1 /******************************************************************************* 2 3 Intel 10 Gigabit PCI Express Linux driver 4 Copyright(c) 1999 - 2016 Intel Corporation. 5 6 This program is free software; you can redistribute it and/or modify it 7 under the terms and conditions of the GNU General Public License, 8 version 2, as published by the Free Software Foundation. 9 10 This program is distributed in the hope it will be useful, but WITHOUT 11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 13 more details. 14 15 You should have received a copy of the GNU General Public License along with 16 this program; if not, write to the Free Software Foundation, Inc., 17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. 18 19 The full GNU General Public License is included in this distribution in 20 the file called "COPYING". 21 22 Contact Information: 23 Linux NICS <linux.nics@intel.com> 24 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net> 25 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497 26 27 *******************************************************************************/ 28 29 #include <linux/types.h> 30 #include <linux/module.h> 31 #include <linux/pci.h> 32 #include <linux/netdevice.h> 33 #include <linux/vmalloc.h> 34 #include <linux/string.h> 35 #include <linux/in.h> 36 #include <linux/interrupt.h> 37 #include <linux/ip.h> 38 #include <linux/tcp.h> 39 #include <linux/sctp.h> 40 #include <linux/pkt_sched.h> 41 #include <linux/ipv6.h> 42 #include <linux/slab.h> 43 #include <net/checksum.h> 44 #include <net/ip6_checksum.h> 45 #include <linux/etherdevice.h> 46 #include <linux/ethtool.h> 47 #include <linux/if.h> 48 #include <linux/if_vlan.h> 49 #include <linux/if_macvlan.h> 50 #include <linux/if_bridge.h> 51 #include <linux/prefetch.h> 52 #include <scsi/fc/fc_fcoe.h> 53 #include <net/vxlan.h> 54 #include <net/pkt_cls.h> 55 #include <net/tc_act/tc_gact.h> 56 #include <net/tc_act/tc_mirred.h> 57 58 #include "ixgbe.h" 59 #include "ixgbe_common.h" 60 #include "ixgbe_dcb_82599.h" 61 #include "ixgbe_sriov.h" 62 #include "ixgbe_model.h" 63 64 char ixgbe_driver_name[] = "ixgbe"; 65 static const char ixgbe_driver_string[] = 66 "Intel(R) 10 Gigabit PCI Express Network Driver"; 67 #ifdef IXGBE_FCOE 68 char ixgbe_default_device_descr[] = 69 "Intel(R) 10 Gigabit Network Connection"; 70 #else 71 static char ixgbe_default_device_descr[] = 72 "Intel(R) 10 Gigabit Network Connection"; 73 #endif 74 #define DRV_VERSION "4.4.0-k" 75 const char ixgbe_driver_version[] = DRV_VERSION; 76 static const char ixgbe_copyright[] = 77 "Copyright (c) 1999-2016 Intel Corporation."; 78 79 static const char ixgbe_overheat_msg[] = "Network adapter has been stopped because it has over heated. Restart the computer. If the problem persists, power off the system and replace the adapter"; 80 81 static const struct ixgbe_info *ixgbe_info_tbl[] = { 82 [board_82598] = &ixgbe_82598_info, 83 [board_82599] = &ixgbe_82599_info, 84 [board_X540] = &ixgbe_X540_info, 85 [board_X550] = &ixgbe_X550_info, 86 [board_X550EM_x] = &ixgbe_X550EM_x_info, 87 [board_x550em_a] = &ixgbe_x550em_a_info, 88 }; 89 90 /* ixgbe_pci_tbl - PCI Device ID Table 91 * 92 * Wildcard entries (PCI_ANY_ID) should come last 93 * Last entry must be all 0s 94 * 95 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID, 96 * Class, Class Mask, private data (not used) } 97 */ 98 static const struct pci_device_id ixgbe_pci_tbl[] = { 99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 }, 100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 }, 101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 }, 102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 }, 103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 }, 104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 }, 105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 }, 106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 }, 107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 }, 108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 }, 109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 }, 110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 }, 111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 }, 112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 }, 113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 }, 114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 }, 115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 }, 116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 }, 117 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 }, 118 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 }, 119 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 }, 120 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 }, 121 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 }, 122 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 }, 123 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 }, 124 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 }, 125 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 }, 126 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 }, 127 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 }, 128 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 }, 129 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550}, 130 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T1), board_X550}, 131 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x}, 132 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x}, 133 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x}, 134 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x}, 135 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR), board_x550em_a }, 136 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR_L), board_x550em_a }, 137 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP_N), board_x550em_a }, 138 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII), board_x550em_a }, 139 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII_L), board_x550em_a }, 140 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP), board_x550em_a }, 141 /* required last entry */ 142 {0, } 143 }; 144 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl); 145 146 #ifdef CONFIG_IXGBE_DCA 147 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event, 148 void *p); 149 static struct notifier_block dca_notifier = { 150 .notifier_call = ixgbe_notify_dca, 151 .next = NULL, 152 .priority = 0 153 }; 154 #endif 155 156 #ifdef CONFIG_PCI_IOV 157 static unsigned int max_vfs; 158 module_param(max_vfs, uint, 0); 159 MODULE_PARM_DESC(max_vfs, 160 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)"); 161 #endif /* CONFIG_PCI_IOV */ 162 163 static unsigned int allow_unsupported_sfp; 164 module_param(allow_unsupported_sfp, uint, 0); 165 MODULE_PARM_DESC(allow_unsupported_sfp, 166 "Allow unsupported and untested SFP+ modules on 82599-based adapters"); 167 168 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK) 169 static int debug = -1; 170 module_param(debug, int, 0); 171 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)"); 172 173 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>"); 174 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver"); 175 MODULE_LICENSE("GPL"); 176 MODULE_VERSION(DRV_VERSION); 177 178 static struct workqueue_struct *ixgbe_wq; 179 180 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev); 181 182 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter, 183 u32 reg, u16 *value) 184 { 185 struct pci_dev *parent_dev; 186 struct pci_bus *parent_bus; 187 188 parent_bus = adapter->pdev->bus->parent; 189 if (!parent_bus) 190 return -1; 191 192 parent_dev = parent_bus->self; 193 if (!parent_dev) 194 return -1; 195 196 if (!pci_is_pcie(parent_dev)) 197 return -1; 198 199 pcie_capability_read_word(parent_dev, reg, value); 200 if (*value == IXGBE_FAILED_READ_CFG_WORD && 201 ixgbe_check_cfg_remove(&adapter->hw, parent_dev)) 202 return -1; 203 return 0; 204 } 205 206 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter) 207 { 208 struct ixgbe_hw *hw = &adapter->hw; 209 u16 link_status = 0; 210 int err; 211 212 hw->bus.type = ixgbe_bus_type_pci_express; 213 214 /* Get the negotiated link width and speed from PCI config space of the 215 * parent, as this device is behind a switch 216 */ 217 err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status); 218 219 /* assume caller will handle error case */ 220 if (err) 221 return err; 222 223 hw->bus.width = ixgbe_convert_bus_width(link_status); 224 hw->bus.speed = ixgbe_convert_bus_speed(link_status); 225 226 return 0; 227 } 228 229 /** 230 * ixgbe_check_from_parent - Determine whether PCIe info should come from parent 231 * @hw: hw specific details 232 * 233 * This function is used by probe to determine whether a device's PCI-Express 234 * bandwidth details should be gathered from the parent bus instead of from the 235 * device. Used to ensure that various locations all have the correct device ID 236 * checks. 237 */ 238 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw) 239 { 240 switch (hw->device_id) { 241 case IXGBE_DEV_ID_82599_SFP_SF_QP: 242 case IXGBE_DEV_ID_82599_QSFP_SF_QP: 243 return true; 244 default: 245 return false; 246 } 247 } 248 249 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter, 250 int expected_gts) 251 { 252 struct ixgbe_hw *hw = &adapter->hw; 253 int max_gts = 0; 254 enum pci_bus_speed speed = PCI_SPEED_UNKNOWN; 255 enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN; 256 struct pci_dev *pdev; 257 258 /* Some devices are not connected over PCIe and thus do not negotiate 259 * speed. These devices do not have valid bus info, and thus any report 260 * we generate may not be correct. 261 */ 262 if (hw->bus.type == ixgbe_bus_type_internal) 263 return; 264 265 /* determine whether to use the parent device */ 266 if (ixgbe_pcie_from_parent(&adapter->hw)) 267 pdev = adapter->pdev->bus->parent->self; 268 else 269 pdev = adapter->pdev; 270 271 if (pcie_get_minimum_link(pdev, &speed, &width) || 272 speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) { 273 e_dev_warn("Unable to determine PCI Express bandwidth.\n"); 274 return; 275 } 276 277 switch (speed) { 278 case PCIE_SPEED_2_5GT: 279 /* 8b/10b encoding reduces max throughput by 20% */ 280 max_gts = 2 * width; 281 break; 282 case PCIE_SPEED_5_0GT: 283 /* 8b/10b encoding reduces max throughput by 20% */ 284 max_gts = 4 * width; 285 break; 286 case PCIE_SPEED_8_0GT: 287 /* 128b/130b encoding reduces throughput by less than 2% */ 288 max_gts = 8 * width; 289 break; 290 default: 291 e_dev_warn("Unable to determine PCI Express bandwidth.\n"); 292 return; 293 } 294 295 e_dev_info("PCI Express bandwidth of %dGT/s available\n", 296 max_gts); 297 e_dev_info("(Speed:%s, Width: x%d, Encoding Loss:%s)\n", 298 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : 299 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : 300 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : 301 "Unknown"), 302 width, 303 (speed == PCIE_SPEED_2_5GT ? "20%" : 304 speed == PCIE_SPEED_5_0GT ? "20%" : 305 speed == PCIE_SPEED_8_0GT ? "<2%" : 306 "Unknown")); 307 308 if (max_gts < expected_gts) { 309 e_dev_warn("This is not sufficient for optimal performance of this card.\n"); 310 e_dev_warn("For optimal performance, at least %dGT/s of bandwidth is required.\n", 311 expected_gts); 312 e_dev_warn("A slot with more lanes and/or higher speed is suggested.\n"); 313 } 314 } 315 316 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter) 317 { 318 if (!test_bit(__IXGBE_DOWN, &adapter->state) && 319 !test_bit(__IXGBE_REMOVING, &adapter->state) && 320 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state)) 321 queue_work(ixgbe_wq, &adapter->service_task); 322 } 323 324 static void ixgbe_remove_adapter(struct ixgbe_hw *hw) 325 { 326 struct ixgbe_adapter *adapter = hw->back; 327 328 if (!hw->hw_addr) 329 return; 330 hw->hw_addr = NULL; 331 e_dev_err("Adapter removed\n"); 332 if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state)) 333 ixgbe_service_event_schedule(adapter); 334 } 335 336 static void ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg) 337 { 338 u32 value; 339 340 /* The following check not only optimizes a bit by not 341 * performing a read on the status register when the 342 * register just read was a status register read that 343 * returned IXGBE_FAILED_READ_REG. It also blocks any 344 * potential recursion. 