 5752 * NOTE: This function is called only when operating in a multiqueue 5753 * receive environment. 5754 **/ 5755 static void igb_update_ring_itr(struct igb_q_vector *q_vector) 5756 { 5757 int new_val = q_vector->itr_val; 5758 int avg_wire_size = 0; 5759 struct igb_adapter *adapter = q_vector->adapter; 5760 unsigned int packets; 5761 5762 /* For non-gigabit speeds, just fix the interrupt rate at 4000 5763 * ints/sec - ITR timer value of 120 ticks. 5764 */ 5765 if (adapter->link_speed != SPEED_1000) { 5766 new_val = IGB_4K_ITR; 5767 goto set_itr_val; 5768 } 5769 5770 packets = q_vector->rx.total_packets; 5771 if (packets) 5772 avg_wire_size = q_vector->rx.total_bytes / packets; 5773 5774 packets = q_vector->tx.total_packets; 5775 if (packets) 5776 avg_wire_size = max_t(u32, avg_wire_size, 5777 q_vector->tx.total_bytes / packets); 5778 5779 /* if avg_wire_size isn't set no work was done */ 5780 if (!avg_wire_size) 5781 goto clear_counts; 5782 5783 /* Add 24 bytes to size to account for CRC, preamble, and gap */ 5784 avg_wire_size += 24; 5785 5786 /* Don't starve jumbo frames */ 5787 avg_wire_size = min(avg_wire_size, 3000); 5788 5789 /* Give a little boost to mid-size frames */ 5790 if ((avg_wire_size > 300) && (avg_wire_size < 1200)) 5791 new_val = avg_wire_size / 3; 5792 else 5793 new_val = avg_wire_size / 2; 5794 5795 /* conservative mode (itr 3) eliminates the lowest_latency setting */ 5796 if (new_val < IGB_20K_ITR && 5797 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) || 5798 (!q_vector->rx.ring && adapter->tx_itr_setting == 3))) 5799 new_val = IGB_20K_ITR; 5800 5801 set_itr_val: 5802 if (new_val != q_vector->itr_val) { 5803 q_vector->itr_val = new_val; 5804 q_vector->set_itr = 1; 5805 } 5806 clear_counts: 5807 q_vector->rx.total_bytes = 0; 5808 q_vector->rx.total_packets = 0; 5809 q_vector->tx.total_bytes = 0; 5810 q_vector->tx.total_packets = 0; 5811 } 5812 5813 /** 5814 * igb_update_itr - update the dynamic ITR value based on statistics 5815 * @q_vector: pointer to q_vector 5816 * @ring_container: ring info to update the itr for 5817 * 5818 * Stores a new ITR value based on packets and byte 5819 * counts during the last interrupt. The advantage of per interrupt 5820 * computation is faster updates and more accurate ITR for the current 5821 * traffic pattern. Constants in this function were computed 5822 * based on theoretical maximum wire speed and thresholds were set based 5823 * on testing data as well as attempting to minimize response time 5824 * while increasing bulk throughput. 5825 * This functionality is controlled by ethtool's coalescing settings. 5826 * NOTE: These calculations are only valid when operating in a single- 5827 * queue environment. 5828 **/ 5829 static void igb_update_itr(struct igb_q_vector *q_vector, 5830 struct igb_ring_container *ring_container) 5831 { 5832 unsigned int packets = ring_container->total_packets; 5833 unsigned int bytes = ring_container->total_bytes; 5834 u8 itrval = ring_container->itr; 5835 5836 /* no packets, exit with status unchanged */ 5837 if (packets == 0) 5838 return; 5839 5840 switch (itrval) { 5841 case lowest_latency: 5842 /* handle TSO and jumbo frames */ 5843 if (bytes/packets > 8000) 5844 itrval = bulk_latency; 5845 else if ((packets < 5) && (bytes > 512)) 5846 itrval = low_latency; 5847 break; 5848 case low_latency: /* 50 usec aka 20000 ints/s */ 5849 if (bytes > 10000) { 5850 /* this if handles the TSO accounting */ 5851 if (bytes/packets > 8000) 5852 itrval = bulk_latency; 5853 else if ((packets < 10) || ((bytes/packets) > 1200)) 5854 itrval = bulk_latency; 5855 else if ((packets > 35)) 5856 itrval = lowest_latency; 5857 } else if (bytes/packets > 2000) { 5858 itrval = bulk_latency; 5859 } else if (packets <= 2 && bytes < 512) { 5860 itrval = lowest_latency; 5861 } 5862 break; 5863 case bulk_latency: /* 250 usec aka 4000 ints/s */ 5864 if (bytes > 25000) { 5865 if (packets > 35) 5866 itrval = low_latency; 5867 } else if (bytes < 1500) { 5868 itrval = low_latency; 5869 } 5870 break; 5871 } 5872 5873 /* clear work counters since we have the values we need */ 5874 ring_container->total_bytes = 0; 5875 ring_container->total_packets = 0; 5876 5877 /* write updated itr to ring container */ 5878 ring_container->itr = itrval; 5879 } 5880 5881 static void igb_set_itr(struct igb_q_vector *q_vector) 5882 { 5883 struct igb_adapter *adapter = q_vector->adapter; 5884 u32 new_itr = q_vector->itr_val; 5885 u8 current_itr = 0; 5886 5887 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */ 5888 if (adapter->link_speed != SPEED_1000) { 5889 current_itr = 0; 5890 new_itr = IGB_4K_ITR; 5891 goto set_itr_now; 5892 } 5893 5894 igb_update_itr(q_vector, &q_vector->tx); 5895 igb_update_itr(q_vector, &q_vector->rx); 5896 5897 current_itr = max(q_vector->rx.itr, q_vector->tx.itr); 5898 5899 /* conservative mode (itr 3) eliminates the lowest_latency setting */ 5900 if (current_itr == lowest_latency && 5901 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) || 5902 (!q_vector->rx.ring && adapter->tx_itr_setting == 3))) 5903 current_itr = low_latency; 5904 5905 switch (current_itr) { 5906 /* counts and packets in update_itr are dependent on these numbers */ 5907 case lowest_latency: 5908 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */ 5909 break; 5910 case low_latency: 5911 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */ 5912 break; 5913 case bulk_latency: 5914 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */ 5915 break; 5916 default: 5917 break; 5918 } 5919 5920 set_itr_now: 5921 if (new_itr != q_vector->itr_val) { 5922 /* this attempts to bias the interrupt rate towards Bulk 5923 * by adding intermediate steps when interrupt rate is 5924 * increasing 5925 */ 5926 new_itr = new_itr > q_vector->itr_val ? 5927 max((new_itr * q_vector->itr_val) / 5928 (new_itr + (q_vector->itr_val >> 2)), 5929 new_itr) : new_itr; 5930 /* Don't write the value here; it resets the adapter's 5931 * internal timer, and causes us to delay far longer than 5932 * we should between interrupts. Instead, we write the ITR 5933 * value at the beginning of the next interrupt so the timing 5934 * ends up being correct. 5935 */ 5936 q_vector->itr_val = new_itr; 5937 q_vector->set_itr = 1; 5938 } 5939 } 5940 5941 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, 5942 struct igb_tx_buffer *first, 5943 u32 vlan_macip_lens, u32 type_tucmd, 5944 u32 mss_l4len_idx) 5945 { 5946 struct e1000_adv_tx_context_desc *context_desc; 5947 u16 i = tx_ring->next_to_use; 5948 struct timespec64 ts; 5949 5950 context_desc = IGB_TX_CTXTDESC(tx_ring, i); 5951 5952 i++; 5953 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; 5954 5955 /* set bits to identify this as an advanced context descriptor */ 5956 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT; 5957 5958 /* For 82575, context index must be unique per ring. */ 5959 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags)) 5960 mss_l4len_idx |= tx_ring->reg_idx << 4; 5961 5962 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens); 5963 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd); 5964 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx); 5965 5966 /* We assume there is always a valid tx time available. Invalid times 5967 * should have been handled by the upper layers. 5968 */ 5969 if (tx_ring->launchtime_enable) { 5970 ts = ktime_to_timespec64(first->skb->tstamp); 5971 skb_txtime_consumed(first->skb); 5972 context_desc->seqnum_seed = cpu_to_le32(ts.tv_nsec / 32); 5973 } else { 5974 context_desc->seqnum_seed = 0; 5975 } 5976 } 5977 5978 static int igb_tso(struct igb_ring *tx_ring, 5979 struct igb_tx_buffer *first, 5980 u8 *hdr_len) 5981 { 5982 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx; 5983 struct sk_buff *skb = first->skb; 5984 union { 5985 struct iphdr *v4; 5986 struct ipv6hdr *v6; 5987 unsigned char *hdr; 5988 } ip; 5989 union { 5990 struct tcphdr *tcp; 5991 struct udphdr *udp; 5992 unsigned char *hdr; 5993 } l4; 5994 u32 paylen, l4_offset; 5995 int err; 5996 5997 if (skb->ip_summed != CHECKSUM_PARTIAL) 5998 return 0; 5999 6000 if (!skb_is_gso(skb)) 6001 return 0; 6002 6003 err = skb_cow_head(skb, 0); 6004 if (err < 0) 6005 return err; 6006 6007 ip.hdr = skb_network_header(skb); 6008 l4.hdr = skb_checksum_start(skb); 6009 6010 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */ 6011 type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ? 6012 E1000_ADVTXD_TUCMD_L4T_UDP : E1000_ADVTXD_TUCMD_L4T_TCP; 6013 6014 /* initialize outer IP header fields */ 6015 if (ip.v4->version == 4) { 6016 unsigned char *csum_start = skb_checksum_start(skb); 6017 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4); 6018 6019 /* IP header will have to cancel out any data that 6020 * is not a part of the outer IP header 6021 */ 6022 ip.v4->check = csum_fold(csum_partial(trans_start, 6023 csum_start - trans_start, 6024 0)); 6025 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4; 6026 6027 ip.v4->tot_len = 0; 6028 first->tx_flags |= IGB_TX_FLAGS_TSO | 6029 IGB_TX_FLAGS_CSUM | 6030 IGB_TX_FLAGS_IPV4; 6031 } else { 6032 ip.v6->payload_len = 0; 6033 first->tx_flags |= IGB_TX_FLAGS_TSO | 6034 IGB_TX_FLAGS_CSUM; 6035 } 6036 6037 /* determine offset of inner transport header */ 6038 l4_offset = l4.hdr - skb->data; 6039 6040 /* remove payload length from inner checksum */ 6041 paylen = skb->len - l4_offset; 6042 if (type_tucmd & E1000_ADVTXD_TUCMD_L4T_TCP) { 6043 /* compute length of segmentation header */ 6044 *hdr_len = (l4.tcp->doff * 4) + l4_offset; 6045 csum_replace_by_diff(&l4.tcp->check, 6046 (__force __wsum)htonl(paylen)); 6047 } else { 6048 /* compute length of segmentation header */ 6049 *hdr_len = sizeof(*l4.udp) + l4_offset; 6050 csum_replace_by_diff(&l4.udp->check, 6051 (__force __wsum)htonl(paylen)); 6052 } 6053 6054 /* update gso size and bytecount with header size */ 6055 first->gso_segs = skb_shinfo(skb)->gso_segs; 6056 first->bytecount += (first->gso_segs - 1) * *hdr_len; 6057 6058 /* MSS L4LEN IDX */ 6059 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT; 6060 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT; 6061 6062 /* VLAN MACLEN IPLEN */ 6063 vlan_macip_lens = l4.hdr - ip.hdr; 6064 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT; 6065 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK; 6066 6067 igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, 6068 type_tucmd, mss_l4len_idx); 6069 6070 return 1; 6071 } 6072 6073 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first) 6074 { 6075 struct sk_buff *skb = first->skb; 6076 u32 vlan_macip_lens = 0; 6077 u32 type_tucmd = 0; 6078 6079 if (skb->ip_summed != CHECKSUM_PARTIAL) { 6080 csum_failed: 6081 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN) && 6082 !tx_ring->launchtime_enable) 6083 return; 6084 goto no_csum; 6085 } 6086 6087 switch (skb->csum_offset) { 6088 case offsetof(struct tcphdr, check): 6089 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP; 6090 fallthrough; 6091 case offsetof(struct udphdr, check): 6092 break; 6093 case offsetof(struct sctphdr, checksum): 6094 /* validate that this is actually an SCTP request */ 6095 if (skb_csum_is_sctp(skb)) { 6096 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP; 6097 break; 6098 } 6099 fallthrough; 6100 default: 6101 skb_checksum_help(skb); 6102 goto csum_failed; 6103 } 6104 6105 /* update TX checksum flag */ 6106 first->tx_flags |= IGB_TX_FLAGS_CSUM; 6107 vlan_macip_lens = skb_checksum_start_offset(skb) - 6108 skb_network_offset(skb); 6109 no_csum: 6110 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT; 6111 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK; 6112 6113 igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, type_tucmd, 0); 6114 } 6115 6116 #define IGB_SET_FLAG(_input, _flag, _result) \ 6117 ((_flag <= _result) ? \ 6118 ((u32)(_input & _flag) * (_result / _flag)) : \ 6119 ((u32)(_input & _flag) / (_flag / _result))) 6120 6121 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags) 6122 { 6123 /* set type for advanced descriptor with frame checksum insertion */ 6124 u32 cmd_type = E1000_ADVTXD_DTYP_DATA | 6125 E1000_ADVTXD_DCMD_DEXT | 6126 E1000_ADVTXD_DCMD_IFCS; 6127 6128 /* set HW vlan bit if vlan is present */ 6129 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN, 6130 (E1000_ADVTXD_DCMD_VLE)); 6131 6132 /* set segmentation bits for TSO */ 6133 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO, 6134 (E1000_ADVTXD_DCMD_TSE)); 6135 6136 /* set timestamp bit if present */ 6137 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP, 6138 (E1000_ADVTXD_MAC_TSTAMP)); 6139 6140 /* insert frame checksum */ 6141 cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS); 6142 6143 return cmd_type; 6144 } 6145 6146 static void igb_tx_olinfo_status(struct igb_ring *tx_ring, 6147 union e1000_adv_tx_desc *tx_desc, 6148 u32 tx_flags, unsigned int paylen) 6149 { 6150 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT; 6151 6152 /* 82575 requires a unique index per ring */ 6153 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags)) 6154 olinfo_status |= tx_ring->reg_idx << 4; 6155 6156 /* insert L4 checksum */ 6157 olinfo_status |= IGB_SET_FLAG(tx_flags, 6158 IGB_TX_FLAGS_CSUM, 6159 (E1000_TXD_POPTS_TXSM << 8)); 6160 6161 /* insert IPv4 checksum */ 6162 olinfo_status |= IGB_SET_FLAG(tx_flags, 6163 IGB_TX_FLAGS_IPV4, 6164 (E1000_TXD_POPTS_IXSM << 8)); 6165 6166 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status); 6167 } 6168 6169 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size) 6170 { 6171 struct net_device *netdev = tx_ring->netdev; 6172 6173 netif_stop_subqueue(netdev, tx_ring->queue_index); 6174 6175 /* Herbert's original patch had: 6176 * smp_mb__after_netif_stop_queue(); 6177 * but since that doesn't exist yet, just open code it. 6178 */ 6179 smp_mb(); 6180 6181 /* We need to check again in a case another CPU has just 6182 * made room available. 6183 */ 6184 if (igb_desc_unused(tx_ring) < size) 6185 return -EBUSY; 6186 6187 /* A reprieve! */ 6188 netif_wake_subqueue(netdev, tx_ring->queue_index); 6189 6190 u64_stats_update_begin(&tx_ring->tx_syncp2); 6191 tx_ring->tx_stats.restart_queue2++; 6192 u64_stats_update_end(&tx_ring->tx_syncp2); 6193 6194 return 0; 6195 } 6196 6197 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size) 6198 { 6199 if (igb_desc_unused(tx_ring) >= size) 6200 return 0; 6201 return __igb_maybe_stop_tx(tx_ring, size); 6202 } 6203 6204 static int igb_tx_map(struct igb_ring *tx_ring, 6205 struct igb_tx_buffer *first, 6206 const u8 hdr_len) 6207 { 6208 struct sk_buff *skb = first->skb; 6209 struct igb_tx_buffer *tx_buffer; 6210 union e1000_adv_tx_desc *tx_desc; 6211 skb_frag_t *frag; 6212 dma_addr_t dma; 6213 unsigned int data_len, size; 6214 u32 tx_flags = first->tx_flags; 6215 u32 cmd_type = igb_tx_cmd_type(skb, tx_flags); 6216 u16 i = tx_ring->next_to_use; 6217 6218 tx_desc = IGB_TX_DESC(tx_ring, i); 6219 6220 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len); 6221 6222 size = skb_headlen(skb); 6223 data_len = skb->data_len; 6224 6225 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE); 6226 6227 tx_buffer = first; 6228 6229 for (frag = &skb_shinfo(skb)->frags[0];; frag++) { 6230 if (dma_mapping_error(tx_ring->dev, dma)) 6231 goto dma_error; 6232 6233 /* record length, and DMA address */ 6234 dma_unmap_len_set(tx_buffer, len, size); 6235 dma_unmap_addr_set(tx_buffer, dma, dma); 6236 6237 tx_desc->read.buffer_addr = cpu_to_le64(dma); 6238 6239 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) { 6240 tx_desc->read.cmd_type_len = 6241 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD); 6242 6243 i++; 6244 tx_desc++; 6245 if (i == tx_ring->count) { 6246 tx_desc = IGB_TX_DESC(tx_ring, 0); 6247 i = 0; 6248 } 6249 tx_desc->read.olinfo_status = 0; 6250 6251 dma += IGB_MAX_DATA_PER_TXD; 6252 size -= IGB_MAX_DATA_PER_TXD; 6253 6254 tx_desc->read.buffer_addr = cpu_to_le64(dma); 6255 } 6256 6257 if (likely(!data_len)) 6258 break; 6259 6260 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size); 6261 6262 i++; 6263 tx_desc++; 6264 if (i == tx_ring->count) { 6265 tx_desc = IGB_TX_DESC(tx_ring, 0); 6266 i = 0; 6267 } 6268 tx_desc->read.olinfo_status = 0; 6269 6270 size = skb_frag_size(frag); 6271 data_len -= size; 6272 6273 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, 6274 size, DMA_TO_DEVICE); 6275 6276 tx_buffer = &tx_ring->tx_buffer_info[i]; 6277 } 6278 6279 /* write last descriptor with RS and EOP bits */ 6280 cmd_type |= size | IGB_TXD_DCMD; 6281 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type); 6282 6283 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount); 6284 6285 /* set the timestamp */ 6286 first->time_stamp = jiffies; 6287 6288 skb_tx_timestamp(skb); 6289 6290 /* Force memory writes to complete before letting h/w know there 6291 * are new descriptors to fetch. (Only applicable for weak-ordered 6292 * memory model archs, such as IA-64). 