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/aer.h> 32 #include <linux/prefetch.h> 33 #include <linux/pm_runtime.h> 34 #include <linux/etherdevice.h> 35 #ifdef CONFIG_IGB_DCA 36 #include <linux/dca.h> 37 #endif 38 #include <linux/i2c.h> 39 #include "igb.h" 40 41 #define MAJ 5 42 #define MIN 6 43 #define BUILD 0 44 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \ 45 __stringify(BUILD) "-k" 46 47 enum queue_mode { 48 QUEUE_MODE_STRICT_PRIORITY, 49 QUEUE_MODE_STREAM_RESERVATION, 50 }; 51 52 enum tx_queue_prio { 53 TX_QUEUE_PRIO_HIGH, 54 TX_QUEUE_PRIO_LOW, 55 }; 56 57 char igb_driver_name[] = "igb"; 58 char igb_driver_version[] = DRV_VERSION; 59 static const char igb_driver_string[] = 60 "Intel(R) Gigabit Ethernet Network Driver"; 61 static const char igb_copyright[] = 62 "Copyright (c) 2007-2014 Intel Corporation."; 63 64 static const struct e1000_info *igb_info_tbl[] = { 65 [board_82575] = &e1000_82575_info, 66 }; 67 68 static const struct pci_device_id igb_pci_tbl[] = { 69 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) }, 70 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) }, 71 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) }, 72 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 }, 73 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 }, 74 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 }, 75 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 }, 76 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 }, 77 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 }, 78 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 }, 79 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 }, 80 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 }, 81 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 }, 82 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 }, 83 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 }, 84 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 }, 85 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 }, 86 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 }, 87 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 }, 88 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 }, 89 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 }, 90 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 }, 91 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 }, 92 { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 }, 93 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 }, 94 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 }, 95 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 }, 96 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 }, 97 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 }, 98 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 }, 99 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 }, 100 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 }, 101 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 }, 102 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 }, 103 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 }, 104 /* required last entry */ 105 {0, } 106 }; 107 108 MODULE_DEVICE_TABLE(pci, igb_pci_tbl); 109 110 static int igb_setup_all_tx_resources(struct igb_adapter *); 111 static int igb_setup_all_rx_resources(struct igb_adapter *); 112 static void igb_free_all_tx_resources(struct igb_adapter *); 113 static void igb_free_all_rx_resources(struct igb_adapter *); 114 static void igb_setup_mrqc(struct igb_adapter *); 115 static int igb_probe(struct pci_dev *, const struct pci_device_id *); 116 static void igb_remove(struct pci_dev *pdev); 117 static int igb_sw_init(struct igb_adapter *); 118 int igb_open(struct net_device *); 119 int igb_close(struct net_device *); 120 static void igb_configure(struct igb_adapter *); 121 static void igb_configure_tx(struct igb_adapter *); 122 static void igb_configure_rx(struct igb_adapter *); 123 static void igb_clean_all_tx_rings(struct igb_adapter *); 124 static void igb_clean_all_rx_rings(struct igb_adapter *); 125 static void igb_clean_tx_ring(struct igb_ring *); 126 static void igb_clean_rx_ring(struct igb_ring *); 127 static void igb_set_rx_mode(struct net_device *); 128 static void igb_update_phy_info(struct timer_list *); 129 static void igb_watchdog(struct timer_list *); 130 static void igb_watchdog_task(struct work_struct *); 131 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *); 132 static void igb_get_stats64(struct net_device *dev, 133 struct rtnl_link_stats64 *stats); 134 static int igb_change_mtu(struct net_device *, int); 135 static int igb_set_mac(struct net_device *, void *); 136 static void igb_set_uta(struct igb_adapter *adapter, bool set); 137 static irqreturn_t igb_intr(int irq, void *); 138 static irqreturn_t igb_intr_msi(int irq, void *); 139 static irqreturn_t igb_msix_other(int irq, void *); 140 static irqreturn_t igb_msix_ring(int irq, void *); 141 #ifdef CONFIG_IGB_DCA 142 static void igb_update_dca(struct igb_q_vector *); 143 static void igb_setup_dca(struct igb_adapter *); 144 #endif /* CONFIG_IGB_DCA */ 145 static int igb_poll(struct napi_struct *, int); 146 static bool igb_clean_tx_irq(struct igb_q_vector *, int); 147 static int igb_clean_rx_irq(struct igb_q_vector *, int); 148 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd); 149 static void igb_tx_timeout(struct net_device *); 150 static void igb_reset_task(struct work_struct *); 151 static void igb_vlan_mode(struct net_device *netdev, 152 netdev_features_t features); 153 static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16); 154 static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16); 155 static void igb_restore_vlan(struct igb_adapter *); 156 static void igb_rar_set_index(struct igb_adapter *, u32); 157 static void igb_ping_all_vfs(struct igb_adapter *); 158 static void igb_msg_task(struct igb_adapter *); 159 static void igb_vmm_control(struct igb_adapter *); 160 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *); 161 static void igb_flush_mac_table(struct igb_adapter *); 162 static int igb_available_rars(struct igb_adapter *, u8); 163 static void igb_set_default_mac_filter(struct igb_adapter *); 164 static int igb_uc_sync(struct net_device *, const unsigned char *); 165 static int igb_uc_unsync(struct net_device *, const unsigned char *); 166 static void igb_restore_vf_multicasts(struct igb_adapter *adapter); 167 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac); 168 static int igb_ndo_set_vf_vlan(struct net_device *netdev, 169 int vf, u16 vlan, u8 qos, __be16 vlan_proto); 170 static int igb_ndo_set_vf_bw(struct net_device *, int, int, int); 171 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf, 172 bool setting); 173 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf, 174 bool setting); 175 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf, 176 struct ifla_vf_info *ivi); 177 static void igb_check_vf_rate_limit(struct igb_adapter *); 178 static void igb_nfc_filter_exit(struct igb_adapter *adapter); 179 static void igb_nfc_filter_restore(struct igb_adapter *adapter); 180 181 #ifdef CONFIG_PCI_IOV 182 static int igb_vf_configure(struct igb_adapter *adapter, int vf); 183 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs); 184 static int igb_disable_sriov(struct pci_dev *dev); 185 static int igb_pci_disable_sriov(struct pci_dev *dev); 186 #endif 187 188 static int igb_suspend(struct device *); 189 static int igb_resume(struct device *); 190 static int igb_runtime_suspend(struct device *dev); 191 static int igb_runtime_resume(struct device *dev); 192 static int igb_runtime_idle(struct device *dev); 193 static const struct dev_pm_ops igb_pm_ops = { 194 SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume) 195 SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume, 196 igb_runtime_idle) 197 }; 198 static void igb_shutdown(struct pci_dev *); 199 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs); 200 #ifdef CONFIG_IGB_DCA 201 static int igb_notify_dca(struct notifier_block *, unsigned long, void *); 202 static struct notifier_block dca_notifier = { 203 .notifier_call = igb_notify_dca, 204 .next = NULL, 205 .priority = 0 206 }; 207 #endif 208 #ifdef CONFIG_PCI_IOV 209 static unsigned int max_vfs; 210 module_param(max_vfs, uint, 0); 211 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function"); 212 #endif /* CONFIG_PCI_IOV */ 213 214 static pci_ers_result_t igb_io_error_detected(struct pci_dev *, 215 pci_channel_state_t); 216 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *); 217 static void igb_io_resume(struct pci_dev *); 218 219 static const struct pci_error_handlers igb_err_handler = { 220 .error_detected = igb_io_error_detected, 221 .slot_reset = igb_io_slot_reset, 222 .resume = igb_io_resume, 223 }; 224 225 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba); 226 227 static struct pci_driver igb_driver = { 228 .name = igb_driver_name, 229 .id_table = igb_pci_tbl, 230 .probe = igb_probe, 231 .remove = igb_remove, 232 #ifdef CONFIG_PM 233 .driver.pm = &igb_pm_ops, 234 #endif 235 .shutdown = igb_shutdown, 236 .sriov_configure = igb_pci_sriov_configure, 237 .err_handler = &igb_err_handler 238 }; 239 240 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>"); 241 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver"); 242 MODULE_LICENSE("GPL v2"); 243 MODULE_VERSION(DRV_VERSION); 244 245 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK) 246 static int debug = -1; 247 module_param(debug, int, 0); 248 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)"); 249 250 struct igb_reg_info { 251 u32 ofs; 252 char *name; 253 }; 254 255 static const struct igb_reg_info igb_reg_info_tbl[] = { 256 257 /* General Registers */ 258 {E1000_CTRL, "CTRL"}, 259 {E1000_STATUS, "STATUS"}, 260 {E1000_CTRL_EXT, "CTRL_EXT"}, 261 262 /* Interrupt Registers */ 263 {E1000_ICR, "ICR"}, 264 265 /* RX Registers */ 266 {E1000_RCTL, "RCTL"}, 267 {E1000_RDLEN(0), "RDLEN"}, 268 {E1000_RDH(0), "RDH"}, 269 {E1000_RDT(0), "RDT"}, 270 {E1000_RXDCTL(0), "RXDCTL"}, 271 {E1000_RDBAL(0), "RDBAL"}, 272 {E1000_RDBAH(0), "RDBAH"}, 273 274 /* TX Registers */ 275 {E1000_TCTL, "TCTL"}, 276 {E1000_TDBAL(0), "TDBAL"}, 277 {E1000_TDBAH(0), "TDBAH"}, 278 {E1000_TDLEN(0), "TDLEN"}, 279 {E1000_TDH(0), "TDH"}, 280 {E1000_TDT(0), "TDT"}, 281 {E1000_TXDCTL(0), "TXDCTL"}, 282 {E1000_TDFH, "TDFH"}, 283 {E1000_TDFT, "TDFT"}, 284 {E1000_TDFHS, "TDFHS"}, 285 {E1000_TDFPC, "TDFPC"}, 286 287 /* List Terminator */ 288 {} 289 }; 290 291 /* igb_regdump - register printout routine */ 292 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo) 293 { 294 int n = 0; 295 char rname[16]; 296 u32 regs[8]; 297 298 switch (reginfo->ofs) { 299 case E1000_RDLEN(0): 300 for (n = 0; n < 4; n++) 301 regs[n] = rd32(E1000_RDLEN(n)); 302 break; 303 case E1000_RDH(0): 304 for (n = 0; n < 4; n++) 305 regs[n] = rd32(E1000_RDH(n)); 306 break; 307 case E1000_RDT(0): 308 for (n = 0; n < 4; n++) 309 regs[n] = rd32(E1000_RDT(n)); 310 break; 311 case E1000_RXDCTL(0): 312 for (n = 0; n < 4; n++) 313 regs[n] = rd32(E1000_RXDCTL(n)); 314 break; 315 case E1000_RDBAL(0): 316 for (n = 0; n < 4; n++) 317 regs[n] = rd32(E1000_RDBAL(n)); 318 break; 319 case E1000_RDBAH(0): 320 for (n = 0; n < 4; n++) 321 regs[n] = rd32(E1000_RDBAH(n)); 322 break; 323 case E1000_TDBAL(0): 324 for (n = 0; n < 4; n++) 325 regs[n] = rd32(E1000_RDBAL(n)); 326 break; 327 case E1000_TDBAH(0): 328 for (n = 0; n < 4; n++) 329 regs[n] = rd32(E1000_TDBAH(n)); 330 break; 331 case E1000_TDLEN(0): 332 for (n = 0; n < 4; n++) 333 regs[n] = rd32(E1000_TDLEN(n)); 334 break; 335 case E1000_TDH(0): 336 for (n = 0; n < 4; n++) 337 regs[n] = rd32(E1000_TDH(n)); 338 break; 339 case E1000_TDT(0): 340 for (n = 0; n < 4; n++) 341 regs[n] = rd32(E1000_TDT(n)); 342 break; 343 case E1000_TXDCTL(0): 344 for (n = 0; n < 4; n++) 345 regs[n] = rd32(E1000_TXDCTL(n)); 346 break; 347 default: 348 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs)); 349 return; 350 } 351 352 snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]"); 353 pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1], 354 regs[2], regs[3]); 355 } 356 357 /* igb_dump - Print registers, Tx-rings and Rx-rings */ 358 static void igb_dump(struct igb_adapter *adapter) 359 { 360 struct net_device *netdev = adapter->netdev; 361 struct e1000_hw *hw = &adapter->hw; 362 struct igb_reg_info *reginfo; 363 struct igb_ring *tx_ring; 364 union e1000_adv_tx_desc *tx_desc; 365 struct my_u0 { u64 a; u64 b; } *u0; 366 struct igb_ring *rx_ring; 367 union e1000_adv_rx_desc *rx_desc; 368 u32 staterr; 369 u16 i, n; 370 371 if (!netif_msg_hw(adapter)) 372 return; 373 374 /* Print netdevice Info */ 375 if (netdev) { 376 dev_info(&adapter->pdev->dev, "Net device Info\n"); 377 pr_info("Device Name state trans_start\n"); 378 pr_info("%-15s %016lX %016lX\n", netdev->name, 379 netdev->state, dev_trans_start(netdev)); 380 } 381 382 /* Print Registers */ 383 dev_info(&adapter->pdev->dev, "Register Dump\n"); 384 pr_info(" Register Name Value\n"); 385 for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl; 386 reginfo->name; reginfo++) { 387 igb_regdump(hw, reginfo); 388 } 389 390 /* Print TX Ring Summary */ 391 if (!netdev || !netif_running(netdev)) 392 goto exit; 393 394 dev_info(&adapter->pdev->dev, "TX Rings Summary\n"); 395 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n"); 396 for (n = 0; n < adapter->num_tx_queues; n++) { 397 struct igb_tx_buffer *buffer_info; 398 tx_ring = adapter->tx_ring[n]; 399 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean]; 400 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n", 401 n, tx_ring->next_to_use, tx_ring->next_to_clean, 402 (u64)dma_unmap_addr(buffer_info, dma), 403 dma_unmap_len(buffer_info, len), 404 buffer_info->next_to_watch, 405 (u64)buffer_info->time_stamp); 406 } 407 408 /* Print TX Rings */ 409 if (!netif_msg_tx_done(adapter)) 410 goto rx_ring_summary; 411 412 dev_info(&adapter->pdev->dev, "TX Rings Dump\n"); 413 414 /* Transmit Descriptor Formats 415 * 416 * Advanced Transmit Descriptor 417 * +--------------------------------------------------------------+ 418 * 0 | Buffer Address [63:0] | 419 * +--------------------------------------------------------------+ 420 * 8 | PAYLEN | PORTS |CC|IDX | STA | DCMD |DTYP|MAC|RSV| DTALEN | 421 * +--------------------------------------------------------------+ 422 * 63 46 45 40 39 38 36 35 32 31 24 15 0 423 */ 424 425 for (n = 0; n < adapter->num_tx_queues; n++) { 426 tx_ring = adapter->tx_ring[n]; 427 pr_info("------------------------------------\n"); 428 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index); 429 pr_info("------------------------------------\n"); 430 pr_info("T [desc] [address 63:0 ] [PlPOCIStDDM Ln] [bi->dma ] leng ntw timestamp bi->skb\n"); 431 432 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) { 433 const char *next_desc; 434 struct igb_tx_buffer *buffer_info; 435 tx_desc = IGB_TX_DESC(tx_ring, i); 436 buffer_info = &tx_ring->tx_buffer_info[i]; 437 u0 = (struct my_u0 *)tx_desc; 438 if (i == tx_ring->next_to_use && 439 i == tx_ring->next_to_clean) 440 next_desc = " NTC/U"; 441 else if (i == tx_ring->next_to_use) 442 next_desc = " NTU"; 443 else if (i == tx_ring->next_to_clean) 444 next_desc = " NTC"; 445 else 446 next_desc = ""; 447 448 pr_info("T [0x%03X] %016llX %016llX %016llX %04X %p %016llX %p%s\n", 449 i, le64_to_cpu(u0->a), 450 le64_to_cpu(u0->b), 451 (u64)dma_unmap_addr(buffer_info, dma), 452 dma_unmap_len(buffer_info, len), 453 buffer_info->next_to_watch, 454 (u64)buffer_info->time_stamp, 455 buffer_info->skb, next_desc); 456 457 if (netif_msg_pktdata(adapter) && buffer_info->skb) 458 print_hex_dump(KERN_INFO, "", 459 DUMP_PREFIX_ADDRESS, 460 16, 1, buffer_info->skb->data, 461 dma_unmap_len(buffer_info, len), 462 true); 463 } 464 } 465 466 /* Print RX Rings Summary */ 467 rx_ring_summary: 468 dev_info(&adapter->pdev->dev, "RX Rings Summary\n"); 469 pr_info("Queue [NTU] [NTC]\n"); 470 for (n = 0; n < adapter->num_rx_queues; n++) { 471 rx_ring = adapter->rx_ring[n]; 472 pr_info(" %5d %5X %5X\n", 473 n, rx_ring->next_to_use, rx_ring->next_to_clean); 474 } 475 476 /* Print RX Rings */ 477 if (!netif_msg_rx_status(adapter)) 478 goto exit; 479 480 dev_info(&adapter->pdev->dev, "RX Rings Dump\n"); 481 482 /* Advanced Receive Descriptor (Read) Format 483 * 63 1 0 484 * +-----------------------------------------------------+ 485 * 0 | Packet Buffer Address [63:1] |A0/NSE| 486 * +----------------------------------------------+------+ 487 * 8 | Header Buffer Address [63:1] | DD | 488 * +-----------------------------------------------------+ 489 * 490 * 491 * Advanced Receive Descriptor (Write-Back) Format 492 * 493 * 63 48 47 32 31 30 21 20 17 16 4 3 0 494 * +------------------------------------------------------+ 495 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS | 496 * | Checksum Ident | | | | Type | Type | 497 * +------------------------------------------------------+ 498 * 8 | VLAN Tag | Length | Extended Error | Extended Status | 499 * +------------------------------------------------------+ 500 * 63 48 47 32 31 20 19 0 501 */ 502 503 for (n = 0; n < adapter->num_rx_queues; n++) { 504 rx_ring = adapter->rx_ring[n]; 505 pr_info("------------------------------------\n"); 506 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index); 507 pr_info("------------------------------------\n"); 508 pr_info("R [desc] [ PktBuf A0] [ HeadBuf DD] [bi->dma ] [bi->skb] <-- Adv Rx Read format\n"); 509 pr_info("RWB[desc] [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n"); 510 511 for (i = 0; i < rx_ring->count; i++) { 512 const char *next_desc; 513 struct igb_rx_buffer *buffer_info; 514 buffer_info = &rx_ring->rx_buffer_info[i]; 515 rx_desc = IGB_RX_DESC(rx_ring, i); 516 u0 = (struct my_u0 *)rx_desc; 517 staterr = le32_to_cpu(rx_desc->wb.upper.status_error); 518 519 if (i == rx_ring->next_to_use) 520 next_desc = " NTU"; 521 else if (i == rx_ring->next_to_clean) 522 next_desc = " NTC"; 523 else 524 next_desc = ""; 525 526 if (staterr & E1000_RXD_STAT_DD) { 527 /* Descriptor Done */ 528 pr_info("%s[0x%03X] %016llX %016llX ---------------- %s\n", 529 "RWB", i, 530 le64_to_cpu(u0->a), 531 le64_to_cpu(u0->b), 532 next_desc); 533 } else { 534 pr_info("%s[0x%03X] %016llX %016llX %016llX %s\n", 535 "R ", i, 536 le64_to_cpu(u0->a), 537 le64_to_cpu(u0->b), 538 (u64)buffer_info->dma, 539 next_desc); 540 541 if (netif_msg_pktdata(adapter) && 542 buffer_info->dma && buffer_info->page) { 543 print_hex_dump(KERN_INFO, "", 544 DUMP_PREFIX_ADDRESS, 545 16, 1, 546 page_address(buffer_info->page) + 547 buffer_info->page_offset, 548 igb_rx_bufsz(rx_ring), true); 549 } 550 } 551 } 552 } 553 554 exit: 555 return; 556 } 557 558 /** 559 * igb_get_i2c_data - Reads the I2C SDA data bit 560 * @hw: pointer to hardware structure 561 * @i2cctl: Current value of I2CCTL register 562 * 563 * Returns the I2C data bit value 564 **/ 565 static int igb_get_i2c_data(void *data) 566 { 567 struct igb_adapter *adapter = (struct igb_adapter *)data; 568 struct e1000_hw *hw = &adapter->hw; 569 s32 i2cctl = rd32(E1000_I2CPARAMS); 570 571 return !!(i2cctl & E1000_I2C_DATA_IN); 572 } 573 574 /** 575 * igb_set_i2c_data - Sets the I2C data bit 576 * @data: pointer to hardware structure 577 * @state: I2C data value (0 or 1) to set 578 * 579 * Sets the I2C data bit 580 **/ 581 static void igb_set_i2c_data(void *data, int state) 582 { 583 struct igb_adapter *adapter = (struct igb_adapter *)data; 584 struct e1000_hw *hw = &adapter->hw; 585 s32 i2cctl = rd32(E1000_I2CPARAMS); 586 587 if (state) 588 i2cctl |= E1000_I2C_DATA_OUT; 589 else 590 i2cctl &= ~E1000_I2C_DATA_OUT; 591 592 i2cctl &= ~E1000_I2C_DATA_OE_N; 593 i2cctl |= E1000_I2C_CLK_OE_N; 594 wr32(E1000_I2CPARAMS, i2cctl); 595 wrfl(); 596 597 } 598 599 /** 600 * igb_set_i2c_clk - Sets the I2C SCL clock 601 * @data: pointer to hardware structure 602 * @state: state to set clock 603 * 604 * Sets the I2C clock line to state 605 **/ 606 static void igb_set_i2c_clk(void *data, int state) 607 { 608 struct igb_adapter *adapter = (struct igb_adapter *)data; 609 struct e1000_hw *hw = &adapter->hw; 610 s32 i2cctl = rd32(E1000_I2CPARAMS); 611 612 if (state) { 613 i2cctl |= E1000_I2C_CLK_OUT; 614 i2cctl &= ~E1000_I2C_CLK_OE_N; 615 } else { 616 i2cctl &= ~E1000_I2C_CLK_OUT; 617 i2cctl &= ~E1000_I2C_CLK_OE_N; 618 } 619 wr32(E1000_I2CPARAMS, i2cctl); 620 wrfl(); 621 } 622 623 /** 624 * igb_get_i2c_clk - Gets the I2C SCL clock state 625 * @data: pointer to hardware structure 626 * 627 * Gets the I2C clock state 628 **/ 629 static int igb_get_i2c_clk(void *data) 630 { 631 struct igb_adapter *adapter = (struct igb_adapter *)data; 632 struct e1000_hw *hw = &adapter->hw; 633 s32 i2cctl = rd32(E1000_I2CPARAMS); 634 635 return !!(i2cctl & E1000_I2C_CLK_IN); 636 } 637 638 static const struct i2c_algo_bit_data igb_i2c_algo = { 639 .setsda = igb_set_i2c_data, 640 .setscl = igb_set_i2c_clk, 641 .getsda = igb_get_i2c_data, 642 .getscl = igb_get_i2c_clk, 643 .udelay = 5, 644 .timeout = 20, 645 }; 646 647 /** 648 * igb_get_hw_dev - return device 649 * @hw: pointer to hardware structure 650 * 651 * used by hardware layer to print debugging information 652 **/ 653 struct net_device *igb_get_hw_dev(struct e1000_hw *hw) 654 { 655 struct igb_adapter *adapter = hw->back; 656 return adapter->netdev; 657 } 658 659 /** 660 * igb_init_module - Driver Registration Routine 661 * 662 * igb_init_module is the first routine called when the driver is 663 * loaded. All it does is register with the PCI subsystem. 664 **/ 665 static int __init igb_init_module(void) 666 { 667 int ret; 668 669 pr_info("%s - version %s\n", 670 igb_driver_string, igb_driver_version); 671 pr_info("%s\n", igb_copyright); 672 673 #ifdef CONFIG_IGB_DCA 674 dca_register_notify(&dca_notifier); 675 #endif 676 ret = pci_register_driver(&igb_driver); 677 return ret; 678 } 679 680 module_init(igb_init_module); 681 682 /** 683 * igb_exit_module - Driver Exit Cleanup Routine 684 * 685 * igb_exit_module is called just before the driver is removed 686 * from memory. 687 **/ 688 static void __exit igb_exit_module(void) 689 { 690 #ifdef CONFIG_IGB_DCA 691 dca_unregister_notify(&dca_notifier); 692 #endif 693 pci_unregister_driver(&igb_driver); 694 } 695 696 module_exit(igb_exit_module); 697 698 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1)) 699 /** 700 * igb_cache_ring_register - Descriptor ring to register mapping 701 * @adapter: board private structure to initialize 702 * 703 * Once we know the feature-set enabled for the device, we'll cache 704 * the register offset the descriptor ring is assigned to. 705 **/ 706 static void igb_cache_ring_register(struct igb_adapter *adapter) 707 { 708 int i = 0, j = 0; 709 u32 rbase_offset = adapter->vfs_allocated_count; 710 711 switch (adapter->hw.mac.type) { 712 case e1000_82576: 713 /* The queues are allocated for virtualization such that VF 0 714 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc. 715 * In order to avoid collision we start at the first free queue 716 * and continue consuming queues in the same sequence 717 */ 718 if (adapter->vfs_allocated_count) { 719 for (; i < adapter->rss_queues; i++) 720 adapter->rx_ring[i]->reg_idx = rbase_offset + 721 Q_IDX_82576(i); 722 } 723 /* Fall through */ 724 case e1000_82575: 725 case e1000_82580: 726 case e1000_i350: 727 case e1000_i354: 728 case e1000_i210: 729 case e1000_i211: 730 /* Fall through */ 731 default: 732 for (; i < adapter->num_rx_queues; i++) 733 adapter->rx_ring[i]->reg_idx = rbase_offset + i; 734 for (; j < adapter->num_tx_queues; j++) 735 adapter->tx_ring[j]->reg_idx = rbase_offset + j; 736 break; 737 } 738 } 739 740 u32 igb_rd32(struct e1000_hw *hw, u32 reg) 741 { 742 struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw); 743 u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr); 744 u32 value = 0; 745 746 if (E1000_REMOVED(hw_addr)) 747 return ~value; 748 749 value = readl(&hw_addr[reg]); 750 751 /* reads should not return all F's */ 752 if (!(~value) && (!reg || !(~readl(hw_addr)))) { 753 struct net_device *netdev = igb->netdev; 754 hw->hw_addr = NULL; 755 netdev_err(netdev, "PCIe link lost\n"); 756 } 757 758 return value; 759 } 760 761 /** 762 * igb_write_ivar - configure ivar for given MSI-X vector 763 * @hw: pointer to the HW structure 764 * @msix_vector: vector number we are allocating to a given ring 765 * @index: row index of IVAR register to write within IVAR table 766 * @offset: column offset of in IVAR, should be multiple of 8 767 * 768 * This function is intended to handle the writing of the IVAR register 769 * for adapters 82576 and newer. The IVAR table consists of 2 columns, 770 * each containing an cause allocation for an Rx and Tx ring, and a 771 * variable number of rows depending on the number of queues supported. 772 **/ 773 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector, 774 int index, int offset) 775 { 776 u32 ivar = array_rd32(E1000_IVAR0, index); 777 778 /* clear any bits that are currently set */ 779 ivar &= ~((u32)0xFF << offset); 780 781 /* write vector and valid bit */ 782 ivar |= (msix_vector | E1000_IVAR_VALID) << offset; 783 784 array_wr32(E1000_IVAR0, index, ivar); 785 } 786 787 #define IGB_N0_QUEUE -1 788 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector) 789 { 790 struct igb_adapter *adapter = q_vector->adapter; 791 struct e1000_hw *hw = &adapter->hw; 792 int rx_queue = IGB_N0_QUEUE; 793 int tx_queue = IGB_N0_QUEUE; 794 u32 msixbm = 0; 795 796 if (q_vector->rx.ring) 797 rx_queue = q_vector->rx.ring->reg_idx; 798 if (q_vector->tx.ring) 799 tx_queue = q_vector->tx.ring->reg_idx; 800 801 switch (hw->mac.type) { 802 case e1000_82575: 803 /* The 82575 assigns vectors using a bitmask, which matches the 804 * bitmask for the EICR/EIMS/EIMC registers. To assign one 805 * or more queues to a vector, we write the appropriate bits 806 * into the MSIXBM register for that vector. 807 */ 808 if (rx_queue > IGB_N0_QUEUE) 809 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue; 810 if (tx_queue > IGB_N0_QUEUE) 811 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue; 812 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0) 813 msixbm |= E1000_EIMS_OTHER; 814 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm); 815 q_vector->eims_value = msixbm; 816 break; 817 case e1000_82576: 818 /* 82576 uses a table that essentially consists of 2 columns 819 * with 8 rows. The ordering is column-major so we use the 820 * lower 3 bits as the row index, and the 4th bit as the 821 * column offset. 822 */ 823 if (rx_queue > IGB_N0_QUEUE) 824 igb_write_ivar(hw, msix_vector, 825 rx_queue & 0x7, 826 (rx_queue & 0x8) << 1); 827 if (tx_queue > IGB_N0_QUEUE) 828 igb_write_ivar(hw, msix_vector, 829 tx_queue & 0x7, 830 ((tx_queue & 0x8) << 1) + 8); 831 q_vector->eims_value = BIT(msix_vector); 832 break; 833 case e1000_82580: 834 case e1000_i350: 835 case e1000_i354: 836 case e1000_i210: 837 case e1000_i211: 838 /* On 82580 and newer adapters the scheme is similar to 82576 839 * however instead of ordering column-major we have things 840 * ordered row-major. So we traverse the table by using 841 * bit 0 as the column offset, and the remaining bits as the 842 * row index. 843 */ 844 if (rx_queue > IGB_N0_QUEUE) 845 igb_write_ivar(hw, msix_vector, 846 rx_queue >> 1, 847 (rx_queue & 0x1) << 4); 848 if (tx_queue > IGB_N0_QUEUE) 849 igb_write_ivar(hw, msix_vector, 850 tx_queue >> 1, 851 ((tx_queue & 0x1) << 4) + 8); 852 q_vector->eims_value = BIT(msix_vector); 853 break; 854 default: 855 BUG(); 856 break; 857 } 858 859 /* add q_vector eims value to global eims_enable_mask */ 860 adapter->eims_enable_mask |= q_vector->eims_value; 861 862 /* configure q_vector to set itr on first interrupt */ 863 q_vector->set_itr = 1; 864 } 865 866 /** 867 * igb_configure_msix - Configure MSI-X hardware 868 * @adapter: board private structure to initialize 869 * 870 * igb_configure_msix sets up the hardware to properly 871 * generate MSI-X interrupts. 872 **/ 873 static void igb_configure_msix(struct igb_adapter *adapter) 874 { 875 u32 tmp; 876 int i, vector = 0; 877 struct e1000_hw *hw = &adapter->hw; 878 879 adapter->eims_enable_mask = 0; 880 881 /* set vector for other causes, i.e. link changes */ 882 switch (hw->mac.type) { 883 case e1000_82575: 884 tmp = rd32(E1000_CTRL_EXT); 885 /* enable MSI-X PBA support*/ 886 tmp |= E1000_CTRL_EXT_PBA_CLR; 887 888 /* Auto-Mask interrupts upon ICR read. */ 889 tmp |= E1000_CTRL_EXT_EIAME; 890 tmp |= E1000_CTRL_EXT_IRCA; 891 892 wr32(E1000_CTRL_EXT, tmp); 893 894 /* enable msix_other interrupt */ 895 array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER); 896 adapter->eims_other = E1000_EIMS_OTHER; 897 898 break; 899 900 case e1000_82576: 901 case e1000_82580: 902 case e1000_i350: 903 case e1000_i354: 904 case e1000_i210: 905 case e1000_i211: 906 /* Turn on MSI-X capability first, or our settings 907 * won't stick. And it will take days to debug. 908 */ 909 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE | 910 E1000_GPIE_PBA | E1000_GPIE_EIAME | 911 E1000_GPIE_NSICR); 912 913 /* enable msix_other interrupt */ 914 adapter->eims_other = BIT(vector); 915 tmp = (vector++ | E1000_IVAR_VALID) << 8; 916 917 wr32(E1000_IVAR_MISC, tmp); 918 break; 919 default: 920 /* do nothing, since nothing else supports MSI-X */ 921 break; 922 } /* switch (hw->mac.type) */ 923 924 adapter->eims_enable_mask |= adapter->eims_other; 925 926 for (i = 0; i < adapter->num_q_vectors; i++) 927 igb_assign_vector(adapter->q_vector[i], vector++); 928 929 wrfl(); 930 } 931 932 /** 933 * igb_request_msix - Initialize MSI-X interrupts 934 * @adapter: board private structure to initialize 935 * 936 * igb_request_msix allocates MSI-X vectors and requests interrupts from the 937 * kernel. 938 **/ 939 static int igb_request_msix(struct igb_adapter *adapter) 940 { 941 struct net_device *netdev = adapter->netdev; 942 int i, err = 0, vector = 0, free_vector = 0; 943 944 err = request_irq(adapter->msix_entries[vector].vector, 945 igb_msix_other, 0, netdev->name, adapter); 946 if (err) 947 goto err_out; 948 949 for (i = 0; i < adapter->num_q_vectors; i++) { 950 struct igb_q_vector *q_vector = adapter->q_vector[i]; 951 952 vector++; 953 954 q_vector->itr_register = adapter->io_addr + E1000_EITR(vector); 955 956 if (q_vector->rx.ring && q_vector->tx.ring) 957 sprintf(q_vector->name, "%s-TxRx-%u", netdev->name, 958 q_vector->rx.ring->queue_index); 959 else if (q_vector->tx.ring) 960 sprintf(q_vector->name, "%s-tx-%u", netdev->name, 961 q_vector->tx.ring->queue_index); 962 else if (q_vector->rx.ring) 963 sprintf(q_vector->name, "%s-rx-%u", netdev->name, 964 q_vector->rx.ring->queue_index); 965 else 966 sprintf(q_vector->name, "%s-unused", netdev->name); 967 968 err = request_irq(adapter->msix_entries[vector].vector, 969 igb_msix_ring, 0, q_vector->name, 970 q_vector); 971 if (err) 972 goto err_free; 973 } 974 975 igb_configure_msix(adapter); 976 return 0; 977 978 err_free: 979 /* free already assigned IRQs */ 980 free_irq(adapter->msix_entries[free_vector++].vector, adapter); 981 982 vector--; 983 for (i = 0; i < vector; i++) { 984 free_irq(adapter->msix_entries[free_vector++].vector, 985 adapter->q_vector[i]); 986 } 987 err_out: 988 return err; 989 } 990 991 /** 992 * igb_free_q_vector - Free memory allocated for specific interrupt vector 993 * @adapter: board private structure to initialize 994 * @v_idx: Index of vector to be freed 995 * 996 * This function frees the memory allocated to the q_vector. 997 **/ 998 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx) 999 { 1000 struct igb_q_vector *q_vector = adapter->q_vector[v_idx]; 1001 1002 adapter->q_vector[v_idx] = NULL; 1003 1004 /* igb_get_stats64() might access the rings on this vector, 1005 * we must wait a grace period before freeing it. 1006 */ 1007 if (q_vector) 1008 kfree_rcu(q_vector, rcu); 1009 } 1010 1011 /** 1012 * igb_reset_q_vector - Reset config for interrupt vector 1013 * @adapter: board private structure to initialize 1014 * @v_idx: Index of vector to be reset 1015 * 1016 * If NAPI is enabled it will delete any references to the 1017 * NAPI struct. This is preparation for igb_free_q_vector. 1018 **/ 1019 static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx) 1020 { 1021 struct igb_q_vector *q_vector = adapter->q_vector[v_idx]; 1022 1023 /* Coming from igb_set_interrupt_capability, the vectors are not yet 1024 * allocated. So, q_vector is NULL so we should stop here. 1025 */ 1026 if (!q_vector) 1027 return; 1028 1029 if (q_vector->tx.ring) 1030 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL; 1031 1032 if (q_vector->rx.ring) 1033 adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL; 1034 1035 netif_napi_del(&q_vector->napi); 1036 1037 } 1038 1039 static void igb_reset_interrupt_capability(struct igb_adapter *adapter) 1040 { 1041 int v_idx = adapter->num_q_vectors; 1042 1043 if (adapter->flags & IGB_FLAG_HAS_MSIX) 1044 pci_disable_msix(adapter->pdev); 1045 else if (adapter->flags & IGB_FLAG_HAS_MSI) 1046 pci_disable_msi(adapter->pdev); 1047 1048 while (v_idx--) 1049 igb_reset_q_vector(adapter, v_idx); 1050 } 1051 1052 /** 1053 * igb_free_q_vectors - Free memory allocated for interrupt vectors 1054 * @adapter: board private structure to initialize 1055 * 1056 * This function frees the memory allocated to the q_vectors. In addition if 1057 * NAPI is enabled it will delete any references to the NAPI struct prior 1058 * to freeing the q_vector. 1059 **/ 1060 static void igb_free_q_vectors(struct igb_adapter *adapter) 1061 { 1062 int v_idx = adapter->num_q_vectors; 1063 1064 adapter->num_tx_queues = 0; 1065 adapter->num_rx_queues = 0; 1066 adapter->num_q_vectors = 0; 1067 1068 while (v_idx--) { 1069 igb_reset_q_vector(adapter, v_idx); 1070 igb_free_q_vector(adapter, v_idx); 1071 } 1072 } 1073 1074 /** 1075 * igb_clear_interrupt_scheme - reset the device to a state of no interrupts 1076 * @adapter: board private structure to initialize 1077 * 1078 * This function resets the device so that it has 0 Rx queues, Tx queues, and 1079 * MSI-X interrupts allocated. 1080 */ 1081 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter) 1082 { 1083 igb_free_q_vectors(adapter); 1084 igb_reset_interrupt_capability(adapter); 1085 } 1086 1087 /** 1088 * igb_set_interrupt_capability - set MSI or MSI-X if supported 1089 * @adapter: board private structure to initialize 1090 * @msix: boolean value of MSIX capability 1091 * 1092 * Attempt to configure interrupts using the best available 1093 * capabilities of the hardware and kernel. 1094 **/ 1095 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix) 1096 { 1097 int err; 1098 int numvecs, i; 1099 1100 if (!msix) 1101 goto msi_only; 1102 adapter->flags |= IGB_FLAG_HAS_MSIX; 1103 1104 /* Number of supported queues. */ 1105 adapter->num_rx_queues = adapter->rss_queues; 1106 if (adapter->vfs_allocated_count) 1107 adapter->num_tx_queues = 1; 1108 else 1109 adapter->num_tx_queues = adapter->rss_queues; 1110 1111 /* start with one vector for every Rx queue */ 1112 numvecs = adapter->num_rx_queues; 1113 1114 /* if Tx handler is separate add 1 for every Tx queue */ 1115 if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS)) 1116 numvecs += adapter->num_tx_queues; 1117 1118 /* store the number of vectors reserved for queues */ 1119 adapter->num_q_vectors = numvecs; 1120 1121 /* add 1 vector for link status interrupts */ 1122 numvecs++; 1123 for (i = 0; i < numvecs; i++) 1124 adapter->msix_entries[i].entry = i; 1125 1126 err = pci_enable_msix_range(adapter->pdev, 1127 adapter->msix_entries, 1128 numvecs, 1129 numvecs); 1130 if (err > 0) 1131 return; 1132 1133 igb_reset_interrupt_capability(adapter); 1134 1135 /* If we can't do MSI-X, try MSI */ 1136 msi_only: 1137 adapter->flags &= ~IGB_FLAG_HAS_MSIX; 1138 #ifdef CONFIG_PCI_IOV 1139 /* disable SR-IOV for non MSI-X configurations */ 1140 if (adapter->vf_data) { 1141 struct e1000_hw *hw = &adapter->hw; 1142 /* disable iov and allow time for transactions to clear */ 1143 pci_disable_sriov(adapter->pdev); 1144 msleep(500); 1145 1146 kfree(adapter->vf_mac_list); 1147 adapter->vf_mac_list = NULL; 1148 kfree(adapter->vf_data); 1149 adapter->vf_data = NULL; 1150 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ); 1151 wrfl(); 1152 msleep(100); 1153 dev_info(&adapter->pdev->dev, "IOV Disabled\n"); 1154 } 1155 #endif 1156 adapter->vfs_allocated_count = 0; 1157 adapter->rss_queues = 1; 1158 adapter->flags |= IGB_FLAG_QUEUE_PAIRS; 1159 adapter->num_rx_queues = 1; 1160 adapter->num_tx_queues = 1; 1161 adapter->num_q_vectors = 1; 1162 if (!pci_enable_msi(adapter->pdev)) 1163 adapter->flags |= IGB_FLAG_HAS_MSI; 1164 } 1165 1166 static void igb_add_ring(struct igb_ring *ring, 1167 struct igb_ring_container *head) 1168 { 1169 head->ring = ring; 1170 head->count++; 1171 } 1172 1173 /** 1174 * igb_alloc_q_vector - Allocate memory for a single interrupt vector 1175 * @adapter: board private structure to initialize 1176 * @v_count: q_vectors allocated on adapter, used for ring interleaving 1177 * @v_idx: index of vector in adapter struct 1178 * @txr_count: total number of Tx rings to allocate 1179 * @txr_idx: index of first Tx ring to allocate 1180 * @rxr_count: total number of Rx rings to allocate 1181 * @rxr_idx: index of first Rx ring to allocate 1182 * 1183 * We allocate one q_vector. If allocation fails we return -ENOMEM. 