345 */ 346 if (reg == IXGBE_STATUS) { 347 ixgbe_remove_adapter(hw); 348 return; 349 } 350 value = ixgbe_read_reg(hw, IXGBE_STATUS); 351 if (value == IXGBE_FAILED_READ_REG) 352 ixgbe_remove_adapter(hw); 353 } 354 355 /** 356 * ixgbe_read_reg - Read from device register 357 * @hw: hw specific details 358 * @reg: offset of register to read 359 * 360 * Returns : value read or IXGBE_FAILED_READ_REG if removed 361 * 362 * This function is used to read device registers. It checks for device 363 * removal by confirming any read that returns all ones by checking the 364 * status register value for all ones. This function avoids reading from 365 * the hardware if a removal was previously detected in which case it 366 * returns IXGBE_FAILED_READ_REG (all ones). 367 */ 368 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg) 369 { 370 u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr); 371 u32 value; 372 373 if (ixgbe_removed(reg_addr)) 374 return IXGBE_FAILED_READ_REG; 375 if (unlikely(hw->phy.nw_mng_if_sel & 376 IXGBE_NW_MNG_IF_SEL_ENABLE_10_100M)) { 377 struct ixgbe_adapter *adapter; 378 int i; 379 380 for (i = 0; i < 200; ++i) { 381 value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY); 382 if (likely(!value)) 383 goto writes_completed; 384 if (value == IXGBE_FAILED_READ_REG) { 385 ixgbe_remove_adapter(hw); 386 return IXGBE_FAILED_READ_REG; 387 } 388 udelay(5); 389 } 390 391 adapter = hw->back; 392 e_warn(hw, "register writes incomplete %08x\n", value); 393 } 394 395 writes_completed: 396 value = readl(reg_addr + reg); 397 if (unlikely(value == IXGBE_FAILED_READ_REG)) 398 ixgbe_check_remove(hw, reg); 399 return value; 400 } 401 402 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev) 403 { 404 u16 value; 405 406 pci_read_config_word(pdev, PCI_VENDOR_ID, &value); 407 if (value == IXGBE_FAILED_READ_CFG_WORD) { 408 ixgbe_remove_adapter(hw); 409 return true; 410 } 411 return false; 412 } 413 414 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg) 415 { 416 struct ixgbe_adapter *adapter = hw->back; 417 u16 value; 418 419 if (ixgbe_removed(hw->hw_addr)) 420 return IXGBE_FAILED_READ_CFG_WORD; 421 pci_read_config_word(adapter->pdev, reg, &value); 422 if (value == IXGBE_FAILED_READ_CFG_WORD && 423 ixgbe_check_cfg_remove(hw, adapter->pdev)) 424 return IXGBE_FAILED_READ_CFG_WORD; 425 return value; 426 } 427 428 #ifdef CONFIG_PCI_IOV 429 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg) 430 { 431 struct ixgbe_adapter *adapter = hw->back; 432 u32 value; 433 434 if (ixgbe_removed(hw->hw_addr)) 435 return IXGBE_FAILED_READ_CFG_DWORD; 436 pci_read_config_dword(adapter->pdev, reg, &value); 437 if (value == IXGBE_FAILED_READ_CFG_DWORD && 438 ixgbe_check_cfg_remove(hw, adapter->pdev)) 439 return IXGBE_FAILED_READ_CFG_DWORD; 440 return value; 441 } 442 #endif /* CONFIG_PCI_IOV */ 443 444 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value) 445 { 446 struct ixgbe_adapter *adapter = hw->back; 447 448 if (ixgbe_removed(hw->hw_addr)) 449 return; 450 pci_write_config_word(adapter->pdev, reg, value); 451 } 452 453 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter) 454 { 455 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state)); 456 457 /* flush memory to make sure state is correct before next watchdog */ 458 smp_mb__before_atomic(); 459 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state); 460 } 461 462 struct ixgbe_reg_info { 463 u32 ofs; 464 char *name; 465 }; 466 467 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = { 468 469 /* General Registers */ 470 {IXGBE_CTRL, "CTRL"}, 471 {IXGBE_STATUS, "STATUS"}, 472 {IXGBE_CTRL_EXT, "CTRL_EXT"}, 473 474 /* Interrupt Registers */ 475 {IXGBE_EICR, "EICR"}, 476 477 /* RX Registers */ 478 {IXGBE_SRRCTL(0), "SRRCTL"}, 479 {IXGBE_DCA_RXCTRL(0), "DRXCTL"}, 480 {IXGBE_RDLEN(0), "RDLEN"}, 481 {IXGBE_RDH(0), "RDH"}, 482 {IXGBE_RDT(0), "RDT"}, 483 {IXGBE_RXDCTL(0), "RXDCTL"}, 484 {IXGBE_RDBAL(0), "RDBAL"}, 485 {IXGBE_RDBAH(0), "RDBAH"}, 486 487 /* TX Registers */ 488 {IXGBE_TDBAL(0), "TDBAL"}, 489 {IXGBE_TDBAH(0), "TDBAH"}, 490 {IXGBE_TDLEN(0), "TDLEN"}, 491 {IXGBE_TDH(0), "TDH"}, 492 {IXGBE_TDT(0), "TDT"}, 493 {IXGBE_TXDCTL(0), "TXDCTL"}, 494 495 /* List Terminator */ 496 { .name = NULL } 497 }; 498 499 500 /* 501 * ixgbe_regdump - register printout routine 502 */ 503 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo) 504 { 505 int i = 0, j = 0; 506 char rname[16]; 507 u32 regs[64]; 508 509 switch (reginfo->ofs) { 510 case IXGBE_SRRCTL(0): 511 for (i = 0; i < 64; i++) 512 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i)); 513 break; 514 case IXGBE_DCA_RXCTRL(0): 515 for (i = 0; i < 64; i++) 516 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i)); 517 break; 518 case IXGBE_RDLEN(0): 519 for (i = 0; i < 64; i++) 520 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i)); 521 break; 522 case IXGBE_RDH(0): 523 for (i = 0; i < 64; i++) 524 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i)); 525 break; 526 case IXGBE_RDT(0): 527 for (i = 0; i < 64; i++) 528 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i)); 529 break; 530 case IXGBE_RXDCTL(0): 531 for (i = 0; i < 64; i++) 532 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i)); 533 break; 534 case IXGBE_RDBAL(0): 535 for (i = 0; i < 64; i++) 536 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i)); 537 break; 538 case IXGBE_RDBAH(0): 539 for (i = 0; i < 64; i++) 540 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i)); 541 break; 542 case IXGBE_TDBAL(0): 543 for (i = 0; i < 64; i++) 544 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i)); 545 break; 546 case IXGBE_TDBAH(0): 547 for (i = 0; i < 64; i++) 548 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i)); 549 break; 550 case IXGBE_TDLEN(0): 551 for (i = 0; i < 64; i++) 552 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i)); 553 break; 554 case IXGBE_TDH(0): 555 for (i = 0; i < 64; i++) 556 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i)); 557 break; 558 case IXGBE_TDT(0): 559 for (i = 0; i < 64; i++) 560 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i)); 561 break; 562 case IXGBE_TXDCTL(0): 563 for (i = 0; i < 64; i++) 564 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i)); 565 break; 566 default: 567 pr_info("%-15s %08x\n", reginfo->name, 568 IXGBE_READ_REG(hw, reginfo->ofs)); 569 return; 570 } 571 572 for (i = 0; i < 8; i++) { 573 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7); 574 pr_err("%-15s", rname); 575 for (j = 0; j < 8; j++) 576 pr_cont(" %08x", regs[i*8+j]); 577 pr_cont("\n"); 578 } 579 580 } 581 582 /* 583 * ixgbe_dump - Print registers, tx-rings and rx-rings 584 */ 585 static void ixgbe_dump(struct ixgbe_adapter *adapter) 586 { 587 struct net_device *netdev = adapter->netdev; 588 struct ixgbe_hw *hw = &adapter->hw; 589 struct ixgbe_reg_info *reginfo; 590 int n = 0; 591 struct ixgbe_ring *tx_ring; 592 struct ixgbe_tx_buffer *tx_buffer; 593 union ixgbe_adv_tx_desc *tx_desc; 594 struct my_u0 { u64 a; u64 b; } *u0; 595 struct ixgbe_ring *rx_ring; 596 union ixgbe_adv_rx_desc *rx_desc; 597 struct ixgbe_rx_buffer *rx_buffer_info; 598 u32 staterr; 599 int i = 0; 600 601 if (!netif_msg_hw(adapter)) 602 return; 603 604 /* Print netdevice Info */ 605 if (netdev) { 606 dev_info(&adapter->pdev->dev, "Net device Info\n"); 607 pr_info("Device Name state " 608 "trans_start last_rx\n"); 609 pr_info("%-15s %016lX %016lX %016lX\n", 610 netdev->name, 611 netdev->state, 612 dev_trans_start(netdev), 613 netdev->last_rx); 614 } 615 616 /* Print Registers */ 617 dev_info(&adapter->pdev->dev, "Register Dump\n"); 618 pr_info(" Register Name Value\n"); 619 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl; 620 reginfo->name; reginfo++) { 621 ixgbe_regdump(hw, reginfo); 622 } 623 624 /* Print TX Ring Summary */ 625 if (!netdev || !netif_running(netdev)) 626 return; 627 628 dev_info(&adapter->pdev->dev, "TX Rings Summary\n"); 629 pr_info(" %s %s %s %s\n", 630 "Queue [NTU] [NTC] [bi(ntc)->dma ]", 631 "leng", "ntw", "timestamp"); 632 for (n = 0; n < adapter->num_tx_queues; n++) { 633 tx_ring = adapter->tx_ring[n]; 634 tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean]; 635 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n", 636 n, tx_ring->next_to_use, tx_ring->next_to_clean, 637 (u64)dma_unmap_addr(tx_buffer, dma), 638 dma_unmap_len(tx_buffer, len), 639 tx_buffer->next_to_watch, 640 (u64)tx_buffer->time_stamp); 641 } 642 643 /* Print TX Rings */ 644 if (!netif_msg_tx_done(adapter)) 645 goto rx_ring_summary; 646 647 dev_info(&adapter->pdev->dev, "TX Rings Dump\n"); 648 649 /* Transmit Descriptor Formats 650 * 651 * 82598 Advanced Transmit Descriptor 652 * +--------------------------------------------------------------+ 653 * 0 | Buffer Address [63:0] | 654 * +--------------------------------------------------------------+ 655 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN | 656 * +--------------------------------------------------------------+ 657 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0 658 * 659 * 82598 Advanced Transmit Descriptor (Write-Back Format) 660 * +--------------------------------------------------------------+ 661 * 0 | RSV [63:0] | 662 * +--------------------------------------------------------------+ 663 * 8 | RSV | STA | NXTSEQ | 664 * +--------------------------------------------------------------+ 665 * 63 36 35 32 31 0 666 * 667 * 82599+ Advanced Transmit Descriptor 668 * +--------------------------------------------------------------+ 669 * 0 | Buffer Address [63:0] | 670 * +--------------------------------------------------------------+ 671 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN | 672 * +--------------------------------------------------------------+ 673 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0 674 * 675 * 82599+ Advanced Transmit Descriptor (Write-Back Format) 676 * +--------------------------------------------------------------+ 677 * 0 | RSV [63:0] | 678 * +--------------------------------------------------------------+ 679 * 8 | RSV | STA | RSV | 680 * +--------------------------------------------------------------+ 681 * 63 36 35 32 31 0 682 */ 683 684 for (n = 0; n < adapter->num_tx_queues; n++) { 685 tx_ring = adapter->tx_ring[n]; 686 pr_info("------------------------------------\n"); 687 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index); 688 pr_info("------------------------------------\n"); 689 pr_info("%s%s %s %s %s %s\n", 690 "T [desc] [address 63:0 ] ", 691 "[PlPOIdStDDt Ln] [bi->dma ] ", 692 "leng", "ntw", "timestamp", "bi->skb"); 693 694 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) { 695 tx_desc = IXGBE_TX_DESC(tx_ring, i); 696 tx_buffer = &tx_ring->tx_buffer_info[i]; 697 u0 = (struct my_u0 *)tx_desc; 698 if (dma_unmap_len(tx_buffer, len) > 0) { 699 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p", 700 i, 701 le64_to_cpu(u0->a), 702 le64_to_cpu(u0->b), 703 (u64)dma_unmap_addr(tx_buffer, dma), 704 dma_unmap_len(tx_buffer, len), 705 tx_buffer->next_to_watch, 706 (u64)tx_buffer->time_stamp, 707 tx_buffer->skb); 708 if (i == tx_ring->next_to_use && 709 i == tx_ring->next_to_clean) 710 pr_cont(" NTC/U\n"); 711 else if (i == tx_ring->next_to_use) 712 pr_cont(" NTU\n"); 713 else if (i == tx_ring->next_to_clean) 714 pr_cont(" NTC\n"); 715 else 716 pr_cont("\n"); 717 718 if (netif_msg_pktdata(adapter) && 719 tx_buffer->skb) 720 print_hex_dump(KERN_INFO, "", 721 DUMP_PREFIX_ADDRESS, 16, 1, 722 tx_buffer->skb->data, 723 dma_unmap_len(tx_buffer, len), 724 true); 725 } 726 } 727 } 728 729 /* Print RX Rings Summary */ 730 rx_ring_summary: 731 dev_info(&adapter->pdev->dev, "RX Rings Summary\n"); 732 pr_info("Queue [NTU] [NTC]\n"); 733 for (n = 0; n < adapter->num_rx_queues; n++) { 734 rx_ring = adapter->rx_ring[n]; 735 pr_info("%5d %5X %5X\n", 736 n, rx_ring->next_to_use, rx_ring->next_to_clean); 737 } 738 739 /* Print RX Rings */ 740 if (!netif_msg_rx_status(adapter)) 741 return; 742 743 dev_info(&adapter->pdev->dev, "RX Rings Dump\n"); 744 745 /* Receive