6293 * 6294 * We also need this memory barrier to make certain all of the 6295 * status bits have been updated before next_to_watch is written. 6296 */ 6297 dma_wmb(); 6298 6299 /* set next_to_watch value indicating a packet is present */ 6300 first->next_to_watch = tx_desc; 6301 6302 i++; 6303 if (i == tx_ring->count) 6304 i = 0; 6305 6306 tx_ring->next_to_use = i; 6307 6308 /* Make sure there is space in the ring for the next send. */ 6309 igb_maybe_stop_tx(tx_ring, DESC_NEEDED); 6310 6311 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) { 6312 writel(i, tx_ring->tail); 6313 } 6314 return 0; 6315 6316 dma_error: 6317 dev_err(tx_ring->dev, "TX DMA map failed\n"); 6318 tx_buffer = &tx_ring->tx_buffer_info[i]; 6319 6320 /* clear dma mappings for failed tx_buffer_info map */ 6321 while (tx_buffer != first) { 6322 if (dma_unmap_len(tx_buffer, len)) 6323 dma_unmap_page(tx_ring->dev, 6324 dma_unmap_addr(tx_buffer, dma), 6325 dma_unmap_len(tx_buffer, len), 6326 DMA_TO_DEVICE); 6327 dma_unmap_len_set(tx_buffer, len, 0); 6328 6329 if (i-- == 0) 6330 i += tx_ring->count; 6331 tx_buffer = &tx_ring->tx_buffer_info[i]; 6332 } 6333 6334 if (dma_unmap_len(tx_buffer, len)) 6335 dma_unmap_single(tx_ring->dev, 6336 dma_unmap_addr(tx_buffer, dma), 6337 dma_unmap_len(tx_buffer, len), 6338 DMA_TO_DEVICE); 6339 dma_unmap_len_set(tx_buffer, len, 0); 6340 6341 dev_kfree_skb_any(tx_buffer->skb); 6342 tx_buffer->skb = NULL; 6343 6344 tx_ring->next_to_use = i; 6345 6346 return -1; 6347 } 6348 6349 int igb_xmit_xdp_ring(struct igb_adapter *adapter, 6350 struct igb_ring *tx_ring, 6351 struct xdp_frame *xdpf) 6352 { 6353 struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf); 6354 u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0; 6355 u16 count, i, index = tx_ring->next_to_use; 6356 struct igb_tx_buffer *tx_head = &tx_ring->tx_buffer_info[index]; 6357 struct igb_tx_buffer *tx_buffer = tx_head; 6358 union e1000_adv_tx_desc *tx_desc = IGB_TX_DESC(tx_ring, index); 6359 u32 len = xdpf->len, cmd_type, olinfo_status; 6360 void *data = xdpf->data; 6361 6362 count = TXD_USE_COUNT(len); 6363 for (i = 0; i < nr_frags; i++) 6364 count += TXD_USE_COUNT(skb_frag_size(&sinfo->frags[i])); 6365 6366 if (igb_maybe_stop_tx(tx_ring, count + 3)) 6367 return IGB_XDP_CONSUMED; 6368 6369 i = 0; 6370 /* record the location of the first descriptor for this packet */ 6371 tx_head->bytecount = xdp_get_frame_len(xdpf); 6372 tx_head->type = IGB_TYPE_XDP; 6373 tx_head->gso_segs = 1; 6374 tx_head->xdpf = xdpf; 6375 6376 olinfo_status = tx_head->bytecount << E1000_ADVTXD_PAYLEN_SHIFT; 6377 /* 82575 requires a unique index per ring */ 6378 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags)) 6379 olinfo_status |= tx_ring->reg_idx << 4; 6380 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status); 6381 6382 for (;;) { 6383 dma_addr_t dma; 6384 6385 dma = dma_map_single(tx_ring->dev, data, len, DMA_TO_DEVICE); 6386 if (dma_mapping_error(tx_ring->dev, dma)) 6387 goto unmap; 6388 6389 /* record length, and DMA address */ 6390 dma_unmap_len_set(tx_buffer, len, len); 6391 dma_unmap_addr_set(tx_buffer, dma, dma); 6392 6393 /* put descriptor type bits */ 6394 cmd_type = E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_DEXT | 6395 E1000_ADVTXD_DCMD_IFCS | len; 6396 6397 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type); 6398 tx_desc->read.buffer_addr = cpu_to_le64(dma); 6399 6400 tx_buffer->protocol = 0; 6401 6402 if (++index == tx_ring->count) 6403 index = 0; 6404 6405 if (i == nr_frags) 6406 break; 6407 6408 tx_buffer = &tx_ring->tx_buffer_info[index]; 6409 tx_desc = IGB_TX_DESC(tx_ring, index); 6410 tx_desc->read.olinfo_status = 0; 6411 6412 data = skb_frag_address(&sinfo->frags[i]); 6413 len = skb_frag_size(&sinfo->frags[i]); 6414 i++; 6415 } 6416 tx_desc->read.cmd_type_len |= cpu_to_le32(IGB_TXD_DCMD); 6417 6418 netdev_tx_sent_queue(txring_txq(tx_ring), tx_head->bytecount); 6419 /* set the timestamp */ 6420 tx_head->time_stamp = jiffies; 6421 6422 /* Avoid any potential race with xdp_xmit and cleanup */ 6423 smp_wmb(); 6424 6425 /* set next_to_watch value indicating a packet is present */ 6426 tx_head->next_to_watch = tx_desc; 6427 tx_ring->next_to_use = index; 6428 6429 /* Make sure there is space in the ring for the next send. */ 6430 igb_maybe_stop_tx(tx_ring, DESC_NEEDED); 6431 6432 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) 6433 writel(index, tx_ring->tail); 6434 6435 return IGB_XDP_TX; 6436 6437 unmap: 6438 for (;;) { 6439 tx_buffer = &tx_ring->tx_buffer_info[index]; 6440 if (dma_unmap_len(tx_buffer, len)) 6441 dma_unmap_page(tx_ring->dev, 6442 dma_unmap_addr(tx_buffer, dma), 6443 dma_unmap_len(tx_buffer, len), 6444 DMA_TO_DEVICE); 6445 dma_unmap_len_set(tx_buffer, len, 0); 6446 if (tx_buffer == tx_head) 6447 break; 6448 6449 if (!index) 6450 index += tx_ring->count; 6451 index--; 6452 } 6453 6454 return IGB_XDP_CONSUMED; 6455 } 6456 6457 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb, 6458 struct igb_ring *tx_ring) 6459 { 6460 struct igb_tx_buffer *first; 6461 int tso; 6462 u32 tx_flags = 0; 6463 unsigned short f; 6464 u16 count = TXD_USE_COUNT(skb_headlen(skb)); 6465 __be16 protocol = vlan_get_protocol(skb); 6466 u8 hdr_len = 0; 6467 6468 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD, 6469 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD, 6470 * + 2 desc gap to keep tail from touching head, 6471 * + 1 desc for context descriptor, 6472 * otherwise try next time 6473 */ 6474 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) 6475 count += TXD_USE_COUNT(skb_frag_size( 6476 &skb_shinfo(skb)->frags[f])); 6477 6478 if (igb_maybe_stop_tx(tx_ring, count + 3)) { 6479 /* this is a hard error */ 6480 return NETDEV_TX_BUSY; 6481 } 6482 6483 /* record the location of the first descriptor for this packet */ 6484 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use]; 6485 first->type = IGB_TYPE_SKB; 6486 first->skb = skb; 6487 first->bytecount = skb->len; 6488 first->gso_segs = 1; 6489 6490 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) { 6491 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev); 6492 6493 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON && 6494 !test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS, 6495 &adapter->state)) { 6496 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS; 6497 tx_flags |= IGB_TX_FLAGS_TSTAMP; 6498 6499 adapter->ptp_tx_skb = skb_get(skb); 6500 adapter->ptp_tx_start = jiffies; 6501 if (adapter->hw.mac.type == e1000_82576) 6502 schedule_work(&adapter->ptp_tx_work); 6503 } else { 6504 adapter->tx_hwtstamp_skipped++; 6505 } 6506 } 6507 6508 if (skb_vlan_tag_present(skb)) { 6509 tx_flags |= IGB_TX_FLAGS_VLAN; 6510 tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT); 6511 } 6512 6513 /* record initial flags and protocol */ 6514 first->tx_flags = tx_flags; 6515 first->protocol = protocol; 6516 6517 tso = igb_tso(tx_ring, first, &hdr_len); 6518 if (tso < 0) 6519 goto out_drop; 6520 else if (!tso) 6521 igb_tx_csum(tx_ring, first); 6522 6523 if (igb_tx_map(tx_ring, first, hdr_len)) 6524 goto cleanup_tx_tstamp; 6525 6526 return NETDEV_TX_OK; 6527 6528 out_drop: 6529 dev_kfree_skb_any(first->skb); 6530 first->skb = NULL; 6531 cleanup_tx_tstamp: 6532 if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP)) { 6533 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev); 6534 6535 dev_kfree_skb_any(adapter->ptp_tx_skb); 6536 adapter->ptp_tx_skb = NULL; 6537 if (adapter->hw.mac.type == e1000_82576) 6538 cancel_work_sync(&adapter->ptp_tx_work); 6539 clear_bit_unlock(__IGB_PTP_TX_IN_PROGRESS, &adapter->state); 6540 } 6541 6542 return NETDEV_TX_OK; 6543 } 6544 6545 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter, 6546 struct sk_buff *skb) 6547 { 6548 unsigned int r_idx = skb->queue_mapping; 6549 6550 if (r_idx >= adapter->num_tx_queues) 6551 r_idx = r_idx % adapter->num_tx_queues; 6552 6553 return adapter->tx_ring[r_idx]; 6554 } 6555 6556 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, 6557 struct net_device *netdev) 6558 { 6559 struct igb_adapter *adapter = netdev_priv(netdev); 6560 6561 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb 6562 * in order to meet this minimum size requirement. 6563 */ 6564 if (skb_put_padto(skb, 17)) 6565 return NETDEV_TX_OK; 6566 6567 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb)); 6568 } 6569 6570 /** 6571 * igb_tx_timeout - Respond to a Tx Hang 6572 * @netdev: network interface device structure 6573 * @txqueue: number of the Tx queue that hung (unused) 6574 **/ 6575 static void igb_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue) 6576 { 6577 struct igb_adapter *adapter = netdev_priv(netdev); 6578 struct e1000_hw *hw = &adapter->hw; 6579 6580 /* Do the reset outside of interrupt context */ 6581 adapter->tx_timeout_count++; 6582 6583 if (hw->mac.type >= e1000_82580) 6584 hw->dev_spec._82575.global_device_reset = true; 6585 6586 schedule_work(&adapter->reset_task); 6587 wr32(E1000_EICS, 6588 (adapter->eims_enable_mask & ~adapter->eims_other)); 6589 } 6590 6591 static void igb_reset_task(struct work_struct *work) 6592 { 6593 struct igb_adapter *adapter; 6594 adapter = container_of(work, struct igb_adapter, reset_task); 6595 6596 rtnl_lock(); 6597 /* If we're already down or resetting, just bail */ 6598 if (test_bit(__IGB_DOWN, &adapter->state) || 6599 test_bit(__IGB_RESETTING, &adapter->state)) { 6600 rtnl_unlock(); 6601 return; 6602 } 6603 6604 igb_dump(adapter); 6605 netdev_err(adapter->netdev, "Reset adapter\n"); 6606 igb_reinit_locked(adapter); 6607 rtnl_unlock(); 6608 } 6609 6610 /** 6611 * igb_get_stats64 - Get System Network Statistics 6612 * @netdev: network interface device structure 6613 * @stats: rtnl_link_stats64 pointer 6614 **/ 6615 static void igb_get_stats64(struct net_device *netdev, 6616 struct rtnl_link_stats64 *stats) 6617 { 6618 struct igb_adapter *adapter = netdev_priv(netdev); 6619 6620 spin_lock(&adapter->stats64_lock); 6621 igb_update_stats(adapter); 6622 memcpy(stats, &adapter->stats64, sizeof(*stats)); 6623 spin_unlock(&adapter->stats64_lock); 6624 } 6625 6626 /** 6627 * igb_change_mtu - Change the Maximum Transfer Unit 6628 * @netdev: network interface device structure 6629 * @new_mtu: new value for maximum frame size 6630 * 6631 * Returns 0 on success, negative on failure 6632 **/ 6633 static int igb_change_mtu(struct net_device *netdev, int new_mtu) 6634 { 6635 struct igb_adapter *adapter = netdev_priv(netdev); 6636 int max_frame = new_mtu + IGB_ETH_PKT_HDR_PAD; 6637 6638 if (adapter->xdp_prog) { 6639 int i; 6640 6641 for (i = 0; i < adapter->num_rx_queues; i++) { 6642 struct igb_ring *ring = adapter->rx_ring[i]; 6643 6644 if (max_frame > igb_rx_bufsz(ring)) { 6645 netdev_warn(adapter->netdev, 6646 "Requested MTU size is not supported with XDP. Max frame size is %d\n", 6647 max_frame); 6648 return -EINVAL; 6649 } 6650 } 6651 } 6652 6653 /* adjust max frame to be at least the size of a standard frame */ 6654 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN)) 6655 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN; 6656 6657 while (test_and_set_bit(__IGB_RESETTING, &adapter->state)) 6658 usleep_range(1000, 2000); 6659 6660 /* igb_down has a dependency on max_frame_size */ 6661 adapter->max_frame_size = max_frame; 6662 6663 if (netif_running(netdev)) 6664 igb_down(adapter); 6665 6666 netdev_dbg(netdev, "changing MTU from %d to %d\n", 6667 netdev->mtu, new_mtu); 6668 netdev->mtu = new_mtu; 6669 6670 if (netif_running(netdev)) 6671 igb_up(adapter); 6672 else 6673 igb_reset(adapter); 6674 6675 clear_bit(__IGB_RESETTING, &adapter->state); 6676 6677 return 0; 6678 } 6679 6680 /** 6681 * igb_update_stats - Update the board statistics counters 6682 * @adapter: board private structure 6683 **/ 6684 void igb_update_stats(struct igb_adapter *adapter) 6685 { 6686 struct rtnl_link_stats64 *net_stats = &adapter->stats64; 6687 struct e1000_hw *hw = &adapter->hw; 6688 struct pci_dev *pdev = adapter->pdev; 6689 u32 reg, mpc; 6690 int i; 6691 u64 bytes, packets; 6692 unsigned int start; 6693 u64 _bytes, _packets; 6694 6695 /* Prevent stats update while adapter is being reset, or if the pci 6696 * connection is down. 6697 */ 6698 if (adapter->link_speed == 0) 6699 return; 6700 if (pci_channel_offline(pdev)) 6701 return; 6702 6703 bytes = 0; 6704 packets = 0; 6705 6706 rcu_read_lock(); 6707 for (i = 0; i < adapter->num_rx_queues; i++) { 6708 struct igb_ring *ring = adapter->rx_ring[i]; 6709 u32 rqdpc = rd32(E1000_RQDPC(i)); 6710 if (hw->mac.type >= e1000_i210) 6711 wr32(E1000_RQDPC(i), 0); 6712 6713 if (rqdpc) { 6714 ring->rx_stats.drops += rqdpc; 6715 net_stats->rx_fifo_errors += rqdpc; 6716 } 6717 6718 do { 6719 start = u64_stats_fetch_begin(&ring->rx_syncp); 6720 _bytes = ring->rx_stats.bytes; 6721 _packets = ring->rx_stats.packets; 6722 } while (u64_stats_fetch_retry(&ring->rx_syncp, start)); 6723 bytes += _bytes; 6724 packets += _packets; 6725 } 6726 6727 net_stats->rx_bytes = bytes; 6728 net_stats->rx_packets = packets; 6729 6730 bytes = 0; 6731 packets = 0; 6732 for (i = 0; i < adapter->num_tx_queues; i++) { 6733 struct igb_ring *ring = adapter->tx_ring[i]; 6734 do { 6735 start = u64_stats_fetch_begin(&ring->tx_syncp); 6736 _bytes = ring->tx_stats.bytes; 6737 _packets = ring->tx_stats.packets; 6738 } while (u64_stats_fetch_retry(&ring->tx_syncp, start)); 6739 bytes += _bytes; 6740 packets += _packets; 6741 } 6742 net_stats->tx_bytes = bytes; 6743 net_stats->tx_packets = packets; 6744 rcu_read_unlock(); 6745 6746 /* read stats registers */ 6747 adapter->stats.crcerrs += rd32(E1000_CRCERRS); 6748 adapter->stats.gprc += rd32(E1000_GPRC); 6749 adapter->stats.gorc += rd32(E1000_GORCL); 6750 rd32(E1000_GORCH); /* clear GORCL */ 6751 adapter->stats.bprc += rd32(E1000_BPRC); 6752 adapter->stats.mprc += rd32(E1000_MPRC); 6753 adapter->stats.roc += rd32(E1000_ROC); 6754 6755 adapter->stats.prc64 += rd32(E1000_PRC64); 6756 adapter->stats.prc127 += rd32(E1000_PRC127); 6757 adapter->stats.prc255 += rd32(E1000_PRC255); 6758 adapter->stats.prc511 += rd32(E1000_PRC511); 6759 adapter->stats.prc1023 += rd32(E1000_PRC1023); 6760 adapter->stats.prc1522 += rd32(E1000_PRC1522); 6761 adapter->stats.symerrs += rd32(E1000_SYMERRS); 6762 adapter->stats.sec += rd32(E1000_SEC); 6763 6764 mpc = rd32(E1000_MPC); 6765 adapter->stats.mpc += mpc; 6766 net_stats->rx_fifo_errors += mpc; 6767 adapter->stats.scc += rd32(E1000_SCC); 6768 adapter->stats.ecol += rd32(E1000_ECOL); 6769 adapter->stats.mcc += rd32(E1000_MCC); 6770 adapter->stats.latecol += rd32(E1000_LATECOL); 6771 adapter->stats.dc += rd32(E1000_DC); 6772 adapter->stats.rlec += rd32(E1000_RLEC); 6773 adapter->stats.xonrxc += rd32(E1000_XONRXC); 6774 adapter->stats.xontxc += rd32(E1000_XONTXC); 6775 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC); 6776 adapter->stats.xofftxc += rd32(E1000_XOFFTXC); 6777 adapter->stats.fcruc += rd32(E1000_FCRUC); 6778 adapter->stats.gptc += rd32(E1000_GPTC); 6779 adapter->stats.gotc += rd32(E1000_GOTCL); 6780 rd32(E1000_GOTCH); /* clear GOTCL */ 6781 adapter->stats.rnbc += rd32(E1000_RNBC); 6782 adapter->stats.ruc += rd32(E1000_RUC); 6783 adapter->stats.rfc += rd32(E1000_RFC); 6784 adapter->stats.rjc += rd32(E1000_RJC); 6785 adapter->stats.tor += rd32(E1000_TORH); 6786 adapter->stats.tot += rd32(E1000_TOTH); 6787 adapter->stats.tpr += rd32(E1000_TPR); 6788 6789 adapter->stats.ptc64 += rd32(E1000_PTC64); 6790 adapter->stats.ptc127 += rd32(E1000_PTC127); 6791 adapter->stats.ptc255 += rd32(E1000_PTC255); 6792 adapter->stats.ptc511 += rd32(E1000_PTC511); 6793 adapter->stats.ptc1023 += rd32(E1000_PTC1023); 6794 adapter->stats.ptc1522 += rd32(E1000_PTC1522); 6795 6796 adapter->stats.mptc += rd32(E1000_MPTC); 6797 adapter->stats.bptc += rd32(E1000_BPTC); 6798 6799 adapter->stats.tpt += rd32(E1000_TPT); 6800 adapter->stats.colc += rd32(E1000_COLC); 6801 6802 adapter->stats.algnerrc += rd32(E1000_ALGNERRC); 6803 /* read internal phy specific stats */ 6804 reg = rd32(E1000_CTRL_EXT); 6805 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) { 6806 adapter->stats.rxerrc += rd32(E1000_RXERRC); 6807 6808 /* this stat has invalid values on i210/i211 */ 6809 if ((hw->mac.type != e1000_i210) && 6810 (hw->mac.type != e1000_i211)) 6811 adapter->stats.tncrs += rd32(E1000_TNCRS); 6812 } 6813 6814 adapter->stats.tsctc += rd32(E1000_TSCTC); 6815 adapter->stats.tsctfc += rd32(E1000_TSCTFC); 6816 6817 adapter->stats.iac += rd32(E1000_IAC); 6818 adapter->stats.icrxoc += rd32(E1000_ICRXOC); 6819 adapter->stats.icrxptc += rd32(E1000_ICRXPTC); 6820 adapter->stats.icrxatc += rd32(E1000_ICRXATC); 6821 adapter->stats.ictxptc += rd32(E1000_ICTXPTC); 6822 adapter->stats.ictxatc += rd32(E1000_ICTXATC); 6823 adapter->stats.ictxqec += rd32(E1000_ICTXQEC); 6824 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC); 6825 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC); 6826 6827 /* Fill out the OS statistics structure */ 6828 net_stats->multicast = adapter->stats.mprc; 6829 net_stats->collisions = adapter->stats.colc; 6830 6831 /* Rx Errors */ 6832 6833 /* RLEC on some newer hardware can be incorrect so build 6834 * our own version based on RUC and ROC 6835 */ 6836 net_stats->rx_errors = adapter->stats.rxerrc + 6837 adapter->stats.crcerrs + adapter->stats.algnerrc + 6838 adapter->stats.ruc + adapter->stats.roc + 6839 adapter->stats.cexterr; 6840 net_stats->rx_length_errors = adapter->stats.ruc + 6841 adapter->stats.roc; 6842 net_stats->rx_crc_errors = adapter->stats.crcerrs; 6843 net_stats->rx_frame_errors = adapter->stats.algnerrc; 6844 net_stats->rx_missed_errors = adapter->stats.mpc; 6845 6846 /* Tx Errors */ 6847 net_stats->tx_errors = adapter->stats.ecol + 6848 adapter->stats.latecol; 6849 net_stats->tx_aborted_errors = adapter->stats.ecol; 6850 net_stats->tx_window_errors = adapter->stats.latecol; 6851 net_stats->tx_carrier_errors = adapter->stats.tncrs; 6852 6853 /* Tx Dropped needs to be maintained elsewhere */ 6854 6855 /* Management Stats */ 6856 adapter->stats.mgptc += rd32(E1000_MGTPTC); 6857 adapter->stats.mgprc += rd32(E1000_MGTPRC); 6858 adapter->stats.mgpdc += rd32(E1000_MGTPDC); 6859 6860 /* OS2BMC Stats */ 6861 reg = rd32(E1000_MANC); 6862 if (reg & E1000_MANC_EN_BMC2OS) { 6863 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC); 6864 adapter->stats.o2bspc += rd32(E1000_O2BSPC); 6865 adapter->stats.b2ospc += rd32(E1000_B2OSPC); 6866 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC); 6867 } 6868 } 6869 6870 static void igb_perout(struct igb_adapter *adapter, int tsintr_tt) 6871 { 6872 int pin = ptp_find_pin(adapter->ptp_clock, PTP_PF_PEROUT, tsintr_tt); 6873 struct e1000_hw *hw = &adapter->hw; 6874 struct timespec64 ts; 6875 u32 tsauxc; 6876 6877 if (pin < 0 || pin >= IGB_N_SDP) 6878 return; 6879 6880 spin_lock(&adapter->tmreg_lock); 6881 6882 if (hw->mac.type == e1000_82580 || 6883 hw->mac.type == e1000_i354 || 6884 hw->mac.type == e1000_i350) { 6885 s64 ns = timespec64_to_ns(&adapter->perout[tsintr_tt].period); 6886 u32 systiml, systimh, level_mask, level, rem; 6887 u64 systim, now; 6888 6889 /* read systim registers in sequence */ 6890 rd32(E1000_SYSTIMR); 6891 systiml = rd32(E1000_SYSTIML); 6892 systimh = rd32(E1000_SYSTIMH); 6893 systim = (((u64)(systimh & 0xFF)) << 32) | ((u64)systiml); 6894 now = timecounter_cyc2time(&adapter->tc, systim); 6895 6896 if (pin < 2) { 6897 level_mask = (tsintr_tt == 1) ? 0x80000 : 0x40000; 6898 level = (rd32(E1000_CTRL) & level_mask) ? 1 : 0; 6899 } else { 6900 level_mask = (tsintr_tt == 1) ? 0x80 : 0x40; 6901 level = (rd32(E1000_CTRL_EXT) & level_mask) ? 1 : 0; 6902 } 6903 6904 div_u64_rem(now, ns, &rem); 6905 systim = systim + (ns - rem); 6906 6907 /* synchronize pin level with rising/falling edges */ 6908 div_u64_rem(now, ns << 1, &rem); 6909 if (rem < ns) { 6910 /* first half of period */ 6911 if (level == 0) { 6912 /* output is already low, skip this period */ 6913 systim += ns; 6914 pr_notice("igb: periodic output on %s missed falling edge\n", 6915 adapter->sdp_config[pin].name); 6916 } 6917 } else { 6918 /* second half of period */ 6919 if (level == 1) { 6920 /* output is already high, skip this period */ 6921 systim += ns; 6922 pr_notice("igb: periodic output on %s missed rising edge\n", 6923 adapter->sdp_config[pin].name); 6924 } 6925 } 6926 6927 /* for this chip family tv_sec is the upper part of the binary value, 6928 * so not seconds 6929 */ 6930 ts.tv_nsec = (u32)systim; 6931 ts.tv_sec = ((u32)(systim >> 32)) & 0xFF; 6932 } else { 6933 ts = timespec64_add(adapter->perout[tsintr_tt].start, 6934 adapter->perout[tsintr_tt].period); 6935 } 6936 6937 /* u32 conversion of tv_sec is safe until y2106 */ 6938 wr32((tsintr_tt == 1) ? E1000_TRGTTIML1 : E1000_TRGTTIML0, ts.tv_nsec); 6939 wr32((tsintr_tt == 1) ? E1000_TRGTTIMH1 : E1000_TRGTTIMH0, (u32)ts.tv_sec); 6940 tsauxc = rd32(E1000_TSAUXC); 6941 tsauxc |= TSAUXC_EN_TT0; 6942 wr32(E1000_TSAUXC, tsauxc); 6943 adapter->perout[tsintr_tt].start = ts; 6944 6945 spin_unlock(&adapter->tmreg_lock); 6946 } 6947 6948 static void igb_extts(struct igb_adapter *adapter, int tsintr_tt) 6949 { 6950 int pin = ptp_find_pin(adapter->ptp_clock, PTP_PF_EXTTS, tsintr_tt); 6951 int auxstmpl = (tsintr_tt == 1) ? E1000_AUXSTMPL1 : E1000_AUXSTMPL0; 6952 int auxstmph = (tsintr_tt == 1) ? E1000_AUXSTMPH1 : E1000_AUXSTMPH0; 6953 struct e1000_hw *hw = &adapter->hw; 6954 struct ptp_clock_event event; 6955 struct timespec64 ts; 6956 unsigned long flags; 6957 6958 if (pin < 0 || pin >= IGB_N_SDP) 6959 return; 6960 6961 if (hw->mac.type == e1000_82580 || 6962 hw->mac.type == e1000_i354 || 6963 hw->mac.type == e1000_i350) { 6964 u64 ns = rd32(auxstmpl); 6965 6966 ns += ((u64)(rd32(auxstmph) & 0xFF)) << 32; 6967 spin_lock_irqsave(&adapter->tmreg_lock, flags); 6968 ns = timecounter_cyc2time(&adapter->tc, ns); 6969 spin_unlock_irqrestore(&adapter->tmreg_lock, flags); 6970 ts = ns_to_timespec64(ns); 6971 } else { 6972 ts.tv_nsec = rd32(auxstmpl); 6973 ts.tv_sec = rd32(auxstmph); 6974 } 6975 6976 event.type = PTP_CLOCK_EXTTS; 6977 event.index = tsintr_tt; 6978 event.timestamp = ts.tv_sec * 1000000000ULL + ts.tv_nsec; 6979 ptp_clock_event(adapter->ptp_clock, &event); 6980 } 6981 6982 static void igb_tsync_interrupt(struct igb_adapter *adapter) 6983 { 6984 struct e1000_hw *hw = &adapter->hw; 6985 u32 ack = 0, tsicr = rd32(E1000_TSICR); 6986 struct ptp_clock_event event; 6987 6988 if (tsicr & TSINTR_SYS_WRAP) { 6989 event.type = PTP_CLOCK_PPS; 6990 if (adapter->ptp_caps.pps) 6991 ptp_clock_event(adapter->ptp_clock, &event); 6992 ack |= TSINTR_SYS_WRAP; 6993 } 6994 6995 if (tsicr & E1000_TSICR_TXTS) { 6996 /* retrieve hardware timestamp */ 6997 schedule_work(&adapter->ptp_tx_work); 6998 ack |= E1000_TSICR_TXTS; 6999 } 7000 7001 if (tsicr & TSINTR_TT0) { 7002 igb_perout(adapter, 0); 7003 ack |= TSINTR_TT0; 7004 } 7005 7006 if (tsicr & TSINTR_TT1) { 7007 igb_perout(adapter, 1); 7008 ack |= TSINTR_TT1; 7009 } 7010 7011 if (tsicr & TSINTR_AUTT0) { 7012 igb_extts(adapter, 0); 7013 ack |= TSINTR_AUTT0; 7014 } 7015 7016 if (tsicr & TSINTR_AUTT1) { 7017 igb_extts(adapter, 1); 7018 ack |= TSINTR_AUTT1; 7019 } 7020 7021 /* acknowledge the interrupts */ 7022 wr32(E1000_TSICR, ack); 7023 } 7024 7025 static irqreturn_t igb_msix_other(int irq, void *data) 7026 { 7027 struct igb_adapter *adapter = data; 7028 struct e1000_hw *hw = &adapter->hw; 7029 u32 icr = rd32(E1000_ICR); 7030 /* reading ICR causes bit 31 of EICR to be cleared */ 7031 7032 if (icr & E1000_ICR_DRSTA) 7033 schedule_work(&adapter->reset_task); 7034 7035 if (icr & E1000_ICR_DOUTSYNC) { 7036 /* HW is reporting DMA is out of sync */ 7037 adapter->stats.doosync++; 7038 /* The DMA Out of Sync is also indication of a spoof event 7039 * in IOV mode. Check the Wrong VM Behavior register to 7040 * see if it is really a spoof event. 7041 */ 7042 igb_check_wvbr(adapter); 7043 } 7044 7045 /* Check for a mailbox event */ 7046 if (icr & E1000_ICR_VMMB) 7047 igb_msg_task(adapter); 7048 7049 if (icr & E1000_ICR_LSC) { 7050 hw->mac.get_link_status = 1; 7051 /* guard against interrupt when we're going down */ 7052 if (!test_bit(__IGB_DOWN, &adapter->state)) 7053 mod_timer(&adapter->watchdog_timer, jiffies + 1); 7054 } 7055 7056 if (icr & E1000_ICR_TS) 7057 igb_tsync_interrupt(adapter); 7058 7059 wr32(E1000_EIMS, adapter->eims_other); 7060 7061 return IRQ_HANDLED; 7062 } 7063 7064 static void igb_write_itr(struct igb_q_vector *q_vector) 7065 { 7066 struct igb_adapter *adapter = q_vector->adapter; 7067 u32 itr_val = q_vector->itr_val & 0x7FFC; 7068 7069 if (!q_vector->set_itr) 7070 return; 7071 7072 if (!itr_val) 7073 itr_val = 0x4; 7074 7075 if (adapter->hw.mac.type == e1000_82575) 7076 itr_val |= itr_val << 16; 7077 else 7078 itr_val |= E1000_EITR_CNT_IGNR; 7079 7080 writel(itr_val, q_vector->itr_register); 7081 q_vector->set_itr = 0; 7082 } 7083 7084 static irqreturn_t igb_msix_ring(int irq, void *data) 7085 { 7086 struct igb_q_vector *q_vector = data; 7087 7088 /* Write the ITR value calculated from the previous interrupt. */ 7089 igb_write_itr(q_vector); 7090 7091 napi_schedule(&q_vector->napi); 7092 7093 return IRQ_HANDLED; 7094 } 7095 7096 #ifdef CONFIG_IGB_DCA 7097 static void igb_update_tx_dca(struct igb_adapter *adapter, 7098 struct igb_ring *tx_ring, 7099 int cpu) 7100 { 7101 struct e1000_hw *hw = &adapter->hw; 7102 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu); 7103 7104 if (hw->mac.type != e1000_82575) 7105 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT; 7106 7107 /* We can enable relaxed ordering for reads, but not writes when 7108 * DCA is enabled. This is due to a known issue in some chipsets 7109 * which will cause the DCA tag to be cleared. 7110 */ 7111 txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN | 7112 E1000_DCA_TXCTRL_DATA_RRO_EN | 7113 E1000_DCA_TXCTRL_DESC_DCA_EN; 7114 7115 wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl); 7116 } 7117 7118 static void igb_update_rx_dca(struct igb_adapter *adapter, 7119 struct igb_ring *rx_ring, 7120 int cpu) 7121 { 7122 struct e1000_hw *hw = &adapter->hw; 7123 u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu); 7124 7125 if (hw->mac.type != e1000_82575) 7126 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT; 7127 7128 /* We can enable relaxed ordering for reads, but not writes when 7129 * DCA is enabled. This is due to a known issue in some chipsets 7130 * which will cause the DCA tag to be cleared. 7131 */ 7132 rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN | 7133 E1000_DCA_RXCTRL_DESC_DCA_EN; 7134 7135 wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl); 7136 } 7137 7138 static void igb_update_dca(struct igb_q_vector *q_vector) 7139 { 7140 struct igb_adapter *adapter = q_vector->adapter; 7141 int cpu = get_cpu(); 7142 7143 if (q_vector->cpu == cpu) 7144 goto out_no_update; 7145 7146 if (q_vector->tx.ring) 7147 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu); 7148 7149 if (q_vector->rx.ring) 7150 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu); 7151 7152 q_vector->cpu = cpu; 7153 out_no_update: 7154 put_cpu(); 7155 } 7156 7157 static void igb_setup_dca(struct igb_adapter *adapter) 7158 { 7159 struct e1000_hw *hw = &adapter->hw; 7160 int i; 7161 7162 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED)) 7163 return; 7164 7165 /* Always use CB2 mode, difference is masked in the CB driver. */ 7166 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2); 7167 7168 for (i = 0; i < adapter->num_q_vectors; i++) { 7169 adapter->q_vector[i]->cpu = -1; 7170 igb_update_dca(adapter->q_vector[i]); 7171 } 7172 } 7173 7174 static int __igb_notify_dca(struct device *dev, void *data) 7175 { 7176 struct net_device *netdev = dev_get_drvdata(dev); 7177 struct igb_adapter *adapter = netdev_priv(netdev); 7178 struct pci_dev *pdev = adapter->pdev; 7179 struct e1000_hw *hw = &adapter->hw; 7180 unsigned long event = *(unsigned long *)data; 7181 7182 switch (event) { 7183 case DCA_PROVIDER_ADD: 7184 /* if already enabled, don't do it again */ 7185 if (adapter->flags & IGB_FLAG_DCA_ENABLED) 7186 break; 7187 if (dca_add_requester(dev) == 0) { 7188 adapter->flags |= IGB_FLAG_DCA_ENABLED; 7189 dev_info(&pdev->dev, "DCA enabled\n"); 7190 igb_setup_dca(adapter); 7191 break; 7192 } 7193 fallthrough; /* since DCA is disabled. */ 7194 case DCA_PROVIDER_REMOVE: 7195 if (adapter->flags & IGB_FLAG_DCA_ENABLED) { 7196 /* without this a class_device is left 7197 * hanging around in the sysfs model 7198 */ 7199 dca_remove_requester(dev); 7200 dev_info(&pdev->dev, "DCA disabled\n"); 7201 adapter->flags &= ~IGB_FLAG_DCA_ENABLED; 7202 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE); 7203 } 7204 break; 7205 } 7206 7207 return 0; 7208 } 7209 7210 static int igb_notify_dca(struct notifier_block *nb, unsigned long event, 7211 void *p) 7212 { 7213 int ret_val; 7214 7215 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event, 7216 __igb_notify_dca); 7217 7218 return ret_val ? NOTIFY_BAD : NOTIFY_DONE; 7219 } 7220 #endif /* CONFIG_IGB_DCA */ 7221 7222 #ifdef CONFIG_PCI_IOV 7223 static int igb_vf_configure(struct igb_adapter *adapter, int vf) 7224 { 7225 unsigned char mac_addr[ETH_ALEN]; 7226 7227 eth_zero_addr(mac_addr); 7228 igb_set_vf_mac(adapter, vf, mac_addr); 7229 7230 /* By default spoof check is enabled for all VFs */ 7231 adapter->vf_data[vf].spoofchk_enabled = true; 7232 7233 /* By default VFs are not trusted */ 7234 adapter->vf_data[vf].trusted = false; 7235 7236 return 0; 7237 } 7238 7239 #endif 7240 static void igb_ping_all_vfs(struct igb_adapter *adapter) 7241 { 7242 struct e1000_hw *hw = &adapter->hw; 7243 u32 ping; 7244 int i; 7245 7246 for (i = 0 ; i < adapter->vfs_allocated_count; i++) { 7247 ping = E1000_PF_CONTROL_MSG; 7248 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS) 7249 ping |= E1000_VT_MSGTYPE_CTS; 7250 igb_write_mbx(hw, &ping, 1, i); 7251 } 7252 } 7253 7254 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf) 7255 { 7256 struct e1000_hw *hw = &adapter->hw; 7257 u32 vmolr = rd32(E1000_VMOLR(vf)); 7258 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 7259 7260 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC | 7261 IGB_VF_FLAG_MULTI_PROMISC); 7262 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME); 7263 7264 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) { 7265 vmolr |= E1000_VMOLR_MPME; 7266 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC; 7267 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST; 7268 } else { 7269 /* if we have hashes and we are clearing a multicast promisc 7270 * flag we need to write the hashes to the MTA as this step 7271 * was previously skipped 7272 */ 7273 if (vf_data->num_vf_mc_hashes > 30) { 7274 vmolr |= E1000_VMOLR_MPME; 7275 } else if (vf_data->num_vf_mc_hashes) { 7276 int j; 7277 7278 vmolr |= E1000_VMOLR_ROMPE; 7279 for (j = 0; j < vf_data->num_vf_mc_hashes; j++) 7280 igb_mta_set(hw, vf_data->vf_mc_hashes[j]); 7281 } 7282 } 7283 7284 wr32(E1000_VMOLR(vf), vmolr); 7285 7286 /* there are flags left unprocessed, likely not supported */ 7287 if (*msgbuf & E1000_VT_MSGINFO_MASK) 7288 return -EINVAL; 7289 7290 return 0; 7291 } 7292 7293 static int igb_set_vf_multicasts(struct igb_adapter *adapter, 7294 u32 *msgbuf, u32 vf) 7295 { 7296 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT; 7297 u16 *hash_list = (u16 *)&msgbuf[1]; 7298 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 7299 int i; 7300 7301 /* salt away the number of multicast addresses assigned 7302 * to this VF for later use to restore when the PF multi cast 7303 * list changes 7304 */ 7305 vf_data->num_vf_mc_hashes = n; 7306 7307 /* only up to 30 hash values supported */ 7308 if (n > 30) 7309 n = 30; 7310 7311 /* store the hashes for later use */ 7312 for (i = 0; i < n; i++) 7313 vf_data->vf_mc_hashes[i] = hash_list[i]; 7314 7315 /* Flush and reset the mta with the new values */ 7316 igb_set_rx_mode(adapter->netdev); 7317 7318 return 0; 7319 } 7320 7321 static void igb_restore_vf_multicasts(struct igb_adapter *adapter) 7322 { 7323 struct e1000_hw *hw = &adapter->hw; 7324 struct vf_data_storage *vf_data; 7325 int i, j; 7326 7327 for (i = 0; i < adapter->vfs_allocated_count; i++) { 7328 u32 vmolr = rd32(E1000_VMOLR(i)); 7329 7330 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME); 7331 7332 vf_data = &adapter->vf_data[i]; 7333 7334 if ((vf_data->num_vf_mc_hashes > 30) || 7335 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) { 7336 vmolr |= E1000_VMOLR_MPME; 7337 } else if (vf_data->num_vf_mc_hashes) { 7338 vmolr |= E1000_VMOLR_ROMPE; 7339 for (j = 0; j < vf_data->num_vf_mc_hashes; j++) 7340 igb_mta_set(hw, vf_data->vf_mc_hashes[j]); 7341 } 7342 wr32(E1000_VMOLR(i), vmolr); 7343 } 7344 } 7345 7346 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf) 7347 { 7348 struct e1000_hw *hw = &adapter->hw; 7349 u32 pool_mask, vlvf_mask, i; 7350 7351 /* create mask for VF and other pools */ 7352 pool_mask = E1000_VLVF_POOLSEL_MASK; 7353 vlvf_mask = BIT(E1000_VLVF_POOLSEL_SHIFT + vf); 7354 7355 /* drop PF from pool bits */ 7356 pool_mask &= ~BIT(E1000_VLVF_POOLSEL_SHIFT + 7357 adapter->vfs_allocated_count); 7358 7359 /* Find the vlan filter for this id */ 7360 for (i = E1000_VLVF_ARRAY_SIZE; i--;) { 7361 u32 vlvf = rd32(E1000_VLVF(i)); 7362 u32 vfta_mask, vid, vfta; 7363 7364 /* remove the vf from the pool */ 7365 if (!