1184 **/ 1185 static int igb_alloc_q_vector(struct igb_adapter *adapter, 1186 int v_count, int v_idx, 1187 int txr_count, int txr_idx, 1188 int rxr_count, int rxr_idx) 1189 { 1190 struct igb_q_vector *q_vector; 1191 struct igb_ring *ring; 1192 int ring_count; 1193 size_t size; 1194 1195 /* igb only supports 1 Tx and/or 1 Rx queue per vector */ 1196 if (txr_count > 1 || rxr_count > 1) 1197 return -ENOMEM; 1198 1199 ring_count = txr_count + rxr_count; 1200 size = struct_size(q_vector, ring, ring_count); 1201 1202 /* allocate q_vector and rings */ 1203 q_vector = adapter->q_vector[v_idx]; 1204 if (!q_vector) { 1205 q_vector = kzalloc(size, GFP_KERNEL); 1206 } else if (size > ksize(q_vector)) { 1207 kfree_rcu(q_vector, rcu); 1208 q_vector = kzalloc(size, GFP_KERNEL); 1209 } else { 1210 memset(q_vector, 0, size); 1211 } 1212 if (!q_vector) 1213 return -ENOMEM; 1214 1215 /* initialize NAPI */ 1216 netif_napi_add(adapter->netdev, &q_vector->napi, 1217 igb_poll, 64); 1218 1219 /* tie q_vector and adapter together */ 1220 adapter->q_vector[v_idx] = q_vector; 1221 q_vector->adapter = adapter; 1222 1223 /* initialize work limits */ 1224 q_vector->tx.work_limit = adapter->tx_work_limit; 1225 1226 /* initialize ITR configuration */ 1227 q_vector->itr_register = adapter->io_addr + E1000_EITR(0); 1228 q_vector->itr_val = IGB_START_ITR; 1229 1230 /* initialize pointer to rings */ 1231 ring = q_vector->ring; 1232 1233 /* intialize ITR */ 1234 if (rxr_count) { 1235 /* rx or rx/tx vector */ 1236 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3) 1237 q_vector->itr_val = adapter->rx_itr_setting; 1238 } else { 1239 /* tx only vector */ 1240 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3) 1241 q_vector->itr_val = adapter->tx_itr_setting; 1242 } 1243 1244 if (txr_count) { 1245 /* assign generic ring traits */ 1246 ring->dev = &adapter->pdev->dev; 1247 ring->netdev = adapter->netdev; 1248 1249 /* configure backlink on ring */ 1250 ring->q_vector = q_vector; 1251 1252 /* update q_vector Tx values */ 1253 igb_add_ring(ring, &q_vector->tx); 1254 1255 /* For 82575, context index must be unique per ring. */ 1256 if (adapter->hw.mac.type == e1000_82575) 1257 set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags); 1258 1259 /* apply Tx specific ring traits */ 1260 ring->count = adapter->tx_ring_count; 1261 ring->queue_index = txr_idx; 1262 1263 ring->cbs_enable = false; 1264 ring->idleslope = 0; 1265 ring->sendslope = 0; 1266 ring->hicredit = 0; 1267 ring->locredit = 0; 1268 1269 u64_stats_init(&ring->tx_syncp); 1270 u64_stats_init(&ring->tx_syncp2); 1271 1272 /* assign ring to adapter */ 1273 adapter->tx_ring[txr_idx] = ring; 1274 1275 /* push pointer to next ring */ 1276 ring++; 1277 } 1278 1279 if (rxr_count) { 1280 /* assign generic ring traits */ 1281 ring->dev = &adapter->pdev->dev; 1282 ring->netdev = adapter->netdev; 1283 1284 /* configure backlink on ring */ 1285 ring->q_vector = q_vector; 1286 1287 /* update q_vector Rx values */ 1288 igb_add_ring(ring, &q_vector->rx); 1289 1290 /* set flag indicating ring supports SCTP checksum offload */ 1291 if (adapter->hw.mac.type >= e1000_82576) 1292 set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags); 1293 1294 /* On i350, i354, i210, and i211, loopback VLAN packets 1295 * have the tag byte-swapped. 1296 */ 1297 if (adapter->hw.mac.type >= e1000_i350) 1298 set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags); 1299 1300 /* apply Rx specific ring traits */ 1301 ring->count = adapter->rx_ring_count; 1302 ring->queue_index = rxr_idx; 1303 1304 u64_stats_init(&ring->rx_syncp); 1305 1306 /* assign ring to adapter */ 1307 adapter->rx_ring[rxr_idx] = ring; 1308 } 1309 1310 return 0; 1311 } 1312 1313 1314 /** 1315 * igb_alloc_q_vectors - Allocate memory for interrupt vectors 1316 * @adapter: board private structure to initialize 1317 * 1318 * We allocate one q_vector per queue interrupt. If allocation fails we 1319 * return -ENOMEM. 1320 **/ 1321 static int igb_alloc_q_vectors(struct igb_adapter *adapter) 1322 { 1323 int q_vectors = adapter->num_q_vectors; 1324 int rxr_remaining = adapter->num_rx_queues; 1325 int txr_remaining = adapter->num_tx_queues; 1326 int rxr_idx = 0, txr_idx = 0, v_idx = 0; 1327 int err; 1328 1329 if (q_vectors >= (rxr_remaining + txr_remaining)) { 1330 for (; rxr_remaining; v_idx++) { 1331 err = igb_alloc_q_vector(adapter, q_vectors, v_idx, 1332 0, 0, 1, rxr_idx); 1333 1334 if (err) 1335 goto err_out; 1336 1337 /* update counts and index */ 1338 rxr_remaining--; 1339 rxr_idx++; 1340 } 1341 } 1342 1343 for (; v_idx < q_vectors; v_idx++) { 1344 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx); 1345 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx); 1346 1347 err = igb_alloc_q_vector(adapter, q_vectors, v_idx, 1348 tqpv, txr_idx, rqpv, rxr_idx); 1349 1350 if (err) 1351 goto err_out; 1352 1353 /* update counts and index */ 1354 rxr_remaining -= rqpv; 1355 txr_remaining -= tqpv; 1356 rxr_idx++; 1357 txr_idx++; 1358 } 1359 1360 return 0; 1361 1362 err_out: 1363 adapter->num_tx_queues = 0; 1364 adapter->num_rx_queues = 0; 1365 adapter->num_q_vectors = 0; 1366 1367 while (v_idx--) 1368 igb_free_q_vector(adapter, v_idx); 1369 1370 return -ENOMEM; 1371 } 1372 1373 /** 1374 * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors 1375 * @adapter: board private structure to initialize 1376 * @msix: boolean value of MSIX capability 1377 * 1378 * This function initializes the interrupts and allocates all of the queues. 1379 **/ 1380 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix) 1381 { 1382 struct pci_dev *pdev = adapter->pdev; 1383 int err; 1384 1385 igb_set_interrupt_capability(adapter, msix); 1386 1387 err = igb_alloc_q_vectors(adapter); 1388 if (err) { 1389 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n"); 1390 goto err_alloc_q_vectors; 1391 } 1392 1393 igb_cache_ring_register(adapter); 1394 1395 return 0; 1396 1397 err_alloc_q_vectors: 1398 igb_reset_interrupt_capability(adapter); 1399 return err; 1400 } 1401 1402 /** 1403 * igb_request_irq - initialize interrupts 1404 * @adapter: board private structure to initialize 1405 * 1406 * Attempts to configure interrupts using the best available 1407 * capabilities of the hardware and kernel. 1408 **/ 1409 static int igb_request_irq(struct igb_adapter *adapter) 1410 { 1411 struct net_device *netdev = adapter->netdev; 1412 struct pci_dev *pdev = adapter->pdev; 1413 int err = 0; 1414 1415 if (adapter->flags & IGB_FLAG_HAS_MSIX) { 1416 err = igb_request_msix(adapter); 1417 if (!err) 1418 goto request_done; 1419 /* fall back to MSI */ 1420 igb_free_all_tx_resources(adapter); 1421 igb_free_all_rx_resources(adapter); 1422 1423 igb_clear_interrupt_scheme(adapter); 1424 err = igb_init_interrupt_scheme(adapter, false); 1425 if (err) 1426 goto request_done; 1427 1428 igb_setup_all_tx_resources(adapter); 1429 igb_setup_all_rx_resources(adapter); 1430 igb_configure(adapter); 1431 } 1432 1433 igb_assign_vector(adapter->q_vector[0], 0); 1434 1435 if (adapter->flags & IGB_FLAG_HAS_MSI) { 1436 err = request_irq(pdev->irq, igb_intr_msi, 0, 1437 netdev->name, adapter); 1438 if (!err) 1439 goto request_done; 1440 1441 /* fall back to legacy interrupts */ 1442 igb_reset_interrupt_capability(adapter); 1443 adapter->flags &= ~IGB_FLAG_HAS_MSI; 1444 } 1445 1446 err = request_irq(pdev->irq, igb_intr, IRQF_SHARED, 1447 netdev->name, adapter); 1448 1449 if (err) 1450 dev_err(&pdev->dev, "Error %d getting interrupt\n", 1451 err); 1452 1453 request_done: 1454 return err; 1455 } 1456 1457 static void igb_free_irq(struct igb_adapter *adapter) 1458 { 1459 if (adapter->flags & IGB_FLAG_HAS_MSIX) { 1460 int vector = 0, i; 1461 1462 free_irq(adapter->msix_entries[vector++].vector, adapter); 1463 1464 for (i = 0; i < adapter->num_q_vectors; i++) 1465 free_irq(adapter->msix_entries[vector++].vector, 1466 adapter->q_vector[i]); 1467 } else { 1468 free_irq(adapter->pdev->irq, adapter); 1469 } 1470 } 1471 1472 /** 1473 * igb_irq_disable - Mask off interrupt generation on the NIC 1474 * @adapter: board private structure 1475 **/ 1476 static void igb_irq_disable(struct igb_adapter *adapter) 1477 { 1478 struct e1000_hw *hw = &adapter->hw; 1479 1480 /* we need to be careful when disabling interrupts. The VFs are also 1481 * mapped into these registers and so clearing the bits can cause 1482 * issues on the VF drivers so we only need to clear what we set 1483 */ 1484 if (adapter->flags & IGB_FLAG_HAS_MSIX) { 1485 u32 regval = rd32(E1000_EIAM); 1486 1487 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask); 1488 wr32(E1000_EIMC, adapter->eims_enable_mask); 1489 regval = rd32(E1000_EIAC); 1490 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask); 1491 } 1492 1493 wr32(E1000_IAM, 0); 1494 wr32(E1000_IMC, ~0); 1495 wrfl(); 1496 if (adapter->flags & IGB_FLAG_HAS_MSIX) { 1497 int i; 1498 1499 for (i = 0; i < adapter->num_q_vectors; i++) 1500 synchronize_irq(adapter->msix_entries[i].vector); 1501 } else { 1502 synchronize_irq(adapter->pdev->irq); 1503 } 1504 } 1505 1506 /** 1507 * igb_irq_enable - Enable default interrupt generation settings 1508 * @adapter: board private structure 1509 **/ 1510 static void igb_irq_enable(struct igb_adapter *adapter) 1511 { 1512 struct e1000_hw *hw = &adapter->hw; 1513 1514 if (adapter->flags & IGB_FLAG_HAS_MSIX) { 1515 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA; 1516 u32 regval = rd32(E1000_EIAC); 1517 1518 wr32(E1000_EIAC, regval | adapter->eims_enable_mask); 1519 regval = rd32(E1000_EIAM); 1520 wr32(E1000_EIAM, regval | adapter->eims_enable_mask); 1521 wr32(E1000_EIMS, adapter->eims_enable_mask); 1522 if (adapter->vfs_allocated_count) { 1523 wr32(E1000_MBVFIMR, 0xFF); 1524 ims |= E1000_IMS_VMMB; 1525 } 1526 wr32(E1000_IMS, ims); 1527 } else { 1528 wr32(E1000_IMS, IMS_ENABLE_MASK | 1529 E1000_IMS_DRSTA); 1530 wr32(E1000_IAM, IMS_ENABLE_MASK | 1531 E1000_IMS_DRSTA); 1532 } 1533 } 1534 1535 static void igb_update_mng_vlan(struct igb_adapter *adapter) 1536 { 1537 struct e1000_hw *hw = &adapter->hw; 1538 u16 pf_id = adapter->vfs_allocated_count; 1539 u16 vid = adapter->hw.mng_cookie.vlan_id; 1540 u16 old_vid = adapter->mng_vlan_id; 1541 1542 if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) { 1543 /* add VID to filter table */ 1544 igb_vfta_set(hw, vid, pf_id, true, true); 1545 adapter->mng_vlan_id = vid; 1546 } else { 1547 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE; 1548 } 1549 1550 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) && 1551 (vid != old_vid) && 1552 !test_bit(old_vid, adapter->active_vlans)) { 1553 /* remove VID from filter table */ 1554 igb_vfta_set(hw, vid, pf_id, false, true); 1555 } 1556 } 1557 1558 /** 1559 * igb_release_hw_control - release control of the h/w to f/w 1560 * @adapter: address of board private structure 1561 * 1562 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit. 1563 * For ASF and Pass Through versions of f/w this means that the 1564 * driver is no longer loaded. 1565 **/ 1566 static void igb_release_hw_control(struct igb_adapter *adapter) 1567 { 1568 struct e1000_hw *hw = &adapter->hw; 1569 u32 ctrl_ext; 1570 1571 /* Let firmware take over control of h/w */ 1572 ctrl_ext = rd32(E1000_CTRL_EXT); 1573 wr32(E1000_CTRL_EXT, 1574 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD); 1575 } 1576 1577 /** 1578 * igb_get_hw_control - get control of the h/w from f/w 1579 * @adapter: address of board private structure 1580 * 1581 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit. 1582 * For ASF and Pass Through versions of f/w this means that 1583 * the driver is loaded. 1584 **/ 1585 static void igb_get_hw_control(struct igb_adapter *adapter) 1586 { 1587 struct e1000_hw *hw = &adapter->hw; 1588 u32 ctrl_ext; 1589 1590 /* Let firmware know the driver has taken over */ 1591 ctrl_ext = rd32(E1000_CTRL_EXT); 1592 wr32(E1000_CTRL_EXT, 1593 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD); 1594 } 1595 1596 static void enable_fqtss(struct igb_adapter *adapter, bool enable) 1597 { 1598 struct net_device *netdev = adapter->netdev; 1599 struct e1000_hw *hw = &adapter->hw; 1600 1601 WARN_ON(hw->mac.type != e1000_i210); 1602 1603 if (enable) 1604 adapter->flags |= IGB_FLAG_FQTSS; 1605 else 1606 adapter->flags &= ~IGB_FLAG_FQTSS; 1607 1608 if (netif_running(netdev)) 1609 schedule_work(&adapter->reset_task); 1610 } 1611 1612 static bool is_fqtss_enabled(struct igb_adapter *adapter) 1613 { 1614 return (adapter->flags & IGB_FLAG_FQTSS) ? true : false; 1615 } 1616 1617 static void set_tx_desc_fetch_prio(struct e1000_hw *hw, int queue, 1618 enum tx_queue_prio prio) 1619 { 1620 u32 val; 1621 1622 WARN_ON(hw->mac.type != e1000_i210); 1623 WARN_ON(queue < 0 || queue > 4); 1624 1625 val = rd32(E1000_I210_TXDCTL(queue)); 1626 1627 if (prio == TX_QUEUE_PRIO_HIGH) 1628 val |= E1000_TXDCTL_PRIORITY; 1629 else 1630 val &= ~E1000_TXDCTL_PRIORITY; 1631 1632 wr32(E1000_I210_TXDCTL(queue), val); 1633 } 1634 1635 static void set_queue_mode(struct e1000_hw *hw, int queue, enum queue_mode mode) 1636 { 1637 u32 val; 1638 1639 WARN_ON(hw->mac.type != e1000_i210); 1640 WARN_ON(queue < 0 || queue > 1); 1641 1642 val = rd32(E1000_I210_TQAVCC(queue)); 1643 1644 if (mode == QUEUE_MODE_STREAM_RESERVATION) 1645 val |= E1000_TQAVCC_QUEUEMODE; 1646 else 1647 val &= ~E1000_TQAVCC_QUEUEMODE; 1648 1649 wr32(E1000_I210_TQAVCC(queue), val); 1650 } 1651 1652 static bool is_any_cbs_enabled(struct igb_adapter *adapter) 1653 { 1654 int i; 1655 1656 for (i = 0; i < adapter->num_tx_queues; i++) { 1657 if (adapter->tx_ring[i]->cbs_enable) 1658 return true; 1659 } 1660 1661 return false; 1662 } 1663 1664 static bool is_any_txtime_enabled(struct igb_adapter *adapter) 1665 { 1666 int i; 1667 1668 for (i = 0; i < adapter->num_tx_queues; i++) { 1669 if (adapter->tx_ring[i]->launchtime_enable) 1670 return true; 1671 } 1672 1673 return false; 1674 } 1675 1676 /** 1677 * igb_config_tx_modes - Configure "Qav Tx mode" features on igb 1678 * @adapter: pointer to adapter struct 1679 * @queue: queue number 1680 * 1681 * Configure CBS and Launchtime for a given hardware queue. 1682 * Parameters are retrieved from the correct Tx ring, so 1683 * igb_save_cbs_params() and igb_save_txtime_params() should be used 1684 * for setting those correctly prior to this function being called. 1685 **/ 1686 static void igb_config_tx_modes(struct igb_adapter *adapter, int queue) 1687 { 1688 struct igb_ring *ring = adapter->tx_ring[queue]; 1689 struct net_device *netdev = adapter->netdev; 1690 struct e1000_hw *hw = &adapter->hw; 1691 u32 tqavcc, tqavctrl; 1692 u16 value; 1693 1694 WARN_ON(hw->mac.type != e1000_i210); 1695 WARN_ON(queue < 0 || queue > 1); 1696 1697 /* If any of the Qav features is enabled, configure queues as SR and 1698 * with HIGH PRIO. If none is, then configure them with LOW PRIO and 1699 * as SP. 1700 */ 1701 if (ring->cbs_enable || ring->launchtime_enable) { 1702 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_HIGH); 1703 set_queue_mode(hw, queue, QUEUE_MODE_STREAM_RESERVATION); 1704 } else { 1705 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_LOW); 1706 set_queue_mode(hw, queue, QUEUE_MODE_STRICT_PRIORITY); 1707 } 1708 1709 /* If CBS is enabled, set DataTranARB and config its parameters. */ 1710 if (ring->cbs_enable || queue == 0) { 1711 /* i210 does not allow the queue 0 to be in the Strict 1712 * Priority mode while the Qav mode is enabled, so, 1713 * instead of disabling strict priority mode, we give 1714 * queue 0 the maximum of credits possible. 1715 * 1716 * See section 8.12.19 of the i210 datasheet, "Note: 1717 * Queue0 QueueMode must be set to 1b when 1718 * TransmitMode is set to Qav." 1719 */ 1720 if (queue == 0 && !ring->cbs_enable) { 1721 /* max "linkspeed" idleslope in kbps */ 1722 ring->idleslope = 1000000; 1723 ring->hicredit = ETH_FRAME_LEN; 1724 } 1725 1726 /* Always set data transfer arbitration to credit-based 1727 * shaper algorithm on TQAVCTRL if CBS is enabled for any of 1728 * the queues. 1729 */ 1730 tqavctrl = rd32(E1000_I210_TQAVCTRL); 1731 tqavctrl |= E1000_TQAVCTRL_DATATRANARB; 1732 wr32(E1000_I210_TQAVCTRL, tqavctrl); 1733 1734 /* According to i210 datasheet section 7.2.7.7, we should set 1735 * the 'idleSlope' field from TQAVCC register following the 1736 * equation: 1737 * 1738 * For 100 Mbps link speed: 1739 * 1740 * value = BW * 0x7735 * 0.2 (E1) 1741 * 1742 * For 1000Mbps link speed: 1743 * 1744 * value = BW * 0x7735 * 2 (E2) 1745 * 1746 * E1 and E2 can be merged into one equation as shown below. 1747 * Note that 'link-speed' is in Mbps. 1748 * 1749 * value = BW * 0x7735 * 2 * link-speed 1750 * -------------- (E3) 1751 * 1000 1752 * 1753 * 'BW' is the percentage bandwidth out of full link speed 1754 * which can be found with the following equation. Note that 1755 * idleSlope here is the parameter from this function which 1756 * is in kbps. 1757 * 1758 * BW = idleSlope 1759 * ----------------- (E4) 1760 * link-speed * 1000 1761 * 1762 * That said, we can come up with a generic equation to 1763 * calculate the value we should set it TQAVCC register by 1764 * replacing 'BW' in E3 by E4. The resulting equation is: 1765 * 1766 * value = idleSlope * 0x7735 * 2 * link-speed 1767 * ----------------- -------------- (E5) 1768 * link-speed * 1000 1000 1769 * 1770 * 'link-speed' is present in both sides of the fraction so 1771 * it is canceled out. The final equation is the following: 1772 * 1773 * value = idleSlope * 61034 1774 * ----------------- (E6) 1775 * 1000000 1776 * 1777 * NOTE: For i210, given the above, we can see that idleslope 1778 * is represented in 16.38431 kbps units by the value at 1779 * the TQAVCC register (1Gbps / 61034), which reduces 1780 * the granularity for idleslope increments. 1781 * For instance, if you want to configure a 2576kbps 1782 * idleslope, the value to be written on the register 1783 * would have to be 157.23. If rounded down, you end 1784 * up with less bandwidth available than originally 1785 * required (~2572 kbps). If rounded up, you end up 1786 * with a higher bandwidth (~2589 kbps). Below the 1787 * approach we take is to always round up the 1788 * calculated value, so the resulting bandwidth might 1789 * be slightly higher for some configurations. 1790 */ 1791 value = DIV_ROUND_UP_ULL(ring->idleslope * 61034ULL, 1000000); 1792 1793 tqavcc = rd32(E1000_I210_TQAVCC(queue)); 1794 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK; 1795 tqavcc |= value; 1796 wr32(E1000_I210_TQAVCC(queue), tqavcc); 1797 1798 wr32(E1000_I210_TQAVHC(queue), 1799 0x80000000 + ring->hicredit * 0x7735); 1800 } else { 1801 1802 /* Set idleSlope to zero. */ 1803 tqavcc = rd32(E1000_I210_TQAVCC(queue)); 1804 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK; 1805 wr32(E1000_I210_TQAVCC(queue), tqavcc); 1806 1807 /* Set hiCredit to zero. */ 1808 wr32(E1000_I210_TQAVHC(queue), 0); 1809 1810 /* If CBS is not enabled for any queues anymore, then return to 1811 * the default state of Data Transmission Arbitration on 1812 * TQAVCTRL. 1813 */ 1814 if (!is_any_cbs_enabled(adapter)) { 1815 tqavctrl = rd32(E1000_I210_TQAVCTRL); 1816 tqavctrl &= ~E1000_TQAVCTRL_DATATRANARB; 1817 wr32(E1000_I210_TQAVCTRL, tqavctrl); 1818 } 1819 } 1820 1821 /* If LaunchTime is enabled, set DataTranTIM. */ 1822 if (ring->launchtime_enable) { 1823 /* Always set DataTranTIM on TQAVCTRL if LaunchTime is enabled 1824 * for any of the SR queues, and configure fetchtime delta. 1825 * XXX NOTE: 1826 * - LaunchTime will be enabled for all SR queues. 1827 * - A fixed offset can be added relative to the launch 1828 * time of all packets if configured at reg LAUNCH_OS0. 1829 * We are keeping it as 0 for now (default value). 1830 */ 1831 tqavctrl = rd32(E1000_I210_TQAVCTRL); 1832 tqavctrl |= E1000_TQAVCTRL_DATATRANTIM | 1833 E1000_TQAVCTRL_FETCHTIME_DELTA; 1834 wr32(E1000_I210_TQAVCTRL, tqavctrl); 1835 } else { 1836 /* If Launchtime is not enabled for any SR queues anymore, 1837 * then clear DataTranTIM on TQAVCTRL and clear fetchtime delta, 1838 * effectively disabling Launchtime. 1839 */ 1840 if (!is_any_txtime_enabled(adapter)) { 1841 tqavctrl = rd32(E1000_I210_TQAVCTRL); 1842 tqavctrl &= ~E1000_TQAVCTRL_DATATRANTIM; 1843 tqavctrl &= ~E1000_TQAVCTRL_FETCHTIME_DELTA; 1844 wr32(E1000_I210_TQAVCTRL, tqavctrl); 1845 } 1846 } 1847 1848 /* XXX: In i210 controller the sendSlope and loCredit parameters from 1849 * CBS are not configurable by software so we don't do any 'controller 1850 * configuration' in respect to these parameters. 1851 */ 1852 1853 netdev_dbg(netdev, "Qav Tx mode: cbs %s, launchtime %s, queue %d idleslope %d sendslope %d hiCredit %d locredit %d\n", 1854 ring->cbs_enable ? "enabled" : "disabled", 1855 ring->launchtime_enable ? "enabled" : "disabled", 1856 queue, 1857 ring->idleslope, ring->sendslope, 1858 ring->hicredit, ring->locredit); 1859 } 1860 1861 static int igb_save_txtime_params(struct igb_adapter *adapter, int queue, 1862 bool enable) 1863 { 1864 struct igb_ring *ring; 1865 1866 if (queue < 0 || queue > adapter->num_tx_queues) 1867 return -EINVAL; 1868 1869 ring = adapter->tx_ring[queue]; 1870 ring->launchtime_enable = enable; 1871 1872 return 0; 1873 } 1874 1875 static int igb_save_cbs_params(struct igb_adapter *adapter, int queue, 1876 bool enable, int idleslope, int sendslope, 1877 int hicredit, int locredit) 1878 { 1879 struct igb_ring *ring; 1880 1881 if (queue < 0 || queue > adapter->num_tx_queues) 1882 return -EINVAL; 1883 1884 ring = adapter->tx_ring[queue]; 1885 1886 ring->cbs_enable = enable; 1887 ring->idleslope = idleslope; 1888 ring->sendslope = sendslope; 1889 ring->hicredit = hicredit; 1890 ring->locredit = locredit; 1891 1892 return 0; 1893 } 1894 1895 /** 1896 * igb_setup_tx_mode - Switch to/from Qav Tx mode when applicable 1897 * @adapter: pointer to adapter struct 1898 * 1899 * Configure TQAVCTRL register switching the controller's Tx mode 1900 * if FQTSS mode is enabled or disabled. Additionally, will issue 1901 * a call to igb_config_tx_modes() per queue so any previously saved 1902 * Tx parameters are applied. 1903 **/ 1904 static void igb_setup_tx_mode(struct igb_adapter *adapter) 1905 { 1906 struct net_device *netdev = adapter->netdev; 1907 struct e1000_hw *hw = &adapter->hw; 1908 u32 val; 1909 1910 /* Only i210 controller supports changing the transmission mode. */ 1911 if (hw->mac.type != e1000_i210) 1912 return; 1913 1914 if (is_fqtss_enabled(adapter)) { 1915 int i, max_queue; 1916 1917 /* Configure TQAVCTRL register: set transmit mode to 'Qav', 1918 * set data fetch arbitration to 'round robin', set SP_WAIT_SR 1919 * so SP queues wait for SR ones. 1920 */ 1921 val = rd32(E1000_I210_TQAVCTRL); 1922 val |= E1000_TQAVCTRL_XMIT_MODE | E1000_TQAVCTRL_SP_WAIT_SR; 1923 val &= ~E1000_TQAVCTRL_DATAFETCHARB; 1924 wr32(E1000_I210_TQAVCTRL, val); 1925 1926 /* Configure Tx and Rx packet buffers sizes as described in 1927 * i210 datasheet section 7.2.7.7. 1928 */ 1929 val = rd32(E1000_TXPBS); 1930 val &= ~I210_TXPBSIZE_MASK; 1931 val |= I210_TXPBSIZE_PB0_8KB | I210_TXPBSIZE_PB1_8KB | 1932 I210_TXPBSIZE_PB2_4KB | I210_TXPBSIZE_PB3_4KB; 1933 wr32(E1000_TXPBS, val); 1934 1935 val = rd32(E1000_RXPBS); 1936 val &= ~I210_RXPBSIZE_MASK; 1937 val |= I210_RXPBSIZE_PB_30KB; 1938 wr32(E1000_RXPBS, val); 1939 1940 /* Section 8.12.9 states that MAX_TPKT_SIZE from DTXMXPKTSZ 1941 * register should not exceed the buffer size programmed in 1942 * TXPBS. The smallest buffer size programmed in TXPBS is 4kB 1943 * so according to the datasheet we should set MAX_TPKT_SIZE to 1944 * 4kB / 64. 1945 * 1946 * However, when we do so, no frame from queue 2 and 3 are 1947 * transmitted. It seems the MAX_TPKT_SIZE should not be great 1948 * or _equal_ to the buffer size programmed in TXPBS. For this 1949 * reason, we set set MAX_ TPKT_SIZE to (4kB - 1) / 64. 1950 */ 1951 val = (4096 - 1) / 64; 1952 wr32(E1000_I210_DTXMXPKTSZ, val); 1953 1954 /* Since FQTSS mode is enabled, apply any CBS configuration 1955 * previously set. If no previous CBS configuration has been 1956 * done, then the initial configuration is applied, which means 1957 * CBS is disabled. 1958 */ 1959 max_queue = (adapter->num_tx_queues < I210_SR_QUEUES_NUM) ? 1960 adapter->num_tx_queues : I210_SR_QUEUES_NUM; 1961 1962 for (i = 0; i < max_queue; i++) { 1963 igb_config_tx_modes(adapter, i); 1964 } 1965 } else { 1966 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT); 1967 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT); 1968 wr32(E1000_I210_DTXMXPKTSZ, I210_DTXMXPKTSZ_DEFAULT); 1969 1970 val = rd32(E1000_I210_TQAVCTRL); 1971 /* According to Section 8.12.21, the other flags we've set when 1972 * enabling FQTSS are not relevant when disabling FQTSS so we 1973 * don't set they here. 1974 */ 1975 val &= ~E1000_TQAVCTRL_XMIT_MODE; 1976 wr32(E1000_I210_TQAVCTRL, val); 1977 } 1978 1979 netdev_dbg(netdev, "FQTSS %s\n", (is_fqtss_enabled(adapter)) ? 1980 "enabled" : "disabled"); 1981 } 1982 1983 /** 1984 * igb_configure - configure the hardware for RX and TX 1985 * @adapter: private board structure 1986 **/ 1987 static void igb_configure(struct igb_adapter *adapter) 1988 { 1989 struct net_device *netdev = adapter->netdev; 1990 int i; 1991 1992 igb_get_hw_control(adapter); 1993 igb_set_rx_mode(netdev); 1994 igb_setup_tx_mode(adapter); 1995 1996 igb_restore_vlan(adapter); 1997 1998 igb_setup_tctl(adapter); 1999 igb_setup_mrqc(adapter); 2000 igb_setup_rctl(adapter); 2001 2002 igb_nfc_filter_restore(adapter); 2003 igb_configure_tx(adapter); 2004 igb_configure_rx(adapter); 2005 2006 igb_rx_fifo_flush_82575(&adapter->hw); 2007 2008 /* call igb_desc_unused which always leaves 2009 * at least 1 descriptor unused to make sure 2010 * next_to_use != next_to_clean 2011 */ 2012 for (i = 0; i < adapter->num_rx_queues; i++) { 2013 struct igb_ring *ring = adapter->rx_ring[i]; 2014 igb_alloc_rx_buffers(ring, igb_desc_unused(ring)); 2015 } 2016 } 2017 2018 /** 2019 * igb_power_up_link - Power up the phy/serdes link 2020 * @adapter: address of board private structure 2021 **/ 2022 void igb_power_up_link(struct igb_adapter *adapter) 2023 { 2024 igb_reset_phy(&adapter->hw); 2025 2026 if (adapter->hw.phy.media_type == e1000_media_type_copper) 2027 igb_power_up_phy_copper(&adapter->hw); 2028 else 2029 igb_power_up_serdes_link_82575(&adapter->hw); 2030 2031 igb_setup_link(&adapter->hw); 2032 } 2033 2034 /** 2035 * igb_power_down_link - Power down the phy/serdes link 2036 * @adapter: address of board private structure 2037 */ 2038 static void igb_power_down_link(struct igb_adapter *adapter) 2039 { 2040 if (adapter->hw.phy.media_type == e1000_media_type_copper) 2041 igb_power_down_phy_copper_82575(&adapter->hw); 2042 else 2043 igb_shutdown_serdes_link_82575(&adapter->hw); 2044 } 2045 2046 /** 2047 * Detect and switch function for Media Auto Sense 2048 * @adapter: address of the board private structure 2049 **/ 2050 static void igb_check_swap_media(struct igb_adapter *adapter) 2051 { 2052 struct e1000_hw *hw = &adapter->hw; 2053 u32 ctrl_ext, connsw; 2054 bool swap_now = false; 2055 2056 ctrl_ext = rd32(E1000_CTRL_EXT); 2057 connsw = rd32(E1000_CONNSW); 2058 2059 /* need to live swap if current media is copper and we have fiber/serdes 2060 * to go to. 2061 */ 2062 2063 if ((hw->phy.media_type == e1000_media_type_copper) && 2064 (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) { 2065 swap_now = true; 2066 } else if (!(connsw & E1000_CONNSW_SERDESD)) { 2067 /* copper signal takes time to appear */ 2068 if (adapter->copper_tries < 4) { 2069 adapter->copper_tries++; 2070 connsw |= E1000_CONNSW_AUTOSENSE_CONF; 2071 wr32(E1000_CONNSW, connsw); 2072 return; 2073 } else { 2074 adapter->copper_tries = 0; 2075 if ((connsw & E1000_CONNSW_PHYSD) && 2076 (!(connsw & E1000_CONNSW_PHY_PDN))) { 2077 swap_now = true; 2078 connsw &= ~E1000_CONNSW_AUTOSENSE_CONF; 2079 wr32(E1000_CONNSW, connsw); 2080 } 2081 } 2082 } 2083 2084 if (!swap_now) 2085 return; 2086 2087 switch (hw->phy.media_type) { 2088 case e1000_media_type_copper: 2089 netdev_info(adapter->netdev, 2090 "MAS: changing media to fiber/serdes\n"); 2091 ctrl_ext |= 2092 E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES; 2093 adapter->flags |= IGB_FLAG_MEDIA_RESET; 2094 adapter->copper_tries = 0; 2095 break; 2096 case e1000_media_type_internal_serdes: 2097 case e1000_media_type_fiber: 2098 netdev_info(adapter->netdev, 2099 "MAS: changing media to copper\n"); 2100 ctrl_ext &= 2101 ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES; 2102 adapter->flags |= IGB_FLAG_MEDIA_RESET; 2103 break; 2104 default: 2105 /* shouldn't get here during regular operation */ 2106 netdev_err(adapter->netdev, 2107 "AMS: Invalid media type found, returning\n"); 2108 break; 2109 } 2110 wr32(E1000_CTRL_EXT, ctrl_ext); 2111 } 2112 2113 /** 2114 * igb_up - Open the interface and prepare it to handle traffic 2115 * @adapter: board private structure 2116 **/ 2117 int igb_up(struct igb_adapter *adapter) 2118 { 2119 struct e1000_hw *hw = &adapter->hw; 2120 int i; 2121 2122 /* hardware has been reset, we need to reload some things */ 2123 igb_configure(adapter); 2124 2125 clear_bit(__IGB_DOWN, &adapter->state); 2126 2127 for (i = 0; i < adapter->num_q_vectors; i++) 2128 napi_enable(&(adapter->q_vector[i]->napi)); 2129 2130 if (adapter->flags & IGB_FLAG_HAS_MSIX) 2131 igb_configure_msix(adapter); 2132 else 2133 igb_assign_vector(adapter->q_vector[0], 0); 2134 2135 /* Clear any pending interrupts. */ 2136 rd32(E1000_TSICR); 2137 rd32(E1000_ICR); 2138 igb_irq_enable(adapter); 2139 2140 /* notify VFs that reset has been completed */ 2141 if (adapter->vfs_allocated_count) { 2142 u32 reg_data = rd32(E1000_CTRL_EXT); 2143 2144 reg_data |= E1000_CTRL_EXT_PFRSTD; 2145 wr32(E1000_CTRL_EXT, reg_data); 2146 } 2147 2148 netif_tx_start_all_queues(adapter->netdev); 2149 2150 /* start the watchdog. */ 2151 hw->mac.get_link_status = 1; 2152 schedule_work(&adapter->watchdog_task); 2153 2154 if ((adapter->flags & IGB_FLAG_EEE) && 2155 (!hw->dev_spec._82575.eee_disable)) 2156 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T; 2157 2158 return 0; 2159 } 2160 2161 void igb_down(struct igb_adapter *adapter) 2162 { 2163 struct net_device *netdev = adapter->netdev; 2164 struct e1000_hw *hw = &adapter->hw; 2165 u32 tctl, rctl; 2166 int i; 2167 2168 /* signal that we're down so the interrupt handler does not 2169 * reschedule our watchdog timer 2170 */ 2171 set_bit(__IGB_DOWN, &adapter->state); 2172 2173 /* disable receives in the hardware */ 2174 rctl = rd32(E1000_RCTL); 2175 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN); 2176 /* flush and sleep below */ 2177 2178 igb_nfc_filter_exit(adapter); 2179 2180 netif_carrier_off(netdev); 2181 netif_tx_stop_all_queues(netdev); 2182 2183 /* disable transmits in the hardware */ 2184 tctl = rd32(E1000_TCTL); 2185 tctl &= ~E1000_TCTL_EN; 2186 wr32(E1000_TCTL, tctl); 2187 /* flush both disables and wait for them to finish */ 2188 wrfl(); 2189 usleep_range(10000, 11000); 2190 2191 igb_irq_disable(adapter); 2192 2193 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE; 2194 2195 for (i = 0; i < adapter->num_q_vectors; i++) { 2196 if (adapter->q_vector[i]) { 2197 napi_synchronize(&adapter->q_vector[i]->napi); 2198 napi_disable(&adapter->q_vector[i]->napi); 2199 } 2200 } 2201 2202 del_timer_sync(&adapter->watchdog_timer); 2203 del_timer_sync(&adapter->phy_info_timer); 2204 2205 /* record the stats before reset*/ 2206 spin_lock(&adapter->stats64_lock); 2207 igb_update_stats(adapter); 2208 spin_unlock(&adapter->stats64_lock); 2209 2210 adapter->link_speed = 0; 2211 adapter->link_duplex = 0; 2212 2213 if (!pci_channel_offline(adapter->pdev)) 2214 igb_reset(adapter); 2215 2216 /* clear VLAN promisc flag so VFTA will be updated if necessary */ 2217 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC; 2218 2219 igb_clean_all_tx_rings(adapter); 2220 igb_clean_all_rx_rings(adapter); 2221 #ifdef CONFIG_IGB_DCA 2222 2223 /* since we reset the hardware DCA settings were cleared */ 2224 igb_setup_dca(adapter); 2225 #endif 2226 } 2227 2228 void igb_reinit_locked(struct igb_adapter *adapter) 2229 { 2230 WARN_ON(in_interrupt()); 2231 while (test_and_set_bit(__IGB_RESETTING, &adapter->state)) 2232 usleep_range(1000, 2000); 2233 igb_down(adapter); 2234 igb_up(adapter); 2235 clear_bit(__IGB_RESETTING, &adapter->state); 2236 } 2237 2238 /** igb_enable_mas - Media Autosense re-enable after swap 2239 * 2240 * @adapter: adapter struct 2241 **/ 2242 static void igb_enable_mas(struct igb_adapter *adapter) 2243 { 2244 struct e1000_hw *hw = &adapter->hw; 2245 u32 connsw = rd32(E1000_CONNSW); 2246 2247 /* configure for SerDes media detect */ 2248 if ((hw->phy.media_type == e1000_media_type_copper) && 2249 (!(connsw & E1000_CONNSW_SERDESD))) { 2250 connsw |= E1000_CONNSW_ENRGSRC; 2251 connsw |= E1000_CONNSW_AUTOSENSE_EN; 2252 wr32(E1000_CONNSW, connsw); 2253 wrfl(); 2254 } 2255 } 2256 2257 void igb_reset(struct igb_adapter *adapter) 2258 { 2259 struct pci_dev *pdev = adapter->pdev; 2260 struct e1000_hw *hw = &adapter->hw; 2261 struct e1000_mac_info *mac = &hw->mac; 2262 struct e1000_fc_info *fc = &hw->fc; 2263 u32 pba, hwm; 2264 2265 /* Repartition Pba for greater than 9k mtu 2266 * To take effect CTRL.RST is required. 2267 */ 2268 switch (mac->type) { 2269 case e1000_i350: 2270 case e1000_i354: 2271 case e1000_82580: 2272 pba = rd32(E1000_RXPBS); 2273 pba = igb_rxpbs_adjust_82580(pba); 2274 break; 2275 case e1000_82576: 2276 pba = rd32(E1000_RXPBS); 2277 pba &= E1000_RXPBS_SIZE_MASK_82576; 2278 break; 2279 case e1000_82575: 2280 case e1000_i210: 2281 case e1000_i211: 2282 default: 2283 pba = E1000_PBA_34K; 2284 break; 2285 } 2286 2287 if (mac->type == e1000_82575) { 2288 u32 min_rx_space, min_tx_space, needed_tx_space; 2289 2290 /* write Rx PBA so that hardware can report correct Tx PBA */ 2291 wr32(E1000_PBA, pba); 2292 2293 /* To maintain wire speed transmits, the Tx FIFO should be 2294 * large enough to accommodate two full transmit packets, 2295 * rounded up to the next 1KB and expressed in KB. Likewise, 2296 * the Rx FIFO should be large enough to accommodate at least 2297 * one full receive packet and is similarly rounded up and 2298 * expressed in KB. 2299 */ 2300 min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024); 2301 2302 /* The Tx FIFO also stores 16 bytes of information about the Tx 2303 * but don't include Ethernet FCS because hardware appends it. 2304 * We only need to round down to the nearest 512 byte block 2305 * count since the value we care about is 2 frames, not 1. 2306 */ 2307 min_tx_space = adapter->max_frame_size; 2308 min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN; 2309 min_tx_space = DIV_ROUND_UP(min_tx_space, 512); 2310 2311 /* upper 16 bits has Tx packet buffer allocation size in KB */ 2312 needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16); 2313 2314 /* If current Tx allocation is less than the min Tx FIFO size, 2315 * and the min Tx FIFO size is less than the current Rx FIFO 2316 * allocation, take space away from current Rx allocation. 2317 */ 2318 if (needed_tx_space < pba) { 2319 pba -= needed_tx_space; 2320 2321 /* if short on Rx space, Rx wins and must trump Tx 2322 * adjustment 2323 */ 2324 if (pba < min_rx_space) 2325 pba = min_rx_space; 2326 } 2327 2328 /* adjust PBA for jumbo frames */ 2329 wr32(E1000_PBA, pba); 2330 } 2331 2332 /* flow control settings 2333 * The high water mark must be low enough to fit one full frame 2334 * after transmitting the pause frame. As such we must have enough 2335 * space to allow for us to complete our current transmit and then 2336 * receive the frame that is in progress from the link partner. 