Descriptor Formats 746 * 747 * 82598 Advanced Receive Descriptor (Read) Format 748 * 63 1 0 749 * +-----------------------------------------------------+ 750 * 0 | Packet Buffer Address [63:1] |A0/NSE| 751 * +----------------------------------------------+------+ 752 * 8 | Header Buffer Address [63:1] | DD | 753 * +-----------------------------------------------------+ 754 * 755 * 756 * 82598 Advanced Receive Descriptor (Write-Back) Format 757 * 758 * 63 48 47 32 31 30 21 20 16 15 4 3 0 759 * +------------------------------------------------------+ 760 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS | 761 * | Packet | IP | | | | Type | Type | 762 * | Checksum | Ident | | | | | | 763 * +------------------------------------------------------+ 764 * 8 | VLAN Tag | Length | Extended Error | Extended Status | 765 * +------------------------------------------------------+ 766 * 63 48 47 32 31 20 19 0 767 * 768 * 82599+ Advanced Receive Descriptor (Read) Format 769 * 63 1 0 770 * +-----------------------------------------------------+ 771 * 0 | Packet Buffer Address [63:1] |A0/NSE| 772 * +----------------------------------------------+------+ 773 * 8 | Header Buffer Address [63:1] | DD | 774 * +-----------------------------------------------------+ 775 * 776 * 777 * 82599+ Advanced Receive Descriptor (Write-Back) Format 778 * 779 * 63 48 47 32 31 30 21 20 17 16 4 3 0 780 * +------------------------------------------------------+ 781 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS | 782 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type | 783 * |/ Flow Dir Flt ID | | | | | | 784 * +------------------------------------------------------+ 785 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP | 786 * +------------------------------------------------------+ 787 * 63 48 47 32 31 20 19 0 788 */ 789 790 for (n = 0; n < adapter->num_rx_queues; n++) { 791 rx_ring = adapter->rx_ring[n]; 792 pr_info("------------------------------------\n"); 793 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index); 794 pr_info("------------------------------------\n"); 795 pr_info("%s%s%s", 796 "R [desc] [ PktBuf A0] ", 797 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ", 798 "<-- Adv Rx Read format\n"); 799 pr_info("%s%s%s", 800 "RWB[desc] [PcsmIpSHl PtRs] ", 801 "[vl er S cks ln] ---------------- [bi->skb ] ", 802 "<-- Adv Rx Write-Back format\n"); 803 804 for (i = 0; i < rx_ring->count; i++) { 805 rx_buffer_info = &rx_ring->rx_buffer_info[i]; 806 rx_desc = IXGBE_RX_DESC(rx_ring, i); 807 u0 = (struct my_u0 *)rx_desc; 808 staterr = le32_to_cpu(rx_desc->wb.upper.status_error); 809 if (staterr & IXGBE_RXD_STAT_DD) { 810 /* Descriptor Done */ 811 pr_info("RWB[0x%03X] %016llX " 812 "%016llX ---------------- %p", i, 813 le64_to_cpu(u0->a), 814 le64_to_cpu(u0->b), 815 rx_buffer_info->skb); 816 } else { 817 pr_info("R [0x%03X] %016llX " 818 "%016llX %016llX %p", i, 819 le64_to_cpu(u0->a), 820 le64_to_cpu(u0->b), 821 (u64)rx_buffer_info->dma, 822 rx_buffer_info->skb); 823 824 if (netif_msg_pktdata(adapter) && 825 rx_buffer_info->dma) { 826 print_hex_dump(KERN_INFO, "", 827 DUMP_PREFIX_ADDRESS, 16, 1, 828 page_address(rx_buffer_info->page) + 829 rx_buffer_info->page_offset, 830 ixgbe_rx_bufsz(rx_ring), true); 831 } 832 } 833 834 if (i == rx_ring->next_to_use) 835 pr_cont(" NTU\n"); 836 else if (i == rx_ring->next_to_clean) 837 pr_cont(" NTC\n"); 838 else 839 pr_cont("\n"); 840 841 } 842 } 843 } 844 845 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter) 846 { 847 u32 ctrl_ext; 848 849 /* Let firmware take over control of h/w */ 850 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT); 851 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT, 852 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD); 853 } 854 855 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter) 856 { 857 u32 ctrl_ext; 858 859 /* Let firmware know the driver has taken over */ 860 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT); 861 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT, 862 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD); 863 } 864 865 /** 866 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors 867 * @adapter: pointer to adapter struct 868 * @direction: 0 for Rx, 1 for Tx, -1 for other causes 869 * @queue: queue to map the corresponding interrupt to 870 * @msix_vector: the vector to map to the corresponding queue 871 * 872 */ 873 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction, 874 u8 queue, u8 msix_vector) 875 { 876 u32 ivar, index; 877 struct ixgbe_hw *hw = &adapter->hw; 878 switch (hw->mac.type) { 879 case ixgbe_mac_82598EB: 880 msix_vector |= IXGBE_IVAR_ALLOC_VAL; 881 if (direction == -1) 882 direction = 0; 883 index = (((direction * 64) + queue) >> 2) & 0x1F; 884 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index)); 885 ivar &= ~(0xFF << (8 * (queue & 0x3))); 886 ivar |= (msix_vector << (8 * (queue & 0x3))); 887 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar); 888 break; 889 case ixgbe_mac_82599EB: 890 case ixgbe_mac_X540: 891 case ixgbe_mac_X550: 892 case ixgbe_mac_X550EM_x: 893 case ixgbe_mac_x550em_a: 894 if (direction == -1) { 895 /* other causes */ 896 msix_vector |= IXGBE_IVAR_ALLOC_VAL; 897 index = ((queue & 1) * 8); 898 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC); 899 ivar &= ~(0xFF << index); 900 ivar |= (msix_vector << index); 901 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar); 902 break; 903 } else { 904 /* tx or rx causes */ 905 msix_vector |= IXGBE_IVAR_ALLOC_VAL; 906 index = ((16 * (queue & 1)) + (8 * direction)); 907 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1)); 908 ivar &= ~(0xFF << index); 909 ivar |= (msix_vector << index); 910 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar); 911 break; 912 } 913 default: 914 break; 915 } 916 } 917 918 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter, 919 u64 qmask) 920 { 921 u32 mask; 922 923 switch (adapter->hw.mac.type) { 924 case ixgbe_mac_82598EB: 925 mask = (IXGBE_EIMS_RTX_QUEUE & qmask); 926 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask); 927 break; 928 case ixgbe_mac_82599EB: 929 case ixgbe_mac_X540: 930 case ixgbe_mac_X550: 931 case ixgbe_mac_X550EM_x: 932 case ixgbe_mac_x550em_a: 933 mask = (qmask & 0xFFFFFFFF); 934 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask); 935 mask = (qmask >> 32); 936 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask); 937 break; 938 default: 939 break; 940 } 941 } 942 943 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring, 944 struct ixgbe_tx_buffer *tx_buffer) 945 { 946 if (tx_buffer->skb) { 947 dev_kfree_skb_any(tx_buffer->skb); 948 if (dma_unmap_len(tx_buffer, len)) 949 dma_unmap_single(ring->dev, 950 dma_unmap_addr(tx_buffer, dma), 951 dma_unmap_len(tx_buffer, len), 952 DMA_TO_DEVICE); 953 } else if (dma_unmap_len(tx_buffer, len)) { 954 dma_unmap_page(ring->dev, 955 dma_unmap_addr(tx_buffer, dma), 956 dma_unmap_len(tx_buffer, len), 957 DMA_TO_DEVICE); 958 } 959 tx_buffer->next_to_watch = NULL; 960 tx_buffer->skb = NULL; 961 dma_unmap_len_set(tx_buffer, len, 0); 962 /* tx_buffer must be completely set up in the transmit path */ 963 } 964 965 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter) 966 { 967 struct ixgbe_hw *hw = &adapter->hw; 968 struct ixgbe_hw_stats *hwstats = &adapter->stats; 969 int i; 970 u32 data; 971 972 if ((hw->fc.current_mode != ixgbe_fc_full) && 973 (hw->fc.current_mode != ixgbe_fc_rx_pause)) 974 return; 975 976 switch (hw->mac.type) { 977 case ixgbe_mac_82598EB: 978 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC); 979 break; 980 default: 981 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT); 982 } 983 hwstats->lxoffrxc += data; 984 985 /* refill credits (no tx hang) if we received xoff */ 986 if (!data) 987 return; 988 989 for (i = 0; i < adapter->num_tx_queues; i++) 990 clear_bit(__IXGBE_HANG_CHECK_ARMED, 991 &adapter->tx_ring[i]->state); 992 } 993 994 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter) 995 { 996 struct ixgbe_hw *hw = &adapter->hw; 997 struct ixgbe_hw_stats *hwstats = &adapter->stats; 998 u32 xoff[8] = {0}; 999 u8 tc; 1000 int i; 1001 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable; 1002 1003 if (adapter->ixgbe_ieee_pfc) 1004 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en); 1005 1006 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) { 1007 ixgbe_update_xoff_rx_lfc(adapter); 1008 return; 1009 } 1010 1011 /* update stats for each tc, only valid with PFC enabled */ 1012 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) { 1013 u32 pxoffrxc; 1014 1015 switch (hw->mac.type) { 1016 case ixgbe_mac_82598EB: 1017 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i)); 1018 break; 1019 default: 1020 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i)); 1021 } 1022 hwstats->pxoffrxc[i] += pxoffrxc; 1023 /* Get the TC for given UP */ 1024 tc = netdev_get_prio_tc_map(adapter->netdev, i); 1025 xoff[tc] += pxoffrxc; 1026 } 1027 1028 /* disarm tx queues that have received xoff frames */ 1029 for (i = 0; i < adapter->num_tx_queues; i++) { 1030 struct ixgbe_ring *tx_ring = adapter->tx_ring[i]; 1031 1032 tc = tx_ring->dcb_tc; 1033 if (xoff[tc]) 1034 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state); 1035 } 1036 } 1037 1038 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring) 1039 { 1040 return ring->stats.packets; 1041 } 1042 1043 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring) 1044 { 1045 struct ixgbe_adapter *adapter; 1046 struct ixgbe_hw *hw; 1047 u32 head, tail; 1048 1049 if (ring->l2_accel_priv) 1050 adapter = ring->l2_accel_priv->real_adapter; 1051 else 1052 adapter = netdev_priv(ring->netdev); 1053 1054 hw = &adapter->hw; 1055 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx)); 1056 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx)); 1057 1058 if (head != tail) 1059 return (head < tail) ? 1060 tail - head : (tail + ring->count - head); 1061 1062 return 0; 1063 } 1064 1065 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring) 1066 { 1067 u32 tx_done = ixgbe_get_tx_completed(tx_ring); 1068 u32 tx_done_old = tx_ring->tx_stats.tx_done_old; 1069 u32 tx_pending = ixgbe_get_tx_pending(tx_ring); 1070 1071 clear_check_for_tx_hang(tx_ring); 1072 1073 /* 1074 * Check for a hung queue, but be thorough. This verifies 1075 * that a transmit has been completed since the previous 1076 * check AND there is at least one packet pending. The 1077 * ARMED bit is set to indicate a potential hang. The 1078 * bit is cleared if a pause frame is received to remove 1079 * false hang detection due to PFC or 802.3x frames. By 1080 * requiring this to fail twice we avoid races with 1081 * pfc clearing the ARMED bit and conditions where we 1082 * run the check_tx_hang logic with a transmit completion 1083 * pending but without time to complete it yet. 1084 */ 1085 if (tx_done_old == tx_done && tx_pending) 1086 /* make sure it is true for two checks in a row */ 1087 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED, 1088 &tx_ring->state); 1089 /* update completed stats and continue */ 1090 tx_ring->tx_stats.tx_done_old = tx_done; 1091 /* reset the countdown */ 1092 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state); 1093 1094 return false; 1095 } 1096 1097 /** 1098 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout 1099 * @adapter: driver private struct 1100 **/ 1101 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter) 1102 { 1103 1104 /* Do the reset outside of interrupt context */ 1105 if (!test_bit(__IXGBE_DOWN, &adapter->state)) { 1106 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED; 1107 e_warn(drv, "initiating reset due to tx timeout\n"); 1108 ixgbe_service_event_schedule(adapter); 1109 } 1110 } 1111 1112 /** 1113 * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate 1114 **/ 1115 static int ixgbe_tx_maxrate(struct net_device *netdev, 1116 int queue_index, u32 maxrate) 1117 { 1118 struct ixgbe_adapter *adapter = netdev_priv(netdev); 1119 struct ixgbe_hw *hw = &adapter->hw; 1120 u32 bcnrc_val = ixgbe_link_mbps(adapter); 1121 1122 if (!maxrate) 1123 return 0; 1124 1125 /* Calculate the rate factor values to set */ 1126 bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT; 1127 bcnrc_val /= maxrate; 1128 1129 /* clear everything but the rate factor */ 1130 bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK | 1131 IXGBE_RTTBCNRC_RF_DEC_MASK; 1132 1133 /* enable the rate scheduler */ 1134 bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA; 1135 1136 IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index); 1137 IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val); 1138 1139 return 0; 1140 } 1141 1142 /** 1143 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes 1144 * @q_vector: structure containing interrupt and ring information 1145 * @tx_ring: tx ring to clean 1146 * @napi_budget: Used to determine if we are in netpoll 1147 **/ 1148 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector, 1149 struct ixgbe_ring *tx_ring, int napi_budget) 1150 { 1151 struct ixgbe_adapter *adapter = q_vector->adapter; 1152 struct ixgbe_tx_buffer *tx_buffer; 1153 union ixgbe_adv_tx_desc *tx_desc; 1154 unsigned int total_bytes = 0, total_packets = 0; 1155 unsigned int budget = q_vector->tx.work_limit; 1156 unsigned int i = tx_ring->next_to_clean; 1157 1158 if (test_bit(__IXGBE_DOWN, &adapter->state)) 1159 return true; 1160 1161 tx_buffer = &tx_ring->tx_buffer_info[i]; 1162 tx_desc = IXGBE_TX_DESC(tx_ring, i); 1163 i -= tx_ring->count; 1164 1165 do { 1166 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch; 1167 1168 /* if next_to_watch is not set then there is no work pending */ 1169 if (!eop_desc) 1170 break; 1171 1172 /* prevent any other reads prior to eop_desc */ 1173 read_barrier_depends(); 1174 1175 /* if DD is not set pending work has not been completed */ 1176 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD))) 1177 break; 1178 1179 /* clear next_to_watch to prevent false hangs */ 1180 tx_buffer->next_to_watch = NULL; 1181 1182 /* update the statistics for this packet */ 1183 total_bytes += tx_buffer->bytecount; 1184 total_packets += tx_buffer->gso_segs; 1185 1186 /* free the skb */ 1187 napi_consume_skb(tx_buffer->skb, napi_budget); 1188 1189 /* unmap skb header data */ 1190 dma_unmap_single(tx_ring->dev, 1191 dma_unmap_addr(tx_buffer, dma), 1192 dma_unmap_len(tx_buffer, len), 1193 DMA_TO_DEVICE); 1194 1195 /* clear tx_buffer data */ 1196 tx_buffer->skb = NULL; 1197 dma_unmap_len_set(tx_buffer, len, 0); 1198 1199 /* unmap remaining buffers */ 1200 while (tx_desc != eop_desc) { 1201 tx_buffer++; 1202 tx_desc++; 1203 i++; 1204 if (unlikely(!i)) { 1205 i -= tx_ring->count; 1206 tx_buffer = tx_ring->tx_buffer_info; 1207 tx_desc = IXGBE_TX_DESC(tx_ring, 0); 1208 } 1209 1210 /* unmap any remaining paged data */ 1211 if (dma_unmap_len(tx_buffer, len)) { 1212 dma_unmap_page(tx_ring->dev, 1213 dma_unmap_addr(tx_buffer, dma), 1214 dma_unmap_len(tx_buffer, len), 1215 DMA_TO_DEVICE); 1216 dma_unmap_len_set(tx_buffer, len, 0); 1217 } 1218 } 1219 1220 /* move us one more past the eop_desc for start of next pkt */ 1221 tx_buffer++; 1222 tx_desc++; 1223 i++; 1224 if (unlikely(!i)) { 1225 i -= tx_ring->count; 1226 tx_buffer = tx_ring->tx_buffer_info; 1227 tx_desc = IXGBE_TX_DESC(tx_ring, 0); 1228 } 1229 1230 /* issue prefetch for next Tx descriptor */ 1231 prefetch(tx_desc); 1232 1233 /* update budget accounting */ 1234 budget--; 1235 } while (likely(budget)); 1236 1237 i += tx_ring->count; 1238 tx_ring->next_to_clean = i; 1239 u64_stats_update_begin(&tx_ring->syncp); 1240 tx_ring->stats.bytes += total_bytes; 1241 tx_ring->stats.packets += total_packets; 1242 u64_stats_update_end(&tx_ring->syncp); 1243 q_vector->tx.total_bytes += total_bytes; 1244 q_vector->tx.total_packets += total_packets; 1245 1246 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) { 1247 /* schedule immediate reset if we believe we hung */ 1248 struct ixgbe_hw *hw = &adapter->hw; 1249 e_err(drv, "Detected Tx Unit Hang\n" 1250 " Tx Queue <%d>\n" 1251 " TDH, TDT <%x>, <%x>\n" 1252 " next_to_use <%x>\n" 1253 " next_to_clean <%x>\n" 1254 "tx_buffer_info[next_to_clean]\n" 1255 " time_stamp <%lx>\n" 1256 " jiffies <%lx>\n", 1257 tx_ring->queue_index, 1258 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)), 1259 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)), 1260 tx_ring->next_to_use, i, 1261 tx_ring->tx_buffer_info[i].time_stamp, jiffies); 1262 1263 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index); 1264 1265 e_info(probe, 1266 "tx hang %d detected on queue %d, resetting adapter\n", 1267 adapter->tx_timeout_count + 1, tx_ring->queue_index); 1268 1269 /* schedule immediate reset if we believe we hung */ 1270 ixgbe_tx_timeout_reset(adapter); 1271 1272 /* the adapter is about to reset, no point in enabling stuff */ 1273 return true; 1274 } 1275 1276 netdev_tx_completed_queue(txring_txq(tx_ring), 1277 total_packets, total_bytes); 1278 1279 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2) 1280 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) && 1281 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) { 1282 /* Make sure that anybody stopping the queue after this 1283 * sees the new next_to_clean. 1284 */ 1285 smp_mb(); 1286 if (__netif_subqueue_stopped(tx_ring->netdev, 1287 tx_ring->queue_index) 1288 && !test_bit(__IXGBE_DOWN, &adapter->state)) { 1289 netif_wake_subqueue(tx_ring->netdev, 1290 tx_ring->queue_index); 1291 ++tx_ring->tx_stats.restart_queue; 1292 } 1293 } 1294 1295 return !!budget; 1296 } 1297 1298 #ifdef CONFIG_IXGBE_DCA 1299 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter, 1300 struct ixgbe_ring *tx_ring, 1301 int cpu) 1302 { 1303 struct ixgbe_hw *hw = &adapter->hw; 1304 u32 txctrl = 0; 1305 u16 reg_offset; 1306 1307 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) 1308 txctrl = dca3_get_tag(tx_ring->dev, cpu); 1309 1310 switch (hw->mac.type) { 1311 case ixgbe_mac_82598EB: 1312 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx); 1313 break; 1314 case ixgbe_mac_82599EB: 1315 case ixgbe_mac_X540: 1316 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx); 1317 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599; 1318 break; 1319 default: 1320 /* for unknown hardware do not write register */ 1321 return; 1322 } 1323 1324 /* 1325 * We can enable relaxed ordering for reads, but not writes when 1326 * DCA is enabled. This is due to a known issue in some chipsets 1327 * which will cause the DCA tag to be cleared. 1328 */ 1329 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN | 1330 IXGBE_DCA_TXCTRL_DATA_RRO_EN | 1331 IXGBE_DCA_TXCTRL_DESC_DCA_EN; 1332 1333 IXGBE_WRITE_REG(hw, reg_offset, txctrl); 1334 } 1335 1336 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter, 1337 struct ixgbe_ring *rx_ring, 1338 int cpu) 1339 { 1340 struct ixgbe_hw *hw = &adapter->hw; 1341 u32 rxctrl = 0; 1342 u8 reg_idx = rx_ring->reg_idx; 1343 1344 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) 1345 rxctrl = dca3_get_tag(rx_ring->dev, cpu); 1346 1347 switch (hw->mac.type) { 1348 case ixgbe_mac_82599EB: 1349 case ixgbe_mac_X540: 1350 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599; 1351 break; 1352 default: 1353 break; 1354 } 1355 1356 /* 1357 * We can enable relaxed ordering for reads, but not writes when 1358 * DCA is enabled. This is due to a known issue in some chipsets 1359 * which will cause the DCA tag to be cleared. 1360 */ 1361 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN | 1362 IXGBE_DCA_RXCTRL_DATA_DCA_EN | 1363 IXGBE_DCA_RXCTRL_DESC_DCA_EN; 1364 1365 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl); 1366 } 1367 1368 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector) 1369 { 1370 struct ixgbe_adapter *adapter = q_vector->adapter; 1371 struct ixgbe_ring *ring; 1372 int cpu = get_cpu(); 1373 1374 if (q_vector->cpu == cpu) 1375 goto out_no_update; 1376 1377 ixgbe_for_each_ring(ring, q_vector->tx) 1378 ixgbe_update_tx_dca(adapter, ring, cpu); 1379 1380 ixgbe_for_each_ring(ring, q_vector->rx) 1381 ixgbe_update_rx_dca(adapter, ring, cpu); 1382 1383 q_vector->cpu = cpu; 1384 out_no_update: 1385 put_cpu(); 1386 } 1387 1388 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter) 1389 { 1390 int i; 1391 1392 /* always use CB2 mode, difference is masked in the CB driver */ 1393 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) 1394 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1395 IXGBE_DCA_CTRL_DCA_MODE_CB2); 1396 else 1397 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1398 IXGBE_DCA_CTRL_DCA_DISABLE); 1399 1400 for (i = 0; i < adapter->num_q_vectors; i++) { 1401 adapter->q_vector[i]->cpu = -1; 1402 ixgbe_update_dca(adapter->q_vector[i]); 1403 } 1404 } 1405 1406 static int __ixgbe_notify_dca(struct device *dev, void *data) 1407 { 1408 struct ixgbe_adapter *adapter = dev_get_drvdata(dev); 1409 unsigned long event = *(unsigned long *)data; 1410 1411 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE)) 1412 return 0; 1413 1414 switch (event) { 1415 case DCA_PROVIDER_ADD: 1416 /* if we're already enabled, don't do it again */ 1417 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) 1418 break; 1419 if (dca_add_requester(dev) == 0) { 1420 adapter->flags |= IXGBE_FLAG_DCA_ENABLED; 1421 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1422 IXGBE_DCA_CTRL_DCA_MODE_CB2); 1423 break; 1424 } 1425 /* Fall Through since DCA is disabled. */ 1426 case DCA_PROVIDER_REMOVE: 1427 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) { 1428 dca_remove_requester(dev); 1429 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED; 1430 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1431 IXGBE_DCA_CTRL_DCA_DISABLE); 1432 } 1433 break; 1434 } 1435 1436 return 0; 1437 } 1438 1439 #endif /* CONFIG_IXGBE_DCA */ 1440 1441 #define IXGBE_RSS_L4_TYPES_MASK \ 1442 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \ 1443 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \ 1444 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \ 1445 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP)) 1446 1447 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring, 1448 union ixgbe_adv_rx_desc *rx_desc, 1449 struct sk_buff *skb) 1450 { 1451 u16 rss_type; 1452 1453 if (!(ring->netdev->features & NETIF_F_RXHASH)) 1454 return; 1455 1456 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) & 1457 IXGBE_RXDADV_RSSTYPE_MASK; 1458 1459 if (!rss_type) 1460 return; 1461 1462 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss), 1463 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ? 1464 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3); 1465 } 1466 1467 #ifdef IXGBE_FCOE 1468 /** 1469 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type 1470 * @ring: structure containing ring specific data 1471 * @rx_desc: advanced rx descriptor 1472 * 1473 * Returns : true if it is FCoE pkt 1474 */ 1475 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring, 1476 union ixgbe_adv_rx_desc *rx_desc) 1477 { 1478 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info; 1479 1480 return test_bit(__IXGBE_RX_FCOE, &ring->state) && 1481 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) == 1482 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE << 1483 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT))); 1484 } 1485 1486 #endif /* IXGBE_FCOE */ 1487 /** 1488 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum 1489 * @ring: structure containing ring specific data 1490 * @rx_desc: current Rx descriptor being processed 1491 * @skb: skb currently being received and modified 1492 **/ 1493 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring, 1494 union ixgbe_adv_rx_desc *rx_desc, 1495 struct sk_buff *skb) 1496 { 1497 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info; 1498 __le16 hdr_info = rx_desc->wb.lower.lo_dword.hs_rss.hdr_info; 1499 bool encap_pkt = false; 1500 1501 skb_checksum_none_assert(skb); 1502 1503 /* Rx csum disabled */ 1504 if (!(ring->netdev->features & NETIF_F_RXCSUM)) 1505 return; 1506 1507 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) && 1508 (hdr_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_TUNNEL >> 16))) { 1509 encap_pkt = true; 1510 skb->encapsulation = 1; 1511 } 1512 1513 /* if IP and error */ 1514 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) && 1515 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) { 1516 ring->rx_stats.csum_err++; 1517 return; 1518 } 1519 1520 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS)) 1521 return; 1522 1523 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) { 1524 /* 1525 * 82599 errata, UDP frames with a 0 checksum can be marked as 1526 * checksum errors. 