(vlvf & vlvf_mask)) 7366 continue; 7367 7368 /* clear out bit from VLVF */ 7369 vlvf ^= vlvf_mask; 7370 7371 /* if other pools are present, just remove ourselves */ 7372 if (vlvf & pool_mask) 7373 goto update_vlvfb; 7374 7375 /* if PF is present, leave VFTA */ 7376 if (vlvf & E1000_VLVF_POOLSEL_MASK) 7377 goto update_vlvf; 7378 7379 vid = vlvf & E1000_VLVF_VLANID_MASK; 7380 vfta_mask = BIT(vid % 32); 7381 7382 /* clear bit from VFTA */ 7383 vfta = adapter->shadow_vfta[vid / 32]; 7384 if (vfta & vfta_mask) 7385 hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask); 7386 update_vlvf: 7387 /* clear pool selection enable */ 7388 if (adapter->flags & IGB_FLAG_VLAN_PROMISC) 7389 vlvf &= E1000_VLVF_POOLSEL_MASK; 7390 else 7391 vlvf = 0; 7392 update_vlvfb: 7393 /* clear pool bits */ 7394 wr32(E1000_VLVF(i), vlvf); 7395 } 7396 } 7397 7398 static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan) 7399 { 7400 u32 vlvf; 7401 int idx; 7402 7403 /* short cut the special case */ 7404 if (vlan == 0) 7405 return 0; 7406 7407 /* Search for the VLAN id in the VLVF entries */ 7408 for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) { 7409 vlvf = rd32(E1000_VLVF(idx)); 7410 if ((vlvf & VLAN_VID_MASK) == vlan) 7411 break; 7412 } 7413 7414 return idx; 7415 } 7416 7417 static void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid) 7418 { 7419 struct e1000_hw *hw = &adapter->hw; 7420 u32 bits, pf_id; 7421 int idx; 7422 7423 idx = igb_find_vlvf_entry(hw, vid); 7424 if (!idx) 7425 return; 7426 7427 /* See if any other pools are set for this VLAN filter 7428 * entry other than the PF. 7429 */ 7430 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT; 7431 bits = ~BIT(pf_id) & E1000_VLVF_POOLSEL_MASK; 7432 bits &= rd32(E1000_VLVF(idx)); 7433 7434 /* Disable the filter so this falls into the default pool. */ 7435 if (!bits) { 7436 if (adapter->flags & IGB_FLAG_VLAN_PROMISC) 7437 wr32(E1000_VLVF(idx), BIT(pf_id)); 7438 else 7439 wr32(E1000_VLVF(idx), 0); 7440 } 7441 } 7442 7443 static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid, 7444 bool add, u32 vf) 7445 { 7446 int pf_id = adapter->vfs_allocated_count; 7447 struct e1000_hw *hw = &adapter->hw; 7448 int err; 7449 7450 /* If VLAN overlaps with one the PF is currently monitoring make 7451 * sure that we are able to allocate a VLVF entry. This may be 7452 * redundant but it guarantees PF will maintain visibility to 7453 * the VLAN. 7454 */ 7455 if (add && test_bit(vid, adapter->active_vlans)) { 7456 err = igb_vfta_set(hw, vid, pf_id, true, false); 7457 if (err) 7458 return err; 7459 } 7460 7461 err = igb_vfta_set(hw, vid, vf, add, false); 7462 7463 if (add && !err) 7464 return err; 7465 7466 /* If we failed to add the VF VLAN or we are removing the VF VLAN 7467 * we may need to drop the PF pool bit in order to allow us to free 7468 * up the VLVF resources. 7469 */ 7470 if (test_bit(vid, adapter->active_vlans) || 7471 (adapter->flags & IGB_FLAG_VLAN_PROMISC)) 7472 igb_update_pf_vlvf(adapter, vid); 7473 7474 return err; 7475 } 7476 7477 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf) 7478 { 7479 struct e1000_hw *hw = &adapter->hw; 7480 7481 if (vid) 7482 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT)); 7483 else 7484 wr32(E1000_VMVIR(vf), 0); 7485 } 7486 7487 static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf, 7488 u16 vlan, u8 qos) 7489 { 7490 int err; 7491 7492 err = igb_set_vf_vlan(adapter, vlan, true, vf); 7493 if (err) 7494 return err; 7495 7496 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf); 7497 igb_set_vmolr(adapter, vf, !vlan); 7498 7499 /* revoke access to previous VLAN */ 7500 if (vlan != adapter->vf_data[vf].pf_vlan) 7501 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan, 7502 false, vf); 7503 7504 adapter->vf_data[vf].pf_vlan = vlan; 7505 adapter->vf_data[vf].pf_qos = qos; 7506 igb_set_vf_vlan_strip(adapter, vf, true); 7507 dev_info(&adapter->pdev->dev, 7508 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf); 7509 if (test_bit(__IGB_DOWN, &adapter->state)) { 7510 dev_warn(&adapter->pdev->dev, 7511 "The VF VLAN has been set, but the PF device is not up.\n"); 7512 dev_warn(&adapter->pdev->dev, 7513 "Bring the PF device up before attempting to use the VF device.\n"); 7514 } 7515 7516 return err; 7517 } 7518 7519 static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf) 7520 { 7521 /* Restore tagless access via VLAN 0 */ 7522 igb_set_vf_vlan(adapter, 0, true, vf); 7523 7524 igb_set_vmvir(adapter, 0, vf); 7525 igb_set_vmolr(adapter, vf, true); 7526 7527 /* Remove any PF assigned VLAN */ 7528 if (adapter->vf_data[vf].pf_vlan) 7529 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan, 7530 false, vf); 7531 7532 adapter->vf_data[vf].pf_vlan = 0; 7533 adapter->vf_data[vf].pf_qos = 0; 7534 igb_set_vf_vlan_strip(adapter, vf, false); 7535 7536 return 0; 7537 } 7538 7539 static int igb_ndo_set_vf_vlan(struct net_device *netdev, int vf, 7540 u16 vlan, u8 qos, __be16 vlan_proto) 7541 { 7542 struct igb_adapter *adapter = netdev_priv(netdev); 7543 7544 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7)) 7545 return -EINVAL; 7546 7547 if (vlan_proto != htons(ETH_P_8021Q)) 7548 return -EPROTONOSUPPORT; 7549 7550 return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) : 7551 igb_disable_port_vlan(adapter, vf); 7552 } 7553 7554 static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf) 7555 { 7556 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT; 7557 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK); 7558 int ret; 7559 7560 if (adapter->vf_data[vf].pf_vlan) 7561 return -1; 7562 7563 /* VLAN 0 is a special case, don't allow it to be removed */ 7564 if (!vid && !add) 7565 return 0; 7566 7567 ret = igb_set_vf_vlan(adapter, vid, !!add, vf); 7568 if (!ret) 7569 igb_set_vf_vlan_strip(adapter, vf, !!vid); 7570 return ret; 7571 } 7572 7573 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf) 7574 { 7575 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 7576 7577 /* clear flags - except flag that indicates PF has set the MAC */ 7578 vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC; 7579 vf_data->last_nack = jiffies; 7580 7581 /* reset vlans for device */ 7582 igb_clear_vf_vfta(adapter, vf); 7583 igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf); 7584 igb_set_vmvir(adapter, vf_data->pf_vlan | 7585 (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf); 7586 igb_set_vmolr(adapter, vf, !vf_data->pf_vlan); 7587 igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan)); 7588 7589 /* reset multicast table array for vf */ 7590 adapter->vf_data[vf].num_vf_mc_hashes = 0; 7591 7592 /* Flush and reset the mta with the new values */ 7593 igb_set_rx_mode(adapter->netdev); 7594 } 7595 7596 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf) 7597 { 7598 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses; 7599 7600 /* clear mac address as we were hotplug removed/added */ 7601 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC)) 7602 eth_zero_addr(vf_mac); 7603 7604 /* process remaining reset events */ 7605 igb_vf_reset(adapter, vf); 7606 } 7607 7608 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf) 7609 { 7610 struct e1000_hw *hw = &adapter->hw; 7611 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses; 7612 u32 reg, msgbuf[3] = {}; 7613 u8 *addr = (u8 *)(&msgbuf[1]); 7614 7615 /* process all the same items cleared in a function level reset */ 7616 igb_vf_reset(adapter, vf); 7617 7618 /* set vf mac address */ 7619 igb_set_vf_mac(adapter, vf, vf_mac); 7620 7621 /* enable transmit and receive for vf */ 7622 reg = rd32(E1000_VFTE); 7623 wr32(E1000_VFTE, reg | BIT(vf)); 7624 reg = rd32(E1000_VFRE); 7625 wr32(E1000_VFRE, reg | BIT(vf)); 7626 7627 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS; 7628 7629 /* reply to reset with ack and vf mac address */ 7630 if (!is_zero_ether_addr(vf_mac)) { 7631 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK; 7632 memcpy(addr, vf_mac, ETH_ALEN); 7633 } else { 7634 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK; 7635 } 7636 igb_write_mbx(hw, msgbuf, 3, vf); 7637 } 7638 7639 static void igb_flush_mac_table(struct igb_adapter *adapter) 7640 { 7641 struct e1000_hw *hw = &adapter->hw; 7642 int i; 7643 7644 for (i = 0; i < hw->mac.rar_entry_count; i++) { 7645 adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE; 7646 eth_zero_addr(adapter->mac_table[i].addr); 7647 adapter->mac_table[i].queue = 0; 7648 igb_rar_set_index(adapter, i); 7649 } 7650 } 7651 7652 static int igb_available_rars(struct igb_adapter *adapter, u8 queue) 7653 { 7654 struct e1000_hw *hw = &adapter->hw; 7655 /* do not count rar entries reserved for VFs MAC addresses */ 7656 int rar_entries = hw->mac.rar_entry_count - 7657 adapter->vfs_allocated_count; 7658 int i, count = 0; 7659 7660 for (i = 0; i < rar_entries; i++) { 7661 /* do not count default entries */ 7662 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT) 7663 continue; 7664 7665 /* do not count "in use" entries for different queues */ 7666 if ((adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE) && 7667 (adapter->mac_table[i].queue != queue)) 7668 continue; 7669 7670 count++; 7671 } 7672 7673 return count; 7674 } 7675 7676 /* Set default MAC address for the PF in the first RAR entry */ 7677 static void igb_set_default_mac_filter(struct igb_adapter *adapter) 7678 { 7679 struct igb_mac_addr *mac_table = &adapter->mac_table[0]; 7680 7681 ether_addr_copy(mac_table->addr, adapter->hw.mac.addr); 7682 mac_table->queue = adapter->vfs_allocated_count; 7683 mac_table->state = IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE; 7684 7685 igb_rar_set_index(adapter, 0); 7686 } 7687 7688 /* If the filter to be added and an already existing filter express 7689 * the same address and address type, it should be possible to only 7690 * override the other configurations, for example the queue to steer 7691 * traffic. 7692 */ 7693 static bool igb_mac_entry_can_be_used(const struct igb_mac_addr *entry, 7694 const u8 *addr, const u8 flags) 7695 { 7696 if (!(entry->state & IGB_MAC_STATE_IN_USE)) 7697 return true; 7698 7699 if ((entry->state & IGB_MAC_STATE_SRC_ADDR) != 7700 (flags & IGB_MAC_STATE_SRC_ADDR)) 7701 return false; 7702 7703 if (!ether_addr_equal(addr, entry->addr)) 7704 return false; 7705 7706 return true; 7707 } 7708 7709 /* Add a MAC filter for 'addr' directing matching traffic to 'queue', 7710 * 'flags' is used to indicate what kind of match is made, match is by 7711 * default for the destination address, if matching by source address 7712 * is desired the flag IGB_MAC_STATE_SRC_ADDR can be used. 7713 */ 7714 static int igb_add_mac_filter_flags(struct igb_adapter *adapter, 7715 const u8 *addr, const u8 queue, 7716 const u8 flags) 7717 { 7718 struct e1000_hw *hw = &adapter->hw; 7719 int rar_entries = hw->mac.rar_entry_count - 7720 adapter->vfs_allocated_count; 7721 int i; 7722 7723 if (is_zero_ether_addr(addr)) 7724 return -EINVAL; 7725 7726 /* Search for the first empty entry in the MAC table. 7727 * Do not touch entries at the end of the table reserved for the VF MAC 7728 * addresses. 7729 */ 7730 for (i = 0; i < rar_entries; i++) { 7731 if (!igb_mac_entry_can_be_used(&adapter->mac_table[i], 7732 addr, flags)) 7733 continue; 7734 7735 ether_addr_copy(adapter->mac_table[i].addr, addr); 7736 adapter->mac_table[i].queue = queue; 7737 adapter->mac_table[i].state |= IGB_MAC_STATE_IN_USE | flags; 7738 7739 igb_rar_set_index(adapter, i); 7740 return i; 7741 } 7742 7743 return -ENOSPC; 7744 } 7745 7746 static int igb_add_mac_filter(struct igb_adapter *adapter, const u8 *addr, 7747 const u8 queue) 7748 { 7749 return igb_add_mac_filter_flags(adapter, addr, queue, 0); 7750 } 7751 7752 /* Remove a MAC filter for 'addr' directing matching traffic to 7753 * 'queue', 'flags' is used to indicate what kind of match need to be 7754 * removed, match is by default for the destination address, if 7755 * matching by source address is to be removed the flag 7756 * IGB_MAC_STATE_SRC_ADDR can be used. 7757 */ 7758 static int igb_del_mac_filter_flags(struct igb_adapter *adapter, 7759 const u8 *addr, const u8 queue, 7760 const u8 flags) 7761 { 7762 struct e1000_hw *hw = &adapter->hw; 7763 int rar_entries = hw->mac.rar_entry_count - 7764 adapter->vfs_allocated_count; 7765 int i; 7766 7767 if (is_zero_ether_addr(addr)) 7768 return -EINVAL; 7769 7770 /* Search for matching entry in the MAC table based on given address 7771 * and queue. Do not touch entries at the end of the table reserved 7772 * for the VF MAC addresses. 7773 */ 7774 for (i = 0; i < rar_entries; i++) { 7775 if (!(adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE)) 7776 continue; 7777 if ((adapter->mac_table[i].state & flags) != flags) 7778 continue; 7779 if (adapter->mac_table[i].queue != queue) 7780 continue; 7781 if (!ether_addr_equal(adapter->mac_table[i].addr, addr)) 7782 continue; 7783 7784 /* When a filter for the default address is "deleted", 7785 * we return it to its initial configuration 7786 */ 7787 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT) { 7788 adapter->mac_table[i].state = 7789 IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE; 7790 adapter->mac_table[i].queue = 7791 adapter->vfs_allocated_count; 7792 } else { 7793 adapter->mac_table[i].state = 0; 7794 adapter->mac_table[i].queue = 0; 7795 eth_zero_addr(adapter->mac_table[i].addr); 7796 } 7797 7798 igb_rar_set_index(adapter, i); 7799 return 0; 7800 } 7801 7802 return -ENOENT; 7803 } 7804 7805 static int igb_del_mac_filter(struct igb_adapter *adapter, const u8 *addr, 7806 const u8 queue) 7807 { 7808 return igb_del_mac_filter_flags(adapter, addr, queue, 0); 7809 } 7810 7811 int igb_add_mac_steering_filter(struct igb_adapter *adapter, 7812 const u8 *addr, u8 queue, u8 flags) 7813 { 7814 struct e1000_hw *hw = &adapter->hw; 7815 7816 /* In theory, this should be supported on 82575 as well, but 7817 * that part wasn't easily accessible during development. 7818 */ 7819 if (hw->mac.type != e1000_i210) 7820 return -EOPNOTSUPP; 7821 7822 return igb_add_mac_filter_flags(adapter, addr, queue, 7823 IGB_MAC_STATE_QUEUE_STEERING | flags); 7824 } 7825 7826 int igb_del_mac_steering_filter(struct igb_adapter *adapter, 7827 const u8 *addr, u8 queue, u8 flags) 7828 { 7829 return igb_del_mac_filter_flags(adapter, addr, queue, 7830 IGB_MAC_STATE_QUEUE_STEERING | flags); 7831 } 7832 7833 static int igb_uc_sync(struct net_device *netdev, const unsigned char *addr) 7834 { 7835 struct igb_adapter *adapter = netdev_priv(netdev); 7836 int ret; 7837 7838 ret = igb_add_mac_filter(adapter, addr, adapter->vfs_allocated_count); 7839 7840 return min_t(int, ret, 0); 7841 } 7842 7843 static int igb_uc_unsync(struct net_device *netdev, const unsigned char *addr) 7844 { 7845 struct igb_adapter *adapter = netdev_priv(netdev); 7846 7847 igb_del_mac_filter(adapter, addr, adapter->vfs_allocated_count); 7848 7849 return 0; 7850 } 7851 7852 static int igb_set_vf_mac_filter(struct igb_adapter *adapter, const int vf, 7853 const u32 info, const u8 *addr) 7854 { 7855 struct pci_dev *pdev = adapter->pdev; 7856 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 7857 struct list_head *pos; 7858 struct vf_mac_filter *entry = NULL; 7859 int ret = 0; 7860 7861 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) && 7862 !vf_data->trusted) { 7863 dev_warn(&pdev->dev, 7864 "VF %d requested MAC filter but is administratively denied\n", 7865 vf); 7866 return -EINVAL; 7867 } 7868 if (!is_valid_ether_addr(addr)) { 7869 dev_warn(&pdev->dev, 7870 "VF %d attempted to set invalid MAC filter\n", 7871 vf); 7872 return -EINVAL; 7873 } 7874 7875 switch (info) { 7876 case E1000_VF_MAC_FILTER_CLR: 7877 /* remove all unicast MAC filters related to the current VF */ 7878 list_for_each(pos, &adapter->vf_macs.l) { 7879 entry = list_entry(pos, struct vf_mac_filter, l); 7880 if (entry->vf == vf) { 7881 entry->vf = -1; 7882 entry->free = true; 7883 igb_del_mac_filter(adapter, entry->vf_mac, vf); 7884 } 7885 } 7886 break; 7887 case E1000_VF_MAC_FILTER_ADD: 7888 /* try to find empty slot in the list */ 7889 list_for_each(pos, &adapter->vf_macs.l) { 7890 entry = list_entry(pos, struct vf_mac_filter, l); 7891 if (entry->free) 7892 break; 7893 } 7894 7895 if (entry && entry->free) { 7896 entry->free = false; 7897 entry->vf = vf; 7898 ether_addr_copy(entry->vf_mac, addr); 7899 7900 ret = igb_add_mac_filter(adapter, addr, vf); 7901 ret = min_t(int, ret, 0); 7902 } else { 7903 ret = -ENOSPC; 7904 } 7905 7906 if (ret == -ENOSPC) 7907 dev_warn(&pdev->dev, 7908 "VF %d has requested MAC filter but there is no space for it\n", 7909 vf); 7910 break; 7911 default: 7912 ret = -EINVAL; 7913 break; 7914 } 7915 7916 return ret; 7917 } 7918 7919 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf) 7920 { 7921 struct pci_dev *pdev = adapter->pdev; 7922 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 7923 u32 info = msg[0] & E1000_VT_MSGINFO_MASK; 7924 7925 /* The VF MAC Address is stored in a packed array of bytes 7926 * starting at the second 32 bit word of the msg array 7927 */ 7928 unsigned char *addr = (unsigned char *)&msg[1]; 7929 int ret = 0; 7930 7931 if (!info) { 7932 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) && 7933 !vf_data->trusted) { 7934 dev_warn(&pdev->dev, 7935 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n", 7936 vf); 7937 return -EINVAL; 7938 } 7939 7940 if (!is_valid_ether_addr(addr)) { 7941 dev_warn(&pdev->dev, 7942 "VF %d attempted to set invalid MAC\n", 7943 vf); 7944 return -EINVAL; 7945 } 7946 7947 ret = igb_set_vf_mac(adapter, vf, addr); 7948 } else { 7949 ret = igb_set_vf_mac_filter(adapter, vf, info, addr); 7950 } 7951 7952 return ret; 7953 } 7954 7955 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf) 7956 { 7957 struct e1000_hw *hw = &adapter->hw; 7958 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 7959 u32 msg = E1000_VT_MSGTYPE_NACK; 7960 7961 /* if device isn't clear to send it shouldn't be reading either */ 7962 if (!