2337 * Set it to: 2338 * - the full Rx FIFO size minus one full Tx plus one full Rx frame 2339 */ 2340 hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE); 2341 2342 fc->high_water = hwm & 0xFFFFFFF0; /* 16-byte granularity */ 2343 fc->low_water = fc->high_water - 16; 2344 fc->pause_time = 0xFFFF; 2345 fc->send_xon = 1; 2346 fc->current_mode = fc->requested_mode; 2347 2348 /* disable receive for all VFs and wait one second */ 2349 if (adapter->vfs_allocated_count) { 2350 int i; 2351 2352 for (i = 0 ; i < adapter->vfs_allocated_count; i++) 2353 adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC; 2354 2355 /* ping all the active vfs to let them know we are going down */ 2356 igb_ping_all_vfs(adapter); 2357 2358 /* disable transmits and receives */ 2359 wr32(E1000_VFRE, 0); 2360 wr32(E1000_VFTE, 0); 2361 } 2362 2363 /* Allow time for pending master requests to run */ 2364 hw->mac.ops.reset_hw(hw); 2365 wr32(E1000_WUC, 0); 2366 2367 if (adapter->flags & IGB_FLAG_MEDIA_RESET) { 2368 /* need to resetup here after media swap */ 2369 adapter->ei.get_invariants(hw); 2370 adapter->flags &= ~IGB_FLAG_MEDIA_RESET; 2371 } 2372 if ((mac->type == e1000_82575) && 2373 (adapter->flags & IGB_FLAG_MAS_ENABLE)) { 2374 igb_enable_mas(adapter); 2375 } 2376 if (hw->mac.ops.init_hw(hw)) 2377 dev_err(&pdev->dev, "Hardware Error\n"); 2378 2379 /* RAR registers were cleared during init_hw, clear mac table */ 2380 igb_flush_mac_table(adapter); 2381 __dev_uc_unsync(adapter->netdev, NULL); 2382 2383 /* Recover default RAR entry */ 2384 igb_set_default_mac_filter(adapter); 2385 2386 /* Flow control settings reset on hardware reset, so guarantee flow 2387 * control is off when forcing speed. 2388 */ 2389 if (!hw->mac.autoneg) 2390 igb_force_mac_fc(hw); 2391 2392 igb_init_dmac(adapter, pba); 2393 #ifdef CONFIG_IGB_HWMON 2394 /* Re-initialize the thermal sensor on i350 devices. */ 2395 if (!test_bit(__IGB_DOWN, &adapter->state)) { 2396 if (mac->type == e1000_i350 && hw->bus.func == 0) { 2397 /* If present, re-initialize the external thermal sensor 2398 * interface. 2399 */ 2400 if (adapter->ets) 2401 mac->ops.init_thermal_sensor_thresh(hw); 2402 } 2403 } 2404 #endif 2405 /* Re-establish EEE setting */ 2406 if (hw->phy.media_type == e1000_media_type_copper) { 2407 switch (mac->type) { 2408 case e1000_i350: 2409 case e1000_i210: 2410 case e1000_i211: 2411 igb_set_eee_i350(hw, true, true); 2412 break; 2413 case e1000_i354: 2414 igb_set_eee_i354(hw, true, true); 2415 break; 2416 default: 2417 break; 2418 } 2419 } 2420 if (!netif_running(adapter->netdev)) 2421 igb_power_down_link(adapter); 2422 2423 igb_update_mng_vlan(adapter); 2424 2425 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */ 2426 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE); 2427 2428 /* Re-enable PTP, where applicable. */ 2429 if (adapter->ptp_flags & IGB_PTP_ENABLED) 2430 igb_ptp_reset(adapter); 2431 2432 igb_get_phy_info(hw); 2433 } 2434 2435 static netdev_features_t igb_fix_features(struct net_device *netdev, 2436 netdev_features_t features) 2437 { 2438 /* Since there is no support for separate Rx/Tx vlan accel 2439 * enable/disable make sure Tx flag is always in same state as Rx. 2440 */ 2441 if (features & NETIF_F_HW_VLAN_CTAG_RX) 2442 features |= NETIF_F_HW_VLAN_CTAG_TX; 2443 else 2444 features &= ~NETIF_F_HW_VLAN_CTAG_TX; 2445 2446 return features; 2447 } 2448 2449 static int igb_set_features(struct net_device *netdev, 2450 netdev_features_t features) 2451 { 2452 netdev_features_t changed = netdev->features ^ features; 2453 struct igb_adapter *adapter = netdev_priv(netdev); 2454 2455 if (changed & NETIF_F_HW_VLAN_CTAG_RX) 2456 igb_vlan_mode(netdev, features); 2457 2458 if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE))) 2459 return 0; 2460 2461 if (!(features & NETIF_F_NTUPLE)) { 2462 struct hlist_node *node2; 2463 struct igb_nfc_filter *rule; 2464 2465 spin_lock(&adapter->nfc_lock); 2466 hlist_for_each_entry_safe(rule, node2, 2467 &adapter->nfc_filter_list, nfc_node) { 2468 igb_erase_filter(adapter, rule); 2469 hlist_del(&rule->nfc_node); 2470 kfree(rule); 2471 } 2472 spin_unlock(&adapter->nfc_lock); 2473 adapter->nfc_filter_count = 0; 2474 } 2475 2476 netdev->features = features; 2477 2478 if (netif_running(netdev)) 2479 igb_reinit_locked(adapter); 2480 else 2481 igb_reset(adapter); 2482 2483 return 0; 2484 } 2485 2486 static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[], 2487 struct net_device *dev, 2488 const unsigned char *addr, u16 vid, 2489 u16 flags, 2490 struct netlink_ext_ack *extack) 2491 { 2492 /* guarantee we can provide a unique filter for the unicast address */ 2493 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) { 2494 struct igb_adapter *adapter = netdev_priv(dev); 2495 int vfn = adapter->vfs_allocated_count; 2496 2497 if (netdev_uc_count(dev) >= igb_available_rars(adapter, vfn)) 2498 return -ENOMEM; 2499 } 2500 2501 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags); 2502 } 2503 2504 #define IGB_MAX_MAC_HDR_LEN 127 2505 #define IGB_MAX_NETWORK_HDR_LEN 511 2506 2507 static netdev_features_t 2508 igb_features_check(struct sk_buff *skb, struct net_device *dev, 2509 netdev_features_t features) 2510 { 2511 unsigned int network_hdr_len, mac_hdr_len; 2512 2513 /* Make certain the headers can be described by a context descriptor */ 2514 mac_hdr_len = skb_network_header(skb) - skb->data; 2515 if (unlikely(mac_hdr_len > IGB_MAX_MAC_HDR_LEN)) 2516 return features & ~(NETIF_F_HW_CSUM | 2517 NETIF_F_SCTP_CRC | 2518 NETIF_F_HW_VLAN_CTAG_TX | 2519 NETIF_F_TSO | 2520 NETIF_F_TSO6); 2521 2522 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb); 2523 if (unlikely(network_hdr_len > IGB_MAX_NETWORK_HDR_LEN)) 2524 return features & ~(NETIF_F_HW_CSUM | 2525 NETIF_F_SCTP_CRC | 2526 NETIF_F_TSO | 2527 NETIF_F_TSO6); 2528 2529 /* We can only support IPV4 TSO in tunnels if we can mangle the 2530 * inner IP ID field, so strip TSO if MANGLEID is not supported. 2531 */ 2532 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) 2533 features &= ~NETIF_F_TSO; 2534 2535 return features; 2536 } 2537 2538 static void igb_offload_apply(struct igb_adapter *adapter, s32 queue) 2539 { 2540 if (!is_fqtss_enabled(adapter)) { 2541 enable_fqtss(adapter, true); 2542 return; 2543 } 2544 2545 igb_config_tx_modes(adapter, queue); 2546 2547 if (!is_any_cbs_enabled(adapter) && !is_any_txtime_enabled(adapter)) 2548 enable_fqtss(adapter, false); 2549 } 2550 2551 static int igb_offload_cbs(struct igb_adapter *adapter, 2552 struct tc_cbs_qopt_offload *qopt) 2553 { 2554 struct e1000_hw *hw = &adapter->hw; 2555 int err; 2556 2557 /* CBS offloading is only supported by i210 controller. */ 2558 if (hw->mac.type != e1000_i210) 2559 return -EOPNOTSUPP; 2560 2561 /* CBS offloading is only supported by queue 0 and queue 1. */ 2562 if (qopt->queue < 0 || qopt->queue > 1) 2563 return -EINVAL; 2564 2565 err = igb_save_cbs_params(adapter, qopt->queue, qopt->enable, 2566 qopt->idleslope, qopt->sendslope, 2567 qopt->hicredit, qopt->locredit); 2568 if (err) 2569 return err; 2570 2571 igb_offload_apply(adapter, qopt->queue); 2572 2573 return 0; 2574 } 2575 2576 #define ETHER_TYPE_FULL_MASK ((__force __be16)~0) 2577 #define VLAN_PRIO_FULL_MASK (0x07) 2578 2579 static int igb_parse_cls_flower(struct igb_adapter *adapter, 2580 struct tc_cls_flower_offload *f, 2581 int traffic_class, 2582 struct igb_nfc_filter *input) 2583 { 2584 struct flow_rule *rule = tc_cls_flower_offload_flow_rule(f); 2585 struct flow_dissector *dissector = rule->match.dissector; 2586 struct netlink_ext_ack *extack = f->common.extack; 2587 2588 if (dissector->used_keys & 2589 ~(BIT(FLOW_DISSECTOR_KEY_BASIC) | 2590 BIT(FLOW_DISSECTOR_KEY_CONTROL) | 2591 BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) | 2592 BIT(FLOW_DISSECTOR_KEY_VLAN))) { 2593 NL_SET_ERR_MSG_MOD(extack, 2594 "Unsupported key used, only BASIC, CONTROL, ETH_ADDRS and VLAN are supported"); 2595 return -EOPNOTSUPP; 2596 } 2597 2598 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) { 2599 struct flow_match_eth_addrs match; 2600 2601 flow_rule_match_eth_addrs(rule, &match); 2602 if (!is_zero_ether_addr(match.mask->dst)) { 2603 if (!is_broadcast_ether_addr(match.mask->dst)) { 2604 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for destination MAC address"); 2605 return -EINVAL; 2606 } 2607 2608 input->filter.match_flags |= 2609 IGB_FILTER_FLAG_DST_MAC_ADDR; 2610 ether_addr_copy(input->filter.dst_addr, match.key->dst); 2611 } 2612 2613 if (!is_zero_ether_addr(match.mask->src)) { 2614 if (!is_broadcast_ether_addr(match.mask->src)) { 2615 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for source MAC address"); 2616 return -EINVAL; 2617 } 2618 2619 input->filter.match_flags |= 2620 IGB_FILTER_FLAG_SRC_MAC_ADDR; 2621 ether_addr_copy(input->filter.src_addr, match.key->src); 2622 } 2623 } 2624 2625 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) { 2626 struct flow_match_basic match; 2627 2628 flow_rule_match_basic(rule, &match); 2629 if (match.mask->n_proto) { 2630 if (match.mask->n_proto != ETHER_TYPE_FULL_MASK) { 2631 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for EtherType filter"); 2632 return -EINVAL; 2633 } 2634 2635 input->filter.match_flags |= IGB_FILTER_FLAG_ETHER_TYPE; 2636 input->filter.etype = match.key->n_proto; 2637 } 2638 } 2639 2640 if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) { 2641 struct flow_match_vlan match; 2642 2643 flow_rule_match_vlan(rule, &match); 2644 if (match.mask->vlan_priority) { 2645 if (match.mask->vlan_priority != VLAN_PRIO_FULL_MASK) { 2646 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for VLAN priority"); 2647 return -EINVAL; 2648 } 2649 2650 input->filter.match_flags |= IGB_FILTER_FLAG_VLAN_TCI; 2651 input->filter.vlan_tci = match.key->vlan_priority; 2652 } 2653 } 2654 2655 input->action = traffic_class; 2656 input->cookie = f->cookie; 2657 2658 return 0; 2659 } 2660 2661 static int igb_configure_clsflower(struct igb_adapter *adapter, 2662 struct tc_cls_flower_offload *cls_flower) 2663 { 2664 struct netlink_ext_ack *extack = cls_flower->common.extack; 2665 struct igb_nfc_filter *filter, *f; 2666 int err, tc; 2667 2668 tc = tc_classid_to_hwtc(adapter->netdev, cls_flower->classid); 2669 if (tc < 0) { 2670 NL_SET_ERR_MSG_MOD(extack, "Invalid traffic class"); 2671 return -EINVAL; 2672 } 2673 2674 filter = kzalloc(sizeof(*filter), GFP_KERNEL); 2675 if (!filter) 2676 return -ENOMEM; 2677 2678 err = igb_parse_cls_flower(adapter, cls_flower, tc, filter); 2679 if (err < 0) 2680 goto err_parse; 2681 2682 spin_lock(&adapter->nfc_lock); 2683 2684 hlist_for_each_entry(f, &adapter->nfc_filter_list, nfc_node) { 2685 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) { 2686 err = -EEXIST; 2687 NL_SET_ERR_MSG_MOD(extack, 2688 "This filter is already set in ethtool"); 2689 goto err_locked; 2690 } 2691 } 2692 2693 hlist_for_each_entry(f, &adapter->cls_flower_list, nfc_node) { 2694 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) { 2695 err = -EEXIST; 2696 NL_SET_ERR_MSG_MOD(extack, 2697 "This filter is already set in cls_flower"); 2698 goto err_locked; 2699 } 2700 } 2701 2702 err = igb_add_filter(adapter, filter); 2703 if (err < 0) { 2704 NL_SET_ERR_MSG_MOD(extack, "Could not add filter to the adapter"); 2705 goto err_locked; 2706 } 2707 2708 hlist_add_head(&filter->nfc_node, &adapter->cls_flower_list); 2709 2710 spin_unlock(&adapter->nfc_lock); 2711 2712 return 0; 2713 2714 err_locked: 2715 spin_unlock(&adapter->nfc_lock); 2716 2717 err_parse: 2718 kfree(filter); 2719 2720 return err; 2721 } 2722 2723 static int igb_delete_clsflower(struct igb_adapter *adapter, 2724 struct tc_cls_flower_offload *cls_flower) 2725 { 2726 struct igb_nfc_filter *filter; 2727 int err; 2728 2729 spin_lock(&adapter->nfc_lock); 2730 2731 hlist_for_each_entry(filter, &adapter->cls_flower_list, nfc_node) 2732 if (filter->cookie == cls_flower->cookie) 2733 break; 2734 2735 if (!filter) { 2736 err = -ENOENT; 2737 goto out; 2738 } 2739 2740 err = igb_erase_filter(adapter, filter); 2741 if (err < 0) 2742 goto out; 2743 2744 hlist_del(&filter->nfc_node); 2745 kfree(filter); 2746 2747 out: 2748 spin_unlock(&adapter->nfc_lock); 2749 2750 return err; 2751 } 2752 2753 static int igb_setup_tc_cls_flower(struct igb_adapter *adapter, 2754 struct tc_cls_flower_offload *cls_flower) 2755 { 2756 switch (cls_flower->command) { 2757 case TC_CLSFLOWER_REPLACE: 2758 return igb_configure_clsflower(adapter, cls_flower); 2759 case TC_CLSFLOWER_DESTROY: 2760 return igb_delete_clsflower(adapter, cls_flower); 2761 case TC_CLSFLOWER_STATS: 2762 return -EOPNOTSUPP; 2763 default: 2764 return -EOPNOTSUPP; 2765 } 2766 } 2767 2768 static int igb_setup_tc_block_cb(enum tc_setup_type type, void *type_data, 2769 void *cb_priv) 2770 { 2771 struct igb_adapter *adapter = cb_priv; 2772 2773 if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data)) 2774 return -EOPNOTSUPP; 2775 2776 switch (type) { 2777 case TC_SETUP_CLSFLOWER: 2778 return igb_setup_tc_cls_flower(adapter, type_data); 2779 2780 default: 2781 return -EOPNOTSUPP; 2782 } 2783 } 2784 2785 static int igb_setup_tc_block(struct igb_adapter *adapter, 2786 struct tc_block_offload *f) 2787 { 2788 if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS) 2789 return -EOPNOTSUPP; 2790 2791 switch (f->command) { 2792 case TC_BLOCK_BIND: 2793 return tcf_block_cb_register(f->block, igb_setup_tc_block_cb, 2794 adapter, adapter, f->extack); 2795 case TC_BLOCK_UNBIND: 2796 tcf_block_cb_unregister(f->block, igb_setup_tc_block_cb, 2797 adapter); 2798 return 0; 2799 default: 2800 return -EOPNOTSUPP; 2801 } 2802 } 2803 2804 static int igb_offload_txtime(struct igb_adapter *adapter, 2805 struct tc_etf_qopt_offload *qopt) 2806 { 2807 struct e1000_hw *hw = &adapter->hw; 2808 int err; 2809 2810 /* Launchtime offloading is only supported by i210 controller. */ 2811 if (hw->mac.type != e1000_i210) 2812 return -EOPNOTSUPP; 2813 2814 /* Launchtime offloading is only supported by queues 0 and 1. */ 2815 if (qopt->queue < 0 || qopt->queue > 1) 2816 return -EINVAL; 2817 2818 err = igb_save_txtime_params(adapter, qopt->queue, qopt->enable); 2819 if (err) 2820 return err; 2821 2822 igb_offload_apply(adapter, qopt->queue); 2823 2824 return 0; 2825 } 2826 2827 static int igb_setup_tc(struct net_device *dev, enum tc_setup_type type, 2828 void *type_data) 2829 { 2830 struct igb_adapter *adapter = netdev_priv(dev); 2831 2832 switch (type) { 2833 case TC_SETUP_QDISC_CBS: 2834 return igb_offload_cbs(adapter, type_data); 2835 case TC_SETUP_BLOCK: 2836 return igb_setup_tc_block(adapter, type_data); 2837 case TC_SETUP_QDISC_ETF: 2838 return igb_offload_txtime(adapter, type_data); 2839 2840 default: 2841 return -EOPNOTSUPP; 2842 } 2843 } 2844 2845 static const struct net_device_ops igb_netdev_ops = { 2846 .ndo_open = igb_open, 2847 .ndo_stop = igb_close, 2848 .ndo_start_xmit = igb_xmit_frame, 2849 .ndo_get_stats64 = igb_get_stats64, 2850 .ndo_set_rx_mode = igb_set_rx_mode, 2851 .ndo_set_mac_address = igb_set_mac, 2852 .ndo_change_mtu = igb_change_mtu, 2853 .ndo_do_ioctl = igb_ioctl, 2854 .ndo_tx_timeout = igb_tx_timeout, 2855 .ndo_validate_addr = eth_validate_addr, 2856 .ndo_vlan_rx_add_vid = igb_vlan_rx_add_vid, 2857 .ndo_vlan_rx_kill_vid = igb_vlan_rx_kill_vid, 2858 .ndo_set_vf_mac = igb_ndo_set_vf_mac, 2859 .ndo_set_vf_vlan = igb_ndo_set_vf_vlan, 2860 .ndo_set_vf_rate = igb_ndo_set_vf_bw, 2861 .ndo_set_vf_spoofchk = igb_ndo_set_vf_spoofchk, 2862 .ndo_set_vf_trust = igb_ndo_set_vf_trust, 2863 .ndo_get_vf_config = igb_ndo_get_vf_config, 2864 .ndo_fix_features = igb_fix_features, 2865 .ndo_set_features = igb_set_features, 2866 .ndo_fdb_add = igb_ndo_fdb_add, 2867 .ndo_features_check = igb_features_check, 2868 .ndo_setup_tc = igb_setup_tc, 2869 }; 2870 2871 /** 2872 * igb_set_fw_version - Configure version string for ethtool 2873 * @adapter: adapter struct 2874 **/ 2875 void igb_set_fw_version(struct igb_adapter *adapter) 2876 { 2877 struct e1000_hw *hw = &adapter->hw; 2878 struct e1000_fw_version fw; 2879 2880 igb_get_fw_version(hw, &fw); 2881 2882 switch (hw->mac.type) { 2883 case e1000_i210: 2884 case e1000_i211: 2885 if (!(igb_get_flash_presence_i210(hw))) { 2886 snprintf(adapter->fw_version, 2887 sizeof(adapter->fw_version), 2888 "%2d.%2d-%d", 2889 fw.invm_major, fw.invm_minor, 2890 fw.invm_img_type); 2891 break; 2892 } 2893 /* fall through */ 2894 default: 2895 /* if option is rom valid, display its version too */ 2896 if (fw.or_valid) { 2897 snprintf(adapter->fw_version, 2898 sizeof(adapter->fw_version), 2899 "%d.%d, 0x%08x, %d.%d.%d", 2900 fw.eep_major, fw.eep_minor, fw.etrack_id, 2901 fw.or_major, fw.or_build, fw.or_patch); 2902 /* no option rom */ 2903 } else if (fw.etrack_id != 0X0000) { 2904 snprintf(adapter->fw_version, 2905 sizeof(adapter->fw_version), 2906 "%d.%d, 0x%08x", 2907 fw.eep_major, fw.eep_minor, fw.etrack_id); 2908 } else { 2909 snprintf(adapter->fw_version, 2910 sizeof(adapter->fw_version), 2911 "%d.%d.%d", 2912 fw.eep_major, fw.eep_minor, fw.eep_build); 2913 } 2914 break; 2915 } 2916 } 2917 2918 /** 2919 * igb_init_mas - init Media Autosense feature if enabled in the NVM 2920 * 2921 * @adapter: adapter struct 2922 **/ 2923 static void igb_init_mas(struct igb_adapter *adapter) 2924 { 2925 struct e1000_hw *hw = &adapter->hw; 2926 u16 eeprom_data; 2927 2928 hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data); 2929 switch (hw->bus.func) { 2930 case E1000_FUNC_0: 2931 if (eeprom_data & IGB_MAS_ENABLE_0) { 2932 adapter->flags |= IGB_FLAG_MAS_ENABLE; 2933 netdev_info(adapter->netdev, 2934 "MAS: Enabling Media Autosense for port %d\n", 2935 hw->bus.func); 2936 } 2937 break; 2938 case E1000_FUNC_1: 2939 if (eeprom_data & IGB_MAS_ENABLE_1) { 2940 adapter->flags |= IGB_FLAG_MAS_ENABLE; 2941 netdev_info(adapter->netdev, 2942 "MAS: Enabling Media Autosense for port %d\n", 2943 hw->bus.func); 2944 } 2945 break; 2946 case E1000_FUNC_2: 2947 if (eeprom_data & IGB_MAS_ENABLE_2) { 2948 adapter->flags |= IGB_FLAG_MAS_ENABLE; 2949 netdev_info(adapter->netdev, 2950 "MAS: Enabling Media Autosense for port %d\n", 2951 hw->bus.func); 2952 } 2953 break; 2954 case E1000_FUNC_3: 2955 if (eeprom_data & IGB_MAS_ENABLE_3) { 2956 adapter->flags |= IGB_FLAG_MAS_ENABLE; 2957 netdev_info(adapter->netdev, 2958 "MAS: Enabling Media Autosense for port %d\n", 2959 hw->bus.func); 2960 } 2961 break; 2962 default: 2963 /* Shouldn't get here */ 2964 netdev_err(adapter->netdev, 2965 "MAS: Invalid port configuration, returning\n"); 2966 break; 2967 } 2968 } 2969 2970 /** 2971 * igb_init_i2c - Init I2C interface 2972 * @adapter: pointer to adapter structure 2973 **/ 2974 static s32 igb_init_i2c(struct igb_adapter *adapter) 2975 { 2976 s32 status = 0; 2977 2978 /* I2C interface supported on i350 devices */ 2979 if (adapter->hw.mac.type != e1000_i350) 2980 return 0; 2981 2982 /* Initialize the i2c bus which is controlled by the registers. 2983 * This bus will use the i2c_algo_bit structue that implements 2984 * the protocol through toggling of the 4 bits in the register. 2985 */ 2986 adapter->i2c_adap.owner = THIS_MODULE; 2987 adapter->i2c_algo = igb_i2c_algo; 2988 adapter->i2c_algo.data = adapter; 2989 adapter->i2c_adap.algo_data = &adapter->i2c_algo; 2990 adapter->i2c_adap.dev.parent = &adapter->pdev->dev; 2991 strlcpy(adapter->i2c_adap.name, "igb BB", 2992 sizeof(adapter->i2c_adap.name)); 2993 status = i2c_bit_add_bus(&adapter->i2c_adap); 2994 return status; 2995 } 2996 2997 /** 2998 * igb_probe - Device Initialization Routine 2999 * @pdev: PCI device information struct 3000 * @ent: entry in igb_pci_tbl 3001 * 3002 * Returns 0 on success, negative on failure 3003 * 3004 * igb_probe initializes an adapter identified by a pci_dev structure. 3005 * The OS initialization, configuring of the adapter private structure, 3006 * and a hardware reset occur. 3007 **/ 3008 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 3009 { 3010 struct net_device *netdev; 3011 struct igb_adapter *adapter; 3012 struct e1000_hw *hw; 3013 u16 eeprom_data = 0; 3014 s32 ret_val; 3015 static int global_quad_port_a; /* global quad port a indication */ 3016 const struct e1000_info *ei = igb_info_tbl[ent->driver_data]; 3017 int err, pci_using_dac; 3018 u8 part_str[E1000_PBANUM_LENGTH]; 3019 3020 /* Catch broken hardware that put the wrong VF device ID in 3021 * the PCIe SR-IOV capability. 3022 */ 3023 if (pdev->is_virtfn) { 3024 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n", 3025 pci_name(pdev), pdev->vendor, pdev->device); 3026 return -EINVAL; 3027 } 3028 3029 err = pci_enable_device_mem(pdev); 3030 if (err) 3031 return err; 3032 3033 pci_using_dac = 0; 3034 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); 3035 if (!err) { 3036 pci_using_dac = 1; 3037 } else { 3038 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); 3039 if (err) { 3040 dev_err(&pdev->dev, 3041 "No usable DMA configuration, aborting\n"); 3042 goto err_dma; 3043 } 3044 } 3045 3046 err = pci_request_mem_regions(pdev, igb_driver_name); 3047 if (err) 3048 goto err_pci_reg; 3049 3050 pci_enable_pcie_error_reporting(pdev); 3051 3052 pci_set_master(pdev); 3053 pci_save_state(pdev); 3054 3055 err = -ENOMEM; 3056 netdev = alloc_etherdev_mq(sizeof(struct igb_adapter), 3057 IGB_MAX_TX_QUEUES); 3058 if (!netdev) 3059 goto err_alloc_etherdev; 3060 3061 SET_NETDEV_DEV(netdev, &pdev->dev); 3062 3063 pci_set_drvdata(pdev, netdev); 3064 adapter = netdev_priv(netdev); 3065 adapter->netdev = netdev; 3066 adapter->pdev = pdev; 3067 hw = &adapter->hw; 3068 hw->back = adapter; 3069 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE); 3070 3071 err = -EIO; 3072 adapter->io_addr = pci_iomap(pdev, 0, 0); 3073 if (!adapter->io_addr) 3074 goto err_ioremap; 3075 /* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */ 3076 hw->hw_addr = adapter->io_addr; 3077 3078 netdev->netdev_ops = &igb_netdev_ops; 3079 igb_set_ethtool_ops(netdev); 3080 netdev->watchdog_timeo = 5 * HZ; 3081 3082 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1); 3083 3084 netdev->mem_start = pci_resource_start(pdev, 0); 3085 netdev->mem_end = pci_resource_end(pdev, 0); 3086 3087 /* PCI config space info */ 3088 hw->vendor_id = pdev->vendor; 3089 hw->device_id = pdev->device; 3090 hw->revision_id = pdev->revision; 3091 hw->subsystem_vendor_id = pdev->subsystem_vendor; 3092 hw->subsystem_device_id = pdev->subsystem_device; 3093 3094 /* Copy the default MAC, PHY and NVM function pointers */ 3095 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops)); 3096 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops)); 3097 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops)); 3098 /* Initialize skew-specific constants */ 3099 err = ei->get_invariants(hw); 3100 if (err) 3101 goto err_sw_init; 3102 3103 /* setup the private structure */ 3104 err = igb_sw_init(adapter); 3105 if (err) 3106 goto err_sw_init; 3107 3108 igb_get_bus_info_pcie(hw); 3109 3110 hw->phy.autoneg_wait_to_complete = false; 3111 3112 /* Copper options */ 3113 if (hw->phy.media_type == e1000_media_type_copper) { 3114 hw->phy.mdix = AUTO_ALL_MODES; 3115 hw->phy.disable_polarity_correction = false; 3116 hw->phy.ms_type = e1000_ms_hw_default; 3117 } 3118 3119 if (igb_check_reset_block(hw)) 3120 dev_info(&pdev->dev, 3121 "PHY reset is blocked due to SOL/IDER session.\n"); 3122 3123 /* features is initialized to 0 in allocation, it might have bits 3124 * set by igb_sw_init so we should use an or instead of an 3125 * assignment. 3126 */ 3127 netdev->features |= NETIF_F_SG | 3128 NETIF_F_TSO | 3129 NETIF_F_TSO6 | 3130 NETIF_F_RXHASH | 3131 NETIF_F_RXCSUM | 3132 NETIF_F_HW_CSUM; 3133 3134 if (hw->mac.type >= e1000_82576) 3135 netdev->features |= NETIF_F_SCTP_CRC; 3136 3137 if (hw->mac.type >= e1000_i350) 3138 netdev->features |= NETIF_F_HW_TC; 3139 3140 #define IGB_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \ 3141 NETIF_F_GSO_GRE_CSUM | \ 3142 NETIF_F_GSO_IPXIP4 | \ 3143 NETIF_F_GSO_IPXIP6 | \ 3144 NETIF_F_GSO_UDP_TUNNEL | \ 3145 NETIF_F_GSO_UDP_TUNNEL_CSUM) 3146 3147 netdev->gso_partial_features = IGB_GSO_PARTIAL_FEATURES; 3148 netdev->features |= NETIF_F_GSO_PARTIAL | IGB_GSO_PARTIAL_FEATURES; 3149 3150 /* copy netdev features into list of user selectable features */ 3151 netdev->hw_features |= netdev->features | 3152 NETIF_F_HW_VLAN_CTAG_RX | 3153 NETIF_F_HW_VLAN_CTAG_TX | 3154 NETIF_F_RXALL; 3155 3156 if (hw->mac.type >= e1000_i350) 3157 netdev->hw_features |= NETIF_F_NTUPLE; 3158 3159 if (pci_using_dac) 3160 netdev->features |= NETIF_F_HIGHDMA; 3161 3162 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID; 3163 netdev->mpls_features |= NETIF_F_HW_CSUM; 3164 netdev->hw_enc_features |= netdev->vlan_features; 3165 3166 /* set this bit last since it cannot be part of vlan_features */ 3167 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER | 3168 NETIF_F_HW_VLAN_CTAG_RX | 3169 NETIF_F_HW_VLAN_CTAG_TX; 3170 3171 netdev->priv_flags |= IFF_SUPP_NOFCS; 3172 3173 netdev->priv_flags |= IFF_UNICAST_FLT; 3174 3175 /* MTU range: 68 - 9216 */ 3176 netdev->min_mtu = ETH_MIN_MTU; 3177 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE; 3178 3179 adapter->en_mng_pt = igb_enable_mng_pass_thru(hw); 3180 3181 /* before reading the NVM, reset the controller to put the device in a 3182 * known good starting state 3183 */ 3184 hw->mac.ops.reset_hw(hw); 3185 3186 /* make sure the NVM is good , i211/i210 parts can have special NVM 3187 * that doesn't contain a checksum 3188 */ 3189 switch (hw->mac.type) { 3190 case e1000_i210: 3191 case e1000_i211: 3192 if (igb_get_flash_presence_i210(hw)) { 3193 if (hw->nvm.ops.validate(hw) < 0) { 3194 dev_err(&pdev->dev, 3195 "The NVM Checksum Is Not Valid\n"); 3196 err = -EIO; 3197 goto err_eeprom; 3198 } 3199 } 3200 break; 3201 default: 3202 if (hw->nvm.ops.validate(hw) < 0) { 3203 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n"); 3204 err = -EIO; 3205 goto err_eeprom; 3206 } 3207 break; 3208 } 3209 3210 if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) { 3211 /* copy the MAC address out of the NVM */ 3212 if (hw->mac.ops.read_mac_addr(hw)) 3213 dev_err(&pdev->dev, "NVM Read Error\n"); 3214 } 3215 3216 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len); 3217 3218 if (!is_valid_ether_addr(netdev->dev_addr)) { 3219 dev_err(&pdev->dev, "Invalid MAC Address\n"); 3220 err = -EIO; 3221 goto err_eeprom; 3222 } 3223 3224 igb_set_default_mac_filter(adapter); 3225 3226 /* get firmware version for ethtool -i */ 3227 igb_set_fw_version(adapter); 3228 3229 /* configure RXPBSIZE and TXPBSIZE */ 3230 if (hw->mac.type == e1000_i210) { 3231 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT); 3232 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT); 3233 } 3234 3235 timer_setup(&adapter->watchdog_timer, igb_watchdog, 0); 3236 timer_setup(&adapter->phy_info_timer, igb_update_phy_info, 0); 3237 3238 INIT_WORK(&adapter->reset_task, igb_reset_task); 3239 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task); 3240 3241 /* Initialize link properties that are user-changeable */ 3242 adapter->fc_autoneg = true; 3243 hw->mac.autoneg = true; 3244 hw->phy.autoneg_advertised = 0x2f; 3245 3246 hw->fc.requested_mode = e1000_fc_default; 3247 hw->fc.current_mode = e1000_fc_default; 3248 3249 igb_validate_mdi_setting(hw); 3250 3251 /* By default, support wake on port A */ 3252 if (hw->bus.func == 0) 3253 adapter->flags |= IGB_FLAG_WOL_SUPPORTED; 3254 3255 /* Check the NVM for wake support on non-port A ports */ 3256 if (hw->mac.type >= e1000_82580) 3257 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A + 3258 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1, 3259 &eeprom_data); 3260 else if (hw->bus.func == 1) 3261 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data); 3262 3263 if (eeprom_data & IGB_EEPROM_APME) 3264 adapter->flags |= IGB_FLAG_WOL_SUPPORTED; 3265 3266 /* now that we have the eeprom settings, apply the special cases where 3267 * the eeprom may be wrong or the board simply won't support wake on 3268 * lan on a particular port 3269 */ 3270 switch (pdev->device) { 3271 case E1000_DEV_ID_82575GB_QUAD_COPPER: 3272 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED; 3273 break; 3274 case E1000_DEV_ID_82575EB_FIBER_SERDES: 3275 case E1000_DEV_ID_82576_FIBER: 3276 case E1000_DEV_ID_82576_SERDES: 3277 /* Wake events only supported on port A for dual fiber 3278 * regardless of eeprom setting 3279 */ 3280 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1) 3281 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED; 3282 break; 3283 case E1000_DEV_ID_82576_QUAD_COPPER: 3284 case E1000_DEV_ID_82576_QUAD_COPPER_ET2: 3285 /* if quad port adapter, disable WoL on all but port A */ 3286 if (global_quad_port_a != 0) 3287 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED; 3288 else 3289 adapter->flags |= IGB_FLAG_QUAD_PORT_A; 3290 /* Reset for multiple quad port adapters */ 3291 if (++global_quad_port_a == 4) 3292 global_quad_port_a = 0; 3293 break; 3294 default: 3295 /* If the device can't wake, don't set software support */ 3296 if (!device_can_wakeup(&adapter->pdev->dev)) 3297 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED; 3298 } 3299 3300 /* initialize the wol settings based on the eeprom settings */ 3301 if (adapter->flags & IGB_FLAG_WOL_SUPPORTED) 3302 adapter->wol |= E1000_WUFC_MAG; 3303 3304 /* Some vendors want WoL disabled by default, but still supported */ 3305 if ((hw->mac.type == e1000_i350) && 3306 (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) { 3307 adapter->flags |= IGB_FLAG_WOL_SUPPORTED; 3308 adapter->wol = 0; 3309 } 3310 3311 /* Some vendors want the ability to Use the EEPROM setting as 3312 * enable/disable only, and not for capability 3313 */ 3314 if (((hw->mac.type == e1000_i350) || 3315 (hw->mac.type == e1000_i354)) && 3316 (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) { 3317 adapter->flags |= IGB_FLAG_WOL_SUPPORTED; 3318 adapter->wol = 0; 3319 } 3320 if (hw->mac.type == e1000_i350) { 3321 if (((pdev->subsystem_device == 0x5001) || 3322 (pdev->subsystem_device == 0x5002)) && 3323 (hw->bus.func == 0)) { 3324 adapter->flags |= IGB_FLAG_WOL_SUPPORTED; 3325 adapter->wol = 0; 3326 } 3327 if (pdev->subsystem_device == 0x1F52) 3328 adapter->flags |= IGB_FLAG_WOL_SUPPORTED; 3329 } 3330 3331 device_set_wakeup_enable(&adapter->pdev->dev, 3332 adapter->flags & IGB_FLAG_WOL_SUPPORTED); 3333 3334 /* reset the hardware with the new settings */ 3335 igb_reset(adapter); 3336 3337 /* Init the I2C interface */ 3338 err = igb_init_i2c(adapter); 3339 if (err) { 3340 dev_err(&pdev->dev, "failed to init i2c interface\n"); 3341 goto err_eeprom; 3342 } 3343 3344 /* let the f/w know that the h/w is now under the control of the 3345 * driver. 3346 */ 3347 igb_get_hw_control(adapter); 3348 3349 strcpy(netdev->name, "eth%d"); 3350 err = register_netdev(netdev); 3351 if (err) 3352 goto err_register; 3353 3354 /* carrier off reporting is important to ethtool even BEFORE open */ 3355 netif_carrier_off(netdev); 3356 3357 #ifdef CONFIG_IGB_DCA 3358 if (dca_add_requester(&pdev->dev) == 0) { 3359 adapter->flags |= IGB_FLAG_DCA_ENABLED; 3360 dev_info(&pdev->dev, "DCA enabled\n"); 3361 igb_setup_dca(adapter); 3362 } 3363 3364 #endif 3365 #ifdef CONFIG_IGB_HWMON 3366 /* Initialize the thermal sensor on i350 devices. */ 3367 if (hw->mac.type == e1000_i350 && hw->bus.func == 0) { 3368 u16 ets_word; 3369 3370 /* Read the NVM to determine if this i350 device supports an 3371 * external thermal sensor. 3372 */ 3373 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word); 3374 if (ets_word != 0x0000 && ets_word != 0xFFFF) 3375 adapter->ets = true; 3376 else 3377 adapter->ets = false; 3378 if (igb_sysfs_init(adapter)) 3379 dev_err(&pdev->dev, 3380 "failed to allocate sysfs resources\n"); 3381 } else { 3382 adapter->ets = false; 3383 } 3384 #endif 3385 /* Check if Media Autosense is enabled */ 3386 adapter->ei = *ei; 3387 if (hw->dev_spec._82575.mas_capable) 3388 igb_init_mas(adapter); 3389 3390 /* do hw tstamp init after resetting */ 3391 igb_ptp_init(adapter); 3392 3393 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n"); 3394 /* print bus type/speed/width info, not applicable to i354 */ 3395 if (hw->mac.type != e1000_i354) { 3396 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n", 3397 netdev->name, 3398 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" : 3399 (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" : 3400 "unknown"), 3401 ((hw->bus.width == e1000_bus_width_pcie_x4) ? 3402 "Width x4" : 3403 (hw->bus.width == e1000_bus_width_pcie_x2) ? 3404 "Width x2" : 3405 (hw->bus.width == e1000_bus_width_pcie_x1) ? 3406 "Width x1" : "unknown"), netdev->dev_addr); 3407 } 3408 3409 if ((hw->mac.type >= e1000_i210 || 3410 igb_get_flash_presence_i210(hw))) { 3411 ret_val = igb_read_part_string(hw, part_str, 3412 E1000_PBANUM_LENGTH); 3413 } else { 3414 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND; 3415 } 3416 3417 if (ret_val) 3418 strcpy(part_str, "Unknown"); 3419 dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str); 3420 dev_info(&pdev->dev, 3421 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n", 3422 (adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" : 3423 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy", 3424 adapter->num_rx_queues, adapter->num_tx_queues); 3425 if (hw->phy.media_type == e1000_media_type_copper) { 3426 switch (hw->mac.type) { 3427 case e1000_i350: 3428 case e1000_i210: 3429 case e1000_i211: 3430 /* Enable EEE for internal copper PHY devices */ 3431 err = igb_set_eee_i350(hw, true, true); 3432 if ((!err) && 3433 (!hw->dev_spec._82575.eee_disable)) { 3434 adapter->eee_advert = 3435 MDIO_EEE_100TX | MDIO_EEE_1000T; 3436 adapter->flags |= IGB_FLAG_EEE; 3437 } 3438 break; 3439 case e1000_i354: 3440 if ((rd32(E1000_CTRL_EXT) & 3441 E1000_CTRL_EXT_LINK_MODE_SGMII)) { 3442 err = igb_set_eee_i354(hw, true, true); 3443 if ((!err) && 3444 (!hw->dev_spec._82575.eee_disable)) { 3445 adapter->eee_advert = 3446 MDIO_EEE_100TX | MDIO_EEE_1000T; 3447 adapter->flags |= IGB_FLAG_EEE; 3448 } 3449 } 3450 break; 3451 default: 3452 break; 3453 } 3454 } 3455 pm_runtime_put_noidle(&pdev->dev); 3456 return 0; 3457 3458 err_register: 3459 igb_release_hw_control(adapter); 3460 memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap)); 3461 err_eeprom: 3462 if (!igb_check_reset_block(hw)) 3463 igb_reset_phy(hw); 3464 3465 if (hw->flash_address) 3466 iounmap(hw->flash_address); 3467 err_sw_init: 3468 kfree(adapter->mac_table); 3469 kfree(adapter->shadow_vfta); 3470 igb_clear_interrupt_scheme(adapter); 3471 #ifdef CONFIG_PCI_IOV 3472 igb_disable_sriov(pdev); 3473 #endif 3474 pci_iounmap(pdev, adapter->io_addr); 3475 err_ioremap: 3476 free_netdev(netdev); 3477 err_alloc_etherdev: 3478 pci_release_mem_regions(pdev); 3479 err_pci_reg: 3480 err_dma: 3481 pci_disable_device(pdev); 3482 return err; 3483 } 3484 3485 #ifdef CONFIG_PCI_IOV 3486 static int igb_disable_sriov(struct pci_dev *pdev) 3487 { 3488 struct net_device *netdev = pci_get_drvdata(pdev); 3489 struct igb_adapter *adapter = netdev_priv(netdev); 3490 struct e1000_hw *hw = &adapter->hw; 3491 3492 /* reclaim resources allocated to VFs */ 3493 if (adapter->vf_data) { 3494 /* disable iov and allow time for transactions to clear */ 3495 if (pci_vfs_assigned(pdev)) { 3496 dev_warn(&pdev->dev, 3497 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n"); 3498 return -EPERM; 3499 } else { 3500 pci_disable_sriov(pdev); 3501 msleep(500); 3502 } 3503 3504 kfree(adapter->vf_mac_list); 3505 adapter->vf_mac_list = NULL; 3506 kfree(adapter->vf_data); 3507 adapter->vf_data = NULL; 3508 adapter->vfs_allocated_count = 0; 3509 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ); 3510 wrfl(); 3511 msleep(100); 3512 dev_info(&pdev->dev, "IOV Disabled\n"); 3513 3514 /* Re-enable DMA Coalescing flag since IOV is turned off */ 3515 adapter->flags |= IGB_FLAG_DMAC; 3516 } 3517 3518 return 0; 3519 } 3520 3521 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs) 3522 { 3523 struct net_device *netdev = pci_get_drvdata(pdev); 3524 struct igb_adapter *adapter = netdev_priv(netdev); 3525 int old_vfs = pci_num_vf(pdev); 3526 struct vf_mac_filter *mac_list; 3527 int err = 0; 3528 int num_vf_mac_filters, i; 3529 3530 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) { 3531 err = -EPERM; 3532 goto out; 3533 } 3534 if (!num_vfs) 3535 goto out; 3536 3537 if (old_vfs) { 3538 dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n", 3539 old_vfs, max_vfs); 3540 adapter->vfs_allocated_count = old_vfs; 3541 } else 3542 adapter->vfs_allocated_count = num_vfs; 3543 3544 adapter->vf_data = kcalloc(adapter->vfs_allocated_count, 3545 sizeof(struct vf_data_storage), GFP_KERNEL); 3546 3547 /* if allocation failed then we do not support SR-IOV */ 3548 if (!adapter->vf_data) { 3549 adapter->vfs_allocated_count = 0; 3550 err = -ENOMEM; 3551 goto out; 3552 } 3553 3554 /* Due to the limited number of RAR entries calculate potential 3555 * number of MAC filters available for the VFs. Reserve entries 3556 * for PF default MAC, PF MAC filters and at least one RAR entry 3557 * for each VF for VF MAC. 3558 */ 3559 num_vf_mac_filters = adapter->hw.mac.rar_entry_count - 3560 (1 + IGB_PF_MAC_FILTERS_RESERVED + 3561 adapter->vfs_allocated_count); 3562 3563 adapter->vf_mac_list = kcalloc(num_vf_mac_filters, 3564 sizeof(struct vf_mac_filter), 3565 GFP_KERNEL); 3566 3567 mac_list = adapter->vf_mac_list; 3568 INIT_LIST_HEAD(&adapter->vf_macs.l); 3569 3570 if (adapter->vf_mac_list) { 3571 /* Initialize list of VF MAC filters */ 3572 for (i = 0; i < num_vf_mac_filters; i++) { 3573 mac_list->vf = -1; 3574 mac_list->free = true; 3575 list_add(&mac_list->l, &adapter->vf_macs.l); 3576 mac_list++; 3577 } 3578 } else { 3579 /* If we could not allocate memory for the VF MAC filters 3580 * we can continue without this feature but warn user. 3581 */ 3582 dev_err(&pdev->dev, 3583 "Unable to allocate memory for VF MAC filter list\n"); 3584 } 3585 3586 /* only call pci_enable_sriov() if no VFs are allocated already */ 3587 if (!old_vfs) { 3588 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count); 3589 if (err) 3590 goto err_out; 3591 } 3592 dev_info(&pdev->dev, "%d VFs allocated\n", 3593 adapter->vfs_allocated_count); 3594 for (i = 0; i < adapter->vfs_allocated_count; i++) 3595 igb_vf_configure(adapter, i); 3596 3597 /* DMA Coalescing is not supported in IOV mode. */ 3598 adapter->flags &= ~IGB_FLAG_DMAC; 3599 goto out; 3600 3601 err_out: 3602 kfree(adapter->vf_mac_list); 3603 adapter->vf_mac_list = NULL; 3604 kfree(adapter->vf_data); 3605 adapter->vf_data = NULL; 3606 adapter->vfs_allocated_count = 0; 3607 out: 3608 return err; 3609 } 3610 3611 #endif 3612 /** 3613 * igb_remove_i2c - Cleanup I2C interface 3614 * @adapter: pointer to adapter structure 3615 **/ 3616 static void igb_remove_i2c(struct igb_adapter *adapter) 3617 { 3618 /* free the adapter bus structure */ 3619 i2c_del_adapter(&adapter->i2c_adap); 3620 } 3621 3622 /** 3623 * igb_remove - Device Removal Routine 3624 * @pdev: PCI device information struct 3625 * 3626 * igb_remove is called by the PCI subsystem to alert the driver 3627 * that it should release a PCI device. The could be caused by a 3628 * Hot-Plug event, or because the driver is going to be removed from 3629 * memory. 