1527 */ 1528 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) && 1529 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state)) 1530 return; 1531 1532 ring->rx_stats.csum_err++; 1533 return; 1534 } 1535 1536 /* It must be a TCP or UDP packet with a valid checksum */ 1537 skb->ip_summed = CHECKSUM_UNNECESSARY; 1538 if (encap_pkt) { 1539 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS)) 1540 return; 1541 1542 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) { 1543 skb->ip_summed = CHECKSUM_NONE; 1544 return; 1545 } 1546 /* If we checked the outer header let the stack know */ 1547 skb->csum_level = 1; 1548 } 1549 } 1550 1551 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring, 1552 struct ixgbe_rx_buffer *bi) 1553 { 1554 struct page *page = bi->page; 1555 dma_addr_t dma; 1556 1557 /* since we are recycling buffers we should seldom need to alloc */ 1558 if (likely(page)) 1559 return true; 1560 1561 /* alloc new page for storage */ 1562 page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring)); 1563 if (unlikely(!page)) { 1564 rx_ring->rx_stats.alloc_rx_page_failed++; 1565 return false; 1566 } 1567 1568 /* map page for use */ 1569 dma = dma_map_page(rx_ring->dev, page, 0, 1570 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE); 1571 1572 /* 1573 * if mapping failed free memory back to system since 1574 * there isn't much point in holding memory we can't use 1575 */ 1576 if (dma_mapping_error(rx_ring->dev, dma)) { 1577 __free_pages(page, ixgbe_rx_pg_order(rx_ring)); 1578 1579 rx_ring->rx_stats.alloc_rx_page_failed++; 1580 return false; 1581 } 1582 1583 bi->dma = dma; 1584 bi->page = page; 1585 bi->page_offset = 0; 1586 1587 return true; 1588 } 1589 1590 /** 1591 * ixgbe_alloc_rx_buffers - Replace used receive buffers 1592 * @rx_ring: ring to place buffers on 1593 * @cleaned_count: number of buffers to replace 1594 **/ 1595 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count) 1596 { 1597 union ixgbe_adv_rx_desc *rx_desc; 1598 struct ixgbe_rx_buffer *bi; 1599 u16 i = rx_ring->next_to_use; 1600 1601 /* nothing to do */ 1602 if (!cleaned_count) 1603 return; 1604 1605 rx_desc = IXGBE_RX_DESC(rx_ring, i); 1606 bi = &rx_ring->rx_buffer_info[i]; 1607 i -= rx_ring->count; 1608 1609 do { 1610 if (!ixgbe_alloc_mapped_page(rx_ring, bi)) 1611 break; 1612 1613 /* 1614 * Refresh the desc even if buffer_addrs didn't change 1615 * because each write-back erases this info. 1616 */ 1617 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset); 1618 1619 rx_desc++; 1620 bi++; 1621 i++; 1622 if (unlikely(!i)) { 1623 rx_desc = IXGBE_RX_DESC(rx_ring, 0); 1624 bi = rx_ring->rx_buffer_info; 1625 i -= rx_ring->count; 1626 } 1627 1628 /* clear the status bits for the next_to_use descriptor */ 1629 rx_desc->wb.upper.status_error = 0; 1630 1631 cleaned_count--; 1632 } while (cleaned_count); 1633 1634 i += rx_ring->count; 1635 1636 if (rx_ring->next_to_use != i) { 1637 rx_ring->next_to_use = i; 1638 1639 /* update next to alloc since we have filled the ring */ 1640 rx_ring->next_to_alloc = i; 1641 1642 /* Force memory writes to complete before letting h/w 1643 * know there are new descriptors to fetch. (Only 1644 * applicable for weak-ordered memory model archs, 1645 * such as IA-64). 1646 */ 1647 wmb(); 1648 writel(i, rx_ring->tail); 1649 } 1650 } 1651 1652 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring, 1653 struct sk_buff *skb) 1654 { 1655 u16 hdr_len = skb_headlen(skb); 1656 1657 /* set gso_size to avoid messing up TCP MSS */ 1658 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len), 1659 IXGBE_CB(skb)->append_cnt); 1660 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; 1661 } 1662 1663 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring, 1664 struct sk_buff *skb) 1665 { 1666 /* if append_cnt is 0 then frame is not RSC */ 1667 if (!IXGBE_CB(skb)->append_cnt) 1668 return; 1669 1670 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt; 1671 rx_ring->rx_stats.rsc_flush++; 1672 1673 ixgbe_set_rsc_gso_size(rx_ring, skb); 1674 1675 /* gso_size is computed using append_cnt so always clear it last */ 1676 IXGBE_CB(skb)->append_cnt = 0; 1677 } 1678 1679 /** 1680 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor 1681 * @rx_ring: rx descriptor ring packet is being transacted on 1682 * @rx_desc: pointer to the EOP Rx descriptor 1683 * @skb: pointer to current skb being populated 1684 * 1685 * This function checks the ring, descriptor, and packet information in 1686 * order to populate the hash, checksum, VLAN, timestamp, protocol, and 1687 * other fields within the skb. 1688 **/ 1689 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring, 1690 union ixgbe_adv_rx_desc *rx_desc, 1691 struct sk_buff *skb) 1692 { 1693 struct net_device *dev = rx_ring->netdev; 1694 u32 flags = rx_ring->q_vector->adapter->flags; 1695 1696 ixgbe_update_rsc_stats(rx_ring, skb); 1697 1698 ixgbe_rx_hash(rx_ring, rx_desc, skb); 1699 1700 ixgbe_rx_checksum(rx_ring, rx_desc, skb); 1701 1702 if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED)) 1703 ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb); 1704 1705 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) && 1706 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) { 1707 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan); 1708 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid); 1709 } 1710 1711 skb_record_rx_queue(skb, rx_ring->queue_index); 1712 1713 skb->protocol = eth_type_trans(skb, dev); 1714 } 1715 1716 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector, 1717 struct sk_buff *skb) 1718 { 1719 skb_mark_napi_id(skb, &q_vector->napi); 1720 if (ixgbe_qv_busy_polling(q_vector)) 1721 netif_receive_skb(skb); 1722 else 1723 napi_gro_receive(&q_vector->napi, skb); 1724 } 1725 1726 /** 1727 * ixgbe_is_non_eop - process handling of non-EOP buffers 1728 * @rx_ring: Rx ring being processed 1729 * @rx_desc: Rx descriptor for current buffer 1730 * @skb: Current socket buffer containing buffer in progress 1731 * 1732 * This function updates next to clean. If the buffer is an EOP buffer 1733 * this function exits returning false, otherwise it will place the 1734 * sk_buff in the next buffer to be chained and return true indicating 1735 * that this is in fact a non-EOP buffer. 1736 **/ 1737 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring, 1738 union ixgbe_adv_rx_desc *rx_desc, 1739 struct sk_buff *skb) 1740 { 1741 u32 ntc = rx_ring->next_to_clean + 1; 1742 1743 /* fetch, update, and store next to clean */ 1744 ntc = (ntc < rx_ring->count) ? ntc : 0; 1745 rx_ring->next_to_clean = ntc; 1746 1747 prefetch(IXGBE_RX_DESC(rx_ring, ntc)); 1748 1749 /* update RSC append count if present */ 1750 if (ring_is_rsc_enabled(rx_ring)) { 1751 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data & 1752 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK); 1753 1754 if (unlikely(rsc_enabled)) { 1755 u32 rsc_cnt = le32_to_cpu(rsc_enabled); 1756 1757 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT; 1758 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1; 1759 1760 /* update ntc based on RSC value */ 1761 ntc = le32_to_cpu(rx_desc->wb.upper.status_error); 1762 ntc &= IXGBE_RXDADV_NEXTP_MASK; 1763 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT; 1764 } 1765 } 1766 1767 /* if we are the last buffer then there is nothing else to do */ 1768 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))) 1769 return false; 1770 1771 /* place skb in next buffer to be received */ 1772 rx_ring->rx_buffer_info[ntc].skb = skb; 1773 rx_ring->rx_stats.non_eop_descs++; 1774 1775 return true; 1776 } 1777 1778 /** 1779 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail 1780 * @rx_ring: rx descriptor ring packet is being transacted on 1781 * @skb: pointer to current skb being adjusted 1782 * 1783 * This function is an ixgbe specific version of __pskb_pull_tail. The 1784 * main difference between this version and the original function is that 1785 * this function can make several assumptions about the state of things 1786 * that allow for significant optimizations versus the standard function. 1787 * As a result we can do things like drop a frag and maintain an accurate 1788 * truesize for the skb. 1789 */ 1790 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring, 1791 struct sk_buff *skb) 1792 { 1793 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0]; 1794 unsigned char *va; 1795 unsigned int pull_len; 1796 1797 /* 1798 * it is valid to use page_address instead of kmap since we are 1799 * working with pages allocated out of the lomem pool per 1800 * alloc_page(GFP_ATOMIC) 1801 */ 1802 va = skb_frag_address(frag); 1803 1804 /* 1805 * we need the header to contain the greater of either ETH_HLEN or 1806 * 60 bytes if the skb->len is less than 60 for skb_pad. 1807 */ 1808 pull_len = eth_get_headlen(va, IXGBE_RX_HDR_SIZE); 1809 1810 /* align pull length to size of long to optimize memcpy performance */ 1811 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long))); 1812 1813 /* update all of the pointers */ 1814 skb_frag_size_sub(frag, pull_len); 1815 frag->page_offset += pull_len; 1816 skb->data_len -= pull_len; 1817 skb->tail += pull_len; 1818 } 1819 1820 /** 1821 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB 1822 * @rx_ring: rx descriptor ring packet is being transacted on 1823 * @skb: pointer to current skb being updated 1824 * 1825 * This function provides a basic DMA sync up for the first fragment of an 1826 * skb. The reason for doing this is that the first fragment cannot be 1827 * unmapped until we have reached the end of packet descriptor for a buffer 1828 * chain. 1829 */ 1830 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring, 1831 struct sk_buff *skb) 1832 { 1833 /* if the page was released unmap it, else just sync our portion */ 1834 if (unlikely(IXGBE_CB(skb)->page_released)) { 1835 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma, 1836 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE); 1837 IXGBE_CB(skb)->page_released = false; 1838 } else { 1839 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0]; 1840 1841 dma_sync_single_range_for_cpu(rx_ring->dev, 1842 IXGBE_CB(skb)->dma, 1843 frag->page_offset, 1844 ixgbe_rx_bufsz(rx_ring), 1845 DMA_FROM_DEVICE); 1846 } 1847 IXGBE_CB(skb)->dma = 0; 1848 } 1849 1850 /** 1851 * ixgbe_cleanup_headers - Correct corrupted or empty headers 1852 * @rx_ring: rx descriptor ring packet is being transacted on 1853 * @rx_desc: pointer to the EOP Rx descriptor 1854 * @skb: pointer to current skb being fixed 1855 * 1856 * Check for corrupted packet headers caused by senders on the local L2 1857 * embedded NIC switch not setting up their Tx Descriptors right. These 1858 * should be very rare. 1859 * 1860 * Also address the case where we are pulling data in on pages only 1861 * and as such no data is present in the skb header. 1862 * 1863 * In addition if skb is not at least 60 bytes we need to pad it so that 1864 * it is large enough to qualify as a valid Ethernet frame. 1865 * 1866 * Returns true if an error was encountered and skb was freed. 1867 **/ 1868 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring, 1869 union ixgbe_adv_rx_desc *rx_desc, 1870 struct sk_buff *skb) 1871 { 1872 struct net_device *netdev = rx_ring->netdev; 1873 1874 /* verify that the packet does not have any known errors */ 1875 if (unlikely(ixgbe_test_staterr(rx_desc, 1876 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) && 1877 !