(vf_data->flags & IGB_VF_FLAG_CTS) && 7963 time_after(jiffies, vf_data->last_nack + (2 * HZ))) { 7964 igb_write_mbx(hw, &msg, 1, vf); 7965 vf_data->last_nack = jiffies; 7966 } 7967 } 7968 7969 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf) 7970 { 7971 struct pci_dev *pdev = adapter->pdev; 7972 u32 msgbuf[E1000_VFMAILBOX_SIZE]; 7973 struct e1000_hw *hw = &adapter->hw; 7974 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 7975 s32 retval; 7976 7977 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf, false); 7978 7979 if (retval) { 7980 /* if receive failed revoke VF CTS stats and restart init */ 7981 dev_err(&pdev->dev, "Error receiving message from VF\n"); 7982 vf_data->flags &= ~IGB_VF_FLAG_CTS; 7983 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ))) 7984 goto unlock; 7985 goto out; 7986 } 7987 7988 /* this is a message we already processed, do nothing */ 7989 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK)) 7990 goto unlock; 7991 7992 /* until the vf completes a reset it should not be 7993 * allowed to start any configuration. 7994 */ 7995 if (msgbuf[0] == E1000_VF_RESET) { 7996 /* unlocks mailbox */ 7997 igb_vf_reset_msg(adapter, vf); 7998 return; 7999 } 8000 8001 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) { 8002 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ))) 8003 goto unlock; 8004 retval = -1; 8005 goto out; 8006 } 8007 8008 switch ((msgbuf[0] & 0xFFFF)) { 8009 case E1000_VF_SET_MAC_ADDR: 8010 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf); 8011 break; 8012 case E1000_VF_SET_PROMISC: 8013 retval = igb_set_vf_promisc(adapter, msgbuf, vf); 8014 break; 8015 case E1000_VF_SET_MULTICAST: 8016 retval = igb_set_vf_multicasts(adapter, msgbuf, vf); 8017 break; 8018 case E1000_VF_SET_LPE: 8019 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf); 8020 break; 8021 case E1000_VF_SET_VLAN: 8022 retval = -1; 8023 if (vf_data->pf_vlan) 8024 dev_warn(&pdev->dev, 8025 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n", 8026 vf); 8027 else 8028 retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf); 8029 break; 8030 default: 8031 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]); 8032 retval = -1; 8033 break; 8034 } 8035 8036 msgbuf[0] |= E1000_VT_MSGTYPE_CTS; 8037 out: 8038 /* notify the VF of the results of what it sent us */ 8039 if (retval) 8040 msgbuf[0] |= E1000_VT_MSGTYPE_NACK; 8041 else 8042 msgbuf[0] |= E1000_VT_MSGTYPE_ACK; 8043 8044 /* unlocks mailbox */ 8045 igb_write_mbx(hw, msgbuf, 1, vf); 8046 return; 8047 8048 unlock: 8049 igb_unlock_mbx(hw, vf); 8050 } 8051 8052 static void igb_msg_task(struct igb_adapter *adapter) 8053 { 8054 struct e1000_hw *hw = &adapter->hw; 8055 unsigned long flags; 8056 u32 vf; 8057 8058 spin_lock_irqsave(&adapter->vfs_lock, flags); 8059 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) { 8060 /* process any reset requests */ 8061 if (!igb_check_for_rst(hw, vf)) 8062 igb_vf_reset_event(adapter, vf); 8063 8064 /* process any messages pending */ 8065 if (!igb_check_for_msg(hw, vf)) 8066 igb_rcv_msg_from_vf(adapter, vf); 8067 8068 /* process any acks */ 8069 if (!igb_check_for_ack(hw, vf)) 8070 igb_rcv_ack_from_vf(adapter, vf); 8071 } 8072 spin_unlock_irqrestore(&adapter->vfs_lock, flags); 8073 } 8074 8075 /** 8076 * igb_set_uta - Set unicast filter table address 8077 * @adapter: board private structure 8078 * @set: boolean indicating if we are setting or clearing bits 8079 * 8080 * The unicast table address is a register array of 32-bit registers. 8081 * The table is meant to be used in a way similar to how the MTA is used 8082 * however due to certain limitations in the hardware it is necessary to 8083 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous 8084 * enable bit to allow vlan tag stripping when promiscuous mode is enabled 8085 **/ 8086 static void igb_set_uta(struct igb_adapter *adapter, bool set) 8087 { 8088 struct e1000_hw *hw = &adapter->hw; 8089 u32 uta = set ? ~0 : 0; 8090 int i; 8091 8092 /* we only need to do this if VMDq is enabled */ 8093 if (!adapter->vfs_allocated_count) 8094 return; 8095 8096 for (i = hw->mac.uta_reg_count; i--;) 8097 array_wr32(E1000_UTA, i, uta); 8098 } 8099 8100 /** 8101 * igb_intr_msi - Interrupt Handler 8102 * @irq: interrupt number 8103 * @data: pointer to a network interface device structure 8104 **/ 8105 static irqreturn_t igb_intr_msi(int irq, void *data) 8106 { 8107 struct igb_adapter *adapter = data; 8108 struct igb_q_vector *q_vector = adapter->q_vector[0]; 8109 struct e1000_hw *hw = &adapter->hw; 8110 /* read ICR disables interrupts using IAM */ 8111 u32 icr = rd32(E1000_ICR); 8112 8113 igb_write_itr(q_vector); 8114 8115 if (icr & E1000_ICR_DRSTA) 8116 schedule_work(&adapter->reset_task); 8117 8118 if (icr & E1000_ICR_DOUTSYNC) { 8119 /* HW is reporting DMA is out of sync */ 8120 adapter->stats.doosync++; 8121 } 8122 8123 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) { 8124 hw->mac.get_link_status = 1; 8125 if (!test_bit(__IGB_DOWN, &adapter->state)) 8126 mod_timer(&adapter->watchdog_timer, jiffies + 1); 8127 } 8128 8129 if (icr & E1000_ICR_TS) 8130 igb_tsync_interrupt(adapter); 8131 8132 napi_schedule(&q_vector->napi); 8133 8134 return IRQ_HANDLED; 8135 } 8136 8137 /** 8138 * igb_intr - Legacy Interrupt Handler 8139 * @irq: interrupt number 8140 * @data: pointer to a network interface device structure 8141 **/ 8142 static irqreturn_t igb_intr(int irq, void *data) 8143 { 8144 struct igb_adapter *adapter = data; 8145 struct igb_q_vector *q_vector = adapter->q_vector[0]; 8146 struct e1000_hw *hw = &adapter->hw; 8147 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No 8148 * need for the IMC write 8149 */ 8150 u32 icr = rd32(E1000_ICR); 8151 8152 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is 8153 * not set, then the adapter didn't send an interrupt 8154 */ 8155 if (!(icr & E1000_ICR_INT_ASSERTED)) 8156 return IRQ_NONE; 8157 8158 igb_write_itr(q_vector); 8159 8160 if (icr & E1000_ICR_DRSTA) 8161 schedule_work(&adapter->reset_task); 8162 8163 if (icr & E1000_ICR_DOUTSYNC) { 8164 /* HW is reporting DMA is out of sync */ 8165 adapter->stats.doosync++; 8166 } 8167 8168 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) { 8169 hw->mac.get_link_status = 1; 8170 /* guard against interrupt when we're going down */ 8171 if (!test_bit(__IGB_DOWN, &adapter->state)) 8172 mod_timer(&adapter->watchdog_timer, jiffies + 1); 8173 } 8174 8175 if (icr & E1000_ICR_TS) 8176 igb_tsync_interrupt(adapter); 8177 8178 napi_schedule(&q_vector->napi); 8179 8180 return IRQ_HANDLED; 8181 } 8182 8183 static void igb_ring_irq_enable(struct igb_q_vector *q_vector) 8184 { 8185 struct igb_adapter *adapter = q_vector->adapter; 8186 struct e1000_hw *hw = &adapter->hw; 8187 8188 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) || 8189 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) { 8190 if ((adapter->num_q_vectors == 1) && !adapter->vf_data) 8191 igb_set_itr(q_vector); 8192 else 8193 igb_update_ring_itr(q_vector); 8194 } 8195 8196 if (!test_bit(__IGB_DOWN, &adapter->state)) { 8197 if (adapter->flags & IGB_FLAG_HAS_MSIX) 8198 wr32(E1000_EIMS, q_vector->eims_value); 8199 else 8200 igb_irq_enable(adapter); 8201 } 8202 } 8203 8204 /** 8205 * igb_poll - NAPI Rx polling callback 8206 * @napi: napi polling structure 8207 * @budget: count of how many packets we should handle 8208 **/ 8209 static int igb_poll(struct napi_struct *napi, int budget) 8210 { 8211 struct igb_q_vector *q_vector = container_of(napi, 8212 struct igb_q_vector, 8213 napi); 8214 bool clean_complete = true; 8215 int work_done = 0; 8216 8217 #ifdef CONFIG_IGB_DCA 8218 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED) 8219 igb_update_dca(q_vector); 8220 #endif 8221 if (q_vector->tx.ring) 8222 clean_complete = igb_clean_tx_irq(q_vector, budget); 8223 8224 if (q_vector->rx.ring) { 8225 int cleaned = igb_clean_rx_irq(q_vector, budget); 8226 8227 work_done += cleaned; 8228 if (cleaned >= budget) 8229 clean_complete = false; 8230 } 8231 8232 /* If all work not completed, return budget and keep polling */ 8233 if (!clean_complete) 8234 return budget; 8235 8236 /* Exit the polling mode, but don't re-enable interrupts if stack might 8237 * poll us due to busy-polling 8238 */ 8239 if (likely(napi_complete_done(napi, work_done))) 8240 igb_ring_irq_enable(q_vector); 8241 8242 return work_done; 8243 } 8244 8245 /** 8246 * igb_clean_tx_irq - Reclaim resources after transmit completes 8247 * @q_vector: pointer to q_vector containing needed info 8248 * @napi_budget: Used to determine if we are in netpoll 8249 * 8250 * returns true if ring is completely cleaned 8251 **/ 8252 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector, int napi_budget) 8253 { 8254 struct igb_adapter *adapter = q_vector->adapter; 8255 struct igb_ring *tx_ring = q_vector->tx.ring; 8256 struct igb_tx_buffer *tx_buffer; 8257 union e1000_adv_tx_desc *tx_desc; 8258 unsigned int total_bytes = 0, total_packets = 0; 8259 unsigned int budget = q_vector->tx.work_limit; 8260 unsigned int i = tx_ring->next_to_clean; 8261 8262 if (test_bit(__IGB_DOWN, &adapter->state)) 8263 return true; 8264 8265 tx_buffer = &tx_ring->tx_buffer_info[i]; 8266 tx_desc = IGB_TX_DESC(tx_ring, i); 8267 i -= tx_ring->count; 8268 8269 do { 8270 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch; 8271 8272 /* if next_to_watch is not set then there is no work pending */ 8273 if (!eop_desc) 8274 break; 8275 8276 /* prevent any other reads prior to eop_desc */ 8277 smp_rmb(); 8278 8279 /* if DD is not set pending work has not been completed */ 8280 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD))) 8281 break; 8282 8283 /* clear next_to_watch to prevent false hangs */ 8284 tx_buffer->next_to_watch = NULL; 8285 8286 /* update the statistics for this packet */ 8287 total_bytes += tx_buffer->bytecount; 8288 total_packets += tx_buffer->gso_segs; 8289 8290 /* free the skb */ 8291 if (tx_buffer->type == IGB_TYPE_SKB) 8292 napi_consume_skb(tx_buffer->skb, napi_budget); 8293 else 8294 xdp_return_frame(tx_buffer->xdpf); 8295 8296 /* unmap skb header data */ 8297 dma_unmap_single(tx_ring->dev, 8298 dma_unmap_addr(tx_buffer, dma), 8299 dma_unmap_len(tx_buffer, len), 8300 DMA_TO_DEVICE); 8301 8302 /* clear tx_buffer data */ 8303 dma_unmap_len_set(tx_buffer, len, 0); 8304 8305 /* clear last DMA location and unmap remaining buffers */ 8306 while (tx_desc != eop_desc) { 8307 tx_buffer++; 8308 tx_desc++; 8309 i++; 8310 if (unlikely(!i)) { 8311 i -= tx_ring->count; 8312 tx_buffer = tx_ring->tx_buffer_info; 8313 tx_desc = IGB_TX_DESC(tx_ring, 0); 8314 } 8315 8316 /* unmap any remaining paged data */ 8317 if (dma_unmap_len(tx_buffer, len)) { 8318 dma_unmap_page(tx_ring->dev, 8319 dma_unmap_addr(tx_buffer, dma), 8320 dma_unmap_len(tx_buffer, len), 8321 DMA_TO_DEVICE); 8322 dma_unmap_len_set(tx_buffer, len, 0); 8323 } 8324 } 8325 8326 /* move us one more past the eop_desc for start of next pkt */ 8327 tx_buffer++; 8328 tx_desc++; 8329 i++; 8330 if (unlikely(!i)) { 8331 i -= tx_ring->count; 8332 tx_buffer = tx_ring->tx_buffer_info; 8333 tx_desc = IGB_TX_DESC(tx_ring, 0); 8334 } 8335 8336 /* issue prefetch for next Tx descriptor */ 8337 prefetch(tx_desc); 8338 8339 /* update budget accounting */ 8340 budget--; 8341 } while (likely(budget)); 8342 8343 netdev_tx_completed_queue(txring_txq(tx_ring), 8344 total_packets, total_bytes); 8345 i += tx_ring->count; 8346 tx_ring->next_to_clean = i; 8347 u64_stats_update_begin(&tx_ring->tx_syncp); 8348 tx_ring->tx_stats.bytes += total_bytes; 8349 tx_ring->tx_stats.packets += total_packets; 8350 u64_stats_update_end(&tx_ring->tx_syncp); 8351 q_vector->tx.total_bytes += total_bytes; 8352 q_vector->tx.total_packets += total_packets; 8353 8354 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) { 8355 struct e1000_hw *hw = &adapter->hw; 8356 8357 /* Detect a transmit hang in hardware, this serializes the 8358 * check with the clearing of time_stamp and movement of i 8359 */ 8360 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags); 8361 if (tx_buffer->next_to_watch && 8362 time_after(jiffies, tx_buffer->time_stamp + 8363 (adapter->tx_timeout_factor * HZ)) && 8364 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) { 8365 8366 /* detected Tx unit hang */ 8367 dev_err(tx_ring->dev, 8368 "Detected Tx Unit Hang\n" 8369 " Tx Queue <%d>\n" 8370 " TDH <%x>\n" 8371 " TDT <%x>\n" 8372 " next_to_use <%x>\n" 8373 " next_to_clean <%x>\n" 8374 "buffer_info[next_to_clean]\n" 8375 " time_stamp <%lx>\n" 8376 " next_to_watch <%p>\n" 8377 " jiffies <%lx>\n" 8378 " desc.status <%x>\n", 8379 tx_ring->queue_index, 8380 rd32(E1000_TDH(tx_ring->reg_idx)), 8381 readl(tx_ring->tail), 8382 tx_ring->next_to_use, 8383 tx_ring->next_to_clean, 8384 tx_buffer->time_stamp, 8385 tx_buffer->next_to_watch, 8386 jiffies, 8387 tx_buffer->next_to_watch->wb.status); 8388 netif_stop_subqueue(tx_ring->netdev, 8389 tx_ring->queue_index); 8390 8391 /* we are about to reset, no point in enabling stuff */ 8392 return true; 8393 } 8394 } 8395 8396 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2) 8397 if (unlikely(total_packets && 8398 netif_carrier_ok(tx_ring->netdev) && 8399 igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) { 8400 /* Make sure that anybody stopping the queue after this 8401 * sees the new next_to_clean. 8402 */ 8403 smp_mb(); 8404 if (__netif_subqueue_stopped(tx_ring->netdev, 8405 tx_ring->queue_index) && 8406 !(test_bit(__IGB_DOWN, &adapter->state))) { 8407 netif_wake_subqueue(tx_ring->netdev, 8408 tx_ring->queue_index); 8409 8410 u64_stats_update_begin(&tx_ring->tx_syncp); 8411 tx_ring->tx_stats.restart_queue++; 8412 u64_stats_update_end(&tx_ring->tx_syncp); 8413 } 8414 } 8415 8416 return !!budget; 8417 } 8418 8419 /** 8420 * igb_reuse_rx_page - page flip buffer and store it back on the ring 8421 * @rx_ring: rx descriptor ring to store buffers on 8422 * @old_buff: donor buffer to have page reused 8423 * 8424 * Synchronizes page for reuse by the adapter 8425 **/ 8426 static void igb_reuse_rx_page(struct igb_ring *rx_ring, 8427 struct igb_rx_buffer *old_buff) 8428 { 8429 struct igb_rx_buffer *new_buff; 8430 u16 nta = rx_ring->next_to_alloc; 8431 8432 new_buff = &rx_ring->rx_buffer_info[nta]; 8433 8434 /* update, and store next to alloc */ 8435 nta++; 8436 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0; 8437 8438 /* Transfer page from old buffer to new buffer. 8439 * Move each member individually to avoid possible store 8440 * forwarding stalls. 8441 */ 8442 new_buff->dma = old_buff->dma; 8443 new_buff->page = old_buff->page; 8444 new_buff->page_offset = old_buff->page_offset; 8445 new_buff->pagecnt_bias = old_buff->pagecnt_bias; 8446 } 8447 8448 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer, 8449 int rx_buf_pgcnt) 8450 { 8451 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias; 8452 struct page *page = rx_buffer->page; 8453 8454 /* avoid re-using remote and pfmemalloc pages */ 8455 if (!dev_page_is_reusable(page)) 8456 return false; 8457 8458 #if (PAGE_SIZE < 8192) 8459 /* if we are only owner of page we can reuse it */ 8460 if (unlikely((rx_buf_pgcnt - pagecnt_bias) > 1)) 8461 return false; 8462 #else 8463 #define IGB_LAST_OFFSET \ 8464 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IGB_RXBUFFER_2048) 8465 8466 if (rx_buffer->page_offset > IGB_LAST_OFFSET) 8467 return false; 8468 #endif 8469 8470 /* If we have drained the page fragment pool we need to update 8471 * the pagecnt_bias and page count so that we fully restock the 8472 * number of references the driver holds. 