3630 **/ 3631 static void igb_remove(struct pci_dev *pdev) 3632 { 3633 struct net_device *netdev = pci_get_drvdata(pdev); 3634 struct igb_adapter *adapter = netdev_priv(netdev); 3635 struct e1000_hw *hw = &adapter->hw; 3636 3637 pm_runtime_get_noresume(&pdev->dev); 3638 #ifdef CONFIG_IGB_HWMON 3639 igb_sysfs_exit(adapter); 3640 #endif 3641 igb_remove_i2c(adapter); 3642 igb_ptp_stop(adapter); 3643 /* The watchdog timer may be rescheduled, so explicitly 3644 * disable watchdog from being rescheduled. 3645 */ 3646 set_bit(__IGB_DOWN, &adapter->state); 3647 del_timer_sync(&adapter->watchdog_timer); 3648 del_timer_sync(&adapter->phy_info_timer); 3649 3650 cancel_work_sync(&adapter->reset_task); 3651 cancel_work_sync(&adapter->watchdog_task); 3652 3653 #ifdef CONFIG_IGB_DCA 3654 if (adapter->flags & IGB_FLAG_DCA_ENABLED) { 3655 dev_info(&pdev->dev, "DCA disabled\n"); 3656 dca_remove_requester(&pdev->dev); 3657 adapter->flags &= ~IGB_FLAG_DCA_ENABLED; 3658 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE); 3659 } 3660 #endif 3661 3662 /* Release control of h/w to f/w. If f/w is AMT enabled, this 3663 * would have already happened in close and is redundant. 3664 */ 3665 igb_release_hw_control(adapter); 3666 3667 #ifdef CONFIG_PCI_IOV 3668 igb_disable_sriov(pdev); 3669 #endif 3670 3671 unregister_netdev(netdev); 3672 3673 igb_clear_interrupt_scheme(adapter); 3674 3675 pci_iounmap(pdev, adapter->io_addr); 3676 if (hw->flash_address) 3677 iounmap(hw->flash_address); 3678 pci_release_mem_regions(pdev); 3679 3680 kfree(adapter->mac_table); 3681 kfree(adapter->shadow_vfta); 3682 free_netdev(netdev); 3683 3684 pci_disable_pcie_error_reporting(pdev); 3685 3686 pci_disable_device(pdev); 3687 } 3688 3689 /** 3690 * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space 3691 * @adapter: board private structure to initialize 3692 * 3693 * This function initializes the vf specific data storage and then attempts to 3694 * allocate the VFs. The reason for ordering it this way is because it is much 3695 * mor expensive time wise to disable SR-IOV than it is to allocate and free 3696 * the memory for the VFs. 3697 **/ 3698 static void igb_probe_vfs(struct igb_adapter *adapter) 3699 { 3700 #ifdef CONFIG_PCI_IOV 3701 struct pci_dev *pdev = adapter->pdev; 3702 struct e1000_hw *hw = &adapter->hw; 3703 3704 /* Virtualization features not supported on i210 family. */ 3705 if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211)) 3706 return; 3707 3708 /* Of the below we really only want the effect of getting 3709 * IGB_FLAG_HAS_MSIX set (if available), without which 3710 * igb_enable_sriov() has no effect. 3711 */ 3712 igb_set_interrupt_capability(adapter, true); 3713 igb_reset_interrupt_capability(adapter); 3714 3715 pci_sriov_set_totalvfs(pdev, 7); 3716 igb_enable_sriov(pdev, max_vfs); 3717 3718 #endif /* CONFIG_PCI_IOV */ 3719 } 3720 3721 unsigned int igb_get_max_rss_queues(struct igb_adapter *adapter) 3722 { 3723 struct e1000_hw *hw = &adapter->hw; 3724 unsigned int max_rss_queues; 3725 3726 /* Determine the maximum number of RSS queues supported. */ 3727 switch (hw->mac.type) { 3728 case e1000_i211: 3729 max_rss_queues = IGB_MAX_RX_QUEUES_I211; 3730 break; 3731 case e1000_82575: 3732 case e1000_i210: 3733 max_rss_queues = IGB_MAX_RX_QUEUES_82575; 3734 break; 3735 case e1000_i350: 3736 /* I350 cannot do RSS and SR-IOV at the same time */ 3737 if (!!adapter->vfs_allocated_count) { 3738 max_rss_queues = 1; 3739 break; 3740 } 3741 /* fall through */ 3742 case e1000_82576: 3743 if (!!adapter->vfs_allocated_count) { 3744 max_rss_queues = 2; 3745 break; 3746 } 3747 /* fall through */ 3748 case e1000_82580: 3749 case e1000_i354: 3750 default: 3751 max_rss_queues = IGB_MAX_RX_QUEUES; 3752 break; 3753 } 3754 3755 return max_rss_queues; 3756 } 3757 3758 static void igb_init_queue_configuration(struct igb_adapter *adapter) 3759 { 3760 u32 max_rss_queues; 3761 3762 max_rss_queues = igb_get_max_rss_queues(adapter); 3763 adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus()); 3764 3765 igb_set_flag_queue_pairs(adapter, max_rss_queues); 3766 } 3767 3768 void igb_set_flag_queue_pairs(struct igb_adapter *adapter, 3769 const u32 max_rss_queues) 3770 { 3771 struct e1000_hw *hw = &adapter->hw; 3772 3773 /* Determine if we need to pair queues. */ 3774 switch (hw->mac.type) { 3775 case e1000_82575: 3776 case e1000_i211: 3777 /* Device supports enough interrupts without queue pairing. */ 3778 break; 3779 case e1000_82576: 3780 case e1000_82580: 3781 case e1000_i350: 3782 case e1000_i354: 3783 case e1000_i210: 3784 default: 3785 /* If rss_queues > half of max_rss_queues, pair the queues in 3786 * order to conserve interrupts due to limited supply. 3787 */ 3788 if (adapter->rss_queues > (max_rss_queues / 2)) 3789 adapter->flags |= IGB_FLAG_QUEUE_PAIRS; 3790 else 3791 adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS; 3792 break; 3793 } 3794 } 3795 3796 /** 3797 * igb_sw_init - Initialize general software structures (struct igb_adapter) 3798 * @adapter: board private structure to initialize 3799 * 3800 * igb_sw_init initializes the Adapter private data structure. 3801 * Fields are initialized based on PCI device information and 3802 * OS network device settings (MTU size). 3803 **/ 3804 static int igb_sw_init(struct igb_adapter *adapter) 3805 { 3806 struct e1000_hw *hw = &adapter->hw; 3807 struct net_device *netdev = adapter->netdev; 3808 struct pci_dev *pdev = adapter->pdev; 3809 3810 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word); 3811 3812 /* set default ring sizes */ 3813 adapter->tx_ring_count = IGB_DEFAULT_TXD; 3814 adapter->rx_ring_count = IGB_DEFAULT_RXD; 3815 3816 /* set default ITR values */ 3817 adapter->rx_itr_setting = IGB_DEFAULT_ITR; 3818 adapter->tx_itr_setting = IGB_DEFAULT_ITR; 3819 3820 /* set default work limits */ 3821 adapter->tx_work_limit = IGB_DEFAULT_TX_WORK; 3822 3823 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN + 3824 VLAN_HLEN; 3825 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN; 3826 3827 spin_lock_init(&adapter->nfc_lock); 3828 spin_lock_init(&adapter->stats64_lock); 3829 #ifdef CONFIG_PCI_IOV 3830 switch (hw->mac.type) { 3831 case e1000_82576: 3832 case e1000_i350: 3833 if (max_vfs > 7) { 3834 dev_warn(&pdev->dev, 3835 "Maximum of 7 VFs per PF, using max\n"); 3836 max_vfs = adapter->vfs_allocated_count = 7; 3837 } else 3838 adapter->vfs_allocated_count = max_vfs; 3839 if (adapter->vfs_allocated_count) 3840 dev_warn(&pdev->dev, 3841 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n"); 3842 break; 3843 default: 3844 break; 3845 } 3846 #endif /* CONFIG_PCI_IOV */ 3847 3848 /* Assume MSI-X interrupts, will be checked during IRQ allocation */ 3849 adapter->flags |= IGB_FLAG_HAS_MSIX; 3850 3851 adapter->mac_table = kcalloc(hw->mac.rar_entry_count, 3852 sizeof(struct igb_mac_addr), 3853 GFP_KERNEL); 3854 if (!adapter->mac_table) 3855 return -ENOMEM; 3856 3857 igb_probe_vfs(adapter); 3858 3859 igb_init_queue_configuration(adapter); 3860 3861 /* Setup and initialize a copy of the hw vlan table array */ 3862 adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32), 3863 GFP_KERNEL); 3864 if (!adapter->shadow_vfta) 3865 return -ENOMEM; 3866 3867 /* This call may decrease the number of queues */ 3868 if (igb_init_interrupt_scheme(adapter, true)) { 3869 dev_err(&pdev->dev, "Unable to allocate memory for queues\n"); 3870 return -ENOMEM; 3871 } 3872 3873 /* Explicitly disable IRQ since the NIC can be in any state. */ 3874 igb_irq_disable(adapter); 3875 3876 if (hw->mac.type >= e1000_i350) 3877 adapter->flags &= ~IGB_FLAG_DMAC; 3878 3879 set_bit(__IGB_DOWN, &adapter->state); 3880 return 0; 3881 } 3882 3883 /** 3884 * igb_open - Called when a network interface is made active 3885 * @netdev: network interface device structure 3886 * 3887 * Returns 0 on success, negative value on failure 3888 * 3889 * The open entry point is called when a network interface is made 3890 * active by the system (IFF_UP). At this point all resources needed 3891 * for transmit and receive operations are allocated, the interrupt 3892 * handler is registered with the OS, the watchdog timer is started, 3893 * and the stack is notified that the interface is ready. 3894 **/ 3895 static int __igb_open(struct net_device *netdev, bool resuming) 3896 { 3897 struct igb_adapter *adapter = netdev_priv(netdev); 3898 struct e1000_hw *hw = &adapter->hw; 3899 struct pci_dev *pdev = adapter->pdev; 3900 int err; 3901 int i; 3902 3903 /* disallow open during test */ 3904 if (test_bit(__IGB_TESTING, &adapter->state)) { 3905 WARN_ON(resuming); 3906 return -EBUSY; 3907 } 3908 3909 if (!resuming) 3910 pm_runtime_get_sync(&pdev->dev); 3911 3912 netif_carrier_off(netdev); 3913 3914 /* allocate transmit descriptors */ 3915 err = igb_setup_all_tx_resources(adapter); 3916 if (err) 3917 goto err_setup_tx; 3918 3919 /* allocate receive descriptors */ 3920 err = igb_setup_all_rx_resources(adapter); 3921 if (err) 3922 goto err_setup_rx; 3923 3924 igb_power_up_link(adapter); 3925 3926 /* before we allocate an interrupt, we must be ready to handle it. 3927 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt 3928 * as soon as we call pci_request_irq, so we have to setup our 3929 * clean_rx handler before we do so. 3930 */ 3931 igb_configure(adapter); 3932 3933 err = igb_request_irq(adapter); 3934 if (err) 3935 goto err_req_irq; 3936 3937 /* Notify the stack of the actual queue counts. */ 3938 err = netif_set_real_num_tx_queues(adapter->netdev, 3939 adapter->num_tx_queues); 3940 if (err) 3941 goto err_set_queues; 3942 3943 err = netif_set_real_num_rx_queues(adapter->netdev, 3944 adapter->num_rx_queues); 3945 if (err) 3946 goto err_set_queues; 3947 3948 /* From here on the code is the same as igb_up() */ 3949 clear_bit(__IGB_DOWN, &adapter->state); 3950 3951 for (i = 0; i < adapter->num_q_vectors; i++) 3952 napi_enable(&(adapter->q_vector[i]->napi)); 3953 3954 /* Clear any pending interrupts. */ 3955 rd32(E1000_TSICR); 3956 rd32(E1000_ICR); 3957 3958 igb_irq_enable(adapter); 3959 3960 /* notify VFs that reset has been completed */ 3961 if (adapter->vfs_allocated_count) { 3962 u32 reg_data = rd32(E1000_CTRL_EXT); 3963 3964 reg_data |= E1000_CTRL_EXT_PFRSTD; 3965 wr32(E1000_CTRL_EXT, reg_data); 3966 } 3967 3968 netif_tx_start_all_queues(netdev); 3969 3970 if (!resuming) 3971 pm_runtime_put(&pdev->dev); 3972 3973 /* start the watchdog. */ 3974 hw->mac.get_link_status = 1; 3975 schedule_work(&adapter->watchdog_task); 3976 3977 return 0; 3978 3979 err_set_queues: 3980 igb_free_irq(adapter); 3981 err_req_irq: 3982 igb_release_hw_control(adapter); 3983 igb_power_down_link(adapter); 3984 igb_free_all_rx_resources(adapter); 3985 err_setup_rx: 3986 igb_free_all_tx_resources(adapter); 3987 err_setup_tx: 3988 igb_reset(adapter); 3989 if (!resuming) 3990 pm_runtime_put(&pdev->dev); 3991 3992 return err; 3993 } 3994 3995 int igb_open(struct net_device *netdev) 3996 { 3997 return __igb_open(netdev, false); 3998 } 3999 4000 /** 4001 * igb_close - Disables a network interface 4002 * @netdev: network interface device structure 4003 * 4004 * Returns 0, this is not allowed to fail 4005 * 4006 * The close entry point is called when an interface is de-activated 4007 * by the OS. The hardware is still under the driver's control, but 4008 * needs to be disabled. A global MAC reset is issued to stop the 4009 * hardware, and all transmit and receive resources are freed. 4010 **/ 4011 static int __igb_close(struct net_device *netdev, bool suspending) 4012 { 4013 struct igb_adapter *adapter = netdev_priv(netdev); 4014 struct pci_dev *pdev = adapter->pdev; 4015 4016 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state)); 4017 4018 if (!suspending) 4019 pm_runtime_get_sync(&pdev->dev); 4020 4021 igb_down(adapter); 4022 igb_free_irq(adapter); 4023 4024 igb_free_all_tx_resources(adapter); 4025 igb_free_all_rx_resources(adapter); 4026 4027 if (!suspending) 4028 pm_runtime_put_sync(&pdev->dev); 4029 return 0; 4030 } 4031 4032 int igb_close(struct net_device *netdev) 4033 { 4034 if (netif_device_present(netdev) || netdev->dismantle) 4035 return __igb_close(netdev, false); 4036 return 0; 4037 } 4038 4039 /** 4040 * igb_setup_tx_resources - allocate Tx resources (Descriptors) 4041 * @tx_ring: tx descriptor ring (for a specific queue) to setup 4042 * 4043 * Return 0 on success, negative on failure 4044 **/ 4045 int igb_setup_tx_resources(struct igb_ring *tx_ring) 4046 { 4047 struct device *dev = tx_ring->dev; 4048 int size; 4049 4050 size = sizeof(struct igb_tx_buffer) * tx_ring->count; 4051 4052 tx_ring->tx_buffer_info = vmalloc(size); 4053 if (!tx_ring->tx_buffer_info) 4054 goto err; 4055 4056 /* round up to nearest 4K */ 4057 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc); 4058 tx_ring->size = ALIGN(tx_ring->size, 4096); 4059 4060 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size, 4061 &tx_ring->dma, GFP_KERNEL); 4062 if (!tx_ring->desc) 4063 goto err; 4064 4065 tx_ring->next_to_use = 0; 4066 tx_ring->next_to_clean = 0; 4067 4068 return 0; 4069 4070 err: 4071 vfree(tx_ring->tx_buffer_info); 4072 tx_ring->tx_buffer_info = NULL; 4073 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n"); 4074 return -ENOMEM; 4075 } 4076 4077 /** 4078 * igb_setup_all_tx_resources - wrapper to allocate Tx resources 4079 * (Descriptors) for all queues 4080 * @adapter: board private structure 4081 * 4082 * Return 0 on success, negative on failure 4083 **/ 4084 static int igb_setup_all_tx_resources(struct igb_adapter *adapter) 4085 { 4086 struct pci_dev *pdev = adapter->pdev; 4087 int i, err = 0; 4088 4089 for (i = 0; i < adapter->num_tx_queues; i++) { 4090 err = igb_setup_tx_resources(adapter->tx_ring[i]); 4091 if (err) { 4092 dev_err(&pdev->dev, 4093 "Allocation for Tx Queue %u failed\n", i); 4094 for (i--; i >= 0; i--) 4095 igb_free_tx_resources(adapter->tx_ring[i]); 4096 break; 4097 } 4098 } 4099 4100 return err; 4101 } 4102 4103 /** 4104 * igb_setup_tctl - configure the transmit control registers 4105 * @adapter: Board private structure 4106 **/ 4107 void igb_setup_tctl(struct igb_adapter *adapter) 4108 { 4109 struct e1000_hw *hw = &adapter->hw; 4110 u32 tctl; 4111 4112 /* disable queue 0 which is enabled by default on 82575 and 82576 */ 4113 wr32(E1000_TXDCTL(0), 0); 4114 4115 /* Program the Transmit Control Register */ 4116 tctl = rd32(E1000_TCTL); 4117 tctl &= ~E1000_TCTL_CT; 4118 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC | 4119 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT); 4120 4121 igb_config_collision_dist(hw); 4122 4123 /* Enable transmits */ 4124 tctl |= E1000_TCTL_EN; 4125 4126 wr32(E1000_TCTL, tctl); 4127 } 4128 4129 /** 4130 * igb_configure_tx_ring - Configure transmit ring after Reset 4131 * @adapter: board private structure 4132 * @ring: tx ring to configure 4133 * 4134 * Configure a transmit ring after a reset. 4135 **/ 4136 void igb_configure_tx_ring(struct igb_adapter *adapter, 4137 struct igb_ring *ring) 4138 { 4139 struct e1000_hw *hw = &adapter->hw; 4140 u32 txdctl = 0; 4141 u64 tdba = ring->dma; 4142 int reg_idx = ring->reg_idx; 4143 4144 wr32(E1000_TDLEN(reg_idx), 4145 ring->count * sizeof(union e1000_adv_tx_desc)); 4146 wr32(E1000_TDBAL(reg_idx), 4147 tdba & 0x00000000ffffffffULL); 4148 wr32(E1000_TDBAH(reg_idx), tdba >> 32); 4149 4150 ring->tail = adapter->io_addr + E1000_TDT(reg_idx); 4151 wr32(E1000_TDH(reg_idx), 0); 4152 writel(0, ring->tail); 4153 4154 txdctl |= IGB_TX_PTHRESH; 4155 txdctl |= IGB_TX_HTHRESH << 8; 4156 txdctl |= IGB_TX_WTHRESH << 16; 4157 4158 /* reinitialize tx_buffer_info */ 4159 memset(ring->tx_buffer_info, 0, 4160 sizeof(struct igb_tx_buffer) * ring->count); 4161 4162 txdctl |= E1000_TXDCTL_QUEUE_ENABLE; 4163 wr32(E1000_TXDCTL(reg_idx), txdctl); 4164 } 4165 4166 /** 4167 * igb_configure_tx - Configure transmit Unit after Reset 4168 * @adapter: board private structure 4169 * 4170 * Configure the Tx unit of the MAC after a reset. 4171 **/ 4172 static void igb_configure_tx(struct igb_adapter *adapter) 4173 { 4174 struct e1000_hw *hw = &adapter->hw; 4175 int i; 4176 4177 /* disable the queues */ 4178 for (i = 0; i < adapter->num_tx_queues; i++) 4179 wr32(E1000_TXDCTL(adapter->tx_ring[i]->reg_idx), 0); 4180 4181 wrfl(); 4182 usleep_range(10000, 20000); 4183 4184 for (i = 0; i < adapter->num_tx_queues; i++) 4185 igb_configure_tx_ring(adapter, adapter->tx_ring[i]); 4186 } 4187 4188 /** 4189 * igb_setup_rx_resources - allocate Rx resources (Descriptors) 4190 * @rx_ring: Rx descriptor ring (for a specific queue) to setup 4191 * 4192 * Returns 0 on success, negative on failure 4193 **/ 4194 int igb_setup_rx_resources(struct igb_ring *rx_ring) 4195 { 4196 struct device *dev = rx_ring->dev; 4197 int size; 4198 4199 size = sizeof(struct igb_rx_buffer) * rx_ring->count; 4200 4201 rx_ring->rx_buffer_info = vmalloc(size); 4202 if (!rx_ring->rx_buffer_info) 4203 goto err; 4204 4205 /* Round up to nearest 4K */ 4206 rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc); 4207 rx_ring->size = ALIGN(rx_ring->size, 4096); 4208 4209 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size, 4210 &rx_ring->dma, GFP_KERNEL); 4211 if (!rx_ring->desc) 4212 goto err; 4213 4214 rx_ring->next_to_alloc = 0; 4215 rx_ring->next_to_clean = 0; 4216 rx_ring->next_to_use = 0; 4217 4218 return 0; 4219 4220 err: 4221 vfree(rx_ring->rx_buffer_info); 4222 rx_ring->rx_buffer_info = NULL; 4223 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n"); 4224 return -ENOMEM; 4225 } 4226 4227 /** 4228 * igb_setup_all_rx_resources - wrapper to allocate Rx resources 4229 * (Descriptors) for all queues 4230 * @adapter: board private structure 4231 * 4232 * Return 0 on success, negative on failure 4233 **/ 4234 static int igb_setup_all_rx_resources(struct igb_adapter *adapter) 4235 { 4236 struct pci_dev *pdev = adapter->pdev; 4237 int i, err = 0; 4238 4239 for (i = 0; i < adapter->num_rx_queues; i++) { 4240 err = igb_setup_rx_resources(adapter->rx_ring[i]); 4241 if (err) { 4242 dev_err(&pdev->dev, 4243 "Allocation for Rx Queue %u failed\n", i); 4244 for (i--; i >= 0; i--) 4245 igb_free_rx_resources(adapter->rx_ring[i]); 4246 break; 4247 } 4248 } 4249 4250 return err; 4251 } 4252 4253 /** 4254 * igb_setup_mrqc - configure the multiple receive queue control registers 4255 * @adapter: Board private structure 4256 **/ 4257 static void igb_setup_mrqc(struct igb_adapter *adapter) 4258 { 4259 struct e1000_hw *hw = &adapter->hw; 4260 u32 mrqc, rxcsum; 4261 u32 j, num_rx_queues; 4262 u32 rss_key[10]; 4263 4264 netdev_rss_key_fill(rss_key, sizeof(rss_key)); 4265 for (j = 0; j < 10; j++) 4266 wr32(E1000_RSSRK(j), rss_key[j]); 4267 4268 num_rx_queues = adapter->rss_queues; 4269 4270 switch (hw->mac.type) { 4271 case e1000_82576: 4272 /* 82576 supports 2 RSS queues for SR-IOV */ 4273 if (adapter->vfs_allocated_count) 4274 num_rx_queues = 2; 4275 break; 4276 default: 4277 break; 4278 } 4279 4280 if (adapter->rss_indir_tbl_init != num_rx_queues) { 4281 for (j = 0; j < IGB_RETA_SIZE; j++) 4282 adapter->rss_indir_tbl[j] = 4283 (j * num_rx_queues) / IGB_RETA_SIZE; 4284 adapter->rss_indir_tbl_init = num_rx_queues; 4285 } 4286 igb_write_rss_indir_tbl(adapter); 4287 4288 /* Disable raw packet checksumming so that RSS hash is placed in 4289 * descriptor on writeback. No need to enable TCP/UDP/IP checksum 4290 * offloads as they are enabled by default 4291 */ 4292 rxcsum = rd32(E1000_RXCSUM); 4293 rxcsum |= E1000_RXCSUM_PCSD; 4294 4295 if (adapter->hw.mac.type >= e1000_82576) 4296 /* Enable Receive Checksum Offload for SCTP */ 4297 rxcsum |= E1000_RXCSUM_CRCOFL; 4298 4299 /* Don't need to set TUOFL or IPOFL, they default to 1 */ 4300 wr32(E1000_RXCSUM, rxcsum); 4301 4302 /* Generate RSS hash based on packet types, TCP/UDP 4303 * port numbers and/or IPv4/v6 src and dst addresses 4304 */ 4305 mrqc = E1000_MRQC_RSS_FIELD_IPV4 | 4306 E1000_MRQC_RSS_FIELD_IPV4_TCP | 4307 E1000_MRQC_RSS_FIELD_IPV6 | 4308 E1000_MRQC_RSS_FIELD_IPV6_TCP | 4309 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX; 4310 4311 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP) 4312 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP; 4313 if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP) 4314 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP; 4315 4316 /* If VMDq is enabled then we set the appropriate mode for that, else 4317 * we default to RSS so that an RSS hash is calculated per packet even 4318 * if we are only using one queue 4319 */ 4320 if (adapter->vfs_allocated_count) { 4321 if (hw->mac.type > e1000_82575) { 4322 /* Set the default pool for the PF's first queue */ 4323 u32 vtctl = rd32(E1000_VT_CTL); 4324 4325 vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK | 4326 E1000_VT_CTL_DISABLE_DEF_POOL); 4327 vtctl |= adapter->vfs_allocated_count << 4328 E1000_VT_CTL_DEFAULT_POOL_SHIFT; 4329 wr32(E1000_VT_CTL, vtctl); 4330 } 4331 if (adapter->rss_queues > 1) 4332 mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ; 4333 else 4334 mrqc |= E1000_MRQC_ENABLE_VMDQ; 4335 } else { 4336 if (hw->mac.type != e1000_i211) 4337 mrqc |= E1000_MRQC_ENABLE_RSS_MQ; 4338 } 4339 igb_vmm_control(adapter); 4340 4341 wr32(E1000_MRQC, mrqc); 4342 } 4343 4344 /** 4345 * igb_setup_rctl - configure the receive control registers 4346 * @adapter: Board private structure 4347 **/ 4348 void igb_setup_rctl(struct igb_adapter *adapter) 4349 { 4350 struct e1000_hw *hw = &adapter->hw; 4351 u32 rctl; 4352 4353 rctl = rd32(E1000_RCTL); 4354 4355 rctl &= ~(3 << E1000_RCTL_MO_SHIFT); 4356 rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC); 4357 4358 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF | 4359 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT); 4360 4361 /* enable stripping of CRC. It's unlikely this will break BMC 4362 * redirection as it did with e1000. Newer features require 4363 * that the HW strips the CRC. 4364 */ 4365 rctl |= E1000_RCTL_SECRC; 4366 4367 /* disable store bad packets and clear size bits. */ 4368 rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256); 4369 4370 /* enable LPE to allow for reception of jumbo frames */ 4371 rctl |= E1000_RCTL_LPE; 4372 4373 /* disable queue 0 to prevent tail write w/o re-config */ 4374 wr32(E1000_RXDCTL(0), 0); 4375 4376 /* Attention!!! For SR-IOV PF driver operations you must enable 4377 * queue drop for all VF and PF queues to prevent head of line blocking 4378 * if an un-trusted VF does not provide descriptors to hardware. 4379 */ 4380 if (adapter->vfs_allocated_count) { 4381 /* set all queue drop enable bits */ 4382 wr32(E1000_QDE, ALL_QUEUES); 4383 } 4384 4385 /* This is useful for sniffing bad packets. */ 4386 if (adapter->netdev->features & NETIF_F_RXALL) { 4387 /* UPE and MPE will be handled by normal PROMISC logic 4388 * in e1000e_set_rx_mode 4389 */ 4390 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */ 4391 E1000_RCTL_BAM | /* RX All Bcast Pkts */ 4392 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */ 4393 4394 rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */ 4395 E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */ 4396 /* Do not mess with E1000_CTRL_VME, it affects transmit as well, 4397 * and that breaks VLANs. 4398 */ 4399 } 4400 4401 wr32(E1000_RCTL, rctl); 4402 } 4403 4404 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size, 4405 int vfn) 4406 { 4407 struct e1000_hw *hw = &adapter->hw; 4408 u32 vmolr; 4409 4410 if (size > MAX_JUMBO_FRAME_SIZE) 4411 size = MAX_JUMBO_FRAME_SIZE; 4412 4413 vmolr = rd32(E1000_VMOLR(vfn)); 4414 vmolr &= ~E1000_VMOLR_RLPML_MASK; 4415 vmolr |= size | E1000_VMOLR_LPE; 4416 wr32(E1000_VMOLR(vfn), vmolr); 4417 4418 return 0; 4419 } 4420 4421 static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter, 4422 int vfn, bool enable) 4423 { 4424 struct e1000_hw *hw = &adapter->hw; 4425 u32 val, reg; 4426 4427 if (hw->mac.type < e1000_82576) 4428 return; 4429 4430 if (hw->mac.type == e1000_i350) 4431 reg = E1000_DVMOLR(vfn); 4432 else 4433 reg = E1000_VMOLR(vfn); 4434 4435 val = rd32(reg); 4436 if (enable) 4437 val |= E1000_VMOLR_STRVLAN; 4438 else 4439 val &= ~(E1000_VMOLR_STRVLAN); 4440 wr32(reg, val); 4441 } 4442 4443 static inline void igb_set_vmolr(struct igb_adapter *adapter, 4444 int vfn, bool aupe) 4445 { 4446 struct e1000_hw *hw = &adapter->hw; 4447 u32 vmolr; 4448 4449 /* This register exists only on 82576 and newer so if we are older then 4450 * we should exit and do nothing 4451 */ 4452 if (hw->mac.type < e1000_82576) 4453 return; 4454 4455 vmolr = rd32(E1000_VMOLR(vfn)); 4456 if (aupe) 4457 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */ 4458 else 4459 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */ 4460 4461 /* clear all bits that might not be set */ 4462 vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE); 4463 4464 if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count) 4465 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */ 4466 /* for VMDq only allow the VFs and pool 0 to accept broadcast and 4467 * multicast packets 4468 */ 4469 if (vfn <= adapter->vfs_allocated_count) 4470 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */ 4471 4472 wr32(E1000_VMOLR(vfn), vmolr); 4473 } 4474 4475 /** 4476 * igb_configure_rx_ring - Configure a receive ring after Reset 4477 * @adapter: board private structure 4478 * @ring: receive ring to be configured 4479 * 4480 * Configure the Rx unit of the MAC after a reset. 4481 **/ 4482 void igb_configure_rx_ring(struct igb_adapter *adapter, 4483 struct igb_ring *ring) 4484 { 4485 struct e1000_hw *hw = &adapter->hw; 4486 union e1000_adv_rx_desc *rx_desc; 4487 u64 rdba = ring->dma; 4488 int reg_idx = ring->reg_idx; 4489 u32 srrctl = 0, rxdctl = 0; 4490 4491 /* disable the queue */ 4492 wr32(E1000_RXDCTL(reg_idx), 0); 4493 4494 /* Set DMA base address registers */ 4495 wr32(E1000_RDBAL(reg_idx), 4496 rdba & 0x00000000ffffffffULL); 4497 wr32(E1000_RDBAH(reg_idx), rdba >> 32); 4498 wr32(E1000_RDLEN(reg_idx), 4499 ring->count * sizeof(union e1000_adv_rx_desc)); 4500 4501 /* initialize head and tail */ 4502 ring->tail = adapter->io_addr + E1000_RDT(reg_idx); 4503 wr32(E1000_RDH(reg_idx), 0); 4504 writel(0, ring->tail); 4505 4506 /* set descriptor configuration */ 4507 srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT; 4508 if (ring_uses_large_buffer(ring)) 4509 srrctl |= IGB_RXBUFFER_3072 >> E1000_SRRCTL_BSIZEPKT_SHIFT; 4510 else 4511 srrctl |= IGB_RXBUFFER_2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT; 4512 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF; 4513 if (hw->mac.type >= e1000_82580) 4514 srrctl |= E1000_SRRCTL_TIMESTAMP; 4515 /* Only set Drop Enable if we are supporting multiple queues */ 4516 if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1) 4517 srrctl |= E1000_SRRCTL_DROP_EN; 4518 4519 wr32(E1000_SRRCTL(reg_idx), srrctl); 4520 4521 /* set filtering for VMDQ pools */ 4522 igb_set_vmolr(adapter, reg_idx & 0x7, true); 4523 4524 rxdctl |= IGB_RX_PTHRESH; 4525 rxdctl |= IGB_RX_HTHRESH << 8; 4526 rxdctl |= IGB_RX_WTHRESH << 16; 4527 4528 /* initialize rx_buffer_info */ 4529 memset(ring->rx_buffer_info, 0, 4530 sizeof(struct igb_rx_buffer) * ring->count); 4531 4532 /* initialize Rx descriptor 0 */ 4533 rx_desc = IGB_RX_DESC(ring, 0); 4534 rx_desc->wb.upper.length = 0; 4535 4536 /* enable receive descriptor fetching */ 4537 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE; 4538 wr32(E1000_RXDCTL(reg_idx), rxdctl); 4539 } 4540 4541 static void igb_set_rx_buffer_len(struct igb_adapter *adapter, 4542 struct igb_ring *rx_ring) 4543 { 4544 /* set build_skb and buffer size flags */ 4545 clear_ring_build_skb_enabled(rx_ring); 4546 clear_ring_uses_large_buffer(rx_ring); 4547 4548 if (adapter->flags & IGB_FLAG_RX_LEGACY) 4549 return; 4550 4551 set_ring_build_skb_enabled(rx_ring); 4552 4553 #if (PAGE_SIZE < 8192) 4554 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB) 4555 return; 4556 4557 set_ring_uses_large_buffer(rx_ring); 4558 #endif 4559 } 4560 4561 /** 4562 * igb_configure_rx - Configure receive Unit after Reset 4563 * @adapter: board private structure 4564 * 4565 * Configure the Rx unit of the MAC after a reset. 4566 **/ 4567 static void igb_configure_rx(struct igb_adapter *adapter) 4568 { 4569 int i; 4570 4571 /* set the correct pool for the PF default MAC address in entry 0 */ 4572 igb_set_default_mac_filter(adapter); 4573 4574 /* Setup the HW Rx Head and Tail Descriptor Pointers and 4575 * the Base and Length of the Rx Descriptor Ring 4576 */ 4577 for (i = 0; i < adapter->num_rx_queues; i++) { 4578 struct igb_ring *rx_ring = adapter->rx_ring[i]; 4579 4580 igb_set_rx_buffer_len(adapter, rx_ring); 4581 igb_configure_rx_ring(adapter, rx_ring); 4582 } 4583 } 4584 4585 /** 4586 * igb_free_tx_resources - Free Tx Resources per Queue 4587 * @tx_ring: Tx descriptor ring for a specific queue 4588 * 4589 * Free all transmit software resources 4590 **/ 4591 void igb_free_tx_resources(struct igb_ring *tx_ring) 4592 { 4593 igb_clean_tx_ring(tx_ring); 4594 4595 vfree(tx_ring->tx_buffer_info); 4596 tx_ring->tx_buffer_info = NULL; 4597 4598 /* if not set, then don't free */ 4599 if (!tx_ring->desc) 4600 return; 4601 4602 dma_free_coherent(tx_ring->dev, tx_ring->size, 4603 tx_ring->desc, tx_ring->dma); 4604 4605 tx_ring->desc = NULL; 4606 } 4607 4608 /** 4609 * igb_free_all_tx_resources - Free Tx Resources for All Queues 4610 * @adapter: board private structure 4611 * 4612 * Free all transmit software resources 4613 **/ 4614 static void igb_free_all_tx_resources(struct igb_adapter *adapter) 4615 { 4616 int i; 4617 4618 for (i = 0; i < adapter->num_tx_queues; i++) 4619 if (adapter->tx_ring[i]) 4620 igb_free_tx_resources(adapter->tx_ring[i]); 4621 } 4622 4623 /** 4624 * igb_clean_tx_ring - Free Tx Buffers 4625 * @tx_ring: ring to be cleaned 4626 **/ 4627 static void igb_clean_tx_ring(struct igb_ring *tx_ring) 4628 { 4629 u16 i = tx_ring->next_to_clean; 4630 struct igb_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i]; 4631 4632 while (i != tx_ring->next_to_use) { 4633 union e1000_adv_tx_desc *eop_desc, *tx_desc; 4634 4635 /* Free all the Tx ring sk_buffs */ 4636 dev_kfree_skb_any(tx_buffer->skb); 4637 4638 /* unmap skb header data */ 4639 dma_unmap_single(tx_ring->dev, 4640 dma_unmap_addr(tx_buffer, dma), 4641 dma_unmap_len(tx_buffer, len), 4642 DMA_TO_DEVICE); 4643 4644 /* check for eop_desc to determine the end of the packet */ 4645 eop_desc = tx_buffer->next_to_watch; 4646 tx_desc = IGB_TX_DESC(tx_ring, i); 4647 4648 /* unmap remaining buffers */ 4649 while (tx_desc != eop_desc) { 4650 tx_buffer++; 4651 tx_desc++; 4652 i++; 4653 if (unlikely(i == tx_ring->count)) { 4654 i = 0; 4655 tx_buffer = tx_ring->tx_buffer_info; 4656 tx_desc = IGB_TX_DESC(tx_ring, 0); 4657 } 4658 4659 /* unmap any remaining paged data */ 4660 if (dma_unmap_len(tx_buffer, len)) 4661 dma_unmap_page(tx_ring->dev, 4662 dma_unmap_addr(tx_buffer, dma), 4663 dma_unmap_len(tx_buffer, len), 4664 DMA_TO_DEVICE); 4665 } 4666 4667 /* move us one more past the eop_desc for start of next pkt */ 4668 tx_buffer++; 4669 i++; 4670 if (unlikely(i == tx_ring->count)) { 4671 i = 0; 4672 tx_buffer = tx_ring->tx_buffer_info; 4673 } 4674 } 4675 4676 /* reset BQL for queue */ 4677 netdev_tx_reset_queue(txring_txq(tx_ring)); 4678 4679 /* reset next_to_use and next_to_clean */ 4680 tx_ring->next_to_use = 0; 4681 tx_ring->next_to_clean = 0; 4682 } 4683 4684 /** 4685 * igb_clean_all_tx_rings - Free Tx Buffers for all queues 4686 * @adapter: board private structure 4687 **/ 4688 static void igb_clean_all_tx_rings(struct igb_adapter *adapter) 4689 { 4690 int i; 4691 4692 for (i = 0; i < adapter->num_tx_queues; i++) 4693 if (adapter->tx_ring[i]) 4694 igb_clean_tx_ring(adapter->tx_ring[i]); 4695 } 4696 4697 /** 4698 * igb_free_rx_resources - Free Rx Resources 4699 * @rx_ring: ring to clean the resources from 4700 * 4701 * Free all receive software resources 4702 **/ 4703 void igb_free_rx_resources(struct igb_ring *rx_ring) 4704 { 4705 igb_clean_rx_ring(rx_ring); 4706 4707 vfree(rx_ring->rx_buffer_info); 4708 rx_ring->rx_buffer_info = NULL; 4709 4710 /* if not set, then don't free */ 4711 if (!rx_ring->desc) 4712 return; 4713 4714 dma_free_coherent(rx_ring->dev, rx_ring->size, 4715 rx_ring->desc, rx_ring->dma); 4716 4717 rx_ring->desc = NULL; 4718 } 4719 4720 /** 4721 * igb_free_all_rx_resources - Free Rx Resources for All Queues 4722 * @adapter: board private structure 4723 * 4724 * Free all receive software resources 4725 **/ 4726 static void igb_free_all_rx_resources(struct igb_adapter *adapter) 4727 { 4728 int i; 4729 4730 for (i = 0; i < adapter->num_rx_queues; i++) 4731 if (adapter->rx_ring[i]) 4732 igb_free_rx_resources(adapter->rx_ring[i]); 4733 } 4734 4735 /** 4736 * igb_clean_rx_ring - Free Rx Buffers per Queue 4737 * @rx_ring: ring to free buffers from 4738 **/ 4739 static void igb_clean_rx_ring(struct igb_ring *rx_ring) 4740 { 4741 u16 i = rx_ring->next_to_clean; 4742 4743 if (rx_ring->skb) 4744 dev_kfree_skb(rx_ring->skb); 4745 rx_ring->skb = NULL; 4746 4747 /* Free all the Rx ring sk_buffs */ 4748 while (i != rx_ring->next_to_alloc) { 4749 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i]; 4750 4751 /* Invalidate cache lines that may have been written to by 4752 * device so that we avoid corrupting memory. 