(netdev->features & NETIF_F_RXALL))) { 1878 dev_kfree_skb_any(skb); 1879 return true; 1880 } 1881 1882 /* place header in linear portion of buffer */ 1883 if (skb_is_nonlinear(skb)) 1884 ixgbe_pull_tail(rx_ring, skb); 1885 1886 #ifdef IXGBE_FCOE 1887 /* do not attempt to pad FCoE Frames as this will disrupt DDP */ 1888 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) 1889 return false; 1890 1891 #endif 1892 /* if eth_skb_pad returns an error the skb was freed */ 1893 if (eth_skb_pad(skb)) 1894 return true; 1895 1896 return false; 1897 } 1898 1899 /** 1900 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring 1901 * @rx_ring: rx descriptor ring to store buffers on 1902 * @old_buff: donor buffer to have page reused 1903 * 1904 * Synchronizes page for reuse by the adapter 1905 **/ 1906 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring, 1907 struct ixgbe_rx_buffer *old_buff) 1908 { 1909 struct ixgbe_rx_buffer *new_buff; 1910 u16 nta = rx_ring->next_to_alloc; 1911 1912 new_buff = &rx_ring->rx_buffer_info[nta]; 1913 1914 /* update, and store next to alloc */ 1915 nta++; 1916 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0; 1917 1918 /* transfer page from old buffer to new buffer */ 1919 *new_buff = *old_buff; 1920 1921 /* sync the buffer for use by the device */ 1922 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma, 1923 new_buff->page_offset, 1924 ixgbe_rx_bufsz(rx_ring), 1925 DMA_FROM_DEVICE); 1926 } 1927 1928 static inline bool ixgbe_page_is_reserved(struct page *page) 1929 { 1930 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page); 1931 } 1932 1933 /** 1934 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff 1935 * @rx_ring: rx descriptor ring to transact packets on 1936 * @rx_buffer: buffer containing page to add 1937 * @rx_desc: descriptor containing length of buffer written by hardware 1938 * @skb: sk_buff to place the data into 1939 * 1940 * This function will add the data contained in rx_buffer->page to the skb. 1941 * This is done either through a direct copy if the data in the buffer is 1942 * less than the skb header size, otherwise it will just attach the page as 1943 * a frag to the skb. 1944 * 1945 * The function will then update the page offset if necessary and return 1946 * true if the buffer can be reused by the adapter. 1947 **/ 1948 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring, 1949 struct ixgbe_rx_buffer *rx_buffer, 1950 union ixgbe_adv_rx_desc *rx_desc, 1951 struct sk_buff *skb) 1952 { 1953 struct page *page = rx_buffer->page; 1954 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length); 1955 #if (PAGE_SIZE < 8192) 1956 unsigned int truesize = ixgbe_rx_bufsz(rx_ring); 1957 #else 1958 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES); 1959 unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) - 1960 ixgbe_rx_bufsz(rx_ring); 1961 #endif 1962 1963 if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) { 1964 unsigned char *va = page_address(page) + rx_buffer->page_offset; 1965 1966 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long))); 1967 1968 /* page is not reserved, we can reuse buffer as-is */ 1969 if (likely(!ixgbe_page_is_reserved(page))) 1970 return true; 1971 1972 /* this page cannot be reused so discard it */ 1973 __free_pages(page, ixgbe_rx_pg_order(rx_ring)); 1974 return false; 1975 } 1976 1977 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, 1978 rx_buffer->page_offset, size, truesize); 1979 1980 /* avoid re-using remote pages */ 1981 if (unlikely(ixgbe_page_is_reserved(page))) 1982 return false; 1983 1984 #if (PAGE_SIZE < 8192) 1985 /* if we are only owner of page we can reuse it */ 1986 if (unlikely(page_count(page) != 1)) 1987 return false; 1988 1989 /* flip page offset to other buffer */ 1990 rx_buffer->page_offset ^= truesize; 1991 #else 1992 /* move offset up to the next cache line */ 1993 rx_buffer->page_offset += truesize; 1994 1995 if (rx_buffer->page_offset > last_offset) 1996 return false; 1997 #endif 1998 1999 /* Even if we own the page, we are not allowed to use atomic_set() 2000 * This would break get_page_unless_zero() users. 2001 */ 2002 page_ref_inc(page); 2003 2004 return true; 2005 } 2006 2007 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring, 2008 union ixgbe_adv_rx_desc *rx_desc) 2009 { 2010 struct ixgbe_rx_buffer *rx_buffer; 2011 struct sk_buff *skb; 2012 struct page *page; 2013 2014 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean]; 2015 page = rx_buffer->page; 2016 prefetchw(page); 2017 2018 skb = rx_buffer->skb; 2019 2020 if (likely(!skb)) { 2021 void *page_addr = page_address(page) + 2022 rx_buffer->page_offset; 2023 2024 /* prefetch first cache line of first page */ 2025 prefetch(page_addr); 2026 #if L1_CACHE_BYTES < 128 2027 prefetch(page_addr + L1_CACHE_BYTES); 2028 #endif 2029 2030 /* allocate a skb to store the frags */ 2031 skb = napi_alloc_skb(&rx_ring->q_vector->napi, 2032 IXGBE_RX_HDR_SIZE); 2033 if (unlikely(!skb)) { 2034 rx_ring->rx_stats.alloc_rx_buff_failed++; 2035 return NULL; 2036 } 2037 2038 /* 2039 * we will be copying header into skb->data in 2040 * pskb_may_pull so it is in our interest to prefetch 2041 * it now to avoid a possible cache miss 2042 */ 2043 prefetchw(skb->data); 2044 2045 /* 2046 * Delay unmapping of the first packet. It carries the 2047 * header information, HW may still access the header 2048 * after the writeback. Only unmap it when EOP is 2049 * reached 2050 */ 2051 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))) 2052 goto dma_sync; 2053 2054 IXGBE_CB(skb)->dma = rx_buffer->dma; 2055 } else { 2056 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) 2057 ixgbe_dma_sync_frag(rx_ring, skb); 2058 2059 dma_sync: 2060 /* we are reusing so sync this buffer for CPU use */ 2061 dma_sync_single_range_for_cpu(rx_ring->dev, 2062 rx_buffer->dma, 2063 rx_buffer->page_offset, 2064 ixgbe_rx_bufsz(rx_ring), 2065 DMA_FROM_DEVICE); 2066 2067 rx_buffer->skb = NULL; 2068 } 2069 2070 /* pull page into skb */ 2071 if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) { 2072 /* hand second half of page back to the ring */ 2073 ixgbe_reuse_rx_page(rx_ring, rx_buffer); 2074 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) { 2075 /* the page has been released from the ring */ 2076 IXGBE_CB(skb)->page_released = true; 2077 } else { 2078 /* we are not reusing the buffer so unmap it */ 2079 dma_unmap_page(rx_ring->dev, rx_buffer->dma, 2080 ixgbe_rx_pg_size(rx_ring), 2081 DMA_FROM_DEVICE); 2082 } 2083 2084 /* clear contents of buffer_info */ 2085 rx_buffer->page = NULL; 2086 2087 return skb; 2088 } 2089 2090 /** 2091 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf 2092 * @q_vector: structure containing interrupt and ring information 2093 * @rx_ring: rx descriptor ring to transact packets on 2094 * @budget: Total limit on number of packets to process 2095 * 2096 * This function provides a "bounce buffer" approach to Rx interrupt 2097 * processing. The advantage to this is that on systems that have 2098 * expensive overhead for IOMMU access this provides a means of avoiding 2099 * it by maintaining the mapping of the page to the syste. 2100 * 2101 * Returns amount of work completed 2102 **/ 2103 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector, 2104 struct ixgbe_ring *rx_ring, 2105 const int budget) 2106 { 2107 unsigned int total_rx_bytes = 0, total_rx_packets = 0; 2108 #ifdef IXGBE_FCOE 2109 struct ixgbe_adapter *adapter = q_vector->adapter; 2110 int ddp_bytes; 2111 unsigned int mss = 0; 2112 #endif /* IXGBE_FCOE */ 2113 u16 cleaned_count = ixgbe_desc_unused(rx_ring); 2114 2115 while (likely(total_rx_packets < budget)) { 2116 union ixgbe_adv_rx_desc *rx_desc; 2117 struct sk_buff *skb; 2118 2119 /* return some buffers to hardware, one at a time is too slow */ 2120 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) { 2121 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count); 2122 cleaned_count = 0; 2123 } 2124 2125 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean); 2126 2127 if (!rx_desc->wb.upper.status_error) 2128 break; 2129 2130 /* This memory barrier is needed to keep us from reading 2131 * any other fields out of the rx_desc until we know the 2132 * descriptor has been written back 2133 */ 2134 dma_rmb(); 2135 2136 /* retrieve a buffer from the ring */ 2137 skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc); 2138 2139 /* exit if we failed to retrieve a buffer */ 2140 if (!skb) 2141 break; 2142 2143 cleaned_count++; 2144 2145 /* place incomplete frames back on ring for completion */ 2146 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb)) 2147 continue; 2148 2149 /* verify the packet layout is correct */ 2150 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb)) 2151 continue; 2152 2153 /* probably a little skewed due to removing CRC */ 2154 total_rx_bytes += skb->len; 2155 2156 /* populate checksum, timestamp, VLAN, and protocol */ 2157 ixgbe_process_skb_fields(rx_ring, rx_desc, skb); 2158 2159 #ifdef IXGBE_FCOE 2160 /* if ddp, not passing to ULD unless for FCP_RSP or error */ 2161 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) { 2162 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb); 2163 /* include DDPed FCoE data */ 2164 if (ddp_bytes > 0) { 2165 if (!mss) { 2166 mss = rx_ring->netdev->mtu - 2167 sizeof(struct fcoe_hdr) - 2168 sizeof(struct fc_frame_header) - 2169 sizeof(struct fcoe_crc_eof); 2170 if (mss > 512) 2171 mss &= ~511; 2172 } 2173 total_rx_bytes += ddp_bytes; 2174 total_rx_packets += DIV_ROUND_UP(ddp_bytes, 2175 mss); 2176 } 2177 if (!ddp_bytes) { 2178 dev_kfree_skb_any(skb); 2179 continue; 2180 } 2181 } 2182 2183 #endif /* IXGBE_FCOE */ 2184 ixgbe_rx_skb(q_vector, skb); 2185 2186 /* update budget accounting */ 2187 total_rx_packets++; 2188 } 2189 2190 u64_stats_update_begin(&rx_ring->syncp); 2191 rx_ring->stats.packets += total_rx_packets; 2192 rx_ring->stats.bytes += total_rx_bytes; 2193 u64_stats_update_end(&rx_ring->syncp); 2194 q_vector->rx.total_packets += total_rx_packets; 2195 q_vector->rx.total_bytes += total_rx_bytes; 2196 2197 return total_rx_packets; 2198 } 2199 2200 #ifdef CONFIG_NET_RX_BUSY_POLL 2201 /* must be called with local_bh_disable()d */ 2202 static int ixgbe_low_latency_recv(struct napi_struct *napi) 2203 { 2204 struct ixgbe_q_vector *q_vector = 2205 container_of(napi, struct ixgbe_q_vector, napi); 2206 struct ixgbe_adapter *adapter = q_vector->adapter; 2207 struct ixgbe_ring *ring; 2208 int found = 0; 2209 2210 if (test_bit(__IXGBE_DOWN, &adapter->state)) 2211 return LL_FLUSH_FAILED; 2212 2213 if (!ixgbe_qv_lock_poll(q_vector)) 2214 return LL_FLUSH_BUSY; 2215 2216 ixgbe_for_each_ring(ring, q_vector->rx) { 2217 found = ixgbe_clean_rx_irq(q_vector, ring, 4); 2218 #ifdef BP_EXTENDED_STATS 2219 if (found) 2220 ring->stats.cleaned += found; 2221 else 2222 ring->stats.misses++; 2223 #endif 2224 if (found) 2225 break; 2226 } 2227 2228 ixgbe_qv_unlock_poll(q_vector); 2229 2230 return found; 2231 } 2232 #endif /* CONFIG_NET_RX_BUSY_POLL */ 2233 2234 /** 2235 * ixgbe_configure_msix - Configure MSI-X hardware 2236 * @adapter: board private structure 2237 * 2238 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X 2239 * interrupts. 2240 **/ 2241 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter) 2242 { 2243 struct ixgbe_q_vector *q_vector; 2244 int v_idx; 2245 u32 mask; 2246 2247 /* Populate MSIX to EITR Select */ 2248 if (adapter->num_vfs > 32) { 2249 u32 eitrsel = BIT(adapter->num_vfs - 32) - 1; 2250 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel); 2251 } 2252 2253 /* 2254 * Populate the IVAR table and set the ITR values to the 2255 * corresponding register. 2256 */ 2257 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) { 2258 struct ixgbe_ring *ring; 2259 q_vector = adapter->q_vector[v_idx]; 2260 2261 ixgbe_for_each_ring(ring, q_vector->rx) 2262 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx); 2263 2264 ixgbe_for_each_ring(ring, q_vector->tx) 2265 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx); 2266 2267 ixgbe_write_eitr(q_vector); 2268 } 2269 2270 switch (adapter->hw.mac.type) { 2271 case ixgbe_mac_82598EB: 2272 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX, 2273 v_idx); 2274 break; 2275 case ixgbe_mac_82599EB: 2276 case ixgbe_mac_X540: 2277 case ixgbe_mac_X550: 2278 case ixgbe_mac_X550EM_x: 2279 case ixgbe_mac_x550em_a: 2280 ixgbe_set_ivar(adapter, -1, 1, v_idx); 2281 break; 2282 default: 2283 break; 2284 } 2285 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950); 2286 2287 /* set up to autoclear timer, and the vectors */ 2288 mask = IXGBE_EIMS_ENABLE_MASK; 2289 mask &= ~(IXGBE_EIMS_OTHER | 2290 IXGBE_EIMS_MAILBOX | 2291 IXGBE_EIMS_LSC); 2292 2293 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask); 2294 } 2295 2296 enum latency_range { 2297 lowest_latency = 0, 2298 low_latency = 1, 2299 bulk_latency = 2, 2300 latency_invalid = 255 2301 }; 2302 2303 /** 2304 * ixgbe_update_itr - update the dynamic ITR value based on statistics 2305 * @q_vector: structure containing interrupt and ring information 2306 * @ring_container: structure containing ring performance data 2307 * 2308 * Stores a new ITR value based on packets and byte 2309 * counts during the last interrupt. The advantage of per interrupt 2310 * computation is faster updates and more accurate ITR for the current 2311 * traffic pattern. Constants in this function were computed 2312 * based on theoretical maximum wire speed and thresholds were set based 2313 * on testing data as well as attempting to minimize response time 2314 * while increasing bulk throughput. 