8473 */ 8474 if (unlikely(pagecnt_bias == 1)) { 8475 page_ref_add(page, USHRT_MAX - 1); 8476 rx_buffer->pagecnt_bias = USHRT_MAX; 8477 } 8478 8479 return true; 8480 } 8481 8482 /** 8483 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff 8484 * @rx_ring: rx descriptor ring to transact packets on 8485 * @rx_buffer: buffer containing page to add 8486 * @skb: sk_buff to place the data into 8487 * @size: size of buffer to be added 8488 * 8489 * This function will add the data contained in rx_buffer->page to the skb. 8490 **/ 8491 static void igb_add_rx_frag(struct igb_ring *rx_ring, 8492 struct igb_rx_buffer *rx_buffer, 8493 struct sk_buff *skb, 8494 unsigned int size) 8495 { 8496 #if (PAGE_SIZE < 8192) 8497 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2; 8498 #else 8499 unsigned int truesize = ring_uses_build_skb(rx_ring) ? 8500 SKB_DATA_ALIGN(IGB_SKB_PAD + size) : 8501 SKB_DATA_ALIGN(size); 8502 #endif 8503 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page, 8504 rx_buffer->page_offset, size, truesize); 8505 #if (PAGE_SIZE < 8192) 8506 rx_buffer->page_offset ^= truesize; 8507 #else 8508 rx_buffer->page_offset += truesize; 8509 #endif 8510 } 8511 8512 static struct sk_buff *igb_construct_skb(struct igb_ring *rx_ring, 8513 struct igb_rx_buffer *rx_buffer, 8514 struct xdp_buff *xdp, 8515 ktime_t timestamp) 8516 { 8517 #if (PAGE_SIZE < 8192) 8518 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2; 8519 #else 8520 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end - 8521 xdp->data_hard_start); 8522 #endif 8523 unsigned int size = xdp->data_end - xdp->data; 8524 unsigned int headlen; 8525 struct sk_buff *skb; 8526 8527 /* prefetch first cache line of first page */ 8528 net_prefetch(xdp->data); 8529 8530 /* allocate a skb to store the frags */ 8531 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN); 8532 if (unlikely(!skb)) 8533 return NULL; 8534 8535 if (timestamp) 8536 skb_hwtstamps(skb)->hwtstamp = timestamp; 8537 8538 /* Determine available headroom for copy */ 8539 headlen = size; 8540 if (headlen > IGB_RX_HDR_LEN) 8541 headlen = eth_get_headlen(skb->dev, xdp->data, IGB_RX_HDR_LEN); 8542 8543 /* align pull length to size of long to optimize memcpy performance */ 8544 memcpy(__skb_put(skb, headlen), xdp->data, ALIGN(headlen, sizeof(long))); 8545 8546 /* update all of the pointers */ 8547 size -= headlen; 8548 if (size) { 8549 skb_add_rx_frag(skb, 0, rx_buffer->page, 8550 (xdp->data + headlen) - page_address(rx_buffer->page), 8551 size, truesize); 8552 #if (PAGE_SIZE < 8192) 8553 rx_buffer->page_offset ^= truesize; 8554 #else 8555 rx_buffer->page_offset += truesize; 8556 #endif 8557 } else { 8558 rx_buffer->pagecnt_bias++; 8559 } 8560 8561 return skb; 8562 } 8563 8564 static struct sk_buff *igb_build_skb(struct igb_ring *rx_ring, 8565 struct igb_rx_buffer *rx_buffer, 8566 struct xdp_buff *xdp, 8567 ktime_t timestamp) 8568 { 8569 #if (PAGE_SIZE < 8192) 8570 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2; 8571 #else 8572 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) + 8573 SKB_DATA_ALIGN(xdp->data_end - 8574 xdp->data_hard_start); 8575 #endif 8576 unsigned int metasize = xdp->data - xdp->data_meta; 8577 struct sk_buff *skb; 8578 8579 /* prefetch first cache line of first page */ 8580 net_prefetch(xdp->data_meta); 8581 8582 /* build an skb around the page buffer */ 8583 skb = napi_build_skb(xdp->data_hard_start, truesize); 8584 if (unlikely(!skb)) 8585 return NULL; 8586 8587 /* update pointers within the skb to store the data */ 8588 skb_reserve(skb, xdp->data - xdp->data_hard_start); 8589 __skb_put(skb, xdp->data_end - xdp->data); 8590 8591 if (metasize) 8592 skb_metadata_set(skb, metasize); 8593 8594 if (timestamp) 8595 skb_hwtstamps(skb)->hwtstamp = timestamp; 8596 8597 /* update buffer offset */ 8598 #if (PAGE_SIZE < 8192) 8599 rx_buffer->page_offset ^= truesize; 8600 #else 8601 rx_buffer->page_offset += truesize; 8602 #endif 8603 8604 return skb; 8605 } 8606 8607 static struct sk_buff *igb_run_xdp(struct igb_adapter *adapter, 8608 struct igb_ring *rx_ring, 8609 struct xdp_buff *xdp) 8610 { 8611 int err, result = IGB_XDP_PASS; 8612 struct bpf_prog *xdp_prog; 8613 u32 act; 8614 8615 xdp_prog = READ_ONCE(rx_ring->xdp_prog); 8616 8617 if (!xdp_prog) 8618 goto xdp_out; 8619 8620 prefetchw(xdp->data_hard_start); /* xdp_frame write */ 8621 8622 act = bpf_prog_run_xdp(xdp_prog, xdp); 8623 switch (act) { 8624 case XDP_PASS: 8625 break; 8626 case XDP_TX: 8627 result = igb_xdp_xmit_back(adapter, xdp); 8628 if (result == IGB_XDP_CONSUMED) 8629 goto out_failure; 8630 break; 8631 case XDP_REDIRECT: 8632 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog); 8633 if (err) 8634 goto out_failure; 8635 result = IGB_XDP_REDIR; 8636 break; 8637 default: 8638 bpf_warn_invalid_xdp_action(adapter->netdev, xdp_prog, act); 8639 fallthrough; 8640 case XDP_ABORTED: 8641 out_failure: 8642 trace_xdp_exception(rx_ring->netdev, xdp_prog, act); 8643 fallthrough; 8644 case XDP_DROP: 8645 result = IGB_XDP_CONSUMED; 8646 break; 8647 } 8648 xdp_out: 8649 return ERR_PTR(-result); 8650 } 8651 8652 static unsigned int igb_rx_frame_truesize(struct igb_ring *rx_ring, 8653 unsigned int size) 8654 { 8655 unsigned int truesize; 8656 8657 #if (PAGE_SIZE < 8192) 8658 truesize = igb_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */ 8659 #else 8660 truesize = ring_uses_build_skb(rx_ring) ? 8661 SKB_DATA_ALIGN(IGB_SKB_PAD + size) + 8662 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) : 8663 SKB_DATA_ALIGN(size); 8664 #endif 8665 return truesize; 8666 } 8667 8668 static void igb_rx_buffer_flip(struct igb_ring *rx_ring, 8669 struct igb_rx_buffer *rx_buffer, 8670 unsigned int size) 8671 { 8672 unsigned int truesize = igb_rx_frame_truesize(rx_ring, size); 8673 #if (PAGE_SIZE < 8192) 8674 rx_buffer->page_offset ^= truesize; 8675 #else 8676 rx_buffer->page_offset += truesize; 8677 #endif 8678 } 8679 8680 static inline void igb_rx_checksum(struct igb_ring *ring, 8681 union e1000_adv_rx_desc *rx_desc, 8682 struct sk_buff *skb) 8683 { 8684 skb_checksum_none_assert(skb); 8685 8686 /* Ignore Checksum bit is set */ 8687 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM)) 8688 return; 8689 8690 /* Rx checksum disabled via ethtool */ 8691 if (!(ring->netdev->features & NETIF_F_RXCSUM)) 8692 return; 8693 8694 /* TCP/UDP checksum error bit is set */ 8695 if (igb_test_staterr(rx_desc, 8696 E1000_RXDEXT_STATERR_TCPE | 8697 E1000_RXDEXT_STATERR_IPE)) { 8698 /* work around errata with sctp packets where the TCPE aka 8699 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc) 8700 * packets, (aka let the stack check the crc32c) 8701 */ 8702 if (!((skb->len == 60) && 8703 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) { 8704 u64_stats_update_begin(&ring->rx_syncp); 8705 ring->rx_stats.csum_err++; 8706 u64_stats_update_end(&ring->rx_syncp); 8707 } 8708 /* let the stack verify checksum errors */ 8709 return; 8710 } 8711 /* It must be a TCP or UDP packet with a valid checksum */ 8712 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS | 8713 E1000_RXD_STAT_UDPCS)) 8714 skb->ip_summed = CHECKSUM_UNNECESSARY; 8715 8716 dev_dbg(ring->dev, "cksum success: bits %08X\n", 8717 le32_to_cpu(rx_desc->wb.upper.status_error)); 8718 } 8719 8720 static inline void igb_rx_hash(struct igb_ring *ring, 8721 union e1000_adv_rx_desc *rx_desc, 8722 struct sk_buff *skb) 8723 { 8724 if (ring->netdev->features & NETIF_F_RXHASH) 8725 skb_set_hash(skb, 8726 le32_to_cpu(rx_desc->wb.lower.hi_dword.rss), 8727 PKT_HASH_TYPE_L3); 8728 } 8729 8730 /** 8731 * igb_is_non_eop - process handling of non-EOP buffers 8732 * @rx_ring: Rx ring being processed 8733 * @rx_desc: Rx descriptor for current buffer 8734 * 8735 * This function updates next to clean. If the buffer is an EOP buffer 8736 * this function exits returning false, otherwise it will place the 8737 * sk_buff in the next buffer to be chained and return true indicating 8738 * that this is in fact a non-EOP buffer. 8739 **/ 8740 static bool igb_is_non_eop(struct igb_ring *rx_ring, 8741 union e1000_adv_rx_desc *rx_desc) 8742 { 8743 u32 ntc = rx_ring->next_to_clean + 1; 8744 8745 /* fetch, update, and store next to clean */ 8746 ntc = (ntc < rx_ring->count) ? ntc : 0; 8747 rx_ring->next_to_clean = ntc; 8748 8749 prefetch(IGB_RX_DESC(rx_ring, ntc)); 8750 8751 if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP))) 8752 return false; 8753 8754 return true; 8755 } 8756 8757 /** 8758 * igb_cleanup_headers - Correct corrupted or empty headers 8759 * @rx_ring: rx descriptor ring packet is being transacted on 8760 * @rx_desc: pointer to the EOP Rx descriptor 8761 * @skb: pointer to current skb being fixed 8762 * 8763 * Address the case where we are pulling data in on pages only 8764 * and as such no data is present in the skb header. 8765 * 8766 * In addition if skb is not at least 60 bytes we need to pad it so that 8767 * it is large enough to qualify as a valid Ethernet frame. 8768 * 8769 * Returns true if an error was encountered and skb was freed. 8770 **/ 8771 static bool igb_cleanup_headers(struct igb_ring *rx_ring, 8772 union e1000_adv_rx_desc *rx_desc, 8773 struct sk_buff *skb) 8774 { 8775 /* XDP packets use error pointer so abort at this point */ 8776 if (IS_ERR(skb)) 8777 return true; 8778 8779 if (unlikely((igb_test_staterr(rx_desc, 8780 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) { 8781 struct net_device *netdev = rx_ring->netdev; 8782 if (!(netdev->features & NETIF_F_RXALL)) { 8783 dev_kfree_skb_any(skb); 8784 return true; 8785 } 8786 } 8787 8788 /* if eth_skb_pad returns an error the skb was freed */ 8789 if (eth_skb_pad(skb)) 8790 return true; 8791 8792 return false; 8793 } 8794 8795 /** 8796 * igb_process_skb_fields - Populate skb header fields from Rx descriptor 8797 * @rx_ring: rx descriptor ring packet is being transacted on 8798 * @rx_desc: pointer to the EOP Rx descriptor 8799 * @skb: pointer to current skb being populated 8800 * 8801 * This function checks the ring, descriptor, and packet information in 8802 * order to populate the hash, checksum, VLAN, timestamp, protocol, and 8803 * other fields within the skb. 8804 **/ 8805 static void igb_process_skb_fields(struct igb_ring *rx_ring, 8806 union e1000_adv_rx_desc *rx_desc, 8807 struct sk_buff *skb) 8808 { 8809 struct net_device *dev = rx_ring->netdev; 8810 8811 igb_rx_hash(rx_ring, rx_desc, skb); 8812 8813 igb_rx_checksum(rx_ring, rx_desc, skb); 8814 8815 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) && 8816 !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) 8817 igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb); 8818 8819 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) && 8820 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) { 8821 u16 vid; 8822 8823 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) && 8824 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags)) 8825 vid = be16_to_cpu((__force __be16)rx_desc->wb.upper.vlan); 8826 else 8827 vid = le16_to_cpu(rx_desc->wb.upper.vlan); 8828 8829 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid); 8830 } 8831 8832 skb_record_rx_queue(skb, rx_ring->queue_index); 8833 8834 skb->protocol = eth_type_trans(skb, rx_ring->netdev); 8835 } 8836 8837 static unsigned int igb_rx_offset(struct igb_ring *rx_ring) 8838 { 8839 return ring_uses_build_skb(rx_ring) ? IGB_SKB_PAD : 0; 8840 } 8841 8842 static struct igb_rx_buffer *igb_get_rx_buffer(struct igb_ring *rx_ring, 8843 const unsigned int size, int *rx_buf_pgcnt) 8844 { 8845 struct igb_rx_buffer *rx_buffer; 8846 8847 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean]; 8848 *rx_buf_pgcnt = 8849 #if (PAGE_SIZE < 8192) 8850 page_count(rx_buffer->page); 8851 #else 8852 0; 8853 #endif 8854 prefetchw(rx_buffer->page); 8855 8856 /* we are reusing so sync this buffer for CPU use */ 8857 dma_sync_single_range_for_cpu(rx_ring->dev, 8858 rx_buffer->dma, 8859 rx_buffer->page_offset, 8860 size, 8861 DMA_FROM_DEVICE); 8862 8863 rx_buffer->pagecnt_bias--; 8864 8865 return rx_buffer; 8866 } 8867 8868 static void igb_put_rx_buffer(struct igb_ring *rx_ring, 8869 struct igb_rx_buffer *rx_buffer, int rx_buf_pgcnt) 8870 { 8871 if (igb_can_reuse_rx_page(rx_buffer, rx_buf_pgcnt)) { 8872 /* hand second half of page back to the ring */ 8873 igb_reuse_rx_page(rx_ring, rx_buffer); 8874 } else { 8875 /* We are not reusing the buffer so unmap it and free 8876 * any references we are holding to it 8877 */ 8878 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma, 8879 igb_rx_pg_size(rx_ring), DMA_FROM_DEVICE, 8880 IGB_RX_DMA_ATTR); 8881 __page_frag_cache_drain(rx_buffer->page, 8882 rx_buffer->pagecnt_bias); 8883 } 8884 8885 /* clear contents of rx_buffer */ 8886 rx_buffer->page = NULL; 8887 } 8888 8889 static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget) 8890 { 8891 struct igb_adapter *adapter = q_vector->adapter; 8892 struct igb_ring *rx_ring = q_vector->rx.ring; 8893 struct sk_buff *skb = rx_ring->skb; 8894 unsigned int total_bytes = 0, total_packets = 0; 8895 u16 cleaned_count = igb_desc_unused(rx_ring); 8896 unsigned int xdp_xmit = 0; 8897 struct xdp_buff xdp; 8898 u32 frame_sz = 0; 8899 int rx_buf_pgcnt; 8900 8901 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */ 8902 #if (PAGE_SIZE < 8192) 8903 frame_sz = igb_rx_frame_truesize(rx_ring, 0); 8904 #endif 8905 xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq); 8906 8907 while (likely(total_packets < budget)) { 8908 union e1000_adv_rx_desc *rx_desc; 8909 struct igb_rx_buffer *rx_buffer; 8910 ktime_t timestamp = 0; 8911 int pkt_offset = 0; 8912 unsigned int size; 8913 void *pktbuf; 8914 8915 /* return some buffers to hardware, one at a time is too slow */ 8916 if (cleaned_count >= IGB_RX_BUFFER_WRITE) { 8917 igb_alloc_rx_buffers(rx_ring, cleaned_count); 8918 cleaned_count = 0; 8919 } 8920 8921 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean); 8922 size = le16_to_cpu(rx_desc->wb.upper.length); 8923 if (!size) 8924 break; 8925 8926 /* This memory barrier is needed to keep us from reading 8927 * any other fields out of the rx_desc until we know the 8928 * descriptor has been written back 8929 */ 8930 dma_rmb(); 8931 8932 rx_buffer = igb_get_rx_buffer(rx_ring, size, &rx_buf_pgcnt); 8933 pktbuf = page_address(rx_buffer->page) + rx_buffer->page_offset; 8934 8935 /* pull rx packet timestamp if available and valid */ 8936 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) { 8937 int ts_hdr_len; 8938 8939 ts_hdr_len = igb_ptp_rx_pktstamp(rx_ring->q_vector, 8940 pktbuf, ×tamp); 8941 8942 pkt_offset += ts_hdr_len; 8943 size -= ts_hdr_len; 8944 } 8945 8946 /* retrieve a buffer from the ring */ 8947 if (!skb) { 8948 unsigned char *hard_start = pktbuf - igb_rx_offset(rx_ring); 8949 unsigned int offset = pkt_offset + igb_rx_offset(rx_ring); 8950 8951 xdp_prepare_buff(&xdp, hard_start, offset, size, true); 8952 xdp_buff_clear_frags_flag(&xdp); 8953 #if (PAGE_SIZE > 4096) 8954 /* At larger PAGE_SIZE, frame_sz depend on len size */ 8955 xdp.frame_sz = igb_rx_frame_truesize(rx_ring, size); 8956 #endif 8957 skb = igb_run_xdp(adapter, rx_ring, &xdp); 8958 } 8959 8960 if (IS_ERR(skb)) { 8961 unsigned int xdp_res = -PTR_ERR(skb); 8962 8963 if (xdp_res & (IGB_XDP_TX | IGB_XDP_REDIR)) { 8964 xdp_xmit |= xdp_res; 8965 igb_rx_buffer_flip(rx_ring, rx_buffer, size); 8966 } else { 8967 rx_buffer->pagecnt_bias++; 8968 } 8969 total_packets++; 8970 total_bytes += size; 8971 } else if (skb) 8972 igb_add_rx_frag(rx_ring, rx_buffer, skb, size); 8973 else if (ring_uses_build_skb(rx_ring)) 8974 skb = igb_build_skb(rx_ring, rx_buffer, &xdp, 8975 timestamp); 8976 else 8977 skb = igb_construct_skb(rx_ring, rx_buffer, 8978 &xdp, timestamp); 8979 8980 /* exit if we failed to retrieve a buffer */ 8981 if (!skb) { 8982 rx_ring->rx_stats.alloc_failed++; 8983 rx_buffer->pagecnt_bias++; 8984 break; 8985 } 8986 8987 igb_put_rx_buffer(rx_ring, rx_buffer, rx_buf_pgcnt); 8988 cleaned_count++; 8989 8990 /* fetch next buffer in frame if non-eop */ 8991 if (igb_is_non_eop(rx_ring, rx_desc)) 8992 continue; 8993 8994 /* verify the packet layout is correct */ 8995 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) { 8996 skb = NULL; 8997 continue; 8998 } 8999 9000 /* probably a little skewed due to removing CRC */ 9001 total_bytes += skb->len; 9002 9003 /* populate checksum, timestamp, VLAN, and protocol */ 9004 igb_process_skb_fields(rx_ring, rx_desc, skb); 9005 9006 napi_gro_receive(&q_vector->napi, skb); 