4753 */ 4754 dma_sync_single_range_for_cpu(rx_ring->dev, 4755 buffer_info->dma, 4756 buffer_info->page_offset, 4757 igb_rx_bufsz(rx_ring), 4758 DMA_FROM_DEVICE); 4759 4760 /* free resources associated with mapping */ 4761 dma_unmap_page_attrs(rx_ring->dev, 4762 buffer_info->dma, 4763 igb_rx_pg_size(rx_ring), 4764 DMA_FROM_DEVICE, 4765 IGB_RX_DMA_ATTR); 4766 __page_frag_cache_drain(buffer_info->page, 4767 buffer_info->pagecnt_bias); 4768 4769 i++; 4770 if (i == rx_ring->count) 4771 i = 0; 4772 } 4773 4774 rx_ring->next_to_alloc = 0; 4775 rx_ring->next_to_clean = 0; 4776 rx_ring->next_to_use = 0; 4777 } 4778 4779 /** 4780 * igb_clean_all_rx_rings - Free Rx Buffers for all queues 4781 * @adapter: board private structure 4782 **/ 4783 static void igb_clean_all_rx_rings(struct igb_adapter *adapter) 4784 { 4785 int i; 4786 4787 for (i = 0; i < adapter->num_rx_queues; i++) 4788 if (adapter->rx_ring[i]) 4789 igb_clean_rx_ring(adapter->rx_ring[i]); 4790 } 4791 4792 /** 4793 * igb_set_mac - Change the Ethernet Address of the NIC 4794 * @netdev: network interface device structure 4795 * @p: pointer to an address structure 4796 * 4797 * Returns 0 on success, negative on failure 4798 **/ 4799 static int igb_set_mac(struct net_device *netdev, void *p) 4800 { 4801 struct igb_adapter *adapter = netdev_priv(netdev); 4802 struct e1000_hw *hw = &adapter->hw; 4803 struct sockaddr *addr = p; 4804 4805 if (!is_valid_ether_addr(addr->sa_data)) 4806 return -EADDRNOTAVAIL; 4807 4808 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len); 4809 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len); 4810 4811 /* set the correct pool for the new PF MAC address in entry 0 */ 4812 igb_set_default_mac_filter(adapter); 4813 4814 return 0; 4815 } 4816 4817 /** 4818 * igb_write_mc_addr_list - write multicast addresses to MTA 4819 * @netdev: network interface device structure 4820 * 4821 * Writes multicast address list to the MTA hash table. 4822 * Returns: -ENOMEM on failure 4823 * 0 on no addresses written 4824 * X on writing X addresses to MTA 4825 **/ 4826 static int igb_write_mc_addr_list(struct net_device *netdev) 4827 { 4828 struct igb_adapter *adapter = netdev_priv(netdev); 4829 struct e1000_hw *hw = &adapter->hw; 4830 struct netdev_hw_addr *ha; 4831 u8 *mta_list; 4832 int i; 4833 4834 if (netdev_mc_empty(netdev)) { 4835 /* nothing to program, so clear mc list */ 4836 igb_update_mc_addr_list(hw, NULL, 0); 4837 igb_restore_vf_multicasts(adapter); 4838 return 0; 4839 } 4840 4841 mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC); 4842 if (!mta_list) 4843 return -ENOMEM; 4844 4845 /* The shared function expects a packed array of only addresses. */ 4846 i = 0; 4847 netdev_for_each_mc_addr(ha, netdev) 4848 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN); 4849 4850 igb_update_mc_addr_list(hw, mta_list, i); 4851 kfree(mta_list); 4852 4853 return netdev_mc_count(netdev); 4854 } 4855 4856 static int igb_vlan_promisc_enable(struct igb_adapter *adapter) 4857 { 4858 struct e1000_hw *hw = &adapter->hw; 4859 u32 i, pf_id; 4860 4861 switch (hw->mac.type) { 4862 case e1000_i210: 4863 case e1000_i211: 4864 case e1000_i350: 4865 /* VLAN filtering needed for VLAN prio filter */ 4866 if (adapter->netdev->features & NETIF_F_NTUPLE) 4867 break; 4868 /* fall through */ 4869 case e1000_82576: 4870 case e1000_82580: 4871 case e1000_i354: 4872 /* VLAN filtering needed for pool filtering */ 4873 if (adapter->vfs_allocated_count) 4874 break; 4875 /* fall through */ 4876 default: 4877 return 1; 4878 } 4879 4880 /* We are already in VLAN promisc, nothing to do */ 4881 if (adapter->flags & IGB_FLAG_VLAN_PROMISC) 4882 return 0; 4883 4884 if (!adapter->vfs_allocated_count) 4885 goto set_vfta; 4886 4887 /* Add PF to all active pools */ 4888 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT; 4889 4890 for (i = E1000_VLVF_ARRAY_SIZE; --i;) { 4891 u32 vlvf = rd32(E1000_VLVF(i)); 4892 4893 vlvf |= BIT(pf_id); 4894 wr32(E1000_VLVF(i), vlvf); 4895 } 4896 4897 set_vfta: 4898 /* Set all bits in the VLAN filter table array */ 4899 for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;) 4900 hw->mac.ops.write_vfta(hw, i, ~0U); 4901 4902 /* Set flag so we don't redo unnecessary work */ 4903 adapter->flags |= IGB_FLAG_VLAN_PROMISC; 4904 4905 return 0; 4906 } 4907 4908 #define VFTA_BLOCK_SIZE 8 4909 static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset) 4910 { 4911 struct e1000_hw *hw = &adapter->hw; 4912 u32 vfta[VFTA_BLOCK_SIZE] = { 0 }; 4913 u32 vid_start = vfta_offset * 32; 4914 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32); 4915 u32 i, vid, word, bits, pf_id; 4916 4917 /* guarantee that we don't scrub out management VLAN */ 4918 vid = adapter->mng_vlan_id; 4919 if (vid >= vid_start && vid < vid_end) 4920 vfta[(vid - vid_start) / 32] |= BIT(vid % 32); 4921 4922 if (!adapter->vfs_allocated_count) 4923 goto set_vfta; 4924 4925 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT; 4926 4927 for (i = E1000_VLVF_ARRAY_SIZE; --i;) { 4928 u32 vlvf = rd32(E1000_VLVF(i)); 4929 4930 /* pull VLAN ID from VLVF */ 4931 vid = vlvf & VLAN_VID_MASK; 4932 4933 /* only concern ourselves with a certain range */ 4934 if (vid < vid_start || vid >= vid_end) 4935 continue; 4936 4937 if (vlvf & E1000_VLVF_VLANID_ENABLE) { 4938 /* record VLAN ID in VFTA */ 4939 vfta[(vid - vid_start) / 32] |= BIT(vid % 32); 4940 4941 /* if PF is part of this then continue */ 4942 if (test_bit(vid, adapter->active_vlans)) 4943 continue; 4944 } 4945 4946 /* remove PF from the pool */ 4947 bits = ~BIT(pf_id); 4948 bits &= rd32(E1000_VLVF(i)); 4949 wr32(E1000_VLVF(i), bits); 4950 } 4951 4952 set_vfta: 4953 /* extract values from active_vlans and write back to VFTA */ 4954 for (i = VFTA_BLOCK_SIZE; i--;) { 4955 vid = (vfta_offset + i) * 32; 4956 word = vid / BITS_PER_LONG; 4957 bits = vid % BITS_PER_LONG; 4958 4959 vfta[i] |= adapter->active_vlans[word] >> bits; 4960 4961 hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]); 4962 } 4963 } 4964 4965 static void igb_vlan_promisc_disable(struct igb_adapter *adapter) 4966 { 4967 u32 i; 4968 4969 /* We are not in VLAN promisc, nothing to do */ 4970 if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC)) 4971 return; 4972 4973 /* Set flag so we don't redo unnecessary work */ 4974 adapter->flags &= ~IGB_FLAG_VLAN_PROMISC; 4975 4976 for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE) 4977 igb_scrub_vfta(adapter, i); 4978 } 4979 4980 /** 4981 * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set 4982 * @netdev: network interface device structure 4983 * 4984 * The set_rx_mode entry point is called whenever the unicast or multicast 4985 * address lists or the network interface flags are updated. This routine is 4986 * responsible for configuring the hardware for proper unicast, multicast, 4987 * promiscuous mode, and all-multi behavior. 4988 **/ 4989 static void igb_set_rx_mode(struct net_device *netdev) 4990 { 4991 struct igb_adapter *adapter = netdev_priv(netdev); 4992 struct e1000_hw *hw = &adapter->hw; 4993 unsigned int vfn = adapter->vfs_allocated_count; 4994 u32 rctl = 0, vmolr = 0, rlpml = MAX_JUMBO_FRAME_SIZE; 4995 int count; 4996 4997 /* Check for Promiscuous and All Multicast modes */ 4998 if (netdev->flags & IFF_PROMISC) { 4999 rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE; 5000 vmolr |= E1000_VMOLR_MPME; 5001 5002 /* enable use of UTA filter to force packets to default pool */ 5003 if (hw->mac.type == e1000_82576) 5004 vmolr |= E1000_VMOLR_ROPE; 5005 } else { 5006 if (netdev->flags & IFF_ALLMULTI) { 5007 rctl |= E1000_RCTL_MPE; 5008 vmolr |= E1000_VMOLR_MPME; 5009 } else { 5010 /* Write addresses to the MTA, if the attempt fails 5011 * then we should just turn on promiscuous mode so 5012 * that we can at least receive multicast traffic 5013 */ 5014 count = igb_write_mc_addr_list(netdev); 5015 if (count < 0) { 5016 rctl |= E1000_RCTL_MPE; 5017 vmolr |= E1000_VMOLR_MPME; 5018 } else if (count) { 5019 vmolr |= E1000_VMOLR_ROMPE; 5020 } 5021 } 5022 } 5023 5024 /* Write addresses to available RAR registers, if there is not 5025 * sufficient space to store all the addresses then enable 5026 * unicast promiscuous mode 5027 */ 5028 if (__dev_uc_sync(netdev, igb_uc_sync, igb_uc_unsync)) { 5029 rctl |= E1000_RCTL_UPE; 5030 vmolr |= E1000_VMOLR_ROPE; 5031 } 5032 5033 /* enable VLAN filtering by default */ 5034 rctl |= E1000_RCTL_VFE; 5035 5036 /* disable VLAN filtering for modes that require it */ 5037 if ((netdev->flags & IFF_PROMISC) || 5038 (netdev->features & NETIF_F_RXALL)) { 5039 /* if we fail to set all rules then just clear VFE */ 5040 if (igb_vlan_promisc_enable(adapter)) 5041 rctl &= ~E1000_RCTL_VFE; 5042 } else { 5043 igb_vlan_promisc_disable(adapter); 5044 } 5045 5046 /* update state of unicast, multicast, and VLAN filtering modes */ 5047 rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE | 5048 E1000_RCTL_VFE); 5049 wr32(E1000_RCTL, rctl); 5050 5051 #if (PAGE_SIZE < 8192) 5052 if (!adapter->vfs_allocated_count) { 5053 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB) 5054 rlpml = IGB_MAX_FRAME_BUILD_SKB; 5055 } 5056 #endif 5057 wr32(E1000_RLPML, rlpml); 5058 5059 /* In order to support SR-IOV and eventually VMDq it is necessary to set 5060 * the VMOLR to enable the appropriate modes. Without this workaround 5061 * we will have issues with VLAN tag stripping not being done for frames 5062 * that are only arriving because we are the default pool 5063 */ 5064 if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350)) 5065 return; 5066 5067 /* set UTA to appropriate mode */ 5068 igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE)); 5069 5070 vmolr |= rd32(E1000_VMOLR(vfn)) & 5071 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE); 5072 5073 /* enable Rx jumbo frames, restrict as needed to support build_skb */ 5074 vmolr &= ~E1000_VMOLR_RLPML_MASK; 5075 #if (PAGE_SIZE < 8192) 5076 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB) 5077 vmolr |= IGB_MAX_FRAME_BUILD_SKB; 5078 else 5079 #endif 5080 vmolr |= MAX_JUMBO_FRAME_SIZE; 5081 vmolr |= E1000_VMOLR_LPE; 5082 5083 wr32(E1000_VMOLR(vfn), vmolr); 5084 5085 igb_restore_vf_multicasts(adapter); 5086 } 5087 5088 static void igb_check_wvbr(struct igb_adapter *adapter) 5089 { 5090 struct e1000_hw *hw = &adapter->hw; 5091 u32 wvbr = 0; 5092 5093 switch (hw->mac.type) { 5094 case e1000_82576: 5095 case e1000_i350: 5096 wvbr = rd32(E1000_WVBR); 5097 if (!wvbr) 5098 return; 5099 break; 5100 default: 5101 break; 5102 } 5103 5104 adapter->wvbr |= wvbr; 5105 } 5106 5107 #define IGB_STAGGERED_QUEUE_OFFSET 8 5108 5109 static void igb_spoof_check(struct igb_adapter *adapter) 5110 { 5111 int j; 5112 5113 if (!adapter->wvbr) 5114 return; 5115 5116 for (j = 0; j < adapter->vfs_allocated_count; j++) { 5117 if (adapter->wvbr & BIT(j) || 5118 adapter->wvbr & BIT(j + IGB_STAGGERED_QUEUE_OFFSET)) { 5119 dev_warn(&adapter->pdev->dev, 5120 "Spoof event(s) detected on VF %d\n", j); 5121 adapter->wvbr &= 5122 ~(BIT(j) | 5123 BIT(j + IGB_STAGGERED_QUEUE_OFFSET)); 5124 } 5125 } 5126 } 5127 5128 /* Need to wait a few seconds after link up to get diagnostic information from 5129 * the phy 5130 */ 5131 static void igb_update_phy_info(struct timer_list *t) 5132 { 5133 struct igb_adapter *adapter = from_timer(adapter, t, phy_info_timer); 5134 igb_get_phy_info(&adapter->hw); 5135 } 5136 5137 /** 5138 * igb_has_link - check shared code for link and determine up/down 5139 * @adapter: pointer to driver private info 5140 **/ 5141 bool igb_has_link(struct igb_adapter *adapter) 5142 { 5143 struct e1000_hw *hw = &adapter->hw; 5144 bool link_active = false; 5145 5146 /* get_link_status is set on LSC (link status) interrupt or 5147 * rx sequence error interrupt. get_link_status will stay 5148 * false until the e1000_check_for_link establishes link 5149 * for copper adapters ONLY 5150 */ 5151 switch (hw->phy.media_type) { 5152 case e1000_media_type_copper: 5153 if (!hw->mac.get_link_status) 5154 return true; 5155 /* fall through */ 5156 case e1000_media_type_internal_serdes: 5157 hw->mac.ops.check_for_link(hw); 5158 link_active = !hw->mac.get_link_status; 5159 break; 5160 default: 5161 case e1000_media_type_unknown: 5162 break; 5163 } 5164 5165 if (((hw->mac.type == e1000_i210) || 5166 (hw->mac.type == e1000_i211)) && 5167 (hw->phy.id == I210_I_PHY_ID)) { 5168 if (!netif_carrier_ok(adapter->netdev)) { 5169 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE; 5170 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) { 5171 adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE; 5172 adapter->link_check_timeout = jiffies; 5173 } 5174 } 5175 5176 return link_active; 5177 } 5178 5179 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event) 5180 { 5181 bool ret = false; 5182 u32 ctrl_ext, thstat; 5183 5184 /* check for thermal sensor event on i350 copper only */ 5185 if (hw->mac.type == e1000_i350) { 5186 thstat = rd32(E1000_THSTAT); 5187 ctrl_ext = rd32(E1000_CTRL_EXT); 5188 5189 if ((hw->phy.media_type == e1000_media_type_copper) && 5190 !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII)) 5191 ret = !!(thstat & event); 5192 } 5193 5194 return ret; 5195 } 5196 5197 /** 5198 * igb_check_lvmmc - check for malformed packets received 5199 * and indicated in LVMMC register 5200 * @adapter: pointer to adapter 5201 **/ 5202 static void igb_check_lvmmc(struct igb_adapter *adapter) 5203 { 5204 struct e1000_hw *hw = &adapter->hw; 5205 u32 lvmmc; 5206 5207 lvmmc = rd32(E1000_LVMMC); 5208 if (lvmmc) { 5209 if (unlikely(net_ratelimit())) { 5210 netdev_warn(adapter->netdev, 5211 "malformed Tx packet detected and dropped, LVMMC:0x%08x\n", 5212 lvmmc); 5213 } 5214 } 5215 } 5216 5217 /** 5218 * igb_watchdog - Timer Call-back 5219 * @data: pointer to adapter cast into an unsigned long 5220 **/ 5221 static void igb_watchdog(struct timer_list *t) 5222 { 5223 struct igb_adapter *adapter = from_timer(adapter, t, watchdog_timer); 5224 /* Do the rest outside of interrupt context */ 5225 schedule_work(&adapter->watchdog_task); 5226 } 5227 5228 static void igb_watchdog_task(struct work_struct *work) 5229 { 5230 struct igb_adapter *adapter = container_of(work, 5231 struct igb_adapter, 5232 watchdog_task); 5233 struct e1000_hw *hw = &adapter->hw; 5234 struct e1000_phy_info *phy = &hw->phy; 5235 struct net_device *netdev = adapter->netdev; 5236 u32 link; 5237 int i; 5238 u32 connsw; 5239 u16 phy_data, retry_count = 20; 5240 5241 link = igb_has_link(adapter); 5242 5243 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) { 5244 if (time_after(jiffies, (adapter->link_check_timeout + HZ))) 5245 adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE; 5246 else 5247 link = false; 5248 } 5249 5250 /* Force link down if we have fiber to swap to */ 5251 if (adapter->flags & IGB_FLAG_MAS_ENABLE) { 5252 if (hw->phy.media_type == e1000_media_type_copper) { 5253 connsw = rd32(E1000_CONNSW); 5254 if (!(connsw & E1000_CONNSW_AUTOSENSE_EN)) 5255 link = 0; 5256 } 5257 } 5258 if (link) { 5259 /* Perform a reset if the media type changed. */ 5260 if (hw->dev_spec._82575.media_changed) { 5261 hw->dev_spec._82575.media_changed = false; 5262 adapter->flags |= IGB_FLAG_MEDIA_RESET; 5263 igb_reset(adapter); 5264 } 5265 /* Cancel scheduled suspend requests. */ 5266 pm_runtime_resume(netdev->dev.parent); 5267 5268 if (!netif_carrier_ok(netdev)) { 5269 u32 ctrl; 5270 5271 hw->mac.ops.get_speed_and_duplex(hw, 5272 &adapter->link_speed, 5273 &adapter->link_duplex); 5274 5275 ctrl = rd32(E1000_CTRL); 5276 /* Links status message must follow this format */ 5277 netdev_info(netdev, 5278 "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n", 5279 netdev->name, 5280 adapter->link_speed, 5281 adapter->link_duplex == FULL_DUPLEX ? 5282 "Full" : "Half", 5283 (ctrl & E1000_CTRL_TFCE) && 5284 (ctrl & E1000_CTRL_RFCE) ? "RX/TX" : 5285 (ctrl & E1000_CTRL_RFCE) ? "RX" : 5286 (ctrl & E1000_CTRL_TFCE) ? "TX" : "None"); 5287 5288 /* disable EEE if enabled */ 5289 if ((adapter->flags & IGB_FLAG_EEE) && 5290 (adapter->link_duplex == HALF_DUPLEX)) { 5291 dev_info(&adapter->pdev->dev, 5292 "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n"); 5293 adapter->hw.dev_spec._82575.eee_disable = true; 5294 adapter->flags &= ~IGB_FLAG_EEE; 5295 } 5296 5297 /* check if SmartSpeed worked */ 5298 igb_check_downshift(hw); 5299 if (phy->speed_downgraded) 5300 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n"); 5301 5302 /* check for thermal sensor event */ 5303 if (igb_thermal_sensor_event(hw, 5304 E1000_THSTAT_LINK_THROTTLE)) 5305 netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n"); 5306 5307 /* adjust timeout factor according to speed/duplex */ 5308 adapter->tx_timeout_factor = 1; 5309 switch (adapter->link_speed) { 5310 case SPEED_10: 5311 adapter->tx_timeout_factor = 14; 5312 break; 5313 case SPEED_100: 5314 /* maybe add some timeout factor ? */ 5315 break; 5316 } 5317 5318 if (adapter->link_speed != SPEED_1000) 5319 goto no_wait; 5320 5321 /* wait for Remote receiver status OK */ 5322 retry_read_status: 5323 if (!igb_read_phy_reg(hw, PHY_1000T_STATUS, 5324 &phy_data)) { 5325 if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) && 5326 retry_count) { 5327 msleep(100); 5328 retry_count--; 5329 goto retry_read_status; 5330 } else if (!retry_count) { 5331 dev_err(&adapter->pdev->dev, "exceed max 2 second\n"); 5332 } 5333 } else { 5334 dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n"); 5335 } 5336 no_wait: 5337 netif_carrier_on(netdev); 5338 5339 igb_ping_all_vfs(adapter); 5340 igb_check_vf_rate_limit(adapter); 5341 5342 /* link state has changed, schedule phy info update */ 5343 if (!test_bit(__IGB_DOWN, &adapter->state)) 5344 mod_timer(&adapter->phy_info_timer, 5345 round_jiffies(jiffies + 2 * HZ)); 5346 } 5347 } else { 5348 if (netif_carrier_ok(netdev)) { 5349 adapter->link_speed = 0; 5350 adapter->link_duplex = 0; 5351 5352 /* check for thermal sensor event */ 5353 if (igb_thermal_sensor_event(hw, 5354 E1000_THSTAT_PWR_DOWN)) { 5355 netdev_err(netdev, "The network adapter was stopped because it overheated\n"); 5356 } 5357 5358 /* Links status message must follow this format */ 5359 netdev_info(netdev, "igb: %s NIC Link is Down\n", 5360 netdev->name); 5361 netif_carrier_off(netdev); 5362 5363 igb_ping_all_vfs(adapter); 5364 5365 /* link state has changed, schedule phy info update */ 5366 if (!test_bit(__IGB_DOWN, &adapter->state)) 5367 mod_timer(&adapter->phy_info_timer, 5368 round_jiffies(jiffies + 2 * HZ)); 5369 5370 /* link is down, time to check for alternate media */ 5371 if (adapter->flags & IGB_FLAG_MAS_ENABLE) { 5372 igb_check_swap_media(adapter); 5373 if (adapter->flags & IGB_FLAG_MEDIA_RESET) { 5374 schedule_work(&adapter->reset_task); 5375 /* return immediately */ 5376 return; 5377 } 5378 } 5379 pm_schedule_suspend(netdev->dev.parent, 5380 MSEC_PER_SEC * 5); 5381 5382 /* also check for alternate media here */ 5383 } else if (!netif_carrier_ok(netdev) && 5384 (adapter->flags & IGB_FLAG_MAS_ENABLE)) { 5385 igb_check_swap_media(adapter); 5386 if (adapter->flags & IGB_FLAG_MEDIA_RESET) { 5387 schedule_work(&adapter->reset_task); 5388 /* return immediately */ 5389 return; 5390 } 5391 } 5392 } 5393 5394 spin_lock(&adapter->stats64_lock); 5395 igb_update_stats(adapter); 5396 spin_unlock(&adapter->stats64_lock); 5397 5398 for (i = 0; i < adapter->num_tx_queues; i++) { 5399 struct igb_ring *tx_ring = adapter->tx_ring[i]; 5400 if (!netif_carrier_ok(netdev)) { 5401 /* We've lost link, so the controller stops DMA, 5402 * but we've got queued Tx work that's never going 5403 * to get done, so reset controller to flush Tx. 5404 * (Do the reset outside of interrupt context). 5405 */ 5406 if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) { 5407 adapter->tx_timeout_count++; 5408 schedule_work(&adapter->reset_task); 5409 /* return immediately since reset is imminent */ 5410 return; 5411 } 5412 } 5413 5414 /* Force detection of hung controller every watchdog period */ 5415 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags); 5416 } 5417 5418 /* Cause software interrupt to ensure Rx ring is cleaned */ 5419 if (adapter->flags & IGB_FLAG_HAS_MSIX) { 5420 u32 eics = 0; 5421 5422 for (i = 0; i < adapter->num_q_vectors; i++) 5423 eics |= adapter->q_vector[i]->eims_value; 5424 wr32(E1000_EICS, eics); 5425 } else { 5426 wr32(E1000_ICS, E1000_ICS_RXDMT0); 5427 } 5428 5429 igb_spoof_check(adapter); 5430 igb_ptp_rx_hang(adapter); 5431 igb_ptp_tx_hang(adapter); 5432 5433 /* Check LVMMC register on i350/i354 only */ 5434 if ((adapter->hw.mac.type == e1000_i350) || 5435 (adapter->hw.mac.type == e1000_i354)) 5436 igb_check_lvmmc(adapter); 5437 5438 /* Reset the timer */ 5439 if (!test_bit(__IGB_DOWN, &adapter->state)) { 5440 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) 5441 mod_timer(&adapter->watchdog_timer, 5442 round_jiffies(jiffies + HZ)); 5443 else 5444 mod_timer(&adapter->watchdog_timer, 5445 round_jiffies(jiffies + 2 * HZ)); 5446 } 5447 } 5448 5449 enum latency_range { 5450 lowest_latency = 0, 5451 low_latency = 1, 5452 bulk_latency = 2, 5453 latency_invalid = 255 5454 }; 5455 5456 /** 5457 * igb_update_ring_itr - update the dynamic ITR value based on packet size 5458 * @q_vector: pointer to q_vector 5459 * 5460 * Stores a new ITR value based on strictly on packet size. This 5461 * algorithm is less sophisticated than that used in igb_update_itr, 5462 * due to the difficulty of synchronizing statistics across multiple 5463 * receive rings. The divisors and thresholds used by this function 5464 * were determined based on theoretical maximum wire speed and testing 5465 * data, in order to minimize response time while increasing bulk 5466 * throughput. 5467 * This functionality is controlled by ethtool's coalescing settings. 5468 * NOTE: This function is called only when operating in a multiqueue 5469 * receive environment. 5470 **/ 5471 static void igb_update_ring_itr(struct igb_q_vector *q_vector) 5472 { 5473 int new_val = q_vector->itr_val; 5474 int avg_wire_size = 0; 5475 struct igb_adapter *adapter = q_vector->adapter; 5476 unsigned int packets; 5477 5478 /* For non-gigabit speeds, just fix the interrupt rate at 4000 5479 * ints/sec - ITR timer value of 120 ticks. 5480 */ 5481 if (adapter->link_speed != SPEED_1000) { 5482 new_val = IGB_4K_ITR; 5483 goto set_itr_val; 5484 } 5485 5486 packets = q_vector->rx.total_packets; 5487 if (packets) 5488 avg_wire_size = q_vector->rx.total_bytes / packets; 5489 5490 packets = q_vector->tx.total_packets; 5491 if (packets) 5492 avg_wire_size = max_t(u32, avg_wire_size, 5493 q_vector->tx.total_bytes / packets); 5494 5495 /* if avg_wire_size isn't set no work was done */ 5496 if (!avg_wire_size) 5497 goto clear_counts; 5498 5499 /* Add 24 bytes to size to account for CRC, preamble, and gap */ 5500 avg_wire_size += 24; 5501 5502 /* Don't starve jumbo frames */ 5503 avg_wire_size = min(avg_wire_size, 3000); 5504 5505 /* Give a little boost to mid-size frames */ 5506 if ((avg_wire_size > 300) && (avg_wire_size < 1200)) 5507 new_val = avg_wire_size / 3; 5508 else 5509 new_val = avg_wire_size / 2; 5510 5511 /* conservative mode (itr 3) eliminates the lowest_latency setting */ 5512 if (new_val < IGB_20K_ITR && 5513 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) || 5514 (!q_vector->rx.ring && adapter->tx_itr_setting == 3))) 5515 new_val = IGB_20K_ITR; 5516 5517 set_itr_val: 5518 if (new_val != q_vector->itr_val) { 5519 q_vector->itr_val = new_val; 5520 q_vector->set_itr = 1; 5521 } 5522 clear_counts: 5523 q_vector->rx.total_bytes = 0; 5524 q_vector->rx.total_packets = 0; 5525 q_vector->tx.total_bytes = 0; 5526 q_vector->tx.total_packets = 0; 5527 } 5528 5529 /** 5530 * igb_update_itr - update the dynamic ITR value based on statistics 5531 * @q_vector: pointer to q_vector 5532 * @ring_container: ring info to update the itr for 5533 * 5534 * Stores a new ITR value based on packets and byte 5535 * counts during the last interrupt. The advantage of per interrupt 5536 * computation is faster updates and more accurate ITR for the current 5537 * traffic pattern. Constants in this function were computed 5538 * based on theoretical maximum wire speed and thresholds were set based 5539 * on testing data as well as attempting to minimize response time 5540 * while increasing bulk throughput. 5541 * This functionality is controlled by ethtool's coalescing settings. 5542 * NOTE: These calculations are only valid when operating in a single- 5543 * queue environment. 5544 **/ 5545 static void igb_update_itr(struct igb_q_vector *q_vector, 5546 struct igb_ring_container *ring_container) 5547 { 5548 unsigned int packets = ring_container->total_packets; 5549 unsigned int bytes = ring_container->total_bytes; 5550 u8 itrval = ring_container->itr; 5551 5552 /* no packets, exit with status unchanged */ 5553 if (packets == 0) 5554 return; 5555 5556 switch (itrval) { 5557 case lowest_latency: 5558 /* handle TSO and jumbo frames */ 5559 if (bytes/packets > 8000) 5560 itrval = bulk_latency; 5561 else if ((packets < 5) && (bytes > 512)) 5562 itrval = low_latency; 5563 break; 5564 case low_latency: /* 50 usec aka 20000 ints/s */ 5565 if (bytes > 10000) { 5566 /* this if handles the TSO accounting */ 5567 if (bytes/packets > 8000) 5568 itrval = bulk_latency; 5569 else if ((packets < 10) || ((bytes/packets) > 1200)) 5570 itrval = bulk_latency; 5571 else if ((packets > 35)) 5572 itrval = lowest_latency; 5573 } else if (bytes/packets > 2000) { 5574 itrval = bulk_latency; 5575 } else if (packets <= 2 && bytes < 512) { 5576 itrval = lowest_latency; 5577 } 5578 break; 5579 case bulk_latency: /* 250 usec aka 4000 ints/s */ 5580 if (bytes > 25000) { 5581 if (packets > 35) 5582 itrval = low_latency; 5583 } else if (bytes < 1500) { 5584 itrval = low_latency; 5585 } 5586 break; 5587 } 5588 5589 /* clear work counters since we have the values we need */ 5590 ring_container->total_bytes = 0; 5591 ring_container->total_packets = 0; 5592 5593 /* write updated itr to ring container */ 5594 ring_container->itr = itrval; 5595 } 5596 5597 static void igb_set_itr(struct igb_q_vector *q_vector) 5598 { 5599 struct igb_adapter *adapter = q_vector->adapter; 5600 u32 new_itr = q_vector->itr_val; 5601 u8 current_itr = 0; 5602 5603 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */ 5604 if (adapter->link_speed != SPEED_1000) { 5605 current_itr = 0; 5606 new_itr = IGB_4K_ITR; 5607 goto set_itr_now; 5608 } 5609 5610 igb_update_itr(q_vector, &q_vector->tx); 5611 igb_update_itr(q_vector, &q_vector->rx); 5612 5613 current_itr = max(q_vector->rx.itr, q_vector->tx.itr); 5614 5615 /* conservative mode (itr 3) eliminates the lowest_latency setting */ 5616 if (current_itr == lowest_latency && 5617 ((q_vector->rx.ring && adapter->rx_itr_setting == 3) || 5618 (!q_vector->rx.ring && adapter->tx_itr_setting == 3))) 5619 current_itr = low_latency; 5620 5621 switch (current_itr) { 5622 /* counts and packets in update_itr are dependent on these numbers */ 5623 case lowest_latency: 5624 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */ 5625 break; 5626 case low_latency: 5627 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */ 5628 break; 5629 case bulk_latency: 5630 new_itr = IGB_4K_ITR; /* 4,000 ints/sec */ 5631 break; 5632 default: 5633 break; 5634 } 5635 5636 set_itr_now: 5637 if (new_itr != q_vector->itr_val) { 5638 /* this attempts to bias the interrupt rate towards Bulk 5639 * by adding intermediate steps when interrupt rate is 5640 * increasing 5641 */ 5642 new_itr = new_itr > q_vector->itr_val ? 5643 max((new_itr * q_vector->itr_val) / 5644 (new_itr + (q_vector->itr_val >> 2)), 5645 new_itr) : new_itr; 5646 /* Don't write the value here; it resets the adapter's 5647 * internal timer, and causes us to delay far longer than 5648 * we should between interrupts. Instead, we write the ITR 5649 * value at the beginning of the next interrupt so the timing 5650 * ends up being correct. 5651 */ 5652 q_vector->itr_val = new_itr; 5653 q_vector->set_itr = 1; 5654 } 5655 } 5656 5657 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, 5658 struct igb_tx_buffer *first, 5659 u32 vlan_macip_lens, u32 type_tucmd, 5660 u32 mss_l4len_idx) 5661 { 5662 struct e1000_adv_tx_context_desc *context_desc; 5663 u16 i = tx_ring->next_to_use; 5664 struct timespec64 ts; 5665 5666 context_desc = IGB_TX_CTXTDESC(tx_ring, i); 5667 5668 i++; 5669 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; 5670 5671 /* set bits to identify this as an advanced context descriptor */ 5672 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT; 5673 5674 /* For 82575, context index must be unique per ring. */ 5675 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags)) 5676 mss_l4len_idx |= tx_ring->reg_idx << 4; 5677 5678 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens); 5679 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd); 5680 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx); 5681 5682 /* We assume there is always a valid tx time available. Invalid times 5683 * should have been handled by the upper layers. 5684 */ 5685 if (tx_ring->launchtime_enable) { 5686 ts = ns_to_timespec64(first->skb->tstamp); 5687 context_desc->seqnum_seed = cpu_to_le32(ts.tv_nsec / 32); 5688 } else { 5689 context_desc->seqnum_seed = 0; 5690 } 5691 } 5692 5693 static int igb_tso(struct igb_ring *tx_ring, 5694 struct igb_tx_buffer *first, 5695 u8 *hdr_len) 5696 { 5697 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx; 5698 struct sk_buff *skb = first->skb; 5699 union { 5700 struct iphdr *v4; 5701 struct ipv6hdr *v6; 5702 unsigned char *hdr; 5703 } ip; 5704 union { 5705 struct tcphdr *tcp; 5706 unsigned char *hdr; 5707 } l4; 5708 u32 paylen, l4_offset; 5709 int err; 5710 5711 if (skb->ip_summed != CHECKSUM_PARTIAL) 5712 return 0; 5713 5714 if (!skb_is_gso(skb)) 5715 return 0; 5716 5717 err = skb_cow_head(skb, 0); 5718 if (err < 0) 5719 return err; 5720 5721 ip.hdr = skb_network_header(skb); 5722 l4.hdr = skb_checksum_start(skb); 5723 5724 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */ 5725 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP; 5726 5727 /* initialize outer IP header fields */ 5728 if (ip.v4->version == 4) { 5729 unsigned char *csum_start = skb_checksum_start(skb); 5730 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4); 5731 5732 /* IP header will have to cancel out any data that 5733 * is not a part of the outer IP header 5734 */ 5735 ip.v4->check = csum_fold(csum_partial(trans_start, 5736 csum_start - trans_start, 5737 0)); 5738 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4; 5739 5740 ip.v4->tot_len = 0; 5741 first->tx_flags |= IGB_TX_FLAGS_TSO | 5742 IGB_TX_FLAGS_CSUM | 5743 IGB_TX_FLAGS_IPV4; 5744 } else { 5745 ip.v6->payload_len = 0; 5746 first->tx_flags |= IGB_TX_FLAGS_TSO | 5747 IGB_TX_FLAGS_CSUM; 5748 } 5749 5750 /* determine offset of inner transport header */ 5751 l4_offset = l4.hdr - skb->data; 5752 5753 /* compute length of segmentation header */ 5754 *hdr_len = (l4.tcp->doff * 4) + l4_offset; 5755 5756 /* remove payload length from inner checksum */ 5757 paylen = skb->len - l4_offset; 5758 csum_replace_by_diff(&l4.tcp->check, htonl(paylen)); 5759 5760 /* update gso size and bytecount with header size */ 5761 first->gso_segs = skb_shinfo(skb)->gso_segs; 5762 first->bytecount += (first->gso_segs - 1) * *hdr_len; 5763 5764 /* MSS L4LEN IDX */ 5765 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT; 5766 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT; 5767 5768 /* VLAN MACLEN IPLEN */ 5769 vlan_macip_lens = l4.hdr - ip.hdr; 5770 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT; 5771 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK; 5772 5773 igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, 5774 type_tucmd, mss_l4len_idx); 5775 5776 return 1; 5777 } 5778 5779 static inline bool igb_ipv6_csum_is_sctp(struct sk_buff *skb) 5780 { 5781 unsigned int offset = 0; 5782 5783 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL); 5784 5785 return offset == skb_checksum_start_offset(skb); 5786 } 5787 5788 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first) 5789 { 5790 struct sk_buff *skb = first->skb; 5791 u32 vlan_macip_lens = 0; 5792 u32 type_tucmd = 0; 5793 5794 if (skb->ip_summed != CHECKSUM_PARTIAL) { 5795 csum_failed: 5796 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN) && 5797 !tx_ring->launchtime_enable) 5798 return; 5799 goto no_csum; 5800 } 5801 5802 switch (skb->csum_offset) { 5803 case offsetof(struct tcphdr, check): 5804 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP; 5805 /* fall through */ 5806 case offsetof(struct udphdr, check): 5807 break; 5808 case offsetof(struct sctphdr, checksum): 5809 /* validate that this is actually an SCTP request */ 5810 if (((first->protocol == htons(ETH_P_IP)) && 5811 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) || 5812 ((first->protocol == htons(ETH_P_IPV6)) && 5813 igb_ipv6_csum_is_sctp(skb))) { 5814 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP; 5815 break; 5816 } 5817 /* fall through */ 5818 default: 5819 skb_checksum_help(skb); 5820 goto csum_failed; 5821 } 5822 5823 /* update TX checksum flag */ 5824 first->tx_flags |= IGB_TX_FLAGS_CSUM; 5825 vlan_macip_lens = skb_checksum_start_offset(skb) - 5826 skb_network_offset(skb); 5827 no_csum: 5828 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT; 5829 vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK; 5830 5831 igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, type_tucmd, 0); 5832 } 5833 5834 #define IGB_SET_FLAG(_input, _flag, _result) \ 5835 ((_flag <= _result) ? \ 5836 ((u32)(_input & _flag) * (_result / _flag)) : \ 5837 ((u32)(_input & _flag) / (_flag / _result))) 5838 5839 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags) 5840 { 5841 /* set type for advanced descriptor with frame checksum insertion */ 5842 u32 cmd_type = E1000_ADVTXD_DTYP_DATA | 5843 E1000_ADVTXD_DCMD_DEXT | 5844 E1000_ADVTXD_DCMD_IFCS; 5845 5846 /* set HW vlan bit if vlan is present */ 5847 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN, 5848 (E1000_ADVTXD_DCMD_VLE)); 5849 5850 /* set segmentation bits for TSO */ 5851 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO, 5852 (E1000_ADVTXD_DCMD_TSE)); 5853 5854 /* set timestamp bit if present */ 5855 cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP, 5856 (E1000_ADVTXD_MAC_TSTAMP)); 5857 5858 /* insert frame checksum */ 5859 cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS); 5860 5861 return cmd_type; 5862 } 5863 5864 static void igb_tx_olinfo_status(struct igb_ring *tx_ring, 5865 union e1000_adv_tx_desc *tx_desc, 5866 u32 tx_flags, unsigned int paylen) 5867 { 5868 u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT; 5869 5870 /* 82575 requires a unique index per ring */ 5871 if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags)) 5872 olinfo_status |= tx_ring->reg_idx << 4; 5873 5874 /* insert L4 checksum */ 5875 olinfo_status |= IGB_SET_FLAG(tx_flags, 5876 IGB_TX_FLAGS_CSUM, 5877 (E1000_TXD_POPTS_TXSM << 8)); 5878 5879 /* insert IPv4 checksum */ 5880 olinfo_status |= IGB_SET_FLAG(tx_flags, 5881 IGB_TX_FLAGS_IPV4, 5882 (E1000_TXD_POPTS_IXSM << 8)); 5883 5884 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status); 5885 } 5886 5887 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size) 5888 { 5889 struct net_device *netdev = tx_ring->netdev; 5890 5891 netif_stop_subqueue(netdev, tx_ring->queue_index); 5892 5893 /* Herbert's original patch had: 5894 * smp_mb__after_netif_stop_queue(); 5895 * but since that doesn't exist yet, just open code it. 5896 */ 5897 smp_mb(); 5898 5899 /* We need to check again in a case another CPU has just 5900 * made room available. 