2315 * this functionality is controlled by the InterruptThrottleRate module 2316 * parameter (see ixgbe_param.c) 2317 **/ 2318 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector, 2319 struct ixgbe_ring_container *ring_container) 2320 { 2321 int bytes = ring_container->total_bytes; 2322 int packets = ring_container->total_packets; 2323 u32 timepassed_us; 2324 u64 bytes_perint; 2325 u8 itr_setting = ring_container->itr; 2326 2327 if (packets == 0) 2328 return; 2329 2330 /* simple throttlerate management 2331 * 0-10MB/s lowest (100000 ints/s) 2332 * 10-20MB/s low (20000 ints/s) 2333 * 20-1249MB/s bulk (12000 ints/s) 2334 */ 2335 /* what was last interrupt timeslice? */ 2336 timepassed_us = q_vector->itr >> 2; 2337 if (timepassed_us == 0) 2338 return; 2339 2340 bytes_perint = bytes / timepassed_us; /* bytes/usec */ 2341 2342 switch (itr_setting) { 2343 case lowest_latency: 2344 if (bytes_perint > 10) 2345 itr_setting = low_latency; 2346 break; 2347 case low_latency: 2348 if (bytes_perint > 20) 2349 itr_setting = bulk_latency; 2350 else if (bytes_perint <= 10) 2351 itr_setting = lowest_latency; 2352 break; 2353 case bulk_latency: 2354 if (bytes_perint <= 20) 2355 itr_setting = low_latency; 2356 break; 2357 } 2358 2359 /* clear work counters since we have the values we need */ 2360 ring_container->total_bytes = 0; 2361 ring_container->total_packets = 0; 2362 2363 /* write updated itr to ring container */ 2364 ring_container->itr = itr_setting; 2365 } 2366 2367 /** 2368 * ixgbe_write_eitr - write EITR register in hardware specific way 2369 * @q_vector: structure containing interrupt and ring information 2370 * 2371 * This function is made to be called by ethtool and by the driver 2372 * when it needs to update EITR registers at runtime. Hardware 2373 * specific quirks/differences are taken care of here. 2374 */ 2375 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector) 2376 { 2377 struct ixgbe_adapter *adapter = q_vector->adapter; 2378 struct ixgbe_hw *hw = &adapter->hw; 2379 int v_idx = q_vector->v_idx; 2380 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR; 2381 2382 switch (adapter->hw.mac.type) { 2383 case ixgbe_mac_82598EB: 2384 /* must write high and low 16 bits to reset counter */ 2385 itr_reg |= (itr_reg << 16); 2386 break; 2387 case ixgbe_mac_82599EB: 2388 case ixgbe_mac_X540: 2389 case ixgbe_mac_X550: 2390 case ixgbe_mac_X550EM_x: 2391 case ixgbe_mac_x550em_a: 2392 /* 2393 * set the WDIS bit to not clear the timer bits and cause an 2394 * immediate assertion of the interrupt 2395 */ 2396 itr_reg |= IXGBE_EITR_CNT_WDIS; 2397 break; 2398 default: 2399 break; 2400 } 2401 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg); 2402 } 2403 2404 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector) 2405 { 2406 u32 new_itr = q_vector->itr; 2407 u8 current_itr; 2408 2409 ixgbe_update_itr(q_vector, &q_vector->tx); 2410 ixgbe_update_itr(q_vector, &q_vector->rx); 2411 2412 current_itr = max(q_vector->rx.itr, q_vector->tx.itr); 2413 2414 switch (current_itr) { 2415 /* counts and packets in update_itr are dependent on these numbers */ 2416 case lowest_latency: 2417 new_itr = IXGBE_100K_ITR; 2418 break; 2419 case low_latency: 2420 new_itr = IXGBE_20K_ITR; 2421 break; 2422 case bulk_latency: 2423 new_itr = IXGBE_12K_ITR; 2424 break; 2425 default: 2426 break; 2427 } 2428 2429 if (new_itr != q_vector->itr) { 2430 /* do an exponential smoothing */ 2431 new_itr = (10 * new_itr * q_vector->itr) / 2432 ((9 * new_itr) + q_vector->itr); 2433 2434 /* save the algorithm value here */ 2435 q_vector->itr = new_itr; 2436 2437 ixgbe_write_eitr(q_vector); 2438 } 2439 } 2440 2441 /** 2442 * ixgbe_check_overtemp_subtask - check for over temperature 2443 * @adapter: pointer to adapter 2444 **/ 2445 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter) 2446 { 2447 struct ixgbe_hw *hw = &adapter->hw; 2448 u32 eicr = adapter->interrupt_event; 2449 2450 if (test_bit(__IXGBE_DOWN, &adapter->state)) 2451 return; 2452 2453 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) && 2454 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT)) 2455 return; 2456 2457 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT; 2458 2459 switch (hw->device_id) { 2460 case IXGBE_DEV_ID_82599_T3_LOM: 2461 /* 2462 * Since the warning interrupt is for both ports 2463 * we don't have to check if: 2464 * - This interrupt wasn't for our port. 2465 * - We may have missed the interrupt so always have to 2466 * check if we got a LSC 2467 */ 2468 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) && 2469 !(eicr & IXGBE_EICR_LSC)) 2470 return; 2471 2472 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) { 2473 u32 speed; 2474 bool link_up = false; 2475 2476 hw->mac.ops.check_link(hw, &speed, &link_up, false); 2477 2478 if (link_up) 2479 return; 2480 } 2481 2482 /* Check if this is not due to overtemp */ 2483 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP) 2484 return; 2485 2486 break; 2487 default: 2488 if (adapter->hw.mac.type >= ixgbe_mac_X540) 2489 return; 2490 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw))) 2491 return; 2492 break; 2493 } 2494 e_crit(drv, "%s\n", ixgbe_overheat_msg); 2495 2496 adapter->interrupt_event = 0; 2497 } 2498 2499 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr) 2500 { 2501 struct ixgbe_hw *hw = &adapter->hw; 2502 2503 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) && 2504 (eicr & IXGBE_EICR_GPI_SDP1(hw))) { 2505 e_crit(probe, "Fan has stopped, replace the adapter\n"); 2506 /* write to clear the interrupt */ 2507 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw)); 2508 } 2509 } 2510 2511 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr) 2512 { 2513 struct ixgbe_hw *hw = &adapter->hw; 2514 2515 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)) 2516 return; 2517 2518 switch (adapter->hw.mac.type) { 2519 case ixgbe_mac_82599EB: 2520 /* 2521 * Need to check link state so complete overtemp check 2522 * on service task 2523 */ 2524 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) || 2525 (eicr & IXGBE_EICR_LSC)) && 2526 (!test_bit(__IXGBE_DOWN, &adapter->state))) { 2527 adapter->interrupt_event = eicr; 2528 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT; 2529 ixgbe_service_event_schedule(adapter); 2530 return; 2531 } 2532 return; 2533 case ixgbe_mac_X540: 2534 if (!(eicr & IXGBE_EICR_TS)) 2535 return; 2536 break; 2537 default: 2538 return; 2539 } 2540 2541 e_crit(drv, "%s\n", ixgbe_overheat_msg); 2542 } 2543 2544 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw) 2545 { 2546 switch (hw->mac.type) { 2547 case ixgbe_mac_82598EB: 2548 if (hw->phy.type == ixgbe_phy_nl) 2549 return true; 2550 return false; 2551 case ixgbe_mac_82599EB: 2552 case ixgbe_mac_X550EM_x: 2553 case ixgbe_mac_x550em_a: 2554 switch (hw->mac.ops.get_media_type(hw)) { 2555 case ixgbe_media_type_fiber: 2556 case ixgbe_media_type_fiber_qsfp: 2557 return true; 2558 default: 2559 return false; 2560 } 2561 default: 2562 return false; 2563 } 2564 } 2565 2566 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr) 2567 { 2568 struct ixgbe_hw *hw = &adapter->hw; 2569 u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw); 2570 2571 if (!ixgbe_is_sfp(hw)) 2572 return; 2573 2574 /* Later MAC's use different SDP */ 2575 if (hw->mac.type >= ixgbe_mac_X540) 2576 eicr_mask = IXGBE_EICR_GPI_SDP0_X540; 2577 2578 if (eicr & eicr_mask) { 2579 /* Clear the interrupt */ 2580 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask); 2581 if (!test_bit(__IXGBE_DOWN, &adapter->state)) { 2582 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET; 2583 adapter->sfp_poll_time = 0; 2584 ixgbe_service_event_schedule(adapter); 2585 } 2586 } 2587 2588 if (adapter->hw.mac.type == ixgbe_mac_82599EB && 2589 (eicr & IXGBE_EICR_GPI_SDP1(hw))) { 2590 /* Clear the interrupt */ 2591 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw)); 2592 if (!test_bit(__IXGBE_DOWN, &adapter->state)) { 2593 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG; 2594 ixgbe_service_event_schedule(adapter); 2595 } 2596 } 2597 } 2598 2599 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter) 2600 { 2601 struct ixgbe_hw *hw = &adapter->hw; 2602 2603 adapter->lsc_int++; 2604 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE; 2605 adapter->link_check_timeout = jiffies; 2606 if (!test_bit(__IXGBE_DOWN, &adapter->state)) { 2607 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC); 2608 IXGBE_WRITE_FLUSH(hw); 2609 ixgbe_service_event_schedule(adapter); 2610 } 2611 } 2612 2613 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter, 2614 u64 qmask) 2615 { 2616 u32 mask; 2617 struct ixgbe_hw *hw = &adapter->hw; 2618 2619 switch (hw->mac.type) { 2620 case ixgbe_mac_82598EB: 2621 mask = (IXGBE_EIMS_RTX_QUEUE & qmask); 2622 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask); 2623 break; 2624 case ixgbe_mac_82599EB: 2625 case ixgbe_mac_X540: 2626 case ixgbe_mac_X550: 2627 case ixgbe_mac_X550EM_x: 2628 case ixgbe_mac_x550em_a: 2629 mask = (qmask & 0xFFFFFFFF); 2630 if (mask) 2631 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask); 2632 mask = (qmask >> 32); 2633 if (mask) 2634 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask); 2635 break; 2636 default: 2637 break; 2638 } 2639 /* skip the flush */ 2640 } 2641 2642 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter, 2643 u64 qmask) 2644 { 2645 u32 mask; 2646 struct ixgbe_hw *hw = &adapter->hw; 2647 2648 switch (hw->mac.type) { 2649 case ixgbe_mac_82598EB: 2650 mask = (IXGBE_EIMS_RTX_QUEUE & qmask); 2651 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask); 2652 break; 2653 case ixgbe_mac_82599EB: 2654 case ixgbe_mac_X540: 2655 case ixgbe_mac_X550: 2656 case ixgbe_mac_X550EM_x: 2657 case ixgbe_mac_x550em_a: 2658 mask = (qmask & 0xFFFFFFFF); 2659 if (mask) 2660 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask); 2661 mask = (qmask >> 32); 2662 if (mask) 2663 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask); 2664 break; 2665 default: 2666 break; 2667 } 2668 /* skip the flush */ 2669 } 2670 2671 /** 2672 * ixgbe_irq_enable - Enable default interrupt generation settings 2673 * @adapter: board private structure 2674 **/ 2675 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues, 2676 bool flush) 2677 { 2678 struct ixgbe_hw *hw = &adapter->hw; 2679 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE); 2680 2681 /* don't reenable LSC while waiting for link */ 2682 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) 2683 mask &= ~IXGBE_EIMS_LSC; 2684 2685 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) 2686 switch (adapter->hw.mac.type) { 2687 case ixgbe_mac_82599EB: 2688 mask |= IXGBE_EIMS_GPI_SDP0(hw); 2689 break; 2690 case ixgbe_mac_X540: 2691 case ixgbe_mac_X550: 2692 case ixgbe_mac_X550EM_x: 2693 case ixgbe_mac_x550em_a: 2694 mask |= IXGBE_EIMS_TS; 2695 break; 2696 default: 2697 break; 2698 } 2699 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) 2700 mask |= IXGBE_EIMS_GPI_SDP1(hw); 2701 switch (adapter->hw.mac.type) { 2702 case ixgbe_mac_82599EB: 2703 mask |= IXGBE_EIMS_GPI_SDP1(hw); 2704 mask |= IXGBE_EIMS_GPI_SDP2(hw); 2705 /* fall through */ 2706 case ixgbe_mac_X540: 2707 case ixgbe_mac_X550: 2708 case ixgbe_mac_X550EM_x: 2709 case ixgbe_mac_x550em_a: 2710 