9007 9008 /* reset skb pointer */ 9009 skb = NULL; 9010 9011 /* update budget accounting */ 9012 total_packets++; 9013 } 9014 9015 /* place incomplete frames back on ring for completion */ 9016 rx_ring->skb = skb; 9017 9018 if (xdp_xmit & IGB_XDP_REDIR) 9019 xdp_do_flush(); 9020 9021 if (xdp_xmit & IGB_XDP_TX) { 9022 struct igb_ring *tx_ring = igb_xdp_tx_queue_mapping(adapter); 9023 9024 igb_xdp_ring_update_tail(tx_ring); 9025 } 9026 9027 u64_stats_update_begin(&rx_ring->rx_syncp); 9028 rx_ring->rx_stats.packets += total_packets; 9029 rx_ring->rx_stats.bytes += total_bytes; 9030 u64_stats_update_end(&rx_ring->rx_syncp); 9031 q_vector->rx.total_packets += total_packets; 9032 q_vector->rx.total_bytes += total_bytes; 9033 9034 if (cleaned_count) 9035 igb_alloc_rx_buffers(rx_ring, cleaned_count); 9036 9037 return total_packets; 9038 } 9039 9040 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring, 9041 struct igb_rx_buffer *bi) 9042 { 9043 struct page *page = bi->page; 9044 dma_addr_t dma; 9045 9046 /* since we are recycling buffers we should seldom need to alloc */ 9047 if (likely(page)) 9048 return true; 9049 9050 /* alloc new page for storage */ 9051 page = dev_alloc_pages(igb_rx_pg_order(rx_ring)); 9052 if (unlikely(!page)) { 9053 rx_ring->rx_stats.alloc_failed++; 9054 return false; 9055 } 9056 9057 /* map page for use */ 9058 dma = dma_map_page_attrs(rx_ring->dev, page, 0, 9059 igb_rx_pg_size(rx_ring), 9060 DMA_FROM_DEVICE, 9061 IGB_RX_DMA_ATTR); 9062 9063 /* if mapping failed free memory back to system since 9064 * there isn't much point in holding memory we can't use 9065 */ 9066 if (dma_mapping_error(rx_ring->dev, dma)) { 9067 __free_pages(page, igb_rx_pg_order(rx_ring)); 9068 9069 rx_ring->rx_stats.alloc_failed++; 9070 return false; 9071 } 9072 9073 bi->dma = dma; 9074 bi->page = page; 9075 bi->page_offset = igb_rx_offset(rx_ring); 9076 page_ref_add(page, USHRT_MAX - 1); 9077 bi->pagecnt_bias = USHRT_MAX; 9078 9079 return true; 9080 } 9081 9082 /** 9083 * igb_alloc_rx_buffers - Replace used receive buffers 9084 * @rx_ring: rx descriptor ring to allocate new receive buffers 9085 * @cleaned_count: count of buffers to allocate 9086 **/ 9087 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count) 9088 { 9089 union e1000_adv_rx_desc *rx_desc; 9090 struct igb_rx_buffer *bi; 9091 u16 i = rx_ring->next_to_use; 9092 u16 bufsz; 9093 9094 /* nothing to do */ 9095 if (!cleaned_count) 9096 return; 9097 9098 rx_desc = IGB_RX_DESC(rx_ring, i); 9099 bi = &rx_ring->rx_buffer_info[i]; 9100 i -= rx_ring->count; 9101 9102 bufsz = igb_rx_bufsz(rx_ring); 9103 9104 do { 9105 if (!igb_alloc_mapped_page(rx_ring, bi)) 9106 break; 9107 9108 /* sync the buffer for use by the device */ 9109 dma_sync_single_range_for_device(rx_ring->dev, bi->dma, 9110 bi->page_offset, bufsz, 9111 DMA_FROM_DEVICE); 9112 9113 /* Refresh the desc even if buffer_addrs didn't change 9114 * because each write-back erases this info. 9115 */ 9116 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset); 9117 9118 rx_desc++; 9119 bi++; 9120 i++; 9121 if (unlikely(!i)) { 9122 rx_desc = IGB_RX_DESC(rx_ring, 0); 9123 bi = rx_ring->rx_buffer_info; 9124 i -= rx_ring->count; 9125 } 9126 9127 /* clear the length for the next_to_use descriptor */ 9128 rx_desc->wb.upper.length = 0; 9129 9130 cleaned_count--; 9131 } while (cleaned_count); 9132 9133 i += rx_ring->count; 9134 9135 if (rx_ring->next_to_use != i) { 9136 /* record the next descriptor to use */ 9137 rx_ring->next_to_use = i; 9138 9139 /* update next to alloc since we have filled the ring */ 9140 rx_ring->next_to_alloc = i; 9141 9142 /* Force memory writes to complete before letting h/w 9143 * know there are new descriptors to fetch. (Only 9144 * applicable for weak-ordered memory model archs, 9145 * such as IA-64). 9146 */ 9147 dma_wmb(); 9148 writel(i, rx_ring->tail); 9149 } 9150 } 9151 9152 /** 9153 * igb_mii_ioctl - 9154 * @netdev: pointer to netdev struct 9155 * @ifr: interface structure 9156 * @cmd: ioctl command to execute 9157 **/ 9158 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd) 9159 { 9160 struct igb_adapter *adapter = netdev_priv(netdev); 9161 struct mii_ioctl_data *data = if_mii(ifr); 9162 9163 if (adapter->hw.phy.media_type != e1000_media_type_copper) 9164 return -EOPNOTSUPP; 9165 9166 switch (cmd) { 9167 case SIOCGMIIPHY: 9168 data->phy_id = adapter->hw.phy.addr; 9169 break; 9170 case SIOCGMIIREG: 9171 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F, 9172 &data->val_out)) 9173 return -EIO; 9174 break; 9175 case SIOCSMIIREG: 9176 default: 9177 return -EOPNOTSUPP; 9178 } 9179 return 0; 9180 } 9181 9182 /** 9183 * igb_ioctl - 9184 * @netdev: pointer to netdev struct 9185 * @ifr: interface structure 9186 * @cmd: ioctl command to execute 9187 **/ 9188 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd) 9189 { 9190 switch (cmd) { 9191 case SIOCGMIIPHY: 9192 case SIOCGMIIREG: 9193 case SIOCSMIIREG: 9194 return igb_mii_ioctl(netdev, ifr, cmd); 9195 case SIOCGHWTSTAMP: 9196 return igb_ptp_get_ts_config(netdev, ifr); 9197 case SIOCSHWTSTAMP: 9198 return igb_ptp_set_ts_config(netdev, ifr); 9199 default: 9200 return -EOPNOTSUPP; 9201 } 9202 } 9203 9204 void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value) 9205 { 9206 struct igb_adapter *adapter = hw->back; 9207 9208 pci_read_config_word(adapter->pdev, reg, value); 9209 } 9210 9211 void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value) 9212 { 9213 struct igb_adapter *adapter = hw->back; 9214 9215 pci_write_config_word(adapter->pdev, reg, *value); 9216 } 9217 9218 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value) 9219 { 9220 struct igb_adapter *adapter = hw->back; 9221 9222 if (pcie_capability_read_word(adapter->pdev, reg, value)) 9223 return -E1000_ERR_CONFIG; 9224 9225 return 0; 9226 } 9227 9228 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value) 9229 { 9230 struct igb_adapter *adapter = hw->back; 9231 9232 if (pcie_capability_write_word(adapter->pdev, reg, *value)) 9233 return -E1000_ERR_CONFIG; 9234 9235 return 0; 9236 } 9237 9238 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features) 9239 { 9240 struct igb_adapter *adapter = netdev_priv(netdev); 9241 struct e1000_hw *hw = &adapter->hw; 9242 u32 ctrl, rctl; 9243 bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX); 9244 9245 if (enable) { 9246 /* enable VLAN tag insert/strip */ 9247 ctrl = rd32(E1000_CTRL); 9248 ctrl |= E1000_CTRL_VME; 9249 wr32(E1000_CTRL, ctrl); 9250 9251 /* Disable CFI check */ 9252 rctl = rd32(E1000_RCTL); 9253 rctl &= ~E1000_RCTL_CFIEN; 9254 wr32(E1000_RCTL, rctl); 9255 } else { 9256 /* disable VLAN tag insert/strip */ 9257 ctrl = rd32(E1000_CTRL); 9258 ctrl &= ~E1000_CTRL_VME; 9259 wr32(E1000_CTRL, ctrl); 9260 } 9261 9262 igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable); 9263 } 9264 9265 static int igb_vlan_rx_add_vid(struct net_device *netdev, 9266 __be16 proto, u16 vid) 9267 { 9268 struct igb_adapter *adapter = netdev_priv(netdev); 9269 struct e1000_hw *hw = &adapter->hw; 9270 int pf_id = adapter->vfs_allocated_count; 9271 9272 /* add the filter since PF can receive vlans w/o entry in vlvf */ 9273 if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC)) 9274 igb_vfta_set(hw, vid, pf_id, true, !!vid); 9275 9276 set_bit(vid, adapter->active_vlans); 9277 9278 return 0; 9279 } 9280 9281 static int igb_vlan_rx_kill_vid(struct net_device *netdev, 9282 __be16 proto, u16 vid) 9283 { 9284 struct igb_adapter *adapter = netdev_priv(netdev); 9285 int pf_id = adapter->vfs_allocated_count; 9286 struct e1000_hw *hw = &adapter->hw; 9287 9288 /* remove VID from filter table */ 9289 if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC)) 9290 igb_vfta_set(hw, vid, pf_id, false, true); 9291 9292 clear_bit(vid, adapter->active_vlans); 9293 9294 return 0; 9295 } 9296 9297 static void igb_restore_vlan(struct igb_adapter *adapter) 9298 { 9299 u16 vid = 1; 9300 9301 igb_vlan_mode(adapter->netdev, adapter->netdev->features); 9302 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0); 9303 9304 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID) 9305 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid); 9306 } 9307 9308 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx) 9309 { 9310 struct pci_dev *pdev = adapter->pdev; 9311 struct e1000_mac_info *mac = &adapter->hw.mac; 9312 9313 mac->autoneg = 0; 9314 9315 /* Make sure dplx is at most 1 bit and lsb of speed is not set 9316 * for the switch() below to work 9317 */ 9318 if ((spd & 1) || (dplx & ~1)) 9319 goto err_inval; 9320 9321 /* Fiber NIC's only allow 1000 gbps Full duplex 9322 * and 100Mbps Full duplex for 100baseFx sfp 9323 */ 9324 if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) { 9325 switch (spd + dplx) { 9326 case SPEED_10 + DUPLEX_HALF: 9327 case SPEED_10 + DUPLEX_FULL: 9328 case SPEED_100 + DUPLEX_HALF: 9329 goto err_inval; 9330 default: 9331 break; 9332 } 9333 } 9334 9335 switch (spd + dplx) { 9336 case SPEED_10 + DUPLEX_HALF: 9337 mac->forced_speed_duplex = ADVERTISE_10_HALF; 9338 break; 9339 case SPEED_10 + DUPLEX_FULL: 9340 mac->forced_speed_duplex = ADVERTISE_10_FULL; 9341 break; 9342 case SPEED_100 + DUPLEX_HALF: 9343 mac->forced_speed_duplex = ADVERTISE_100_HALF; 9344 break; 9345 case SPEED_100 + DUPLEX_FULL: 9346 mac->forced_speed_duplex = ADVERTISE_100_FULL; 9347 break; 9348 case SPEED_1000 + DUPLEX_FULL: 9349 mac->autoneg = 1; 9350 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL; 9351 break; 9352 case SPEED_1000 + DUPLEX_HALF: /* not supported */ 9353 default: 9354 goto err_inval; 9355 } 9356 9357 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */ 9358 adapter->hw.phy.mdix = AUTO_ALL_MODES; 9359 9360 return 0; 9361 9362 err_inval: 9363 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n"); 9364 return -EINVAL; 9365 } 9366 9367 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake, 9368 bool runtime) 9369 { 9370 struct net_device *netdev = pci_get_drvdata(pdev); 9371 struct igb_adapter *adapter = netdev_priv(netdev); 9372 struct e1000_hw *hw = &adapter->hw; 9373 u32 ctrl, rctl, status; 9374 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol; 9375 bool wake; 9376 9377 rtnl_lock(); 9378 netif_device_detach(netdev); 9379 9380 if (netif_running(netdev)) 9381 __igb_close(netdev, true); 9382 9383 igb_ptp_suspend(adapter); 9384 9385 igb_clear_interrupt_scheme(adapter); 9386 rtnl_unlock(); 9387 9388 status = rd32(E1000_STATUS); 9389 if (status & E1000_STATUS_LU) 9390 wufc &= ~E1000_WUFC_LNKC; 9391 9392 if (wufc) { 9393 igb_setup_rctl(adapter); 9394 igb_set_rx_mode(netdev); 9395 9396 /* turn on all-multi mode if wake on multicast is enabled */ 9397 if (wufc & E1000_WUFC_MC) { 9398 rctl = rd32(E1000_RCTL); 9399 rctl |= E1000_RCTL_MPE; 9400 wr32(E1000_RCTL, rctl); 9401 } 9402 9403 ctrl = rd32(E1000_CTRL); 9404 ctrl |= E1000_CTRL_ADVD3WUC; 9405 wr32(E1000_CTRL, ctrl); 9406 9407 /* Allow time for pending master requests to run */ 9408 igb_disable_pcie_master(hw); 9409 9410 wr32(E1000_WUC, E1000_WUC_PME_EN); 9411 wr32(E1000_WUFC, wufc); 9412 } else { 9413 wr32(E1000_WUC, 0); 9414 wr32(E1000_WUFC, 0); 9415 } 9416 9417 wake = wufc || adapter->en_mng_pt; 9418 if (!wake) 9419 igb_power_down_link(adapter); 9420 else 9421 igb_power_up_link(adapter); 9422 9423 if (enable_wake) 9424 *enable_wake = wake; 9425 9426 /* Release control of h/w to f/w. If f/w is AMT enabled, this 9427 * would have already happened in close and is redundant. 9428 */ 9429 igb_release_hw_control(adapter); 9430 9431 pci_disable_device(pdev); 9432 9433 return 0; 9434 } 9435 9436 static void igb_deliver_wake_packet(struct net_device *netdev) 9437 { 9438 struct igb_adapter *adapter = netdev_priv(netdev); 9439 struct e1000_hw *hw = &adapter->hw; 9440 struct sk_buff *skb; 9441 u32 wupl; 9442 9443 wupl = rd32(E1000_WUPL) & E1000_WUPL_MASK; 9444 9445 /* WUPM stores only the first 128 bytes of the wake packet. 9446 * Read the packet only if we have the whole thing. 9447 */ 9448 if ((wupl == 0) || (wupl > E1000_WUPM_BYTES)) 9449 return; 9450 9451 skb = netdev_alloc_skb_ip_align(netdev, E1000_WUPM_BYTES); 9452 if (!skb) 9453 return; 9454 9455 skb_put(skb, wupl); 9456 9457 /* Ensure reads are 32-bit aligned */ 9458 wupl = roundup(wupl, 4); 9459 9460 memcpy_fromio(skb->data, hw->hw_addr + E1000_WUPM_REG(0), wupl); 9461 9462 skb->protocol = eth_type_trans(skb, netdev); 9463 netif_rx(skb); 9464 } 9465 9466 static int __maybe_unused igb_suspend(struct device *dev) 9467 { 9468 return __igb_shutdown(to_pci_dev(dev), NULL, 0); 9469 } 9470 9471 static int __maybe_unused __igb_resume(struct device *dev, bool rpm) 9472 { 9473 struct pci_dev *pdev = to_pci_dev(dev); 9474 struct net_device *netdev = pci_get_drvdata(pdev); 9475 struct igb_adapter *adapter = netdev_priv(netdev); 9476 struct e1000_hw *hw = &adapter->hw; 9477 u32 err, val; 9478 9479 pci_set_power_state(pdev, PCI_D0); 9480 pci_restore_state(pdev); 9481 pci_save_state(pdev); 9482 9483 if (!pci_device_is_present(pdev)) 9484 return -ENODEV; 9485 err = pci_enable_device_mem(pdev); 9486 if (err) { 9487 dev_err(&pdev->dev, 9488 "igb: Cannot enable PCI device from suspend\n"); 9489 return err; 9490 } 9491 pci_set_master(pdev); 9492 9493 pci_enable_wake(pdev, PCI_D3hot, 0); 9494 pci_enable_wake(pdev, PCI_D3cold, 0); 9495 9496 if (igb_init_interrupt_scheme(adapter, true)) { 9497 dev_err(&pdev->dev, "Unable to allocate memory for queues\n"); 9498 return -ENOMEM; 9499 } 9500 9501 igb_reset(adapter); 9502 9503 /* let the f/w know that the h/w is now under the control of the 9504 * driver. 9505 */ 9506 igb_get_hw_control(adapter); 9507 9508 val = rd32(E1000_WUS); 9509 if (val & WAKE_PKT_WUS) 9510 igb_deliver_wake_packet(netdev); 9511 9512 wr32(E1000_WUS, ~0); 9513 9514 if (!rpm) 9515 rtnl_lock(); 9516 if (!err && netif_running(netdev)) 9517 err = __igb_open(netdev, true); 9518 9519 if (!err) 9520 netif_device_attach(netdev); 9521 if (!rpm) 9522 rtnl_unlock(); 9523 9524 return err; 9525 } 9526 9527 static int __maybe_unused igb_resume(struct device *dev) 9528 { 9529 return __igb_resume(dev, false); 9530 } 9531 9532 static int __maybe_unused igb_runtime_idle(struct device *dev) 9533 { 9534 struct net_device *netdev = dev_get_drvdata(dev); 9535 struct igb_adapter *adapter = netdev_priv(netdev); 9536 9537 if (!igb_has_link(adapter)) 9538 pm_schedule_suspend(dev, MSEC_PER_SEC * 5); 9539 9540 return -EBUSY; 9541 } 9542 9543 static int __maybe_unused igb_runtime_suspend(struct device *dev) 9544 { 9545 return __igb_shutdown(to_pci_dev(dev), NULL, 1); 9546 } 9547 9548 static int __maybe_unused igb_runtime_resume(struct device *dev) 9549 { 9550 return __igb_resume(dev, true); 9551 } 9552 9553 static void igb_shutdown(struct pci_dev *pdev) 9554 { 9555 bool wake; 9556 9557 __igb_shutdown(pdev, &wake, 0); 9558 9559 if (system_state == SYSTEM_POWER_OFF) { 9560 pci_wake_from_d3(pdev, wake); 9561 pci_set_power_state(pdev, PCI_D3hot); 9562 } 9563 } 9564 9565 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs) 9566 { 9567 #ifdef CONFIG_PCI_IOV 9568 int err; 9569 9570 if (num_vfs == 0) { 9571 return igb_disable_sriov(dev, true); 9572 } else { 9573 err = igb_enable_sriov(dev, num_vfs, true); 9574 return err ? err : num_vfs; 9575 } 9576 #endif 9577 return 0; 9578 } 9579 9580 /** 9581 * igb_io_error_detected - called when PCI error is detected 9582 * @pdev: Pointer to PCI device 9583 * @state: The current pci connection state 9584 * 9585 * This function is called after a PCI bus error affecting 9586 * this device has been detected. 9587 **/ 9588 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev, 9589 pci_channel_state_t state) 9590 { 9591 struct net_device *netdev = pci_get_drvdata(pdev); 9592 struct igb_adapter *adapter = netdev_priv(netdev); 9593 9594 if (state == pci_channel_io_normal) { 9595 dev_warn(&pdev->dev, "Non-correctable non-fatal error reported.\n"); 9596 return PCI_ERS_RESULT_CAN_RECOVER; 9597 } 9598 9599 netif_device_detach(netdev); 9600 9601 if (state == pci_channel_io_perm_failure) 9602 return PCI_ERS_RESULT_DISCONNECT; 9603 9604 if (netif_running(netdev)) 9605 igb_down(adapter); 9606 pci_disable_device(pdev); 9607 9608 /* Request a slot reset. */ 9609 return PCI_ERS_RESULT_NEED_RESET; 9610 } 9611 9612 /** 9613 * igb_io_slot_reset - called after the pci bus has been reset. 9614 * @pdev: Pointer to PCI device 9615 * 9616 * Restart the card from scratch, as if from a cold-boot. Implementation 9617 * resembles the first-half of the __igb_resume routine. 9618 **/ 9619 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev) 9620 { 9621 struct net_device *netdev = pci_get_drvdata(pdev); 9622 struct igb_adapter *adapter = netdev_priv(netdev); 9623 struct e1000_hw *hw = &adapter->hw; 9624 pci_ers_result_t result; 9625 9626 if (pci_enable_device_mem(pdev)) { 9627 dev_err(&pdev->dev, 9628 "Cannot re-enable PCI device after reset.