5901 */ 5902 if (igb_desc_unused(tx_ring) < size) 5903 return -EBUSY; 5904 5905 /* A reprieve! */ 5906 netif_wake_subqueue(netdev, tx_ring->queue_index); 5907 5908 u64_stats_update_begin(&tx_ring->tx_syncp2); 5909 tx_ring->tx_stats.restart_queue2++; 5910 u64_stats_update_end(&tx_ring->tx_syncp2); 5911 5912 return 0; 5913 } 5914 5915 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size) 5916 { 5917 if (igb_desc_unused(tx_ring) >= size) 5918 return 0; 5919 return __igb_maybe_stop_tx(tx_ring, size); 5920 } 5921 5922 static int igb_tx_map(struct igb_ring *tx_ring, 5923 struct igb_tx_buffer *first, 5924 const u8 hdr_len) 5925 { 5926 struct sk_buff *skb = first->skb; 5927 struct igb_tx_buffer *tx_buffer; 5928 union e1000_adv_tx_desc *tx_desc; 5929 struct skb_frag_struct *frag; 5930 dma_addr_t dma; 5931 unsigned int data_len, size; 5932 u32 tx_flags = first->tx_flags; 5933 u32 cmd_type = igb_tx_cmd_type(skb, tx_flags); 5934 u16 i = tx_ring->next_to_use; 5935 5936 tx_desc = IGB_TX_DESC(tx_ring, i); 5937 5938 igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len); 5939 5940 size = skb_headlen(skb); 5941 data_len = skb->data_len; 5942 5943 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE); 5944 5945 tx_buffer = first; 5946 5947 for (frag = &skb_shinfo(skb)->frags[0];; frag++) { 5948 if (dma_mapping_error(tx_ring->dev, dma)) 5949 goto dma_error; 5950 5951 /* record length, and DMA address */ 5952 dma_unmap_len_set(tx_buffer, len, size); 5953 dma_unmap_addr_set(tx_buffer, dma, dma); 5954 5955 tx_desc->read.buffer_addr = cpu_to_le64(dma); 5956 5957 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) { 5958 tx_desc->read.cmd_type_len = 5959 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD); 5960 5961 i++; 5962 tx_desc++; 5963 if (i == tx_ring->count) { 5964 tx_desc = IGB_TX_DESC(tx_ring, 0); 5965 i = 0; 5966 } 5967 tx_desc->read.olinfo_status = 0; 5968 5969 dma += IGB_MAX_DATA_PER_TXD; 5970 size -= IGB_MAX_DATA_PER_TXD; 5971 5972 tx_desc->read.buffer_addr = cpu_to_le64(dma); 5973 } 5974 5975 if (likely(!data_len)) 5976 break; 5977 5978 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size); 5979 5980 i++; 5981 tx_desc++; 5982 if (i == tx_ring->count) { 5983 tx_desc = IGB_TX_DESC(tx_ring, 0); 5984 i = 0; 5985 } 5986 tx_desc->read.olinfo_status = 0; 5987 5988 size = skb_frag_size(frag); 5989 data_len -= size; 5990 5991 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, 5992 size, DMA_TO_DEVICE); 5993 5994 tx_buffer = &tx_ring->tx_buffer_info[i]; 5995 } 5996 5997 /* write last descriptor with RS and EOP bits */ 5998 cmd_type |= size | IGB_TXD_DCMD; 5999 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type); 6000 6001 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount); 6002 6003 /* set the timestamp */ 6004 first->time_stamp = jiffies; 6005 6006 skb_tx_timestamp(skb); 6007 6008 /* Force memory writes to complete before letting h/w know there 6009 * are new descriptors to fetch. (Only applicable for weak-ordered 6010 * memory model archs, such as IA-64). 6011 * 6012 * We also need this memory barrier to make certain all of the 6013 * status bits have been updated before next_to_watch is written. 6014 */ 6015 dma_wmb(); 6016 6017 /* set next_to_watch value indicating a packet is present */ 6018 first->next_to_watch = tx_desc; 6019 6020 i++; 6021 if (i == tx_ring->count) 6022 i = 0; 6023 6024 tx_ring->next_to_use = i; 6025 6026 /* Make sure there is space in the ring for the next send. */ 6027 igb_maybe_stop_tx(tx_ring, DESC_NEEDED); 6028 6029 if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) { 6030 writel(i, tx_ring->tail); 6031 } 6032 return 0; 6033 6034 dma_error: 6035 dev_err(tx_ring->dev, "TX DMA map failed\n"); 6036 tx_buffer = &tx_ring->tx_buffer_info[i]; 6037 6038 /* clear dma mappings for failed tx_buffer_info map */ 6039 while (tx_buffer != first) { 6040 if (dma_unmap_len(tx_buffer, len)) 6041 dma_unmap_page(tx_ring->dev, 6042 dma_unmap_addr(tx_buffer, dma), 6043 dma_unmap_len(tx_buffer, len), 6044 DMA_TO_DEVICE); 6045 dma_unmap_len_set(tx_buffer, len, 0); 6046 6047 if (i-- == 0) 6048 i += tx_ring->count; 6049 tx_buffer = &tx_ring->tx_buffer_info[i]; 6050 } 6051 6052 if (dma_unmap_len(tx_buffer, len)) 6053 dma_unmap_single(tx_ring->dev, 6054 dma_unmap_addr(tx_buffer, dma), 6055 dma_unmap_len(tx_buffer, len), 6056 DMA_TO_DEVICE); 6057 dma_unmap_len_set(tx_buffer, len, 0); 6058 6059 dev_kfree_skb_any(tx_buffer->skb); 6060 tx_buffer->skb = NULL; 6061 6062 tx_ring->next_to_use = i; 6063 6064 return -1; 6065 } 6066 6067 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb, 6068 struct igb_ring *tx_ring) 6069 { 6070 struct igb_tx_buffer *first; 6071 int tso; 6072 u32 tx_flags = 0; 6073 unsigned short f; 6074 u16 count = TXD_USE_COUNT(skb_headlen(skb)); 6075 __be16 protocol = vlan_get_protocol(skb); 6076 u8 hdr_len = 0; 6077 6078 /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD, 6079 * + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD, 6080 * + 2 desc gap to keep tail from touching head, 6081 * + 1 desc for context descriptor, 6082 * otherwise try next time 6083 */ 6084 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) 6085 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size); 6086 6087 if (igb_maybe_stop_tx(tx_ring, count + 3)) { 6088 /* this is a hard error */ 6089 return NETDEV_TX_BUSY; 6090 } 6091 6092 /* record the location of the first descriptor for this packet */ 6093 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use]; 6094 first->skb = skb; 6095 first->bytecount = skb->len; 6096 first->gso_segs = 1; 6097 6098 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) { 6099 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev); 6100 6101 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON && 6102 !test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS, 6103 &adapter->state)) { 6104 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS; 6105 tx_flags |= IGB_TX_FLAGS_TSTAMP; 6106 6107 adapter->ptp_tx_skb = skb_get(skb); 6108 adapter->ptp_tx_start = jiffies; 6109 if (adapter->hw.mac.type == e1000_82576) 6110 schedule_work(&adapter->ptp_tx_work); 6111 } else { 6112 adapter->tx_hwtstamp_skipped++; 6113 } 6114 } 6115 6116 if (skb_vlan_tag_present(skb)) { 6117 tx_flags |= IGB_TX_FLAGS_VLAN; 6118 tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT); 6119 } 6120 6121 /* record initial flags and protocol */ 6122 first->tx_flags = tx_flags; 6123 first->protocol = protocol; 6124 6125 tso = igb_tso(tx_ring, first, &hdr_len); 6126 if (tso < 0) 6127 goto out_drop; 6128 else if (!tso) 6129 igb_tx_csum(tx_ring, first); 6130 6131 if (igb_tx_map(tx_ring, first, hdr_len)) 6132 goto cleanup_tx_tstamp; 6133 6134 return NETDEV_TX_OK; 6135 6136 out_drop: 6137 dev_kfree_skb_any(first->skb); 6138 first->skb = NULL; 6139 cleanup_tx_tstamp: 6140 if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP)) { 6141 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev); 6142 6143 dev_kfree_skb_any(adapter->ptp_tx_skb); 6144 adapter->ptp_tx_skb = NULL; 6145 if (adapter->hw.mac.type == e1000_82576) 6146 cancel_work_sync(&adapter->ptp_tx_work); 6147 clear_bit_unlock(__IGB_PTP_TX_IN_PROGRESS, &adapter->state); 6148 } 6149 6150 return NETDEV_TX_OK; 6151 } 6152 6153 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter, 6154 struct sk_buff *skb) 6155 { 6156 unsigned int r_idx = skb->queue_mapping; 6157 6158 if (r_idx >= adapter->num_tx_queues) 6159 r_idx = r_idx % adapter->num_tx_queues; 6160 6161 return adapter->tx_ring[r_idx]; 6162 } 6163 6164 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, 6165 struct net_device *netdev) 6166 { 6167 struct igb_adapter *adapter = netdev_priv(netdev); 6168 6169 /* The minimum packet size with TCTL.PSP set is 17 so pad the skb 6170 * in order to meet this minimum size requirement. 6171 */ 6172 if (skb_put_padto(skb, 17)) 6173 return NETDEV_TX_OK; 6174 6175 return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb)); 6176 } 6177 6178 /** 6179 * igb_tx_timeout - Respond to a Tx Hang 6180 * @netdev: network interface device structure 6181 **/ 6182 static void igb_tx_timeout(struct net_device *netdev) 6183 { 6184 struct igb_adapter *adapter = netdev_priv(netdev); 6185 struct e1000_hw *hw = &adapter->hw; 6186 6187 /* Do the reset outside of interrupt context */ 6188 adapter->tx_timeout_count++; 6189 6190 if (hw->mac.type >= e1000_82580) 6191 hw->dev_spec._82575.global_device_reset = true; 6192 6193 schedule_work(&adapter->reset_task); 6194 wr32(E1000_EICS, 6195 (adapter->eims_enable_mask & ~adapter->eims_other)); 6196 } 6197 6198 static void igb_reset_task(struct work_struct *work) 6199 { 6200 struct igb_adapter *adapter; 6201 adapter = container_of(work, struct igb_adapter, reset_task); 6202 6203 igb_dump(adapter); 6204 netdev_err(adapter->netdev, "Reset adapter\n"); 6205 igb_reinit_locked(adapter); 6206 } 6207 6208 /** 6209 * igb_get_stats64 - Get System Network Statistics 6210 * @netdev: network interface device structure 6211 * @stats: rtnl_link_stats64 pointer 6212 **/ 6213 static void igb_get_stats64(struct net_device *netdev, 6214 struct rtnl_link_stats64 *stats) 6215 { 6216 struct igb_adapter *adapter = netdev_priv(netdev); 6217 6218 spin_lock(&adapter->stats64_lock); 6219 igb_update_stats(adapter); 6220 memcpy(stats, &adapter->stats64, sizeof(*stats)); 6221 spin_unlock(&adapter->stats64_lock); 6222 } 6223 6224 /** 6225 * igb_change_mtu - Change the Maximum Transfer Unit 6226 * @netdev: network interface device structure 6227 * @new_mtu: new value for maximum frame size 6228 * 6229 * Returns 0 on success, negative on failure 6230 **/ 6231 static int igb_change_mtu(struct net_device *netdev, int new_mtu) 6232 { 6233 struct igb_adapter *adapter = netdev_priv(netdev); 6234 struct pci_dev *pdev = adapter->pdev; 6235 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN; 6236 6237 /* adjust max frame to be at least the size of a standard frame */ 6238 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN)) 6239 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN; 6240 6241 while (test_and_set_bit(__IGB_RESETTING, &adapter->state)) 6242 usleep_range(1000, 2000); 6243 6244 /* igb_down has a dependency on max_frame_size */ 6245 adapter->max_frame_size = max_frame; 6246 6247 if (netif_running(netdev)) 6248 igb_down(adapter); 6249 6250 dev_info(&pdev->dev, "changing MTU from %d to %d\n", 6251 netdev->mtu, new_mtu); 6252 netdev->mtu = new_mtu; 6253 6254 if (netif_running(netdev)) 6255 igb_up(adapter); 6256 else 6257 igb_reset(adapter); 6258 6259 clear_bit(__IGB_RESETTING, &adapter->state); 6260 6261 return 0; 6262 } 6263 6264 /** 6265 * igb_update_stats - Update the board statistics counters 6266 * @adapter: board private structure 6267 **/ 6268 void igb_update_stats(struct igb_adapter *adapter) 6269 { 6270 struct rtnl_link_stats64 *net_stats = &adapter->stats64; 6271 struct e1000_hw *hw = &adapter->hw; 6272 struct pci_dev *pdev = adapter->pdev; 6273 u32 reg, mpc; 6274 int i; 6275 u64 bytes, packets; 6276 unsigned int start; 6277 u64 _bytes, _packets; 6278 6279 /* Prevent stats update while adapter is being reset, or if the pci 6280 * connection is down. 6281 */ 6282 if (adapter->link_speed == 0) 6283 return; 6284 if (pci_channel_offline(pdev)) 6285 return; 6286 6287 bytes = 0; 6288 packets = 0; 6289 6290 rcu_read_lock(); 6291 for (i = 0; i < adapter->num_rx_queues; i++) { 6292 struct igb_ring *ring = adapter->rx_ring[i]; 6293 u32 rqdpc = rd32(E1000_RQDPC(i)); 6294 if (hw->mac.type >= e1000_i210) 6295 wr32(E1000_RQDPC(i), 0); 6296 6297 if (rqdpc) { 6298 ring->rx_stats.drops += rqdpc; 6299 net_stats->rx_fifo_errors += rqdpc; 6300 } 6301 6302 do { 6303 start = u64_stats_fetch_begin_irq(&ring->rx_syncp); 6304 _bytes = ring->rx_stats.bytes; 6305 _packets = ring->rx_stats.packets; 6306 } while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start)); 6307 bytes += _bytes; 6308 packets += _packets; 6309 } 6310 6311 net_stats->rx_bytes = bytes; 6312 net_stats->rx_packets = packets; 6313 6314 bytes = 0; 6315 packets = 0; 6316 for (i = 0; i < adapter->num_tx_queues; i++) { 6317 struct igb_ring *ring = adapter->tx_ring[i]; 6318 do { 6319 start = u64_stats_fetch_begin_irq(&ring->tx_syncp); 6320 _bytes = ring->tx_stats.bytes; 6321 _packets = ring->tx_stats.packets; 6322 } while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start)); 6323 bytes += _bytes; 6324 packets += _packets; 6325 } 6326 net_stats->tx_bytes = bytes; 6327 net_stats->tx_packets = packets; 6328 rcu_read_unlock(); 6329 6330 /* read stats registers */ 6331 adapter->stats.crcerrs += rd32(E1000_CRCERRS); 6332 adapter->stats.gprc += rd32(E1000_GPRC); 6333 adapter->stats.gorc += rd32(E1000_GORCL); 6334 rd32(E1000_GORCH); /* clear GORCL */ 6335 adapter->stats.bprc += rd32(E1000_BPRC); 6336 adapter->stats.mprc += rd32(E1000_MPRC); 6337 adapter->stats.roc += rd32(E1000_ROC); 6338 6339 adapter->stats.prc64 += rd32(E1000_PRC64); 6340 adapter->stats.prc127 += rd32(E1000_PRC127); 6341 adapter->stats.prc255 += rd32(E1000_PRC255); 6342 adapter->stats.prc511 += rd32(E1000_PRC511); 6343 adapter->stats.prc1023 += rd32(E1000_PRC1023); 6344 adapter->stats.prc1522 += rd32(E1000_PRC1522); 6345 adapter->stats.symerrs += rd32(E1000_SYMERRS); 6346 adapter->stats.sec += rd32(E1000_SEC); 6347 6348 mpc = rd32(E1000_MPC); 6349 adapter->stats.mpc += mpc; 6350 net_stats->rx_fifo_errors += mpc; 6351 adapter->stats.scc += rd32(E1000_SCC); 6352 adapter->stats.ecol += rd32(E1000_ECOL); 6353 adapter->stats.mcc += rd32(E1000_MCC); 6354 adapter->stats.latecol += rd32(E1000_LATECOL); 6355 adapter->stats.dc += rd32(E1000_DC); 6356 adapter->stats.rlec += rd32(E1000_RLEC); 6357 adapter->stats.xonrxc += rd32(E1000_XONRXC); 6358 adapter->stats.xontxc += rd32(E1000_XONTXC); 6359 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC); 6360 adapter->stats.xofftxc += rd32(E1000_XOFFTXC); 6361 adapter->stats.fcruc += rd32(E1000_FCRUC); 6362 adapter->stats.gptc += rd32(E1000_GPTC); 6363 adapter->stats.gotc += rd32(E1000_GOTCL); 6364 rd32(E1000_GOTCH); /* clear GOTCL */ 6365 adapter->stats.rnbc += rd32(E1000_RNBC); 6366 adapter->stats.ruc += rd32(E1000_RUC); 6367 adapter->stats.rfc += rd32(E1000_RFC); 6368 adapter->stats.rjc += rd32(E1000_RJC); 6369 adapter->stats.tor += rd32(E1000_TORH); 6370 adapter->stats.tot += rd32(E1000_TOTH); 6371 adapter->stats.tpr += rd32(E1000_TPR); 6372 6373 adapter->stats.ptc64 += rd32(E1000_PTC64); 6374 adapter->stats.ptc127 += rd32(E1000_PTC127); 6375 adapter->stats.ptc255 += rd32(E1000_PTC255); 6376 adapter->stats.ptc511 += rd32(E1000_PTC511); 6377 adapter->stats.ptc1023 += rd32(E1000_PTC1023); 6378 adapter->stats.ptc1522 += rd32(E1000_PTC1522); 6379 6380 adapter->stats.mptc += rd32(E1000_MPTC); 6381 adapter->stats.bptc += rd32(E1000_BPTC); 6382 6383 adapter->stats.tpt += rd32(E1000_TPT); 6384 adapter->stats.colc += rd32(E1000_COLC); 6385 6386 adapter->stats.algnerrc += rd32(E1000_ALGNERRC); 6387 /* read internal phy specific stats */ 6388 reg = rd32(E1000_CTRL_EXT); 6389 if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) { 6390 adapter->stats.rxerrc += rd32(E1000_RXERRC); 6391 6392 /* this stat has invalid values on i210/i211 */ 6393 if ((hw->mac.type != e1000_i210) && 6394 (hw->mac.type != e1000_i211)) 6395 adapter->stats.tncrs += rd32(E1000_TNCRS); 6396 } 6397 6398 adapter->stats.tsctc += rd32(E1000_TSCTC); 6399 adapter->stats.tsctfc += rd32(E1000_TSCTFC); 6400 6401 adapter->stats.iac += rd32(E1000_IAC); 6402 adapter->stats.icrxoc += rd32(E1000_ICRXOC); 6403 adapter->stats.icrxptc += rd32(E1000_ICRXPTC); 6404 adapter->stats.icrxatc += rd32(E1000_ICRXATC); 6405 adapter->stats.ictxptc += rd32(E1000_ICTXPTC); 6406 adapter->stats.ictxatc += rd32(E1000_ICTXATC); 6407 adapter->stats.ictxqec += rd32(E1000_ICTXQEC); 6408 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC); 6409 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC); 6410 6411 /* Fill out the OS statistics structure */ 6412 net_stats->multicast = adapter->stats.mprc; 6413 net_stats->collisions = adapter->stats.colc; 6414 6415 /* Rx Errors */ 6416 6417 /* RLEC on some newer hardware can be incorrect so build 6418 * our own version based on RUC and ROC 6419 */ 6420 net_stats->rx_errors = adapter->stats.rxerrc + 6421 adapter->stats.crcerrs + adapter->stats.algnerrc + 6422 adapter->stats.ruc + adapter->stats.roc + 6423 adapter->stats.cexterr; 6424 net_stats->rx_length_errors = adapter->stats.ruc + 6425 adapter->stats.roc; 6426 net_stats->rx_crc_errors = adapter->stats.crcerrs; 6427 net_stats->rx_frame_errors = adapter->stats.algnerrc; 6428 net_stats->rx_missed_errors = adapter->stats.mpc; 6429 6430 /* Tx Errors */ 6431 net_stats->tx_errors = adapter->stats.ecol + 6432 adapter->stats.latecol; 6433 net_stats->tx_aborted_errors = adapter->stats.ecol; 6434 net_stats->tx_window_errors = adapter->stats.latecol; 6435 net_stats->tx_carrier_errors = adapter->stats.tncrs; 6436 6437 /* Tx Dropped needs to be maintained elsewhere */ 6438 6439 /* Management Stats */ 6440 adapter->stats.mgptc += rd32(E1000_MGTPTC); 6441 adapter->stats.mgprc += rd32(E1000_MGTPRC); 6442 adapter->stats.mgpdc += rd32(E1000_MGTPDC); 6443 6444 /* OS2BMC Stats */ 6445 reg = rd32(E1000_MANC); 6446 if (reg & E1000_MANC_EN_BMC2OS) { 6447 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC); 6448 adapter->stats.o2bspc += rd32(E1000_O2BSPC); 6449 adapter->stats.b2ospc += rd32(E1000_B2OSPC); 6450 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC); 6451 } 6452 } 6453 6454 static void igb_tsync_interrupt(struct igb_adapter *adapter) 6455 { 6456 struct e1000_hw *hw = &adapter->hw; 6457 struct ptp_clock_event event; 6458 struct timespec64 ts; 6459 u32 ack = 0, tsauxc, sec, nsec, tsicr = rd32(E1000_TSICR); 6460 6461 if (tsicr & TSINTR_SYS_WRAP) { 6462 event.type = PTP_CLOCK_PPS; 6463 if (adapter->ptp_caps.pps) 6464 ptp_clock_event(adapter->ptp_clock, &event); 6465 ack |= TSINTR_SYS_WRAP; 6466 } 6467 6468 if (tsicr & E1000_TSICR_TXTS) { 6469 /* retrieve hardware timestamp */ 6470 schedule_work(&adapter->ptp_tx_work); 6471 ack |= E1000_TSICR_TXTS; 6472 } 6473 6474 if (tsicr & TSINTR_TT0) { 6475 spin_lock(&adapter->tmreg_lock); 6476 ts = timespec64_add(adapter->perout[0].start, 6477 adapter->perout[0].period); 6478 /* u32 conversion of tv_sec is safe until y2106 */ 6479 wr32(E1000_TRGTTIML0, ts.tv_nsec); 6480 wr32(E1000_TRGTTIMH0, (u32)ts.tv_sec); 6481 tsauxc = rd32(E1000_TSAUXC); 6482 tsauxc |= TSAUXC_EN_TT0; 6483 wr32(E1000_TSAUXC, tsauxc); 6484 adapter->perout[0].start = ts; 6485 spin_unlock(&adapter->tmreg_lock); 6486 ack |= TSINTR_TT0; 6487 } 6488 6489 if (tsicr & TSINTR_TT1) { 6490 spin_lock(&adapter->tmreg_lock); 6491 ts = timespec64_add(adapter->perout[1].start, 6492 adapter->perout[1].period); 6493 wr32(E1000_TRGTTIML1, ts.tv_nsec); 6494 wr32(E1000_TRGTTIMH1, (u32)ts.tv_sec); 6495 tsauxc = rd32(E1000_TSAUXC); 6496 tsauxc |= TSAUXC_EN_TT1; 6497 wr32(E1000_TSAUXC, tsauxc); 6498 adapter->perout[1].start = ts; 6499 spin_unlock(&adapter->tmreg_lock); 6500 ack |= TSINTR_TT1; 6501 } 6502 6503 if (tsicr & TSINTR_AUTT0) { 6504 nsec = rd32(E1000_AUXSTMPL0); 6505 sec = rd32(E1000_AUXSTMPH0); 6506 event.type = PTP_CLOCK_EXTTS; 6507 event.index = 0; 6508 event.timestamp = sec * 1000000000ULL + nsec; 6509 ptp_clock_event(adapter->ptp_clock, &event); 6510 ack |= TSINTR_AUTT0; 6511 } 6512 6513 if (tsicr & TSINTR_AUTT1) { 6514 nsec = rd32(E1000_AUXSTMPL1); 6515 sec = rd32(E1000_AUXSTMPH1); 6516 event.type = PTP_CLOCK_EXTTS; 6517 event.index = 1; 6518 event.timestamp = sec * 1000000000ULL + nsec; 6519 ptp_clock_event(adapter->ptp_clock, &event); 6520 ack |= TSINTR_AUTT1; 6521 } 6522 6523 /* acknowledge the interrupts */ 6524 wr32(E1000_TSICR, ack); 6525 } 6526 6527 static irqreturn_t igb_msix_other(int irq, void *data) 6528 { 6529 struct igb_adapter *adapter = data; 6530 struct e1000_hw *hw = &adapter->hw; 6531 u32 icr = rd32(E1000_ICR); 6532 /* reading ICR causes bit 31 of EICR to be cleared */ 6533 6534 if (icr & E1000_ICR_DRSTA) 6535 schedule_work(&adapter->reset_task); 6536 6537 if (icr & E1000_ICR_DOUTSYNC) { 6538 /* HW is reporting DMA is out of sync */ 6539 adapter->stats.doosync++; 6540 /* The DMA Out of Sync is also indication of a spoof event 6541 * in IOV mode. Check the Wrong VM Behavior register to 6542 * see if it is really a spoof event. 6543 */ 6544 igb_check_wvbr(adapter); 6545 } 6546 6547 /* Check for a mailbox event */ 6548 if (icr & E1000_ICR_VMMB) 6549 igb_msg_task(adapter); 6550 6551 if (icr & E1000_ICR_LSC) { 6552 hw->mac.get_link_status = 1; 6553 /* guard against interrupt when we're going down */ 6554 if (!test_bit(__IGB_DOWN, &adapter->state)) 6555 mod_timer(&adapter->watchdog_timer, jiffies + 1); 6556 } 6557 6558 if (icr & E1000_ICR_TS) 6559 igb_tsync_interrupt(adapter); 6560 6561 wr32(E1000_EIMS, adapter->eims_other); 6562 6563 return IRQ_HANDLED; 6564 } 6565 6566 static void igb_write_itr(struct igb_q_vector *q_vector) 6567 { 6568 struct igb_adapter *adapter = q_vector->adapter; 6569 u32 itr_val = q_vector->itr_val & 0x7FFC; 6570 6571 if (!q_vector->set_itr) 6572 return; 6573 6574 if (!itr_val) 6575 itr_val = 0x4; 6576 6577 if (adapter->hw.mac.type == e1000_82575) 6578 itr_val |= itr_val << 16; 6579 else 6580 itr_val |= E1000_EITR_CNT_IGNR; 6581 6582 writel(itr_val, q_vector->itr_register); 6583 q_vector->set_itr = 0; 6584 } 6585 6586 static irqreturn_t igb_msix_ring(int irq, void *data) 6587 { 6588 struct igb_q_vector *q_vector = data; 6589 6590 /* Write the ITR value calculated from the previous interrupt. */ 6591 igb_write_itr(q_vector); 6592 6593 napi_schedule(&q_vector->napi); 6594 6595 return IRQ_HANDLED; 6596 } 6597 6598 #ifdef CONFIG_IGB_DCA 6599 static void igb_update_tx_dca(struct igb_adapter *adapter, 6600 struct igb_ring *tx_ring, 6601 int cpu) 6602 { 6603 struct e1000_hw *hw = &adapter->hw; 6604 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu); 6605 6606 if (hw->mac.type != e1000_82575) 6607 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT; 6608 6609 /* We can enable relaxed ordering for reads, but not writes when 6610 * DCA is enabled. This is due to a known issue in some chipsets 6611 * which will cause the DCA tag to be cleared. 6612 */ 6613 txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN | 6614 E1000_DCA_TXCTRL_DATA_RRO_EN | 6615 E1000_DCA_TXCTRL_DESC_DCA_EN; 6616 6617 wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl); 6618 } 6619 6620 static void igb_update_rx_dca(struct igb_adapter *adapter, 6621 struct igb_ring *rx_ring, 6622 int cpu) 6623 { 6624 struct e1000_hw *hw = &adapter->hw; 6625 u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu); 6626 6627 if (hw->mac.type != e1000_82575) 6628 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT; 6629 6630 /* We can enable relaxed ordering for reads, but not writes when 6631 * DCA is enabled. This is due to a known issue in some chipsets 6632 * which will cause the DCA tag to be cleared. 6633 */ 6634 rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN | 6635 E1000_DCA_RXCTRL_DESC_DCA_EN; 6636 6637 wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl); 6638 } 6639 6640 static void igb_update_dca(struct igb_q_vector *q_vector) 6641 { 6642 struct igb_adapter *adapter = q_vector->adapter; 6643 int cpu = get_cpu(); 6644 6645 if (q_vector->cpu == cpu) 6646 goto out_no_update; 6647 6648 if (q_vector->tx.ring) 6649 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu); 6650 6651 if (q_vector->rx.ring) 6652 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu); 6653 6654 q_vector->cpu = cpu; 6655 out_no_update: 6656 put_cpu(); 6657 } 6658 6659 static void igb_setup_dca(struct igb_adapter *adapter) 6660 { 6661 struct e1000_hw *hw = &adapter->hw; 6662 int i; 6663 6664 if (!(adapter->flags & IGB_FLAG_DCA_ENABLED)) 6665 return; 6666 6667 /* Always use CB2 mode, difference is masked in the CB driver. */ 6668 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2); 6669 6670 for (i = 0; i < adapter->num_q_vectors; i++) { 6671 adapter->q_vector[i]->cpu = -1; 6672 igb_update_dca(adapter->q_vector[i]); 6673 } 6674 } 6675 6676 static int __igb_notify_dca(struct device *dev, void *data) 6677 { 6678 struct net_device *netdev = dev_get_drvdata(dev); 6679 struct igb_adapter *adapter = netdev_priv(netdev); 6680 struct pci_dev *pdev = adapter->pdev; 6681 struct e1000_hw *hw = &adapter->hw; 6682 unsigned long event = *(unsigned long *)data; 6683 6684 switch (event) { 6685 case DCA_PROVIDER_ADD: 6686 /* if already enabled, don't do it again */ 6687 if (adapter->flags & IGB_FLAG_DCA_ENABLED) 6688 break; 6689 if (dca_add_requester(dev) == 0) { 6690 adapter->flags |= IGB_FLAG_DCA_ENABLED; 6691 dev_info(&pdev->dev, "DCA enabled\n"); 6692 igb_setup_dca(adapter); 6693 break; 6694 } 6695 /* Fall Through since DCA is disabled. */ 6696 case DCA_PROVIDER_REMOVE: 6697 if (adapter->flags & IGB_FLAG_DCA_ENABLED) { 6698 /* without this a class_device is left 6699 * hanging around in the sysfs model 6700 */ 6701 dca_remove_requester(dev); 6702 dev_info(&pdev->dev, "DCA disabled\n"); 6703 adapter->flags &= ~IGB_FLAG_DCA_ENABLED; 6704 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE); 6705 } 6706 break; 6707 } 6708 6709 return 0; 6710 } 6711 6712 static int igb_notify_dca(struct notifier_block *nb, unsigned long event, 6713 void *p) 6714 { 6715 int ret_val; 6716 6717 ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event, 6718 __igb_notify_dca); 6719 6720 return ret_val ? NOTIFY_BAD : NOTIFY_DONE; 6721 } 6722 #endif /* CONFIG_IGB_DCA */ 6723 6724 #ifdef CONFIG_PCI_IOV 6725 static int igb_vf_configure(struct igb_adapter *adapter, int vf) 6726 { 6727 unsigned char mac_addr[ETH_ALEN]; 6728 6729 eth_zero_addr(mac_addr); 6730 igb_set_vf_mac(adapter, vf, mac_addr); 6731 6732 /* By default spoof check is enabled for all VFs */ 6733 adapter->vf_data[vf].spoofchk_enabled = true; 6734 6735 /* By default VFs are not trusted */ 6736 adapter->vf_data[vf].trusted = false; 6737 6738 return 0; 6739 } 6740 6741 #endif 6742 static void igb_ping_all_vfs(struct igb_adapter *adapter) 6743 { 6744 struct e1000_hw *hw = &adapter->hw; 6745 u32 ping; 6746 int i; 6747 6748 for (i = 0 ; i < adapter->vfs_allocated_count; i++) { 6749 ping = E1000_PF_CONTROL_MSG; 6750 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS) 6751 ping |= E1000_VT_MSGTYPE_CTS; 6752 igb_write_mbx(hw, &ping, 1, i); 6753 } 6754 } 6755 6756 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf) 6757 { 6758 struct e1000_hw *hw = &adapter->hw; 6759 u32 vmolr = rd32(E1000_VMOLR(vf)); 6760 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 6761 6762 vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC | 6763 IGB_VF_FLAG_MULTI_PROMISC); 6764 vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME); 6765 6766 if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) { 6767 vmolr |= E1000_VMOLR_MPME; 6768 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC; 6769 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST; 6770 } else { 6771 /* if we have hashes and we are clearing a multicast promisc 6772 * flag we need to write the hashes to the MTA as this step 6773 * was previously skipped 6774 */ 6775 if (vf_data->num_vf_mc_hashes > 30) { 6776 vmolr |= E1000_VMOLR_MPME; 6777 } else if (vf_data->num_vf_mc_hashes) { 6778 int j; 6779 6780 vmolr |= E1000_VMOLR_ROMPE; 6781 for (j = 0; j < vf_data->num_vf_mc_hashes; j++) 6782 igb_mta_set(hw, vf_data->vf_mc_hashes[j]); 6783 } 6784 } 6785 6786 wr32(E1000_VMOLR(vf), vmolr); 6787 6788 /* there are flags left unprocessed, likely not supported */ 6789 if (*msgbuf & E1000_VT_MSGINFO_MASK) 6790 return -EINVAL; 6791 6792 return 0; 6793 } 6794 6795 static int igb_set_vf_multicasts(struct igb_adapter *adapter, 6796 u32 *msgbuf, u32 vf) 6797 { 6798 int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT; 6799 u16 *hash_list = (u16 *)&msgbuf[1]; 6800 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 6801 int i; 6802 6803 /* salt away the number of multicast addresses assigned 6804 * to this VF for later use to restore when the PF multi cast 6805 * list changes 6806 */ 6807 vf_data->num_vf_mc_hashes = n; 6808 6809 /* only up to 30 hash values supported */ 6810 if (n > 30) 6811 n = 30; 6812 6813 /* store the hashes for later use */ 6814 for (i = 0; i < n; i++) 6815 vf_data->vf_mc_hashes[i] = hash_list[i]; 6816 6817 /* Flush and reset the mta with the new values */ 6818 igb_set_rx_mode(adapter->netdev); 6819 6820 return 0; 6821 } 6822 6823 static void igb_restore_vf_multicasts(struct igb_adapter *adapter) 6824 { 6825 struct e1000_hw *hw = &adapter->hw; 6826 struct vf_data_storage *vf_data; 6827 int i, j; 6828 6829 for (i = 0; i < adapter->vfs_allocated_count; i++) { 6830 u32 vmolr = rd32(E1000_VMOLR(i)); 6831 6832 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME); 6833 6834 vf_data = &adapter->vf_data[i]; 6835 6836 if ((vf_data->num_vf_mc_hashes > 30) || 6837 (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) { 6838 vmolr |= E1000_VMOLR_MPME; 6839 } else if (vf_data->num_vf_mc_hashes) { 6840 vmolr |= E1000_VMOLR_ROMPE; 6841 for (j = 0; j < vf_data->num_vf_mc_hashes; j++) 6842 igb_mta_set(hw, vf_data->vf_mc_hashes[j]); 6843 } 6844 wr32(E1000_VMOLR(i), vmolr); 6845 } 6846 } 6847 6848 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf) 6849 { 6850 struct e1000_hw *hw = &adapter->hw; 6851 u32 pool_mask, vlvf_mask, i; 6852 6853 /* create mask for VF and other pools */ 6854 pool_mask = E1000_VLVF_POOLSEL_MASK; 6855 vlvf_mask = BIT(E1000_VLVF_POOLSEL_SHIFT + vf); 6856 6857 /* drop PF from pool bits */ 6858 pool_mask &= ~BIT(E1000_VLVF_POOLSEL_SHIFT + 6859 adapter->vfs_allocated_count); 6860 6861 /* Find the vlan filter for this id */ 6862 for (i = E1000_VLVF_ARRAY_SIZE; i--;) { 6863 u32 vlvf = rd32(E1000_VLVF(i)); 6864 u32 vfta_mask, vid, vfta; 6865 6866 /* remove the vf from the pool */ 6867 if (!(vlvf & vlvf_mask)) 6868 continue; 6869 6870 /* clear out bit from VLVF */ 6871 vlvf ^= vlvf_mask; 6872 6873 /* if other pools are present, just remove ourselves */ 6874 if (vlvf & pool_mask) 6875 goto update_vlvfb; 6876 6877 /* if PF is present, leave VFTA */ 6878 if (vlvf & E1000_VLVF_POOLSEL_MASK) 6879 goto update_vlvf; 6880 6881 vid = vlvf & E1000_VLVF_VLANID_MASK; 6882 vfta_mask = BIT(vid % 32); 6883 6884 /* clear bit from VFTA */ 6885 vfta = adapter->shadow_vfta[vid / 32]; 6886 if (vfta & vfta_mask) 6887 hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask); 6888 update_vlvf: 6889 /* clear pool selection enable */ 6890 if (adapter->flags & IGB_FLAG_VLAN_PROMISC) 6891 vlvf &= E1000_VLVF_POOLSEL_MASK; 6892 else 6893 vlvf = 0; 6894 update_vlvfb: 6895 /* clear pool bits */ 6896 wr32(E1000_VLVF(i), vlvf); 6897 } 6898 } 6899 6900 static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan) 6901 { 6902 u32 vlvf; 6903 int idx; 6904 6905 /* short cut the special case */ 6906 if (vlan == 0) 6907 return 0; 6908 6909 /* Search for the VLAN id in the VLVF entries */ 6910 for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) { 6911 vlvf = rd32(E1000_VLVF(idx)); 6912 if ((vlvf & VLAN_VID_MASK) == vlan) 6913 break; 6914 } 6915 6916 return idx; 6917 } 6918 6919 static void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid) 6920 { 6921 struct e1000_hw *hw = &adapter->hw; 6922 u32 bits, pf_id; 6923 int idx; 6924 6925 idx = igb_find_vlvf_entry(hw, vid); 6926 if (!idx) 6927 return; 6928 6929 /* See if any other pools are set for this VLAN filter 6930 * entry other than the PF. 6931 */ 6932 pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT; 6933 bits = ~BIT(pf_id) & E1000_VLVF_POOLSEL_MASK; 6934 bits &= rd32(E1000_VLVF(idx)); 6935 6936 /* Disable the filter so this falls into the default pool. */ 6937 if (!bits) { 6938 if (adapter->flags & IGB_FLAG_VLAN_PROMISC) 6939 wr32(E1000_VLVF(idx), BIT(pf_id)); 6940 else 6941 wr32(E1000_VLVF(idx), 0); 6942 } 6943 } 6944 6945 static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid, 6946 bool add, u32 vf) 6947 { 6948 int pf_id = adapter->vfs_allocated_count; 6949 struct e1000_hw *hw = &adapter->hw; 6950 int err; 6951 6952 /* If VLAN overlaps with one the PF is currently monitoring make 6953 * sure that we are able to allocate a VLVF entry. This may be 6954 * redundant but it guarantees PF will maintain visibility to 6955 * the VLAN. 6956 */ 6957 if (add && test_bit(vid, adapter->active_vlans)) { 6958 err = igb_vfta_set(hw, vid, pf_id, true, false); 6959 if (err) 6960 return err; 6961 } 6962 6963 err = igb_vfta_set(hw, vid, vf, add, false); 6964 6965 if (add && !err) 6966 return err; 6967 6968 /* If we failed to add the VF VLAN or we are removing the VF VLAN 6969 * we may need to drop the PF pool bit in order to allow us to free 6970 * up the VLVF resources. 6971 */ 6972 if (test_bit(vid, adapter->active_vlans) || 6973 (adapter->flags & IGB_FLAG_VLAN_PROMISC)) 6974 igb_update_pf_vlvf(adapter, vid); 6975 6976 return err; 6977 } 6978 6979 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf) 6980 { 6981 struct e1000_hw *hw = &adapter->hw; 6982 6983 if (vid) 6984 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT)); 6985 else 6986 wr32(E1000_VMVIR(vf), 0); 6987 } 6988 6989 static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf, 6990 u16 vlan, u8 qos) 6991 { 6992 int err; 6993 6994 err = igb_set_vf_vlan(adapter, vlan, true, vf); 6995 if (err) 6996 return err; 6997 6998 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf); 6999 igb_set_vmolr(adapter, vf, !vlan); 7000 7001 /* revoke access to previous VLAN */ 7002 if (vlan != adapter->vf_data[vf].pf_vlan) 7003 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan, 7004 false, vf); 7005 7006 adapter->vf_data[vf].pf_vlan = vlan; 7007 adapter->vf_data[vf].pf_qos = qos; 7008 igb_set_vf_vlan_strip(adapter, vf, true); 7009 dev_info(&adapter->pdev->dev, 7010 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf); 7011 if (test_bit(__IGB_DOWN, &adapter->state)) { 7012 dev_warn(&adapter->pdev->dev, 7013 "The VF VLAN has been set, but the PF device is not up.\n"); 7014 dev_warn(&adapter->pdev->dev, 7015 "Bring the PF device up before attempting to use the VF device.\n"); 7016 } 7017 7018 return err; 7019 } 7020 7021 static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf) 7022 { 7023 /* Restore tagless access via VLAN 0 */ 7024 igb_set_vf_vlan(adapter, 0, true, vf); 7025 7026 igb_set_vmvir(adapter, 0, vf); 7027 igb_set_vmolr(adapter, vf, true); 7028 7029 /* Remove any PF assigned VLAN */ 7030 if (adapter->vf_data[vf].pf_vlan) 7031 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan, 7032 false, vf); 7033 7034 adapter->vf_data[vf].pf_vlan = 0; 7035 adapter->vf_data[vf].pf_qos = 0; 7036 igb_set_vf_vlan_strip(adapter, vf, false); 7037 7038 return 0; 7039 } 7040 7041 static int igb_ndo_set_vf_vlan(struct net_device *netdev, int vf, 7042 u16 vlan, u8 qos, __be16 vlan_proto) 7043 { 7044 struct igb_adapter *adapter = netdev_priv(netdev); 7045 7046 if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7)) 7047 return -EINVAL; 7048 7049 if (vlan_proto != htons(ETH_P_8021Q)) 7050 return -EPROTONOSUPPORT; 7051 7052 return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) : 7053 igb_disable_port_vlan(adapter, vf); 7054 } 7055 7056 static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf) 7057 { 7058 int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT; 7059 int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK); 7060 int ret; 7061 7062 if (adapter->vf_data[vf].pf_vlan) 7063 return -1; 7064 7065 /* VLAN 0 is a special case, don't allow it to be removed */ 7066 if (!vid && !add) 7067 return 0; 7068 7069 ret = igb_set_vf_vlan(adapter, vid, !!add, vf); 7070 if (!ret) 7071 igb_set_vf_vlan_strip(adapter, vf, !!vid); 7072 return ret; 7073 } 7074 7075 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf) 7076 { 7077 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 7078 7079 /* clear flags - except flag that indicates PF has set the MAC */ 7080 vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC; 7081 vf_data->last_nack = jiffies; 7082 7083 /* reset vlans for device */ 7084 igb_clear_vf_vfta(adapter, vf); 7085 igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf); 7086 igb_set_vmvir(adapter, vf_data->pf_vlan | 7087 (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf); 7088 igb_set_vmolr(adapter, vf, !vf_data->pf_vlan); 7089 igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan)); 7090 7091 /* reset multicast table array for vf */ 7092 adapter->vf_data[vf].num_vf_mc_hashes = 0; 7093 7094 /* Flush and reset the mta with the new values */ 7095 igb_set_rx_mode(adapter->netdev); 7096 } 7097 7098 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf) 7099 { 7100 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses; 7101 7102 /* clear mac address as we were hotplug removed/added */ 7103 if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC)) 7104 eth_zero_addr(vf_mac); 7105 7106 /* process remaining reset events */ 7107 igb_vf_reset(adapter, vf); 7108 } 7109 7110 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf) 7111 { 7112 struct e1000_hw *hw = &adapter->hw; 7113 unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses; 7114 u32 reg, msgbuf[3]; 7115 u8 *addr = (u8 *)(&msgbuf[1]); 7116 7117 /* process all the same items cleared in a function level reset */ 7118 igb_vf_reset(adapter, vf); 7119 7120 /* set vf mac address */ 7121 igb_set_vf_mac(adapter, vf, vf_mac); 7122 7123 /* enable transmit and receive for vf */ 7124 reg = rd32(E1000_VFTE); 7125 wr32(E1000_VFTE, reg | BIT(vf)); 7126 reg = rd32(E1000_VFRE); 7127 wr32(E1000_VFRE, reg | BIT(vf)); 7128 7129 adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS; 7130 7131 /* reply to reset with ack and vf mac address */ 7132 if (!is_zero_ether_addr(vf_mac)) { 7133 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK; 7134 memcpy(addr, vf_mac, ETH_ALEN); 7135 } else { 7136 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK; 7137 } 7138 igb_write_mbx(hw, msgbuf, 3, vf); 7139 } 7140 7141 static void igb_flush_mac_table(struct igb_adapter *adapter) 7142 { 7143 struct e1000_hw *hw = &adapter->hw; 7144 int i; 7145 7146 for (i = 0; i < hw->mac.rar_entry_count; i++) { 7147 adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE; 7148 memset(adapter->mac_table[i].addr, 0, ETH_ALEN); 7149 adapter->mac_table[i].queue = 0; 7150 igb_rar_set_index(adapter, i); 7151 } 7152 } 7153 7154 static int igb_available_rars(struct igb_adapter *adapter, u8 queue) 7155 { 7156 struct e1000_hw *hw = &adapter->hw; 7157 /* do not count rar entries reserved for VFs MAC addresses */ 7158 int rar_entries = hw->mac.rar_entry_count - 7159 adapter->vfs_allocated_count; 7160 int i, count = 0; 7161 7162 for (i = 0; i < rar_entries; i++) { 7163 /* do not count default entries */ 7164 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT) 7165 continue; 7166 7167 /* do not count "in use" entries for different queues */ 7168 if ((adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE) && 7169 (adapter->mac_table[i].queue != queue)) 7170 continue; 7171 7172 count++; 7173 } 7174 7175 return count; 7176 } 7177 7178 /* Set default MAC address for the PF in the first RAR entry */ 7179 static void igb_set_default_mac_filter(struct igb_adapter *adapter) 7180 { 7181 struct igb_mac_addr *mac_table = &adapter->mac_table[0]; 7182 7183 ether_addr_copy(mac_table->addr, adapter->hw.mac.addr); 7184 mac_table->queue = adapter->vfs_allocated_count; 7185 mac_table->state = IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE; 7186 7187 igb_rar_set_index(adapter, 0); 7188 } 7189 7190 /* If the filter to be added and an already existing filter express 7191 * the same address and address type, it should be possible to only 7192 * override the other configurations, for example the queue to steer 7193 * traffic. 7194 */ 7195 static bool igb_mac_entry_can_be_used(const struct igb_mac_addr *entry, 7196 const u8 *addr, const u8 flags) 7197 { 7198 if (!(entry->state & IGB_MAC_STATE_IN_USE)) 7199 return true; 7200 7201 if ((entry->state & IGB_MAC_STATE_SRC_ADDR) != 7202 (flags & IGB_MAC_STATE_SRC_ADDR)) 7203 return false; 7204 7205 if (!ether_addr_equal(addr, entry->addr)) 7206 return false; 7207 7208 return true; 7209 } 7210 7211 /* Add a MAC filter for 'addr' directing matching traffic to 'queue', 7212 * 'flags' is used to indicate what kind of match is made, match is by 7213 * default for the destination address, if matching by source address 7214 * is desired the flag IGB_MAC_STATE_SRC_ADDR can be used. 7215 */ 7216 static int igb_add_mac_filter_flags(struct igb_adapter *adapter, 7217 const u8 *addr, const u8 queue, 7218 const u8 flags) 7219 { 7220 struct e1000_hw *hw = &adapter->hw; 7221 int rar_entries = hw->mac.rar_entry_count - 7222 adapter->vfs_allocated_count; 7223 int i; 7224 7225 if (is_zero_ether_addr(addr)) 7226 return -EINVAL; 7227 7228 /* Search for the first empty entry in the MAC table. 7229 * Do not touch entries at the end of the table reserved for the VF MAC 7230 * addresses. 7231 */ 7232 for (i = 0; i < rar_entries; i++) { 7233 if (!igb_mac_entry_can_be_used(&adapter->mac_table[i], 7234 addr, flags)) 7235 continue; 7236 7237 ether_addr_copy(adapter->mac_table[i].addr, addr); 7238 adapter->mac_table[i].queue = queue; 7239 adapter->mac_table[i].state |= IGB_MAC_STATE_IN_USE | flags; 7240 7241 igb_rar_set_index(adapter, i); 7242 return i; 7243 } 7244 7245 return -ENOSPC; 7246 } 7247 7248 static int igb_add_mac_filter(struct igb_adapter *adapter, const u8 *addr, 7249 const u8 queue) 7250 { 7251 return igb_add_mac_filter_flags(adapter, addr, queue, 0); 7252 } 7253 7254 /* Remove a MAC filter for 'addr' directing matching traffic to 7255 * 'queue', 'flags' is used to indicate what kind of match need to be 7256 * removed, match is by default for the destination address, if 7257 * matching by source address is to be removed the flag 7258 * IGB_MAC_STATE_SRC_ADDR can be used. 7259 */ 7260 static int igb_del_mac_filter_flags(struct igb_adapter *adapter, 7261 const u8 *addr, const u8 queue, 7262 const u8 flags) 7263 { 7264 struct e1000_hw *hw = &adapter->hw; 7265 int rar_entries = hw->mac.rar_entry_count - 7266 adapter->vfs_allocated_count; 7267 int i; 7268 7269 if (is_zero_ether_addr(addr)) 7270 return -EINVAL; 7271 7272 /* Search for matching entry in the MAC table based on given address 7273 * and queue. Do not touch entries at the end of the table reserved 7274 * for the VF MAC addresses. 7275 */ 7276 for (i = 0; i < rar_entries; i++) { 7277 if (!(adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE)) 7278 continue; 7279 if ((adapter->mac_table[i].state & flags) != flags) 7280 continue; 7281 if (adapter->mac_table[i].queue != queue) 7282 continue; 7283 if (!ether_addr_equal(adapter->mac_table[i].addr, addr)) 7284 continue; 7285 7286 /* When a filter for the default address is "deleted", 7287 * we return it to its initial configuration 7288 */ 7289 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT) { 7290 adapter->mac_table[i].state = 7291 IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE; 7292 adapter->mac_table[i].queue = 7293 adapter->vfs_allocated_count; 7294 } else { 7295 adapter->mac_table[i].state = 0; 7296 adapter->mac_table[i].queue = 0; 7297 memset(adapter->mac_table[i].addr, 0, ETH_ALEN); 7298 } 7299 7300 igb_rar_set_index(adapter, i); 7301 return 0; 7302 } 7303 7304 return -ENOENT; 7305 } 7306 7307 static int igb_del_mac_filter(struct igb_adapter *adapter, const u8 *addr, 7308 const u8 queue) 7309 { 7310 return igb_del_mac_filter_flags(adapter, addr, queue, 0); 7311 } 7312 7313 int igb_add_mac_steering_filter(struct igb_adapter *adapter, 7314 const u8 *addr, u8 queue, u8 flags) 7315 { 7316 struct e1000_hw *hw = &adapter->hw; 7317 7318 /* In theory, this should be supported on 82575 as well, but 7319 * that part wasn't easily accessible during development. 7320 */ 7321 if (hw->mac.type != e1000_i210) 7322 return -EOPNOTSUPP; 7323 7324 return igb_add_mac_filter_flags(adapter, addr, queue, 7325 IGB_MAC_STATE_QUEUE_STEERING | flags); 7326 } 7327 7328 int igb_del_mac_steering_filter(struct igb_adapter *adapter, 7329 const u8 *addr, u8 queue, u8 flags) 7330 { 7331 return igb_del_mac_filter_flags(adapter, addr, queue, 7332 IGB_MAC_STATE_QUEUE_STEERING | flags); 7333 } 7334 7335 static int igb_uc_sync(struct net_device *netdev, const unsigned char *addr) 7336 { 7337 struct igb_adapter *adapter = netdev_priv(netdev); 7338 int ret; 7339 7340 ret = igb_add_mac_filter(adapter, addr, adapter->vfs_allocated_count); 7341 7342 return min_t(int, ret, 0); 7343 } 7344 7345 static int igb_uc_unsync(struct net_device *netdev, const unsigned char *addr) 7346 { 7347 struct igb_adapter *adapter = netdev_priv(netdev); 7348 7349 igb_del_mac_filter(adapter, addr, adapter->vfs_allocated_count); 7350 7351 return 0; 7352 } 7353 7354 static int igb_set_vf_mac_filter(struct igb_adapter *adapter, const int vf, 7355 const u32 info, const u8 *addr) 7356 { 7357 struct pci_dev *pdev = adapter->pdev; 7358 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 7359 struct list_head *pos; 7360 struct vf_mac_filter *entry = NULL; 7361 int ret = 0; 7362 7363 switch (info) { 7364 case E1000_VF_MAC_FILTER_CLR: 7365 /* remove all unicast MAC filters related to the current VF */ 7366 list_for_each(pos, &adapter->vf_macs.l) { 7367 entry = list_entry(pos, struct vf_mac_filter, l); 7368 if (entry->vf == vf) { 7369 entry->vf = -1; 7370 entry->free = true; 7371 igb_del_mac_filter(adapter, entry->vf_mac, vf); 7372 } 7373 } 7374 break; 7375 case E1000_VF_MAC_FILTER_ADD: 7376 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) && 7377 !vf_data->trusted) { 7378 dev_warn(&pdev->dev, 7379 "VF %d requested MAC filter but is administratively denied\n", 7380 vf); 7381 return -EINVAL; 7382 } 7383 if (!is_valid_ether_addr(addr)) { 7384 dev_warn(&pdev->dev, 7385 "VF %d attempted to set invalid MAC filter\n", 7386 vf); 7387 return -EINVAL; 7388 } 7389 7390 /* try to find empty slot in the list */ 7391 list_for_each(pos, &adapter->vf_macs.l) { 7392 entry = list_entry(pos, struct vf_mac_filter, l); 7393 if (entry->free) 7394 break; 7395 } 7396 7397 if (entry && entry->free) { 7398 entry->free = false; 7399 entry->vf = vf; 7400 ether_addr_copy(entry->vf_mac, addr); 7401 7402 ret = igb_add_mac_filter(adapter, addr, vf); 7403 ret = min_t(int, ret, 0); 7404 } else { 7405 ret = -ENOSPC; 7406 } 7407 7408 if (ret == -ENOSPC) 7409 dev_warn(&pdev->dev, 7410 "VF %d has requested MAC filter but there is no space for it\n", 7411 vf); 7412 break; 7413 default: 7414 ret = -EINVAL; 7415 break; 7416 } 7417 7418 return ret; 7419 } 7420 7421 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf) 7422 { 7423 struct pci_dev *pdev = adapter->pdev; 7424 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 7425 u32 info = msg[0] & E1000_VT_MSGINFO_MASK; 7426 7427 /* The VF MAC Address is stored in a packed array of bytes 7428 * starting at the second 32 bit word of the msg array 7429 */ 7430 unsigned char *addr = (unsigned char *)&msg[1]; 7431 int ret = 0; 7432 7433 if (!info) { 7434 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) && 7435 !vf_data->trusted) { 7436 dev_warn(&pdev->dev, 7437 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n", 7438 vf); 7439 return -EINVAL; 7440 } 7441 7442 if (!is_valid_ether_addr(addr)) { 7443 dev_warn(&pdev->dev, 7444 "VF %d attempted to set invalid MAC\n", 7445 vf); 7446 return -EINVAL; 7447 } 7448 7449 ret = igb_set_vf_mac(adapter, vf, addr); 7450 } else { 7451 ret = igb_set_vf_mac_filter(adapter, vf, info, addr); 7452 } 7453 7454 return ret; 7455 } 7456 7457 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf) 7458 { 7459 struct e1000_hw *hw = &adapter->hw; 7460 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 7461 u32 msg = E1000_VT_MSGTYPE_NACK; 7462 7463 /* if device isn't clear to send it shouldn't be reading either */ 7464 if (!(vf_data->flags & IGB_VF_FLAG_CTS) && 7465 time_after(jiffies, vf_data->last_nack + (2 * HZ))) { 7466 igb_write_mbx(hw, &msg, 1, vf); 7467 vf_data->last_nack = jiffies; 7468 } 7469 } 7470 7471 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf) 7472 { 7473 struct pci_dev *pdev = adapter->pdev; 7474 u32 msgbuf[E1000_VFMAILBOX_SIZE]; 7475 struct e1000_hw *hw = &adapter->hw; 7476 struct vf_data_storage *vf_data = &adapter->vf_data[vf]; 7477 s32 retval; 7478 7479 retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf, false); 7480 7481 if (retval) { 7482 /* if receive failed revoke VF CTS stats and restart init */ 7483 dev_err(&pdev->dev, "Error receiving message from VF\n"); 7484 vf_data->flags &= ~IGB_VF_FLAG_CTS; 7485 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ))) 7486 goto unlock; 7487 goto out; 7488 } 7489 7490 /* this is a message we already processed, do nothing */ 7491 if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK)) 7492 goto unlock; 7493 7494 /* until the vf completes a reset it should not be 7495 * allowed to start any configuration. 7496 */ 7497 if (msgbuf[0] == E1000_VF_RESET) { 7498 /* unlocks mailbox */ 7499 igb_vf_reset_msg(adapter, vf); 7500 return; 7501 } 7502 7503 if (!(vf_data->flags & IGB_VF_FLAG_CTS)) { 7504 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ))) 7505 goto unlock; 7506 retval = -1; 7507 goto out; 7508 } 7509 7510 switch ((msgbuf[0] & 0xFFFF)) { 7511 case E1000_VF_SET_MAC_ADDR: 7512 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf); 7513 break; 7514 case E1000_VF_SET_PROMISC: 7515 retval = igb_set_vf_promisc(adapter, msgbuf, vf); 7516 break; 7517 case E1000_VF_SET_MULTICAST: 7518 retval = igb_set_vf_multicasts(adapter, msgbuf, vf); 7519 break; 7520 case E1000_VF_SET_LPE: 7521 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf); 7522 break; 7523 case E1000_VF_SET_VLAN: 7524 retval = -1; 7525 if (vf_data->pf_vlan) 7526 dev_warn(&pdev->dev, 7527 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n", 7528 vf); 7529 else 7530 retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf); 7531 break; 7532 default: 7533 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]); 7534 retval = -1; 7535 break; 7536 } 7537 7538 msgbuf[0] |= E1000_VT_MSGTYPE_CTS; 7539 out: 7540 /* notify the VF of the results of what it sent us */ 7541 if (retval) 7542 msgbuf[0] |= E1000_VT_MSGTYPE_NACK; 7543 else 7544 msgbuf[0] |= E1000_VT_MSGTYPE_ACK; 7545 7546 /* unlocks mailbox */ 7547 igb_write_mbx(hw, msgbuf, 1, vf); 7548 return; 7549 7550 unlock: 7551 igb_unlock_mbx(hw, vf); 7552 } 7553 7554 static void igb_msg_task(struct igb_adapter *adapter) 7555 { 7556 struct e1000_hw *hw = &adapter->hw; 7557 u32 vf; 7558 7559 for (vf = 0; vf < adapter->vfs_allocated_count; vf++) { 7560 /* process any reset requests */ 7561 if (!igb_check_for_rst(hw, vf)) 7562 igb_vf_reset_event(adapter, vf); 7563 7564 /* process any messages pending */ 7565 if (!igb_check_for_msg(hw, vf)) 7566 igb_rcv_msg_from_vf(adapter, vf); 7567 7568 /* process any acks */ 7569 if (!igb_check_for_ack(hw, vf)) 7570 igb_rcv_ack_from_vf(adapter, vf); 7571 } 7572 } 7573 7574 /** 7575 * igb_set_uta - Set unicast filter table address 7576 * @adapter: board private structure 7577 * @set: boolean indicating if we are setting or clearing bits 7578 * 7579 * The unicast table address is a register array of 32-bit registers. 7580 * The table is meant to be used in a way similar to how the MTA is used 7581 * however due to certain limitations in the hardware it is necessary to 7582 * set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous 7583 * enable bit to allow vlan tag stripping when promiscuous mode is enabled 7584 **/ 7585 static void igb_set_uta(struct igb_adapter *adapter, bool set) 7586 { 7587 struct e1000_hw *hw = &adapter->hw; 7588 u32 uta = set ? ~0 : 0; 7589 int i; 7590 7591 /* we only need to do this if VMDq is enabled */ 7592 if (!adapter->vfs_allocated_count) 7593 return; 7594 7595 for (i = hw->mac.uta_reg_count; i--;) 7596 array_wr32(E1000_UTA, i, uta); 7597 } 7598 7599 /** 7600 * igb_intr_msi - Interrupt Handler 7601 * @irq: interrupt number 7602 * @data: pointer to a network interface device structure 7603 **/ 7604 static irqreturn_t igb_intr_msi(int irq, void *data) 7605 { 7606 struct igb_adapter *adapter = data; 7607 struct igb_q_vector *q_vector = adapter->q_vector[0]; 7608 struct e1000_hw *hw = &adapter->hw; 7609 /* read ICR disables interrupts using IAM */ 7610 u32 icr = rd32(E1000_ICR); 7611 7612 igb_write_itr(q_vector); 7613 7614 if (icr & E1000_ICR_DRSTA) 7615 schedule_work(&adapter->reset_task); 7616 7617 if (icr & E1000_ICR_DOUTSYNC) { 7618 /* HW is reporting DMA is out of sync */ 7619 adapter->stats.doosync++; 7620 } 7621 7622 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) { 7623 hw->mac.get_link_status = 1; 7624 if (!test_bit(__IGB_DOWN, &adapter->state)) 7625 mod_timer(&adapter->watchdog_timer, jiffies + 1); 7626 } 7627 7628 if (icr & E1000_ICR_TS) 7629 igb_tsync_interrupt(adapter); 7630 7631 napi_schedule(&q_vector->napi); 7632 7633 return IRQ_HANDLED; 7634 } 7635 7636 /** 7637 * igb_intr - Legacy Interrupt Handler 7638 * @irq: interrupt number 7639 * @data: pointer to a network interface device structure 7640 **/ 7641 static irqreturn_t igb_intr(int irq, void *data) 7642 { 7643 struct igb_adapter *adapter = data; 7644 struct igb_q_vector *q_vector = adapter->q_vector[0]; 7645 struct e1000_hw *hw = &adapter->hw; 7646 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No 7647 * need for the IMC write 7648 */ 7649 u32 icr = rd32(E1000_ICR); 7650 7651 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is 7652 * not set, then the adapter didn't send an interrupt 7653 */ 7654 if (!(icr & E1000_ICR_INT_ASSERTED)) 7655 return IRQ_NONE; 7656 7657 igb_write_itr(q_vector); 7658 7659 if (icr & E1000_ICR_DRSTA) 7660 schedule_work(&adapter->reset_task); 7661 7662 if (icr & E1000_ICR_DOUTSYNC) { 7663 /* HW is reporting DMA is out of sync */ 7664 adapter->stats.doosync++; 7665 } 7666 7667 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) { 7668 hw->mac.get_link_status = 1; 7669 /* guard against interrupt when we're going down */ 7670 if (!test_bit(__IGB_DOWN, &adapter->state)) 7671 mod_timer(&adapter->watchdog_timer, jiffies + 1); 7672 } 7673 7674 if (icr & E1000_ICR_TS) 7675 igb_tsync_interrupt(adapter); 7676 7677 napi_schedule(&q_vector->napi); 7678 7679 return IRQ_HANDLED; 7680 } 7681 7682 static void igb_ring_irq_enable(struct igb_q_vector *q_vector) 7683 { 7684 struct igb_adapter *adapter = q_vector->adapter; 7685 struct e1000_hw *hw = &adapter->hw; 7686 7687 if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) || 7688 (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) { 7689 if ((adapter->num_q_vectors == 1) && !adapter->vf_data) 7690 igb_set_itr(q_vector); 7691 else 7692 igb_update_ring_itr(q_vector); 7693 } 7694 7695 if (!test_bit(__IGB_DOWN, &adapter->state)) { 7696 if (adapter->flags & IGB_FLAG_HAS_MSIX) 7697 wr32(E1000_EIMS, q_vector->eims_value); 7698 else 7699 igb_irq_enable(adapter); 7700 } 7701 } 7702 7703 /** 7704 * igb_poll - NAPI Rx polling callback 7705 * @napi: napi polling structure 7706 * @budget: count of how many packets we should handle 7707 **/ 7708 static int igb_poll(struct napi_struct *napi, int budget) 7709 { 7710 struct igb_q_vector *q_vector = container_of(napi, 7711 struct igb_q_vector, 7712 napi); 7713 bool clean_complete = true; 7714 int work_done = 0; 7715 7716 #ifdef CONFIG_IGB_DCA 7717 if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED) 7718 igb_update_dca(q_vector); 7719 #endif 7720 if (q_vector->tx.ring) 7721 clean_complete = igb_clean_tx_irq(q_vector, budget); 7722 7723 if (q_vector->rx.ring) { 7724 int cleaned = igb_clean_rx_irq(q_vector, budget); 7725 7726 work_done += cleaned; 7727 if (cleaned >= budget) 7728 clean_complete = false; 7729 } 7730 7731 /* If all work not completed, return budget and keep polling */ 7732 if (!clean_complete) 7733 return budget; 7734 7735 /* Exit the polling mode, but don't re-enable interrupts if stack might 7736 * poll us due to busy-polling 7737 */ 7738 if (likely(napi_complete_done(napi, work_done))) 7739 igb_ring_irq_enable(q_vector); 7740 7741 return min(work_done, budget - 1); 7742 } 7743 7744 /** 7745 * igb_clean_tx_irq - Reclaim resources after transmit completes 7746 * @q_vector: pointer to q_vector containing needed info 7747 * @napi_budget: Used to determine if we are in netpoll 7748 * 7749 * returns true if ring is completely cleaned 7750 **/ 7751 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector, int napi_budget) 7752 { 7753 struct igb_adapter *adapter = q_vector->adapter; 7754 struct igb_ring *tx_ring = q_vector->tx.ring; 7755 struct igb_tx_buffer *tx_buffer; 7756 union e1000_adv_tx_desc *tx_desc; 7757 unsigned int total_bytes = 0, total_packets = 0; 7758 unsigned int budget = q_vector->tx.work_limit; 7759 unsigned int i = tx_ring->next_to_clean; 7760 7761 if (test_bit(__IGB_DOWN, &adapter->state)) 7762 return true; 7763 7764 tx_buffer = &tx_ring->tx_buffer_info[i]; 7765 tx_desc = IGB_TX_DESC(tx_ring, i); 7766 i -= tx_ring->count; 7767 7768 do { 7769 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch; 7770 7771 /* if next_to_watch is not set then there is no work pending */ 7772 if (!eop_desc) 7773 break; 7774 7775 /* prevent any other reads prior to eop_desc */ 7776 smp_rmb(); 7777 7778 /* if DD is not set pending work has not been completed */ 7779 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD))) 7780 break; 7781 7782 /* clear next_to_watch to prevent false hangs */ 7783 tx_buffer->next_to_watch = NULL; 7784 7785 /* update the statistics for this packet */ 7786 total_bytes += tx_buffer->bytecount; 7787 total_packets += tx_buffer->gso_segs; 7788 7789 /* free the skb */ 7790 napi_consume_skb(tx_buffer->skb, napi_budget); 7791 7792 /* unmap skb header data */ 7793 dma_unmap_single(tx_ring->dev, 7794 dma_unmap_addr(tx_buffer, dma), 7795 dma_unmap_len(tx_buffer, len), 7796 DMA_TO_DEVICE); 7797 7798 /* clear tx_buffer data */ 7799 dma_unmap_len_set(tx_buffer, len, 0); 7800 7801 /* clear last DMA location and unmap remaining buffers */ 7802 while (tx_desc != eop_desc) { 7803 tx_buffer++; 7804 tx_desc++; 7805 i++; 7806 if (unlikely(!i)) { 7807 i -= tx_ring->count; 7808 tx_buffer = tx_ring->tx_buffer_info; 7809 tx_desc = IGB_TX_DESC(tx_ring, 0); 7810 } 7811 7812 /* unmap any remaining paged data */ 7813 if (dma_unmap_len(tx_buffer, len)) { 7814 dma_unmap_page(tx_ring->dev, 7815 dma_unmap_addr(tx_buffer, dma), 7816 dma_unmap_len(tx_buffer, len), 7817 DMA_TO_DEVICE); 7818 dma_unmap_len_set(tx_buffer, len, 0); 7819 } 7820 } 7821 7822 /* move us one more past the eop_desc for start of next pkt */ 7823 tx_buffer++; 7824 tx_desc++; 7825 i++; 7826 if (unlikely(!i)) { 7827 i -= tx_ring->count; 7828 tx_buffer = tx_ring->tx_buffer_info; 7829 tx_desc = IGB_TX_DESC(tx_ring, 0); 7830 } 7831 7832 /* issue prefetch for next Tx descriptor */ 7833 prefetch(tx_desc); 7834 7835 /* update budget accounting */ 7836 budget--; 7837 } while (likely(budget)); 7838 7839 netdev_tx_completed_queue(txring_txq(tx_ring), 7840 total_packets, total_bytes); 7841 i += tx_ring->count; 7842 tx_ring->next_to_clean = i; 7843 u64_stats_update_begin(&tx_ring->tx_syncp); 7844 tx_ring->tx_stats.bytes += total_bytes; 7845 tx_ring->tx_stats.packets += total_packets; 7846 u64_stats_update_end(&tx_ring->tx_syncp); 7847 q_vector->tx.total_bytes += total_bytes; 7848 q_vector->tx.total_packets += total_packets; 7849 7850 if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) { 7851 struct e1000_hw *hw = &adapter->hw; 7852 7853 /* Detect a transmit hang in hardware, this serializes the 7854 * check with the clearing of time_stamp and movement of i 7855 */ 7856 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags); 7857 if (tx_buffer->next_to_watch && 7858 time_after(jiffies, tx_buffer->time_stamp + 7859 (adapter->tx_timeout_factor * HZ)) && 7860 !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) { 7861 7862 /* detected Tx unit hang */ 7863 dev_err(tx_ring->dev, 7864 "Detected Tx Unit Hang\n" 7865 " Tx Queue <%d>\n" 7866 " TDH <%x>\n" 7867 " TDT <%x>\n" 7868 " next_to_use <%x>\n" 7869 " next_to_clean <%x>\n" 7870 "buffer_info[next_to_clean]\n" 7871 " time_stamp <%lx>\n" 7872 " next_to_watch <%p>\n" 7873 " jiffies <%lx>\n" 7874 " desc.status <%x>\n", 7875 tx_ring->queue_index, 7876 rd32(E1000_TDH(tx_ring->reg_idx)), 7877 readl(tx_ring->tail), 7878 tx_ring->next_to_use, 7879 tx_ring->next_to_clean, 7880 tx_buffer->time_stamp, 7881 tx_buffer->next_to_watch, 7882 jiffies, 7883 tx_buffer->next_to_watch->wb.status); 7884 netif_stop_subqueue(tx_ring->netdev, 7885 tx_ring->queue_index); 7886 7887 /* we are about to reset, no point in enabling stuff */ 7888 return true; 7889 } 7890 } 7891 7892 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2) 7893 if (unlikely(total_packets && 7894 netif_carrier_ok(tx_ring->netdev) && 7895 igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) { 7896 /* Make sure that anybody stopping the queue after this 7897 * sees the new next_to_clean. 7898 */ 7899 smp_mb(); 7900 if (__netif_subqueue_stopped(tx_ring->netdev, 7901 tx_ring->queue_index) && 7902 !(test_bit(__IGB_DOWN, &adapter->state))) { 7903 netif_wake_subqueue(tx_ring->netdev, 7904 tx_ring->queue_index); 7905 7906 u64_stats_update_begin(&tx_ring->tx_syncp); 7907 tx_ring->tx_stats.restart_queue++; 7908 u64_stats_update_end(&tx_ring->tx_syncp); 7909 } 7910 } 7911 7912 return !!budget; 7913 } 7914 7915 /** 7916 * igb_reuse_rx_page - page flip buffer and store it back on the ring 7917 * @rx_ring: rx descriptor ring to store buffers on 7918 * @old_buff: donor buffer to have page reused 7919 * 7920 * Synchronizes page for reuse by the adapter 7921 **/ 7922 static void igb_reuse_rx_page(struct igb_ring *rx_ring, 7923 struct igb_rx_buffer *old_buff) 7924 { 7925 struct igb_rx_buffer *new_buff; 7926 u16 nta = rx_ring->next_to_alloc; 7927 7928 new_buff = &rx_ring->rx_buffer_info[nta]; 7929 7930 /* update, and store next to alloc */ 7931 nta++; 7932 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0; 7933 7934 /* Transfer page from old buffer to new buffer. 7935 * Move each member individually to avoid possible store 7936 * forwarding stalls. 7937 */ 7938 new_buff->dma = old_buff->dma; 7939 new_buff->page = old_buff->page; 7940 new_buff->page_offset = old_buff->page_offset; 7941 new_buff->pagecnt_bias = old_buff->pagecnt_bias; 7942 } 7943 7944 static inline bool igb_page_is_reserved(struct page *page) 7945 { 7946 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page); 7947 } 7948 7949 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer) 7950 { 7951 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias; 7952 struct page *page = rx_buffer->page; 7953 7954 /* avoid re-using remote pages */ 7955 if (unlikely(igb_page_is_reserved(page))) 7956 return false; 7957 7958 #if (PAGE_SIZE < 8192) 7959 /* if we are only owner of page we can reuse it */ 7960 if (unlikely((page_ref_count(page) - pagecnt_bias) > 1)) 7961 return false; 7962 #else 7963 #define IGB_LAST_OFFSET \ 7964 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IGB_RXBUFFER_2048) 7965 7966 if (rx_buffer->page_offset > IGB_LAST_OFFSET) 7967 return false; 7968 #endif 7969 7970 /* If we have drained the page fragment pool we need to update 7971 * the pagecnt_bias and page count so that we fully restock the 7972 * number of references the driver holds. 7973 */ 7974 if (unlikely(!pagecnt_bias)) { 7975 page_ref_add(page, USHRT_MAX); 7976 rx_buffer->pagecnt_bias = USHRT_MAX; 7977 } 7978 7979 return true; 7980 } 7981 7982 /** 7983 * igb_add_rx_frag - Add contents of Rx buffer to sk_buff 7984 * @rx_ring: rx descriptor ring to transact packets on 7985 * @rx_buffer: buffer containing page to add 7986 * @skb: sk_buff to place the data into 7987 * @size: size of buffer to be added 7988 * 7989 * This function will add the data contained in rx_buffer->page to the skb. 7990 **/ 7991 static void igb_add_rx_frag(struct igb_ring *rx_ring, 7992 struct igb_rx_buffer *rx_buffer, 7993 struct sk_buff *skb, 7994 unsigned int size) 7995 { 7996 #if (PAGE_SIZE < 8192) 7997 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2; 7998 #else 7999 unsigned int truesize = ring_uses_build_skb(rx_ring) ? 8000 SKB_DATA_ALIGN(IGB_SKB_PAD + size) : 8001 SKB_DATA_ALIGN(size); 8002 #endif 8003 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page, 8004 rx_buffer->page_offset, size, truesize); 8005 #if (PAGE_SIZE < 8192) 8006 rx_buffer->page_offset ^= truesize; 8007 #else 8008 rx_buffer->page_offset += truesize; 8009 #endif 8010 } 8011 8012 static struct sk_buff *igb_construct_skb(struct igb_ring *rx_ring, 8013 struct igb_rx_buffer *rx_buffer, 8014 union e1000_adv_rx_desc *rx_desc, 8015 unsigned int size) 8016 { 8017 void *va = page_address(rx_buffer->page) + rx_buffer->page_offset; 8018 #if (PAGE_SIZE < 8192) 8019 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2; 8020 #else 8021 unsigned int truesize = SKB_DATA_ALIGN(size); 8022 #endif 8023 unsigned int headlen; 8024 struct sk_buff *skb; 8025 8026 /* prefetch first cache line of first page */ 8027 prefetch(va); 8028 #if L1_CACHE_BYTES < 128 8029 prefetch(va + L1_CACHE_BYTES); 8030 #endif 8031 8032 /* allocate a skb to store the frags */ 8033 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN); 8034 if (unlikely(!skb)) 8035 return NULL; 8036 8037 if (unlikely(igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))) { 8038 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb); 8039 va += IGB_TS_HDR_LEN; 8040 size -= IGB_TS_HDR_LEN; 8041 } 8042 8043 /* Determine available headroom for copy */ 8044 headlen = size; 8045 if (headlen > IGB_RX_HDR_LEN) 8046 headlen = eth_get_headlen(va, IGB_RX_HDR_LEN); 8047 8048 /* align pull length to size of long to optimize memcpy performance */ 8049 memcpy(__skb_put(skb, headlen), va, ALIGN(headlen, sizeof(long))); 8050 8051 /* update all of the pointers */ 8052 size -= headlen; 8053 if (size) { 8054 skb_add_rx_frag(skb, 0, rx_buffer->page, 8055 (va + headlen) - page_address(rx_buffer->page), 8056 size, truesize); 8057 #if (PAGE_SIZE < 8192) 8058 rx_buffer->page_offset ^= truesize; 8059 #else 8060 rx_buffer->page_offset += truesize; 8061 #endif 8062 } else { 8063 rx_buffer->pagecnt_bias++; 8064 } 8065 8066 return skb; 8067 } 8068 8069 static struct sk_buff *igb_build_skb(struct igb_ring *rx_ring, 8070 struct igb_rx_buffer *rx_buffer, 8071 union e1000_adv_rx_desc *rx_desc, 8072 unsigned int size) 8073 { 8074 void *va = page_address(rx_buffer->page) + rx_buffer->page_offset; 8075 #if (PAGE_SIZE < 8192) 8076 unsigned int truesize = igb_rx_pg_size(rx_ring) / 2; 8077 #else 8078 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) + 8079 SKB_DATA_ALIGN(IGB_SKB_PAD + size); 8080 #endif 8081 struct sk_buff *skb; 8082 8083 /* prefetch first cache line of first page */ 8084 prefetch(va); 8085 #if L1_CACHE_BYTES < 128 8086 prefetch(va + L1_CACHE_BYTES); 8087 #endif 8088 8089 /* build an skb around the page buffer */ 8090 skb = build_skb(va - IGB_SKB_PAD, truesize); 8091 if (unlikely(!skb)) 8092 return NULL; 8093 8094 /* update pointers within the skb to store the data */ 8095 skb_reserve(skb, IGB_SKB_PAD); 8096 __skb_put(skb, size); 8097 8098 /* pull timestamp out of packet data */ 8099 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) { 8100 igb_ptp_rx_pktstamp(rx_ring->q_vector, skb->data, skb); 8101 __skb_pull(skb, IGB_TS_HDR_LEN); 8102 } 8103 8104 /* update buffer offset */ 8105 #if (PAGE_SIZE < 8192) 8106 rx_buffer->page_offset ^= truesize; 8107 #else 8108 rx_buffer->page_offset += truesize; 8109 #endif 8110 8111 return skb; 8112 } 8113 8114 static inline void igb_rx_checksum(struct igb_ring *ring, 8115 union e1000_adv_rx_desc *rx_desc, 8116 struct sk_buff *skb) 8117 { 8118 skb_checksum_none_assert(skb); 8119 8120 /* Ignore Checksum bit is set */ 8121 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM)) 8122 return; 8123 8124 /* Rx checksum disabled via ethtool */ 8125 if (!(ring->netdev->features & NETIF_F_RXCSUM)) 8126 return; 8127 8128 /* TCP/UDP checksum error bit is set */ 8129 if (igb_test_staterr(rx_desc, 8130 E1000_RXDEXT_STATERR_TCPE | 8131 E1000_RXDEXT_STATERR_IPE)) { 8132 /* work around errata with sctp packets where the TCPE aka 8133 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc) 8134 * packets, (aka let the stack check the crc32c) 8135 */ 8136 if (!((skb->len == 60) && 8137 test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) { 8138 u64_stats_update_begin(&ring->rx_syncp); 8139 ring->rx_stats.csum_err++; 8140 u64_stats_update_end(&ring->rx_syncp); 8141 } 8142 /* let the stack verify checksum errors */ 8143 return; 8144 } 8145 /* It must be a TCP or UDP packet with a valid checksum */ 8146 if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS | 8147 E1000_RXD_STAT_UDPCS)) 8148 skb->ip_summed = CHECKSUM_UNNECESSARY; 8149 8150 dev_dbg(ring->dev, "cksum success: bits %08X\n", 8151 le32_to_cpu(rx_desc->wb.upper.status_error)); 8152 } 8153 8154 static inline void igb_rx_hash(struct igb_ring *ring, 8155 union e1000_adv_rx_desc *rx_desc, 8156 struct sk_buff *skb) 8157 { 8158 if (ring->netdev->features & NETIF_F_RXHASH) 8159 skb_set_hash(skb, 8160 le32_to_cpu(rx_desc->wb.lower.hi_dword.rss), 8161 PKT_HASH_TYPE_L3); 8162 } 8163 8164 /** 8165 * igb_is_non_eop - process handling of non-EOP buffers 8166 * @rx_ring: Rx ring being processed 8167 * @rx_desc: Rx descriptor for current buffer 8168 * @skb: current socket buffer containing buffer in progress 8169 * 8170 * This function updates next to clean. If the buffer is an EOP buffer 8171 * this function exits returning false, otherwise it will place the 8172 * sk_buff in the next buffer to be chained and return true indicating 8173 * that this is in fact a non-EOP buffer. 8174 **/ 8175 static bool igb_is_non_eop(struct igb_ring *rx_ring, 8176 union e1000_adv_rx_desc *rx_desc) 8177 { 8178 u32 ntc = rx_ring->next_to_clean + 1; 8179 8180 /* fetch, update, and store next to clean */ 8181 ntc = (ntc < rx_ring->count) ? ntc : 0; 8182 rx_ring->next_to_clean = ntc; 8183 8184 prefetch(IGB_RX_DESC(rx_ring, ntc)); 8185 8186 if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP))) 8187 return false; 8188 8189 return true; 8190 } 8191 8192 /** 8193 * igb_cleanup_headers - Correct corrupted or empty headers 8194 * @rx_ring: rx descriptor ring packet is being transacted on 8195 * @rx_desc: pointer to the EOP Rx descriptor 8196 * @skb: pointer to current skb being fixed 8197 * 8198 * Address the case where we are pulling data in on pages only 8199 * and as such no data is present in the skb header. 8200 * 8201 * In addition if skb is not at least 60 bytes we need to pad it so that 8202 * it is large enough to qualify as a valid Ethernet frame. 8203 * 8204 * Returns true if an error was encountered and skb was freed. 8205 **/ 8206 static bool igb_cleanup_headers(struct igb_ring *rx_ring, 8207 union e1000_adv_rx_desc *rx_desc, 8208 struct sk_buff *skb) 8209 { 8210 if (unlikely((igb_test_staterr(rx_desc, 8211 E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) { 8212 struct net_device *netdev = rx_ring->netdev; 8213 if (!(netdev->features & NETIF_F_RXALL)) { 8214 dev_kfree_skb_any(skb); 8215 return true; 8216 } 8217 } 8218 8219 /* if eth_skb_pad returns an error the skb was freed */ 8220 if (eth_skb_pad(skb)) 8221 return true; 8222 8223 return false; 8224 } 8225 8226 /** 8227 * igb_process_skb_fields - Populate skb header fields from Rx descriptor 8228 * @rx_ring: rx descriptor ring packet is being transacted on 8229 * @rx_desc: pointer to the EOP Rx descriptor 8230 * @skb: pointer to current skb being populated 8231 * 8232 * This function checks the ring, descriptor, and packet information in 8233 * order to populate the hash, checksum, VLAN, timestamp, protocol, and 8234 * other fields within the skb. 8235 **/ 8236 static void igb_process_skb_fields(struct igb_ring *rx_ring, 8237 union e1000_adv_rx_desc *rx_desc, 8238 struct sk_buff *skb) 8239 { 8240 struct net_device *dev = rx_ring->netdev; 8241 8242 igb_rx_hash(rx_ring, rx_desc, skb); 8243 8244 igb_rx_checksum(rx_ring, rx_desc, skb); 8245 8246 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) && 8247 !