\n"); 9629 result = PCI_ERS_RESULT_DISCONNECT; 9630 } else { 9631 pci_set_master(pdev); 9632 pci_restore_state(pdev); 9633 pci_save_state(pdev); 9634 9635 pci_enable_wake(pdev, PCI_D3hot, 0); 9636 pci_enable_wake(pdev, PCI_D3cold, 0); 9637 9638 /* In case of PCI error, adapter lose its HW address 9639 * so we should re-assign it here. 9640 */ 9641 hw->hw_addr = adapter->io_addr; 9642 9643 igb_reset(adapter); 9644 wr32(E1000_WUS, ~0); 9645 result = PCI_ERS_RESULT_RECOVERED; 9646 } 9647 9648 return result; 9649 } 9650 9651 /** 9652 * igb_io_resume - called when traffic can start flowing again. 9653 * @pdev: Pointer to PCI device 9654 * 9655 * This callback is called when the error recovery driver tells us that 9656 * its OK to resume normal operation. Implementation resembles the 9657 * second-half of the __igb_resume routine. 9658 */ 9659 static void igb_io_resume(struct pci_dev *pdev) 9660 { 9661 struct net_device *netdev = pci_get_drvdata(pdev); 9662 struct igb_adapter *adapter = netdev_priv(netdev); 9663 9664 if (netif_running(netdev)) { 9665 if (igb_up(adapter)) { 9666 dev_err(&pdev->dev, "igb_up failed after reset\n"); 9667 return; 9668 } 9669 } 9670 9671 netif_device_attach(netdev); 9672 9673 /* let the f/w know that the h/w is now under the control of the 9674 * driver. 9675 */ 9676 igb_get_hw_control(adapter); 9677 } 9678 9679 /** 9680 * igb_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table 9681 * @adapter: Pointer to adapter structure 9682 * @index: Index of the RAR entry which need to be synced with MAC table 9683 **/ 9684 static void igb_rar_set_index(struct igb_adapter *adapter, u32 index) 9685 { 9686 struct e1000_hw *hw = &adapter->hw; 9687 u32 rar_low, rar_high; 9688 u8 *addr = adapter->mac_table[index].addr; 9689 9690 /* HW expects these to be in network order when they are plugged 9691 * into the registers which are little endian. In order to guarantee 9692 * that ordering we need to do an leXX_to_cpup here in order to be 9693 * ready for the byteswap that occurs with writel 9694 */ 9695 rar_low = le32_to_cpup((__le32 *)(addr)); 9696 rar_high = le16_to_cpup((__le16 *)(addr + 4)); 9697 9698 /* Indicate to hardware the Address is Valid. */ 9699 if (adapter->mac_table[index].state & IGB_MAC_STATE_IN_USE) { 9700 if (is_valid_ether_addr(addr)) 9701 rar_high |= E1000_RAH_AV; 9702 9703 if (adapter->mac_table[index].state & IGB_MAC_STATE_SRC_ADDR) 9704 rar_high |= E1000_RAH_ASEL_SRC_ADDR; 9705 9706 switch (hw->mac.type) { 9707 case e1000_82575: 9708 case e1000_i210: 9709 if (adapter->mac_table[index].state & 9710 IGB_MAC_STATE_QUEUE_STEERING) 9711 rar_high |= E1000_RAH_QSEL_ENABLE; 9712 9713 rar_high |= E1000_RAH_POOL_1 * 9714 adapter->mac_table[index].queue; 9715 break; 9716 default: 9717 rar_high |= E1000_RAH_POOL_1 << 9718 adapter->mac_table[index].queue; 9719 break; 9720 } 9721 } 9722 9723 wr32(E1000_RAL(index), rar_low); 9724 wrfl(); 9725 wr32(E1000_RAH(index), rar_high); 9726 wrfl(); 9727 } 9728 9729 static int igb_set_vf_mac(struct igb_adapter *adapter, 9730 int vf, unsigned char *mac_addr) 9731 { 9732 struct e1000_hw *hw = &adapter->hw; 9733 /* VF MAC addresses start at end of receive addresses and moves 9734 * towards the first, as a result a collision should not be possible 9735 */ 9736 int rar_entry = hw->mac.rar_entry_count - (vf + 1); 9737 unsigned char *vf_mac_addr = adapter->vf_data[vf].vf_mac_addresses; 9738 9739 ether_addr_copy(vf_mac_addr, mac_addr); 9740 ether_addr_copy(adapter->mac_table[rar_entry].addr, mac_addr); 9741 adapter->mac_table[rar_entry].queue = vf; 9742 adapter->mac_table[rar_entry].state |= IGB_MAC_STATE_IN_USE; 9743 igb_rar_set_index(adapter, rar_entry); 9744 9745 return 0; 9746 } 9747 9748 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac) 9749 { 9750 struct igb_adapter *adapter = netdev_priv(netdev); 9751 9752 if (vf >= adapter->vfs_allocated_count) 9753 return -EINVAL; 9754 9755 /* Setting the VF MAC to 0 reverts the IGB_VF_FLAG_PF_SET_MAC 9756 * flag and allows to overwrite the MAC via VF netdev. This 9757 * is necessary to allow libvirt a way to restore the original 9758 * MAC after unbinding vfio-pci and reloading igbvf after shutting 9759 * down a VM. 9760 */ 9761 if (is_zero_ether_addr(mac)) { 9762 adapter->vf_data[vf].flags &= ~IGB_VF_FLAG_PF_SET_MAC; 9763 dev_info(&adapter->pdev->dev, 9764 "remove administratively set MAC on VF %d\n", 9765 vf); 9766 } else if (is_valid_ether_addr(mac)) { 9767 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC; 9768 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", 9769 mac, vf); 9770 dev_info(&adapter->pdev->dev, 9771 "Reload the VF driver to make this change effective."); 9772 /* Generate additional warning if PF is down */ 9773 if (test_bit(__IGB_DOWN, &adapter->state)) { 9774 dev_warn(&adapter->pdev->dev, 9775 "The VF MAC address has been set, but the PF device is not up.\n"); 9776 dev_warn(&adapter->pdev->dev, 9777 "Bring the PF device up before attempting to use the VF device.\n"); 9778 } 9779 } else { 9780 return -EINVAL; 9781 } 9782 return igb_set_vf_mac(adapter, vf, mac); 9783 } 9784 9785 static int igb_link_mbps(int internal_link_speed) 9786 { 9787 switch (internal_link_speed) { 9788 case SPEED_100: 9789 return 100; 9790 case SPEED_1000: 9791 return 1000; 9792 default: 9793 return 0; 9794 } 9795 } 9796 9797 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate, 9798 int link_speed) 9799 { 9800 int rf_dec, rf_int; 9801 u32 bcnrc_val; 9802 9803 if (tx_rate != 0) { 9804 /* Calculate the rate factor values to set */ 9805 rf_int = link_speed / tx_rate; 9806 rf_dec = (link_speed - (rf_int * tx_rate)); 9807 rf_dec = (rf_dec * BIT(E1000_RTTBCNRC_RF_INT_SHIFT)) / 9808 tx_rate; 9809 9810 bcnrc_val = E1000_RTTBCNRC_RS_ENA; 9811 bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) & 9812 E1000_RTTBCNRC_RF_INT_MASK); 9813 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK); 9814 } else { 9815 bcnrc_val = 0; 9816 } 9817 9818 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */ 9819 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM 9820 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported. 9821 */ 9822 wr32(E1000_RTTBCNRM, 0x14); 9823 wr32(E1000_RTTBCNRC, bcnrc_val); 9824 } 9825 9826 static void igb_check_vf_rate_limit(struct igb_adapter *adapter) 9827 { 9828 int actual_link_speed, i; 9829 bool reset_rate = false; 9830 9831 /* VF TX rate limit was not set or not supported */ 9832 if ((adapter->vf_rate_link_speed == 0) || 9833 (adapter->hw.mac.type != e1000_82576)) 9834 return; 9835 9836 actual_link_speed = igb_link_mbps(adapter->link_speed); 9837 if (actual_link_speed != adapter->vf_rate_link_speed) { 9838 reset_rate = true; 9839 adapter->vf_rate_link_speed = 0; 9840 dev_info(&adapter->pdev->dev, 9841 "Link speed has been changed. VF Transmit rate is disabled\n"); 9842 } 9843 9844 for (i = 0; i < adapter->vfs_allocated_count; i++) { 9845 if (reset_rate) 9846 adapter->vf_data[i].tx_rate = 0; 9847 9848 igb_set_vf_rate_limit(&adapter->hw, i, 9849 adapter->vf_data[i].tx_rate, 9850 actual_link_speed); 9851 } 9852 } 9853 9854 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, 9855 int min_tx_rate, int max_tx_rate) 9856 { 9857 struct igb_adapter *adapter = netdev_priv(netdev); 9858 struct e1000_hw *hw = &adapter->hw; 9859 int actual_link_speed; 9860 9861 if (hw->mac.type != e1000_82576) 9862 return -EOPNOTSUPP; 9863 9864 if (min_tx_rate) 9865 return -EINVAL; 9866 9867 actual_link_speed = igb_link_mbps(adapter->link_speed); 9868 if ((vf >= adapter->vfs_allocated_count) || 9869 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) || 9870 (max_tx_rate < 0) || 9871 (max_tx_rate > actual_link_speed)) 9872 return -EINVAL; 9873 9874 adapter->vf_rate_link_speed = actual_link_speed; 9875 adapter->vf_data[vf].tx_rate = (u16)max_tx_rate; 9876 igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed); 9877 9878 return 0; 9879 } 9880 9881 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf, 9882 bool setting) 9883 { 9884 struct igb_adapter *adapter = netdev_priv(netdev); 9885 struct e1000_hw *hw = &adapter->hw; 9886 u32 reg_val, reg_offset; 9887 9888 if (!adapter->vfs_allocated_count) 9889 return -EOPNOTSUPP; 9890 9891 if (vf >= adapter->vfs_allocated_count) 9892 return -EINVAL; 9893 9894 reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC; 9895 reg_val = rd32(reg_offset); 9896 if (setting) 9897 reg_val |= (BIT(vf) | 9898 BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)); 9899 else 9900 reg_val &= ~(BIT(vf) | 9901 BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)); 9902 wr32(reg_offset, reg_val); 9903 9904 adapter->vf_data[vf].spoofchk_enabled = setting; 9905 return 0; 9906 } 9907 9908 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf, bool setting) 9909 { 9910 struct igb_adapter *adapter = netdev_priv(netdev); 9911 9912 if (vf >= adapter->vfs_allocated_count) 9913 return -EINVAL; 9914 if (adapter->vf_data[vf].trusted == setting) 9915 return 0; 9916 9917 adapter->vf_data[vf].trusted = setting; 9918 9919 dev_info(&adapter->pdev->dev, "VF %u is %strusted\n", 9920 vf, setting ? "" : "not "); 9921 return 0; 9922 } 9923 9924 static int igb_ndo_get_vf_config(struct net_device *netdev, 9925 int vf, struct ifla_vf_info *ivi) 9926 { 9927 struct igb_adapter *adapter = netdev_priv(netdev); 9928 if (vf >= adapter->vfs_allocated_count) 9929 return -EINVAL; 9930 ivi->vf = vf; 9931 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN); 9932 ivi->max_tx_rate = adapter->vf_data[vf].tx_rate; 9933 ivi->min_tx_rate = 0; 9934 ivi->vlan = adapter->vf_data[vf].pf_vlan; 9935 ivi->qos = adapter->vf_data[vf].pf_qos; 9936 ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled; 9937 ivi->trusted = adapter->vf_data[vf].trusted; 9938 return 0; 9939 } 9940 9941 static void igb_vmm_control(struct igb_adapter *adapter) 9942 { 9943 struct e1000_hw *hw = &adapter->hw; 9944 u32 reg; 9945 9946 switch (hw->mac.type) { 9947 case e1000_82575: 9948 case e1000_i210: 9949 case e1000_i211: 9950 case e1000_i354: 9951 default: 9952 /* replication is not supported for 82575 */ 9953 return; 9954 case e1000_82576: 9955 /* notify HW that the MAC is adding vlan tags */ 9956 reg = rd32(E1000_DTXCTL); 9957 reg |= E1000_DTXCTL_VLAN_ADDED; 9958 wr32(E1000_DTXCTL, reg); 9959 fallthrough; 9960 case e1000_82580: 9961 /* enable replication vlan tag stripping */ 9962 reg = rd32(E1000_RPLOLR); 9963 reg |= E1000_RPLOLR_STRVLAN; 9964 wr32(E1000_RPLOLR, reg); 9965 fallthrough; 9966 case e1000_i350: 9967 /* none of the above registers are supported by i350 */ 9968 break; 9969 } 9970 9971 if (adapter->vfs_allocated_count) { 9972 igb_vmdq_set_loopback_pf(hw, true); 9973 igb_vmdq_set_replication_pf(hw, true); 9974 igb_vmdq_set_anti_spoofing_pf(hw, true, 9975 adapter->vfs_allocated_count); 9976 } else { 9977 igb_vmdq_set_loopback_pf(hw, false); 9978 igb_vmdq_set_replication_pf(hw, false); 9979 } 9980 } 9981 9982 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba) 9983 { 9984 struct e1000_hw *hw = &adapter->hw; 9985 u32 dmac_thr; 9986 u16 hwm; 9987 u32 reg; 9988 9989 if (hw->mac.type > e1000_82580) { 9990 if (adapter->flags & IGB_FLAG_DMAC) { 9991 /* force threshold to 0. */ 9992 wr32(E1000_DMCTXTH, 0); 9993 9994 /* DMA Coalescing high water mark needs to be greater 9995 * than the Rx threshold. Set hwm to PBA - max frame 9996 * size in 16B units, capping it at PBA - 6KB. 9997 */ 9998 hwm = 64 * (pba - 6); 9999 reg = rd32(E1000_FCRTC); 10000 reg &= ~E1000_FCRTC_RTH_COAL_MASK; 10001 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT) 10002 & E1000_FCRTC_RTH_COAL_MASK); 10003 wr32(E1000_FCRTC, reg); 10004 10005 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max 10006 * frame size, capping it at PBA - 10KB. 10007 */ 10008 dmac_thr = pba - 10; 10009 reg = rd32(E1000_DMACR); 10010 reg &= ~E1000_DMACR_DMACTHR_MASK; 10011 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT) 10012 & E1000_DMACR_DMACTHR_MASK); 10013 10014 /* transition to L0x or L1 if available..*/ 10015 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK); 10016 10017 /* watchdog timer= +-1000 usec in 32usec intervals */ 10018 reg |= (1000 >> 5); 10019 10020 /* Disable BMC-to-OS Watchdog Enable */ 10021 if (hw->mac.type != e1000_i354) 10022 reg &= ~E1000_DMACR_DC_BMC2OSW_EN; 10023 wr32(E1000_DMACR, reg); 10024 10025 /* no lower threshold to disable 10026 * coalescing(smart fifb)-UTRESH=0 10027 */ 10028 wr32(E1000_DMCRTRH, 0); 10029 10030 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4); 10031 10032 wr32(E1000_DMCTLX, reg); 10033 10034 /* free space in tx packet buffer to wake from 10035 * DMA coal 10036 */ 10037 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE - 10038 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6); 10039 } 10040 10041 if (hw->mac.type >= e1000_i210 || 10042 (adapter->flags & IGB_FLAG_DMAC)) { 10043 reg = rd32(E1000_PCIEMISC); 10044 reg |= E1000_PCIEMISC_LX_DECISION; 10045 wr32(E1000_PCIEMISC, reg); 10046 } /* endif adapter->dmac is not disabled */ 10047 } else if (hw->mac.type == e1000_82580) { 10048 u32 reg = rd32(E1000_PCIEMISC); 10049 10050 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION); 10051 wr32(E1000_DMACR, 0); 10052 } 10053 } 10054 10055 /** 10056 * igb_read_i2c_byte - Reads 8 bit word over I2C 10057 * @hw: pointer to hardware structure 10058 * @byte_offset: byte offset to read 10059 * @dev_addr: device address 10060 * @data: value read 10061 * 10062 * Performs byte read operation over I2C interface at 10063 * a specified device address. 10064 **/ 10065 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset, 10066 u8 dev_addr, u8 *data) 10067 { 10068 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw); 10069 struct i2c_client *this_client = adapter->i2c_client; 10070 s32 status; 10071 u16 swfw_mask = 0; 10072 10073 if (!this_client) 10074 return E1000_ERR_I2C; 10075 10076 swfw_mask = E1000_SWFW_PHY0_SM; 10077 10078 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask)) 10079 return E1000_ERR_SWFW_SYNC; 10080 10081 status = i2c_smbus_read_byte_data(this_client, byte_offset); 10082 hw->mac.ops.release_swfw_sync(hw, swfw_mask); 10083 10084 if (status < 0) 10085 return E1000_ERR_I2C; 10086 else { 10087 *data = status; 10088 return 0; 10089 } 10090 } 10091 10092 /** 10093 * igb_write_i2c_byte - Writes 8 bit word over I2C 10094 * @hw: pointer to hardware structure 10095 * @byte_offset: byte offset to write 10096 * @dev_addr: device address 10097 * @data: value to write 10098 * 10099 * Performs byte write operation over I2C interface at 10100 * a specified device address. 10101 **/ 10102 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset, 10103 u8 dev_addr, u8 data) 10104 { 10105 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw); 10106 struct i2c_client *this_client = adapter->i2c_client; 10107 s32 status; 10108 u16 swfw_mask = E1000_SWFW_PHY0_SM; 10109 10110 if (!this_client) 10111 return E1000_ERR_I2C; 10112 10113 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask)) 10114 return E1000_ERR_SWFW_SYNC; 10115 status = i2c_smbus_write_byte_data(this_client, byte_offset, data); 10116 hw->mac.ops.release_swfw_sync(hw, swfw_mask); 10117 10118 if (status) 10119 return E1000_ERR_I2C; 10120 else 10121 return 0; 10122 10123 } 10124 10125 int igb_reinit_queues(struct igb_adapter *adapter) 10126 { 10127 struct net_device *netdev = adapter->netdev; 10128 struct pci_dev *pdev = adapter->pdev; 10129 int err = 0; 10130 10131 if (netif_running(netdev)) 10132 igb_close(netdev); 10133 10134 igb_reset_interrupt_capability(adapter); 10135 10136 if (igb_init_interrupt_scheme(adapter, true)) { 10137 dev_err(&pdev->dev, "Unable to allocate memory for queues\n"); 10138 return -ENOMEM; 10139 } 10140 10141 if (netif_running(netdev)) 10142 err = igb_open(netdev); 10143 10144 return err; 10145 } 10146 10147 static void igb_nfc_filter_exit(struct igb_adapter *adapter) 10148 { 10149 struct igb_nfc_filter *rule; 10150 10151 spin_lock(&adapter->nfc_lock); 10152 10153 hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node) 10154 igb_erase_filter(adapter, rule); 10155 10156 hlist_for_each_entry(rule, &adapter->cls_flower_list, nfc_node) 10157 igb_erase_filter(adapter, rule); 10158 10159 spin_unlock(&adapter->nfc_lock); 10160 } 10161 10162 static void igb_nfc_filter_restore(struct igb_adapter *adapter) 10163 { 10164 struct igb_nfc_filter *rule; 10165 10166 spin_lock(&adapter->nfc_lock); 10167 10168 hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node) 10169 igb_add_filter(adapter, rule); 10170 10171 spin_unlock(&adapter->nfc_lock); 10172 } 10173 /* igb_main.c */ 10174