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) 8248 igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb); 8249 8250 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) && 8251 igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) { 8252 u16 vid; 8253 8254 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) && 8255 test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags)) 8256 vid = be16_to_cpu(rx_desc->wb.upper.vlan); 8257 else 8258 vid = le16_to_cpu(rx_desc->wb.upper.vlan); 8259 8260 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid); 8261 } 8262 8263 skb_record_rx_queue(skb, rx_ring->queue_index); 8264 8265 skb->protocol = eth_type_trans(skb, rx_ring->netdev); 8266 } 8267 8268 static struct igb_rx_buffer *igb_get_rx_buffer(struct igb_ring *rx_ring, 8269 const unsigned int size) 8270 { 8271 struct igb_rx_buffer *rx_buffer; 8272 8273 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean]; 8274 prefetchw(rx_buffer->page); 8275 8276 /* we are reusing so sync this buffer for CPU use */ 8277 dma_sync_single_range_for_cpu(rx_ring->dev, 8278 rx_buffer->dma, 8279 rx_buffer->page_offset, 8280 size, 8281 DMA_FROM_DEVICE); 8282 8283 rx_buffer->pagecnt_bias--; 8284 8285 return rx_buffer; 8286 } 8287 8288 static void igb_put_rx_buffer(struct igb_ring *rx_ring, 8289 struct igb_rx_buffer *rx_buffer) 8290 { 8291 if (igb_can_reuse_rx_page(rx_buffer)) { 8292 /* hand second half of page back to the ring */ 8293 igb_reuse_rx_page(rx_ring, rx_buffer); 8294 } else { 8295 /* We are not reusing the buffer so unmap it and free 8296 * any references we are holding to it 8297 */ 8298 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma, 8299 igb_rx_pg_size(rx_ring), DMA_FROM_DEVICE, 8300 IGB_RX_DMA_ATTR); 8301 __page_frag_cache_drain(rx_buffer->page, 8302 rx_buffer->pagecnt_bias); 8303 } 8304 8305 /* clear contents of rx_buffer */ 8306 rx_buffer->page = NULL; 8307 } 8308 8309 static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget) 8310 { 8311 struct igb_ring *rx_ring = q_vector->rx.ring; 8312 struct sk_buff *skb = rx_ring->skb; 8313 unsigned int total_bytes = 0, total_packets = 0; 8314 u16 cleaned_count = igb_desc_unused(rx_ring); 8315 8316 while (likely(total_packets < budget)) { 8317 union e1000_adv_rx_desc *rx_desc; 8318 struct igb_rx_buffer *rx_buffer; 8319 unsigned int size; 8320 8321 /* return some buffers to hardware, one at a time is too slow */ 8322 if (cleaned_count >= IGB_RX_BUFFER_WRITE) { 8323 igb_alloc_rx_buffers(rx_ring, cleaned_count); 8324 cleaned_count = 0; 8325 } 8326 8327 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean); 8328 size = le16_to_cpu(rx_desc->wb.upper.length); 8329 if (!size) 8330 break; 8331 8332 /* This memory barrier is needed to keep us from reading 8333 * any other fields out of the rx_desc until we know the 8334 * descriptor has been written back 8335 */ 8336 dma_rmb(); 8337 8338 rx_buffer = igb_get_rx_buffer(rx_ring, size); 8339 8340 /* retrieve a buffer from the ring */ 8341 if (skb) 8342 igb_add_rx_frag(rx_ring, rx_buffer, skb, size); 8343 else if (ring_uses_build_skb(rx_ring)) 8344 skb = igb_build_skb(rx_ring, rx_buffer, rx_desc, size); 8345 else 8346 skb = igb_construct_skb(rx_ring, rx_buffer, 8347 rx_desc, size); 8348 8349 /* exit if we failed to retrieve a buffer */ 8350 if (!skb) { 8351 rx_ring->rx_stats.alloc_failed++; 8352 rx_buffer->pagecnt_bias++; 8353 break; 8354 } 8355 8356 igb_put_rx_buffer(rx_ring, rx_buffer); 8357 cleaned_count++; 8358 8359 /* fetch next buffer in frame if non-eop */ 8360 if (igb_is_non_eop(rx_ring, rx_desc)) 8361 continue; 8362 8363 /* verify the packet layout is correct */ 8364 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) { 8365 skb = NULL; 8366 continue; 8367 } 8368 8369 /* probably a little skewed due to removing CRC */ 8370 total_bytes += skb->len; 8371 8372 /* populate checksum, timestamp, VLAN, and protocol */ 8373 igb_process_skb_fields(rx_ring, rx_desc, skb); 8374 8375 napi_gro_receive(&q_vector->napi, skb); 8376 8377 /* reset skb pointer */ 8378 skb = NULL; 8379 8380 /* update budget accounting */ 8381 total_packets++; 8382 } 8383 8384 /* place incomplete frames back on ring for completion */ 8385 rx_ring->skb = skb; 8386 8387 u64_stats_update_begin(&rx_ring->rx_syncp); 8388 rx_ring->rx_stats.packets += total_packets; 8389 rx_ring->rx_stats.bytes += total_bytes; 8390 u64_stats_update_end(&rx_ring->rx_syncp); 8391 q_vector->rx.total_packets += total_packets; 8392 q_vector->rx.total_bytes += total_bytes; 8393 8394 if (cleaned_count) 8395 igb_alloc_rx_buffers(rx_ring, cleaned_count); 8396 8397 return total_packets; 8398 } 8399 8400 static inline unsigned int igb_rx_offset(struct igb_ring *rx_ring) 8401 { 8402 return ring_uses_build_skb(rx_ring) ? IGB_SKB_PAD : 0; 8403 } 8404 8405 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring, 8406 struct igb_rx_buffer *bi) 8407 { 8408 struct page *page = bi->page; 8409 dma_addr_t dma; 8410 8411 /* since we are recycling buffers we should seldom need to alloc */ 8412 if (likely(page)) 8413 return true; 8414 8415 /* alloc new page for storage */ 8416 page = dev_alloc_pages(igb_rx_pg_order(rx_ring)); 8417 if (unlikely(!page)) { 8418 rx_ring->rx_stats.alloc_failed++; 8419 return false; 8420 } 8421 8422 /* map page for use */ 8423 dma = dma_map_page_attrs(rx_ring->dev, page, 0, 8424 igb_rx_pg_size(rx_ring), 8425 DMA_FROM_DEVICE, 8426 IGB_RX_DMA_ATTR); 8427 8428 /* if mapping failed free memory back to system since 8429 * there isn't much point in holding memory we can't use 8430 */ 8431 if (dma_mapping_error(rx_ring->dev, dma)) { 8432 __free_pages(page, igb_rx_pg_order(rx_ring)); 8433 8434 rx_ring->rx_stats.alloc_failed++; 8435 return false; 8436 } 8437 8438 bi->dma = dma; 8439 bi->page = page; 8440 bi->page_offset = igb_rx_offset(rx_ring); 8441 bi->pagecnt_bias = 1; 8442 8443 return true; 8444 } 8445 8446 /** 8447 * igb_alloc_rx_buffers - Replace used receive buffers; packet split 8448 * @adapter: address of board private structure 8449 **/ 8450 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count) 8451 { 8452 union e1000_adv_rx_desc *rx_desc; 8453 struct igb_rx_buffer *bi; 8454 u16 i = rx_ring->next_to_use; 8455 u16 bufsz; 8456 8457 /* nothing to do */ 8458 if (!cleaned_count) 8459 return; 8460 8461 rx_desc = IGB_RX_DESC(rx_ring, i); 8462 bi = &rx_ring->rx_buffer_info[i]; 8463 i -= rx_ring->count; 8464 8465 bufsz = igb_rx_bufsz(rx_ring); 8466 8467 do { 8468 if (!igb_alloc_mapped_page(rx_ring, bi)) 8469 break; 8470 8471 /* sync the buffer for use by the device */ 8472 dma_sync_single_range_for_device(rx_ring->dev, bi->dma, 8473 bi->page_offset, bufsz, 8474 DMA_FROM_DEVICE); 8475 8476 /* Refresh the desc even if buffer_addrs didn't change 8477 * because each write-back erases this info. 8478 */ 8479 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset); 8480 8481 rx_desc++; 8482 bi++; 8483 i++; 8484 if (unlikely(!i)) { 8485 rx_desc = IGB_RX_DESC(rx_ring, 0); 8486 bi = rx_ring->rx_buffer_info; 8487 i -= rx_ring->count; 8488 } 8489 8490 /* clear the length for the next_to_use descriptor */ 8491 rx_desc->wb.upper.length = 0; 8492 8493 cleaned_count--; 8494 } while (cleaned_count); 8495 8496 i += rx_ring->count; 8497 8498 if (rx_ring->next_to_use != i) { 8499 /* record the next descriptor to use */ 8500 rx_ring->next_to_use = i; 8501 8502 /* update next to alloc since we have filled the ring */ 8503 rx_ring->next_to_alloc = i; 8504 8505 /* Force memory writes to complete before letting h/w 8506 * know there are new descriptors to fetch. (Only 8507 * applicable for weak-ordered memory model archs, 8508 * such as IA-64). 8509 */ 8510 dma_wmb(); 8511 writel(i, rx_ring->tail); 8512 } 8513 } 8514 8515 /** 8516 * igb_mii_ioctl - 8517 * @netdev: 8518 * @ifreq: 8519 * @cmd: 8520 **/ 8521 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd) 8522 { 8523 struct igb_adapter *adapter = netdev_priv(netdev); 8524 struct mii_ioctl_data *data = if_mii(ifr); 8525 8526 if (adapter->hw.phy.media_type != e1000_media_type_copper) 8527 return -EOPNOTSUPP; 8528 8529 switch (cmd) { 8530 case SIOCGMIIPHY: 8531 data->phy_id = adapter->hw.phy.addr; 8532 break; 8533 case SIOCGMIIREG: 8534 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F, 8535 &data->val_out)) 8536 return -EIO; 8537 break; 8538 case SIOCSMIIREG: 8539 default: 8540 return -EOPNOTSUPP; 8541 } 8542 return 0; 8543 } 8544 8545 /** 8546 * igb_ioctl - 8547 * @netdev: 8548 * @ifreq: 8549 * @cmd: 8550 **/ 8551 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd) 8552 { 8553 switch (cmd) { 8554 case SIOCGMIIPHY: 8555 case SIOCGMIIREG: 8556 case SIOCSMIIREG: 8557 return igb_mii_ioctl(netdev, ifr, cmd); 8558 case SIOCGHWTSTAMP: 8559 return igb_ptp_get_ts_config(netdev, ifr); 8560 case SIOCSHWTSTAMP: 8561 return igb_ptp_set_ts_config(netdev, ifr); 8562 default: 8563 return -EOPNOTSUPP; 8564 } 8565 } 8566 8567 void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value) 8568 { 8569 struct igb_adapter *adapter = hw->back; 8570 8571 pci_read_config_word(adapter->pdev, reg, value); 8572 } 8573 8574 void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value) 8575 { 8576 struct igb_adapter *adapter = hw->back; 8577 8578 pci_write_config_word(adapter->pdev, reg, *value); 8579 } 8580 8581 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value) 8582 { 8583 struct igb_adapter *adapter = hw->back; 8584 8585 if (pcie_capability_read_word(adapter->pdev, reg, value)) 8586 return -E1000_ERR_CONFIG; 8587 8588 return 0; 8589 } 8590 8591 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value) 8592 { 8593 struct igb_adapter *adapter = hw->back; 8594 8595 if (pcie_capability_write_word(adapter->pdev, reg, *value)) 8596 return -E1000_ERR_CONFIG; 8597 8598 return 0; 8599 } 8600 8601 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features) 8602 { 8603 struct igb_adapter *adapter = netdev_priv(netdev); 8604 struct e1000_hw *hw = &adapter->hw; 8605 u32 ctrl, rctl; 8606 bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX); 8607 8608 if (enable) { 8609 /* enable VLAN tag insert/strip */ 8610 ctrl = rd32(E1000_CTRL); 8611 ctrl |= E1000_CTRL_VME; 8612 wr32(E1000_CTRL, ctrl); 8613 8614 /* Disable CFI check */ 8615 rctl = rd32(E1000_RCTL); 8616 rctl &= ~E1000_RCTL_CFIEN; 8617 wr32(E1000_RCTL, rctl); 8618 } else { 8619 /* disable VLAN tag insert/strip */ 8620 ctrl = rd32(E1000_CTRL); 8621 ctrl &= ~E1000_CTRL_VME; 8622 wr32(E1000_CTRL, ctrl); 8623 } 8624 8625 igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable); 8626 } 8627 8628 static int igb_vlan_rx_add_vid(struct net_device *netdev, 8629 __be16 proto, u16 vid) 8630 { 8631 struct igb_adapter *adapter = netdev_priv(netdev); 8632 struct e1000_hw *hw = &adapter->hw; 8633 int pf_id = adapter->vfs_allocated_count; 8634 8635 /* add the filter since PF can receive vlans w/o entry in vlvf */ 8636 if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC)) 8637 igb_vfta_set(hw, vid, pf_id, true, !!vid); 8638 8639 set_bit(vid, adapter->active_vlans); 8640 8641 return 0; 8642 } 8643 8644 static int igb_vlan_rx_kill_vid(struct net_device *netdev, 8645 __be16 proto, u16 vid) 8646 { 8647 struct igb_adapter *adapter = netdev_priv(netdev); 8648 int pf_id = adapter->vfs_allocated_count; 8649 struct e1000_hw *hw = &adapter->hw; 8650 8651 /* remove VID from filter table */ 8652 if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC)) 8653 igb_vfta_set(hw, vid, pf_id, false, true); 8654 8655 clear_bit(vid, adapter->active_vlans); 8656 8657 return 0; 8658 } 8659 8660 static void igb_restore_vlan(struct igb_adapter *adapter) 8661 { 8662 u16 vid = 1; 8663 8664 igb_vlan_mode(adapter->netdev, adapter->netdev->features); 8665 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0); 8666 8667 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID) 8668 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid); 8669 } 8670 8671 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx) 8672 { 8673 struct pci_dev *pdev = adapter->pdev; 8674 struct e1000_mac_info *mac = &adapter->hw.mac; 8675 8676 mac->autoneg = 0; 8677 8678 /* Make sure dplx is at most 1 bit and lsb of speed is not set 8679 * for the switch() below to work 8680 */ 8681 if ((spd & 1) || (dplx & ~1)) 8682 goto err_inval; 8683 8684 /* Fiber NIC's only allow 1000 gbps Full duplex 8685 * and 100Mbps Full duplex for 100baseFx sfp 8686 */ 8687 if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) { 8688 switch (spd + dplx) { 8689 case SPEED_10 + DUPLEX_HALF: 8690 case SPEED_10 + DUPLEX_FULL: 8691 case SPEED_100 + DUPLEX_HALF: 8692 goto err_inval; 8693 default: 8694 break; 8695 } 8696 } 8697 8698 switch (spd + dplx) { 8699 case SPEED_10 + DUPLEX_HALF: 8700 mac->forced_speed_duplex = ADVERTISE_10_HALF; 8701 break; 8702 case SPEED_10 + DUPLEX_FULL: 8703 mac->forced_speed_duplex = ADVERTISE_10_FULL; 8704 break; 8705 case SPEED_100 + DUPLEX_HALF: 8706 mac->forced_speed_duplex = ADVERTISE_100_HALF; 8707 break; 8708 case SPEED_100 + DUPLEX_FULL: 8709 mac->forced_speed_duplex = ADVERTISE_100_FULL; 8710 break; 8711 case SPEED_1000 + DUPLEX_FULL: 8712 mac->autoneg = 1; 8713 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL; 8714 break; 8715 case SPEED_1000 + DUPLEX_HALF: /* not supported */ 8716 default: 8717 goto err_inval; 8718 } 8719 8720 /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */ 8721 adapter->hw.phy.mdix = AUTO_ALL_MODES; 8722 8723 return 0; 8724 8725 err_inval: 8726 dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n"); 8727 return -EINVAL; 8728 } 8729 8730 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake, 8731 bool runtime) 8732 { 8733 struct net_device *netdev = pci_get_drvdata(pdev); 8734 struct igb_adapter *adapter = netdev_priv(netdev); 8735 struct e1000_hw *hw = &adapter->hw; 8736 u32 ctrl, rctl, status; 8737 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol; 8738 bool wake; 8739 8740 rtnl_lock(); 8741 netif_device_detach(netdev); 8742 8743 if (netif_running(netdev)) 8744 __igb_close(netdev, true); 8745 8746 igb_ptp_suspend(adapter); 8747 8748 igb_clear_interrupt_scheme(adapter); 8749 rtnl_unlock(); 8750 8751 status = rd32(E1000_STATUS); 8752 if (status & E1000_STATUS_LU) 8753 wufc &= ~E1000_WUFC_LNKC; 8754 8755 if (wufc) { 8756 igb_setup_rctl(adapter); 8757 igb_set_rx_mode(netdev); 8758 8759 /* turn on all-multi mode if wake on multicast is enabled */ 8760 if (wufc & E1000_WUFC_MC) { 8761 rctl = rd32(E1000_RCTL); 8762 rctl |= E1000_RCTL_MPE; 8763 wr32(E1000_RCTL, rctl); 8764 } 8765 8766 ctrl = rd32(E1000_CTRL); 8767 ctrl |= E1000_CTRL_ADVD3WUC; 8768 wr32(E1000_CTRL, ctrl); 8769 8770 /* Allow time for pending master requests to run */ 8771 igb_disable_pcie_master(hw); 8772 8773 wr32(E1000_WUC, E1000_WUC_PME_EN); 8774 wr32(E1000_WUFC, wufc); 8775 } else { 8776 wr32(E1000_WUC, 0); 8777 wr32(E1000_WUFC, 0); 8778 } 8779 8780 wake = wufc || adapter->en_mng_pt; 8781 if (!wake) 8782 igb_power_down_link(adapter); 8783 else 8784 igb_power_up_link(adapter); 8785 8786 if (enable_wake) 8787 *enable_wake = wake; 8788 8789 /* Release control of h/w to f/w. If f/w is AMT enabled, this 8790 * would have already happened in close and is redundant. 8791 */ 8792 igb_release_hw_control(adapter); 8793 8794 pci_disable_device(pdev); 8795 8796 return 0; 8797 } 8798 8799 static void igb_deliver_wake_packet(struct net_device *netdev) 8800 { 8801 struct igb_adapter *adapter = netdev_priv(netdev); 8802 struct e1000_hw *hw = &adapter->hw; 8803 struct sk_buff *skb; 8804 u32 wupl; 8805 8806 wupl = rd32(E1000_WUPL) & E1000_WUPL_MASK; 8807 8808 /* WUPM stores only the first 128 bytes of the wake packet. 8809 * Read the packet only if we have the whole thing. 8810 */ 8811 if ((wupl == 0) || (wupl > E1000_WUPM_BYTES)) 8812 return; 8813 8814 skb = netdev_alloc_skb_ip_align(netdev, E1000_WUPM_BYTES); 8815 if (!skb) 8816 return; 8817 8818 skb_put(skb, wupl); 8819 8820 /* Ensure reads are 32-bit aligned */ 8821 wupl = roundup(wupl, 4); 8822 8823 memcpy_fromio(skb->data, hw->hw_addr + E1000_WUPM_REG(0), wupl); 8824 8825 skb->protocol = eth_type_trans(skb, netdev); 8826 netif_rx(skb); 8827 } 8828 8829 static int __maybe_unused igb_suspend(struct device *dev) 8830 { 8831 return __igb_shutdown(to_pci_dev(dev), NULL, 0); 8832 } 8833 8834 static int __maybe_unused igb_resume(struct device *dev) 8835 { 8836 struct pci_dev *pdev = to_pci_dev(dev); 8837 struct net_device *netdev = pci_get_drvdata(pdev); 8838 struct igb_adapter *adapter = netdev_priv(netdev); 8839 struct e1000_hw *hw = &adapter->hw; 8840 u32 err, val; 8841 8842 pci_set_power_state(pdev, PCI_D0); 8843 pci_restore_state(pdev); 8844 pci_save_state(pdev); 8845 8846 if (!pci_device_is_present(pdev)) 8847 return -ENODEV; 8848 err = pci_enable_device_mem(pdev); 8849 if (err) { 8850 dev_err(&pdev->dev, 8851 "igb: Cannot enable PCI device from suspend\n"); 8852 return err; 8853 } 8854 pci_set_master(pdev); 8855 8856 pci_enable_wake(pdev, PCI_D3hot, 0); 8857 pci_enable_wake(pdev, PCI_D3cold, 0); 8858 8859 if (igb_init_interrupt_scheme(adapter, true)) { 8860 dev_err(&pdev->dev, "Unable to allocate memory for queues\n"); 8861 return -ENOMEM; 8862 } 8863 8864 igb_reset(adapter); 8865 8866 /* let the f/w know that the h/w is now under the control of the 8867 * driver. 8868 */ 8869 igb_get_hw_control(adapter); 8870 8871 val = rd32(E1000_WUS); 8872 if (val & WAKE_PKT_WUS) 8873 igb_deliver_wake_packet(netdev); 8874 8875 wr32(E1000_WUS, ~0); 8876 8877 rtnl_lock(); 8878 if (!err && netif_running(netdev)) 8879 err = __igb_open(netdev, true); 8880 8881 if (!err) 8882 netif_device_attach(netdev); 8883 rtnl_unlock(); 8884 8885 return err; 8886 } 8887 8888 static int __maybe_unused igb_runtime_idle(struct device *dev) 8889 { 8890 struct pci_dev *pdev = to_pci_dev(dev); 8891 struct net_device *netdev = pci_get_drvdata(pdev); 8892 struct igb_adapter *adapter = netdev_priv(netdev); 8893 8894 if (!igb_has_link(adapter)) 8895 pm_schedule_suspend(dev, MSEC_PER_SEC * 5); 8896 8897 return -EBUSY; 8898 } 8899 8900 static int __maybe_unused igb_runtime_suspend(struct device *dev) 8901 { 8902 return __igb_shutdown(to_pci_dev(dev), NULL, 1); 8903 } 8904 8905 static int __maybe_unused igb_runtime_resume(struct device *dev) 8906 { 8907 return igb_resume(dev); 8908 } 8909 8910 static void igb_shutdown(struct pci_dev *pdev) 8911 { 8912 bool wake; 8913 8914 __igb_shutdown(pdev, &wake, 0); 8915 8916 if (system_state == SYSTEM_POWER_OFF) { 8917 pci_wake_from_d3(pdev, wake); 8918 pci_set_power_state(pdev, PCI_D3hot); 8919 } 8920 } 8921 8922 #ifdef CONFIG_PCI_IOV 8923 static int igb_sriov_reinit(struct pci_dev *dev) 8924 { 8925 struct net_device *netdev = pci_get_drvdata(dev); 8926 struct igb_adapter *adapter = netdev_priv(netdev); 8927 struct pci_dev *pdev = adapter->pdev; 8928 8929 rtnl_lock(); 8930 8931 if (netif_running(netdev)) 8932 igb_close(netdev); 8933 else 8934 igb_reset(adapter); 8935 8936 igb_clear_interrupt_scheme(adapter); 8937 8938 igb_init_queue_configuration(adapter); 8939 8940 if (igb_init_interrupt_scheme(adapter, true)) { 8941 rtnl_unlock(); 8942 dev_err(&pdev->dev, "Unable to allocate memory for queues\n"); 8943 return -ENOMEM; 8944 } 8945 8946 if (netif_running(netdev)) 8947 igb_open(netdev); 8948 8949 rtnl_unlock(); 8950 8951 return 0; 8952 } 8953 8954 static int igb_pci_disable_sriov(struct pci_dev *dev) 8955 { 8956 int err = igb_disable_sriov(dev); 8957 8958 if (!err) 8959 err = igb_sriov_reinit(dev); 8960 8961 return err; 8962 } 8963 8964 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs) 8965 { 8966 int err = igb_enable_sriov(dev, num_vfs); 8967 8968 if (err) 8969 goto out; 8970 8971 err = igb_sriov_reinit(dev); 8972 if (!err) 8973 return num_vfs; 8974 8975 out: 8976 return err; 8977 } 8978 8979 #endif 8980 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs) 8981 { 8982 #ifdef CONFIG_PCI_IOV 8983 if (num_vfs == 0) 8984 return igb_pci_disable_sriov(dev); 8985 else 8986 return igb_pci_enable_sriov(dev, num_vfs); 8987 #endif 8988 return 0; 8989 } 8990 8991 /** 8992 * igb_io_error_detected - called when PCI error is detected 8993 * @pdev: Pointer to PCI device 8994 * @state: The current pci connection state 8995 * 8996 * This function is called after a PCI bus error affecting 8997 * this device has been detected. 8998 **/ 8999 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev, 9000 pci_channel_state_t state) 9001 { 9002 struct net_device *netdev = pci_get_drvdata(pdev); 9003 struct igb_adapter *adapter = netdev_priv(netdev); 9004 9005 netif_device_detach(netdev); 9006 9007 if (state == pci_channel_io_perm_failure) 9008 return PCI_ERS_RESULT_DISCONNECT; 9009 9010 if (netif_running(netdev)) 9011 igb_down(adapter); 9012 pci_disable_device(pdev); 9013 9014 /* Request a slot slot reset. */ 9015 return PCI_ERS_RESULT_NEED_RESET; 9016 } 9017 9018 /** 9019 * igb_io_slot_reset - called after the pci bus has been reset. 9020 * @pdev: Pointer to PCI device 9021 * 9022 * Restart the card from scratch, as if from a cold-boot. Implementation 9023 * resembles the first-half of the igb_resume routine. 9024 **/ 9025 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev) 9026 { 9027 struct net_device *netdev = pci_get_drvdata(pdev); 9028 struct igb_adapter *adapter = netdev_priv(netdev); 9029 struct e1000_hw *hw = &adapter->hw; 9030 pci_ers_result_t result; 9031 9032 if (pci_enable_device_mem(pdev)) { 9033 dev_err(&pdev->dev, 9034 "Cannot re-enable PCI device after reset.\n"); 9035 result = PCI_ERS_RESULT_DISCONNECT; 9036 } else { 9037 pci_set_master(pdev); 9038 pci_restore_state(pdev); 9039 pci_save_state(pdev); 9040 9041 pci_enable_wake(pdev, PCI_D3hot, 0); 9042 pci_enable_wake(pdev, PCI_D3cold, 0); 9043 9044 /* In case of PCI error, adapter lose its HW address 9045 * so we should re-assign it here. 9046 */ 9047 hw->hw_addr = adapter->io_addr; 9048 9049 igb_reset(adapter); 9050 wr32(E1000_WUS, ~0); 9051 result = PCI_ERS_RESULT_RECOVERED; 9052 } 9053 9054 return result; 9055 } 9056 9057 /** 9058 * igb_io_resume - called when traffic can start flowing again. 9059 * @pdev: Pointer to PCI device 9060 * 9061 * This callback is called when the error recovery driver tells us that 9062 * its OK to resume normal operation. Implementation resembles the 9063 * second-half of the igb_resume routine. 9064 */ 9065 static void igb_io_resume(struct pci_dev *pdev) 9066 { 9067 struct net_device *netdev = pci_get_drvdata(pdev); 9068 struct igb_adapter *adapter = netdev_priv(netdev); 9069 9070 if (netif_running(netdev)) { 9071 if (igb_up(adapter)) { 9072 dev_err(&pdev->dev, "igb_up failed after reset\n"); 9073 return; 9074 } 9075 } 9076 9077 netif_device_attach(netdev); 9078 9079 /* let the f/w know that the h/w is now under the control of the 9080 * driver. 9081 */ 9082 igb_get_hw_control(adapter); 9083 } 9084 9085 /** 9086 * igb_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table 9087 * @adapter: Pointer to adapter structure 9088 * @index: Index of the RAR entry which need to be synced with MAC table 9089 **/ 9090 static void igb_rar_set_index(struct igb_adapter *adapter, u32 index) 9091 { 9092 struct e1000_hw *hw = &adapter->hw; 9093 u32 rar_low, rar_high; 9094 u8 *addr = adapter->mac_table[index].addr; 9095 9096 /* HW expects these to be in network order when they are plugged 9097 * into the registers which are little endian. In order to guarantee 9098 * that ordering we need to do an leXX_to_cpup here in order to be 9099 * ready for the byteswap that occurs with writel 9100 */ 9101 rar_low = le32_to_cpup((__le32 *)(addr)); 9102 rar_high = le16_to_cpup((__le16 *)(addr + 4)); 9103 9104 /* Indicate to hardware the Address is Valid. */ 9105 if (adapter->mac_table[index].state & IGB_MAC_STATE_IN_USE) { 9106 if (is_valid_ether_addr(addr)) 9107 rar_high |= E1000_RAH_AV; 9108 9109 if (adapter->mac_table[index].state & IGB_MAC_STATE_SRC_ADDR) 9110 rar_high |= E1000_RAH_ASEL_SRC_ADDR; 9111 9112 switch (hw->mac.type) { 9113 case e1000_82575: 9114 case e1000_i210: 9115 if (adapter->mac_table[index].state & 9116 IGB_MAC_STATE_QUEUE_STEERING) 9117 rar_high |= E1000_RAH_QSEL_ENABLE; 9118 9119 rar_high |= E1000_RAH_POOL_1 * 9120 adapter->mac_table[index].queue; 9121 break; 9122 default: 9123 rar_high |= E1000_RAH_POOL_1 << 9124 adapter->mac_table[index].queue; 9125 break; 9126 } 9127 } 9128 9129 wr32(E1000_RAL(index), rar_low); 9130 wrfl(); 9131 wr32(E1000_RAH(index), rar_high); 9132 wrfl(); 9133 } 9134 9135 static int igb_set_vf_mac(struct igb_adapter *adapter, 9136 int vf, unsigned char *mac_addr) 9137 { 9138 struct e1000_hw *hw = &adapter->hw; 9139 /* VF MAC addresses start at end of receive addresses and moves 9140 * towards the first, as a result a collision should not be possible 9141 */ 9142 int rar_entry = hw->mac.rar_entry_count - (vf + 1); 9143 unsigned char *vf_mac_addr = adapter->vf_data[vf].vf_mac_addresses; 9144 9145 ether_addr_copy(vf_mac_addr, mac_addr); 9146 ether_addr_copy(adapter->mac_table[rar_entry].addr, mac_addr); 9147 adapter->mac_table[rar_entry].queue = vf; 9148 adapter->mac_table[rar_entry].state |= IGB_MAC_STATE_IN_USE; 9149 igb_rar_set_index(adapter, rar_entry); 9150 9151 return 0; 9152 } 9153 9154 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac) 9155 { 9156 struct igb_adapter *adapter = netdev_priv(netdev); 9157 9158 if (vf >= adapter->vfs_allocated_count) 9159 return -EINVAL; 9160 9161 /* Setting the VF MAC to 0 reverts the IGB_VF_FLAG_PF_SET_MAC 9162 * flag and allows to overwrite the MAC via VF netdev. This 9163 * is necessary to allow libvirt a way to restore the original 9164 * MAC after unbinding vfio-pci and reloading igbvf after shutting 9165 * down a VM. 9166 */ 9167 if (is_zero_ether_addr(mac)) { 9168 adapter->vf_data[vf].flags &= ~IGB_VF_FLAG_PF_SET_MAC; 9169 dev_info(&adapter->pdev->dev, 9170 "remove administratively set MAC on VF %d\n", 9171 vf); 9172 } else if (is_valid_ether_addr(mac)) { 9173 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC; 9174 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", 9175 mac, vf); 9176 dev_info(&adapter->pdev->dev, 9177 "Reload the VF driver to make this change effective."); 9178 /* Generate additional warning if PF is down */ 9179 if (test_bit(__IGB_DOWN, &adapter->state)) { 9180 dev_warn(&adapter->pdev->dev, 9181 "The VF MAC address has been set, but the PF device is not up.\n"); 9182 dev_warn(&adapter->pdev->dev, 9183 "Bring the PF device up before attempting to use the VF device.\n"); 9184 } 9185 } else { 9186 return -EINVAL; 9187 } 9188 return igb_set_vf_mac(adapter, vf, mac); 9189 } 9190 9191 static int igb_link_mbps(int internal_link_speed) 9192 { 9193 switch (internal_link_speed) { 9194 case SPEED_100: 9195 return 100; 9196 case SPEED_1000: 9197 return 1000; 9198 default: 9199 return 0; 9200 } 9201 } 9202 9203 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate, 9204 int link_speed) 9205 { 9206 int rf_dec, rf_int; 9207 u32 bcnrc_val; 9208 9209 if (tx_rate != 0) { 9210 /* Calculate the rate factor values to set */ 9211 rf_int = link_speed / tx_rate; 9212 rf_dec = (link_speed - (rf_int * tx_rate)); 9213 rf_dec = (rf_dec * BIT(E1000_RTTBCNRC_RF_INT_SHIFT)) / 9214 tx_rate; 9215 9216 bcnrc_val = E1000_RTTBCNRC_RS_ENA; 9217 bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) & 9218 E1000_RTTBCNRC_RF_INT_MASK); 9219 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK); 9220 } else { 9221 bcnrc_val = 0; 9222 } 9223 9224 wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */ 9225 /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM 9226 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported. 9227 */ 9228 wr32(E1000_RTTBCNRM, 0x14); 9229 wr32(E1000_RTTBCNRC, bcnrc_val); 9230 } 9231 9232 static void igb_check_vf_rate_limit(struct igb_adapter *adapter) 9233 { 9234 int actual_link_speed, i; 9235 bool reset_rate = false; 9236 9237 /* VF TX rate limit was not set or not supported */ 9238 if ((adapter->vf_rate_link_speed == 0) || 9239 (adapter->hw.mac.type != e1000_82576)) 9240 return; 9241 9242 actual_link_speed = igb_link_mbps(adapter->link_speed); 9243 if (actual_link_speed != adapter->vf_rate_link_speed) { 9244 reset_rate = true; 9245 adapter->vf_rate_link_speed = 0; 9246 dev_info(&adapter->pdev->dev, 9247 "Link speed has been changed. VF Transmit rate is disabled\n"); 9248 } 9249 9250 for (i = 0; i < adapter->vfs_allocated_count; i++) { 9251 if (reset_rate) 9252 adapter->vf_data[i].tx_rate = 0; 9253 9254 igb_set_vf_rate_limit(&adapter->hw, i, 9255 adapter->vf_data[i].tx_rate, 9256 actual_link_speed); 9257 } 9258 } 9259 9260 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, 9261 int min_tx_rate, int max_tx_rate) 9262 { 9263 struct igb_adapter *adapter = netdev_priv(netdev); 9264 struct e1000_hw *hw = &adapter->hw; 9265 int actual_link_speed; 9266 9267 if (hw->mac.type != e1000_82576) 9268 return -EOPNOTSUPP; 9269 9270 if (min_tx_rate) 9271 return -EINVAL; 9272 9273 actual_link_speed = igb_link_mbps(adapter->link_speed); 9274 if ((vf >= adapter->vfs_allocated_count) || 9275 (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) || 9276 (max_tx_rate < 0) || 9277 (max_tx_rate > actual_link_speed)) 9278 return -EINVAL; 9279 9280 adapter->vf_rate_link_speed = actual_link_speed; 9281 adapter->vf_data[vf].tx_rate = (u16)max_tx_rate; 9282 igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed); 9283 9284 return 0; 9285 } 9286 9287 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf, 9288 bool setting) 9289 { 9290 struct igb_adapter *adapter = netdev_priv(netdev); 9291 struct e1000_hw *hw = &adapter->hw; 9292 u32 reg_val, reg_offset; 9293 9294 if (!adapter->vfs_allocated_count) 9295 return -EOPNOTSUPP; 9296 9297 if (vf >= adapter->vfs_allocated_count) 9298 return -EINVAL; 9299 9300 reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC; 9301 reg_val = rd32(reg_offset); 9302 if (setting) 9303 reg_val |= (BIT(vf) | 9304 BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)); 9305 else 9306 reg_val &= ~(BIT(vf) | 9307 BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)); 9308 wr32(reg_offset, reg_val); 9309 9310 adapter->vf_data[vf].spoofchk_enabled = setting; 9311 return 0; 9312 } 9313 9314 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf, bool setting) 9315 { 9316 struct igb_adapter *adapter = netdev_priv(netdev); 9317 9318 if (vf >= adapter->vfs_allocated_count) 9319 return -EINVAL; 9320 if (adapter->vf_data[vf].trusted == setting) 9321 return 0; 9322 9323 adapter->vf_data[vf].trusted = setting; 9324 9325 dev_info(&adapter->pdev->dev, "VF %u is %strusted\n", 9326 vf, setting ? "" : "not "); 9327 return 0; 9328 } 9329 9330 static int igb_ndo_get_vf_config(struct net_device *netdev, 9331 int vf, struct ifla_vf_info *ivi) 9332 { 9333 struct igb_adapter *adapter = netdev_priv(netdev); 9334 if (vf >= adapter->vfs_allocated_count) 9335 return -EINVAL; 9336 ivi->vf = vf; 9337 memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN); 9338 ivi->max_tx_rate = adapter->vf_data[vf].tx_rate; 9339 ivi->min_tx_rate = 0; 9340 ivi->vlan = adapter->vf_data[vf].pf_vlan; 9341 ivi->qos = adapter->vf_data[vf].pf_qos; 9342 ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled; 9343 ivi->trusted = adapter->vf_data[vf].trusted; 9344 return 0; 9345 } 9346 9347 static void igb_vmm_control(struct igb_adapter *adapter) 9348 { 9349 struct e1000_hw *hw = &adapter->hw; 9350 u32 reg; 9351 9352 switch (hw->mac.type) { 9353 case e1000_82575: 9354 case e1000_i210: 9355 case e1000_i211: 9356 case e1000_i354: 9357 default: 9358 /* replication is not supported for 82575 */ 9359 return; 9360 case e1000_82576: 9361 /* notify HW that the MAC is adding vlan tags */ 9362 reg = rd32(E1000_DTXCTL); 9363 reg |= E1000_DTXCTL_VLAN_ADDED; 9364 wr32(E1000_DTXCTL, reg); 9365 /* Fall through */ 9366 case e1000_82580: 9367 /* enable replication vlan tag stripping */ 9368 reg = rd32(E1000_RPLOLR); 9369 reg |= E1000_RPLOLR_STRVLAN; 9370 wr32(E1000_RPLOLR, reg); 9371 /* Fall through */ 9372 case e1000_i350: 9373 /* none of the above registers are supported by i350 */ 9374 break; 9375 } 9376 9377 if (adapter->vfs_allocated_count) { 9378 igb_vmdq_set_loopback_pf(hw, true); 9379 igb_vmdq_set_replication_pf(hw, true); 9380 igb_vmdq_set_anti_spoofing_pf(hw, true, 9381 adapter->vfs_allocated_count); 9382 } else { 9383 igb_vmdq_set_loopback_pf(hw, false); 9384 igb_vmdq_set_replication_pf(hw, false); 9385 } 9386 } 9387 9388 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba) 9389 { 9390 struct e1000_hw *hw = &adapter->hw; 9391 u32 dmac_thr; 9392 u16 hwm; 9393 9394 if (hw->mac.type > e1000_82580) { 9395 if (adapter->flags & IGB_FLAG_DMAC) { 9396 u32 reg; 9397 9398 /* force threshold to 0. */ 9399 wr32(E1000_DMCTXTH, 0); 9400 9401 /* DMA Coalescing high water mark needs to be greater 9402 * than the Rx threshold. Set hwm to PBA - max frame 9403 * size in 16B units, capping it at PBA - 6KB. 9404 */ 9405 hwm = 64 * (pba - 6); 9406 reg = rd32(E1000_FCRTC); 9407 reg &= ~E1000_FCRTC_RTH_COAL_MASK; 9408 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT) 9409 & E1000_FCRTC_RTH_COAL_MASK); 9410 wr32(E1000_FCRTC, reg); 9411 9412 /* Set the DMA Coalescing Rx threshold to PBA - 2 * max 9413 * frame size, capping it at PBA - 10KB. 9414 */ 9415 dmac_thr = pba - 10; 9416 reg = rd32(E1000_DMACR); 9417 reg &= ~E1000_DMACR_DMACTHR_MASK; 9418 reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT) 9419 & E1000_DMACR_DMACTHR_MASK); 9420 9421 /* transition to L0x or L1 if available..*/ 9422 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK); 9423 9424 /* watchdog timer= +-1000 usec in 32usec intervals */ 9425 reg |= (1000 >> 5); 9426 9427 /* Disable BMC-to-OS Watchdog Enable */ 9428 if (hw->mac.type != e1000_i354) 9429 reg &= ~E1000_DMACR_DC_BMC2OSW_EN; 9430 9431 wr32(E1000_DMACR, reg); 9432 9433 /* no lower threshold to disable 9434 * coalescing(smart fifb)-UTRESH=0 9435 */ 9436 wr32(E1000_DMCRTRH, 0); 9437 9438 reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4); 9439 9440 wr32(E1000_DMCTLX, reg); 9441 9442 /* free space in tx packet buffer to wake from 9443 * DMA coal 9444 */ 9445 wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE - 9446 (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6); 9447 9448 /* make low power state decision controlled 9449 * by DMA coal 9450 */ 9451 reg = rd32(E1000_PCIEMISC); 9452 reg &= ~E1000_PCIEMISC_LX_DECISION; 9453 wr32(E1000_PCIEMISC, reg); 9454 } /* endif adapter->dmac is not disabled */ 9455 } else if (hw->mac.type == e1000_82580) { 9456 u32 reg = rd32(E1000_PCIEMISC); 9457 9458 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION); 9459 wr32(E1000_DMACR, 0); 9460 } 9461 } 9462 9463 /** 9464 * igb_read_i2c_byte - Reads 8 bit word over I2C 9465 * @hw: pointer to hardware structure 9466 * @byte_offset: byte offset to read 9467 * @dev_addr: device address 9468 * @data: value read 9469 * 9470 * Performs byte read operation over I2C interface at 9471 * a specified device address. 9472 **/ 9473 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset, 9474 u8 dev_addr, u8 *data) 9475 { 9476 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw); 9477 struct i2c_client *this_client = adapter->i2c_client; 9478 s32 status; 9479 u16 swfw_mask = 0; 9480 9481 if (!this_client) 9482 return E1000_ERR_I2C; 9483 9484 swfw_mask = E1000_SWFW_PHY0_SM; 9485 9486 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask)) 9487 return E1000_ERR_SWFW_SYNC; 9488 9489 status = i2c_smbus_read_byte_data(this_client, byte_offset); 9490 hw->mac.ops.release_swfw_sync(hw, swfw_mask); 9491 9492 if (status < 0) 9493 return E1000_ERR_I2C; 9494 else { 9495 *data = status; 9496 return 0; 9497 } 9498 } 9499 9500 /** 9501 * igb_write_i2c_byte - Writes 8 bit word over I2C 9502 * @hw: pointer to hardware structure 9503 * @byte_offset: byte offset to write 9504 * @dev_addr: device address 9505 * @data: value to write 9506 * 9507 * Performs byte write operation over I2C interface at 9508 * a specified device address. 9509 **/ 9510 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset, 9511 u8 dev_addr, u8 data) 9512 { 9513 struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw); 9514 struct i2c_client *this_client = adapter->i2c_client; 9515 s32 status; 9516 u16 swfw_mask = E1000_SWFW_PHY0_SM; 9517 9518 if (!this_client) 9519 return E1000_ERR_I2C; 9520 9521 if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask)) 9522 return E1000_ERR_SWFW_SYNC; 9523 status = i2c_smbus_write_byte_data(this_client, byte_offset, data); 9524 hw->mac.ops.release_swfw_sync(hw, swfw_mask); 9525 9526 if (status) 9527 return E1000_ERR_I2C; 9528 else 9529 return 0; 9530 9531 } 9532 9533 int igb_reinit_queues(struct igb_adapter *adapter) 9534 { 9535 struct net_device *netdev = adapter->netdev; 9536 struct pci_dev *pdev = adapter->pdev; 9537 int err = 0; 9538 9539 if (netif_running(netdev)) 9540 igb_close(netdev); 9541 9542 igb_reset_interrupt_capability(adapter); 9543 9544 if (igb_init_interrupt_scheme(adapter, true)) { 9545 dev_err(&pdev->dev, "Unable to allocate memory for queues\n"); 9546 return -ENOMEM; 9547 } 9548 9549 if (netif_running(netdev)) 9550 err = igb_open(netdev); 9551 9552 return err; 9553 } 9554 9555 static void igb_nfc_filter_exit(struct igb_adapter *adapter) 9556 { 9557 struct igb_nfc_filter *rule; 9558 9559 spin_lock(&adapter->nfc_lock); 9560 9561 hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node) 9562 igb_erase_filter(adapter, rule); 9563 9564 hlist_for_each_entry(rule, &adapter->cls_flower_list, nfc_node) 9565 igb_erase_filter(adapter, rule); 9566 9567 spin_unlock(&adapter->nfc_lock); 9568 } 9569 9570 static void igb_nfc_filter_restore(struct igb_adapter *adapter) 9571 { 9572 struct igb_nfc_filter *rule; 9573 9574 spin_lock(&adapter->nfc_lock); 9575 9576 hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node) 9577 igb_add_filter(adapter, rule); 9578 9579 spin_unlock(&adapter->nfc_lock); 9580 } 9581 /* igb_main.c */ 9582