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP || 2711 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP || 2712 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N) 2713 mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw); 2714 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t) 2715 mask |= IXGBE_EICR_GPI_SDP0_X540; 2716 mask |= IXGBE_EIMS_ECC; 2717 mask |= IXGBE_EIMS_MAILBOX; 2718 break; 2719 default: 2720 break; 2721 } 2722 2723 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) && 2724 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT)) 2725 mask |= IXGBE_EIMS_FLOW_DIR; 2726 2727 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask); 2728 if (queues) 2729 ixgbe_irq_enable_queues(adapter, ~0); 2730 if (flush) 2731 IXGBE_WRITE_FLUSH(&adapter->hw); 2732 } 2733 2734 static irqreturn_t ixgbe_msix_other(int irq, void *data) 2735 { 2736 struct ixgbe_adapter *adapter = data; 2737 struct ixgbe_hw *hw = &adapter->hw; 2738 u32 eicr; 2739 2740 /* 2741 * Workaround for Silicon errata. Use clear-by-write instead 2742 * of clear-by-read. Reading with EICS will return the 2743 * interrupt causes without clearing, which later be done 2744 * with the write to EICR. 2745 */ 2746 eicr = IXGBE_READ_REG(hw, IXGBE_EICS); 2747 2748 /* The lower 16bits of the EICR register are for the queue interrupts 2749 * which should be masked here in order to not accidentally clear them if 2750 * the bits are high when ixgbe_msix_other is called. There is a race 2751 * condition otherwise which results in possible performance loss 2752 * especially if the ixgbe_msix_other interrupt is triggering 2753 * consistently (as it would when PPS is turned on for the X540 device) 2754 */ 2755 eicr &= 0xFFFF0000; 2756 2757 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr); 2758 2759 if (eicr & IXGBE_EICR_LSC) 2760 ixgbe_check_lsc(adapter); 2761 2762 if (eicr & IXGBE_EICR_MAILBOX) 2763 ixgbe_msg_task(adapter); 2764 2765 switch (hw->mac.type) { 2766 case ixgbe_mac_82599EB: 2767 case ixgbe_mac_X540: 2768 case ixgbe_mac_X550: 2769 case ixgbe_mac_X550EM_x: 2770 case ixgbe_mac_x550em_a: 2771 if (hw->phy.type == ixgbe_phy_x550em_ext_t && 2772 (eicr & IXGBE_EICR_GPI_SDP0_X540)) { 2773 adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT; 2774 ixgbe_service_event_schedule(adapter); 2775 IXGBE_WRITE_REG(hw, IXGBE_EICR, 2776 IXGBE_EICR_GPI_SDP0_X540); 2777 } 2778 if (eicr & IXGBE_EICR_ECC) { 2779 e_info(link, "Received ECC Err, initiating reset\n"); 2780 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED; 2781 ixgbe_service_event_schedule(adapter); 2782 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC); 2783 } 2784 /* Handle Flow Director Full threshold interrupt */ 2785 if (eicr & IXGBE_EICR_FLOW_DIR) { 2786 int reinit_count = 0; 2787 int i; 2788 for (i = 0; i < adapter->num_tx_queues; i++) { 2789 struct ixgbe_ring *ring = adapter->tx_ring[i]; 2790 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE, 2791 &ring->state)) 2792 reinit_count++; 2793 } 2794 if (reinit_count) { 2795 /* no more flow director interrupts until after init */ 2796 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR); 2797 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT; 2798 ixgbe_service_event_schedule(adapter); 2799 } 2800 } 2801 ixgbe_check_sfp_event(adapter, eicr); 2802 ixgbe_check_overtemp_event(adapter, eicr); 2803 break; 2804 default: 2805 break; 2806 } 2807 2808 ixgbe_check_fan_failure(adapter, eicr); 2809 2810 if (unlikely(eicr & IXGBE_EICR_TIMESYNC)) 2811 ixgbe_ptp_check_pps_event(adapter); 2812 2813 /* re-enable the original interrupt state, no lsc, no queues */ 2814 if (!test_bit(__IXGBE_DOWN, &adapter->state)) 2815 ixgbe_irq_enable(adapter, false, false); 2816 2817 return IRQ_HANDLED; 2818 } 2819 2820 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data) 2821 { 2822 struct ixgbe_q_vector *q_vector = data; 2823 2824 /* EIAM disabled interrupts (on this vector) for us */ 2825 2826 if (q_vector->rx.ring || q_vector->tx.ring) 2827 napi_schedule_irqoff(&q_vector->napi); 2828 2829 return IRQ_HANDLED; 2830 } 2831 2832 /** 2833 * ixgbe_poll - NAPI Rx polling callback 2834 * @napi: structure for representing this polling device 2835 * @budget: how many packets driver is allowed to clean 2836 * 2837 * This function is used for legacy and MSI, NAPI mode 2838 **/ 2839 int ixgbe_poll(struct napi_struct *napi, int budget) 2840 { 2841 struct ixgbe_q_vector *q_vector = 2842 container_of(napi, struct ixgbe_q_vector, napi); 2843 struct ixgbe_adapter *adapter = q_vector->adapter; 2844 struct ixgbe_ring *ring; 2845 int per_ring_budget, work_done = 0; 2846 bool clean_complete = true; 2847 2848 #ifdef CONFIG_IXGBE_DCA 2849 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) 2850 ixgbe_update_dca(q_vector); 2851 #endif 2852 2853 ixgbe_for_each_ring(ring, q_vector->tx) { 2854 if (!ixgbe_clean_tx_irq(q_vector, ring, budget)) 2855 clean_complete = false; 2856 } 2857 2858 /* Exit if we are called by netpoll or busy polling is active */ 2859 if ((budget <= 0) || !ixgbe_qv_lock_napi(q_vector)) 2860 return budget; 2861 2862 /* attempt to distribute budget to each queue fairly, but don't allow 2863 * the budget to go below 1 because we'll exit polling */ 2864 if (q_vector->rx.count > 1) 2865 per_ring_budget = max(budget/q_vector->rx.count, 1); 2866 else 2867 per_ring_budget = budget; 2868 2869 ixgbe_for_each_ring(ring, q_vector->rx) { 2870 int cleaned = ixgbe_clean_rx_irq(q_vector, ring, 2871 per_ring_budget); 2872 2873 work_done += cleaned; 2874 if (cleaned >= per_ring_budget) 2875 clean_complete = false; 2876 } 2877 2878 ixgbe_qv_unlock_napi(q_vector); 2879 /* If all work not completed, return budget and keep polling */ 2880 if (!clean_complete) 2881 return budget; 2882 2883 /* all work done, exit the polling mode */ 2884 napi_complete_done(napi, work_done); 2885 if (adapter->rx_itr_setting & 1) 2886 ixgbe_set_itr(q_vector); 2887 if (!test_bit(__IXGBE_DOWN, &adapter->state)) 2888 ixgbe_irq_enable_queues(adapter, BIT_ULL(q_vector->v_idx)); 2889 2890 return 0; 2891 } 2892 2893 /** 2894 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts 2895 * @adapter: board private structure 2896 * 2897 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests 2898 * interrupts from the kernel. 2899 **/ 2900 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter) 2901 { 2902 struct net_device *netdev = adapter->netdev; 2903 int vector, err; 2904 int ri = 0, ti = 0; 2905 2906 for (vector = 0; vector < adapter->num_q_vectors; vector++) { 2907 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector]; 2908 struct msix_entry *entry = &adapter->msix_entries[vector]; 2909 2910 if (q_vector->tx.ring && q_vector->rx.ring) { 2911 snprintf(q_vector->name, sizeof(q_vector->name) - 1, 2912 "%s-%s-%d", netdev->name, "TxRx", ri++); 2913 ti++; 2914 } else if (q_vector->rx.ring) { 2915 snprintf(q_vector->name, sizeof(q_vector->name) - 1, 2916 "%s-%s-%d", netdev->name, "rx", ri++); 2917 } else if (q_vector->tx.ring) { 2918 snprintf(q_vector->name, sizeof(q_vector->name) - 1, 2919 "%s-%s-%d", netdev->name, "tx", ti++); 2920 } else { 2921 /* skip this unused q_vector */ 2922 continue; 2923 } 2924 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0, 2925 q_vector->name, q_vector); 2926 if (err) { 2927 e_err(probe, "request_irq failed for MSIX interrupt " 2928 "Error: %d\n", err); 2929 goto free_queue_irqs; 2930 } 2931 /* If Flow Director is enabled, set interrupt affinity */ 2932 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) { 2933 /* assign the mask for this irq */ 2934 irq_set_affinity_hint(entry->vector, 2935 &q_vector->affinity_mask); 2936 } 2937 } 2938 2939 err = request_irq(adapter->msix_entries[vector].vector, 2940 ixgbe_msix_other, 0, netdev->name, adapter); 2941 if (err) { 2942 e_err(probe, "request_irq for msix_other failed: %d\n", err); 2943 goto free_queue_irqs; 2944 } 2945 2946 return 0; 2947 2948 free_queue_irqs: 2949 while (vector) { 2950 vector--; 2951 irq_set_affinity_hint(adapter->msix_entries[vector].vector, 2952 NULL); 2953 free_irq(adapter->msix_entries[vector].vector, 2954 adapter->q_vector[vector]); 2955 } 2956 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED; 2957 pci_disable_msix(adapter->pdev); 2958 kfree(adapter->msix_entries); 2959 adapter->msix_entries = NULL; 2960 return err; 2961 } 2962 2963 /** 2964 * ixgbe_intr - legacy mode Interrupt Handler 2965 * @irq: interrupt number 2966 * @data: pointer to a network interface device structure 2967 **/ 2968 static irqreturn_t ixgbe_intr(int irq, void *data) 2969 { 2970 struct ixgbe_adapter *adapter = data; 2971 struct ixgbe_hw *hw = &adapter->hw; 2972 struct ixgbe_q_vector *q_vector = adapter->q_vector[0]; 2973 u32 eicr; 2974 2975 /* 2976 * Workaround for silicon errata #26 on 82598. Mask the interrupt 2977 * before the read of EICR. 2978 */ 2979 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK); 2980 2981 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read 2982 * therefore no explicit interrupt disable is necessary */ 2983 eicr = IXGBE_READ_REG(hw, IXGBE_EICR); 2984 if (!eicr) { 2985 /* 2986 * shared interrupt alert! 2987 * make sure interrupts are enabled because the read will 2988 * have disabled interrupts due to EIAM 2989 * finish the workaround of silicon errata on 82598. Unmask 2990 * the interrupt that we masked before the EICR read. 2991 */ 2992 if (!test_bit(__IXGBE_DOWN, &adapter->state)) 2993 ixgbe_irq_enable(adapter, true, true); 2994 return IRQ_NONE; /* Not our interrupt */ 2995 } 2996 2997 if (eicr & IXGBE_EICR_LSC) 2998 ixgbe_check_lsc(adapter); 2999 3000 switch (hw->mac.type) { 3001 case ixgbe_mac_82599EB: 3002 ixgbe_check_sfp_event(adapter, eicr); 3003 /* Fall through */ 3004 case ixgbe_mac_X540: 3005 case ixgbe_mac_X550: 3006 case ixgbe_mac_X550EM_x: 3007 case ixgbe_mac_x550em_a: 3008 if (eicr & IXGBE_EICR_ECC) { 3009 e_info(link, "Received ECC Err, initiating reset\n"); 3010 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED; 3011 ixgbe_service_event_schedule(adapter); 3012 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC); 3013 } 3014 ixgbe_check_overtemp_event(adapter, eicr); 3015 break; 3016 default: 3017 break; 3018 } 3019 3020 ixgbe_check_fan_failure(adapter, eicr); 3021 if (unlikely(eicr & IXGBE_EICR_TIMESYNC)) 3022 ixgbe_ptp_check_pps_event(adapter); 3023 3024 /* would disable interrupts here but EIAM disabled it */ 3025 napi_schedule_irqoff(&q_vector->napi); 3026 3027 /* 3028 * re-enable link(maybe) and non-queue interrupts, no flush. 3029 * ixgbe_poll will re-enable the queue interrupts 3030 */ 3031 if (!test_bit(__IXGBE_DOWN, &adapter->state)) 3032 ixgbe_irq_enable(adapter, false, false); 3033 3034 return IRQ_HANDLED; 3035 } 3036 3037 /** 3038 * ixgbe_request_irq - initialize interrupts 3039 * @adapter: board private structure 3040 * 3041 * Attempts to configure interrupts using the best available 3042 * capabilities of the hardware and kernel. 3043 **/ 3044 static int ixgbe_request_irq(struct ixgbe_adapter *adapter) 3045 { 3046 struct net_device *netdev = adapter->netdev; 3047 int err; 3048 3049 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) 3050 err = ixgbe_request_msix_irqs(adapter); 3051 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) 3052 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0, 3053 netdev->name, adapter); 3054 else 3055 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED, 3056 netdev->name, adapter); 3057 3058 if (err) 3059 e_err(probe, "request_irq failed, Error %d\n", err); 3060 3061 return err; 3062 } 3063 3064 static void ixgbe_free_irq(struct ixgbe_adapter *adapter) 3065 { 3066 int vector; 3067 3068 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) { 3069 free_irq(adapter->pdev->irq, adapter); 3070 return; 3071 } 3072 3073 for (vector = 0; vector < adapter->num_q_vectors; vector++) { 3074 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector]; 3075 struct msix_entry *entry = &adapter->msix_entries[vector]; 3076 3077 /* free only the irqs that were actually requested */ 3078 if (!q_vector->rx.ring && !q_vector->tx.ring) 3079 continue; 3080 3081 /* clear the affinity_mask in the IRQ descriptor */ 3082 irq_set_affinity_hint(entry->vector, NULL); 3083 3084 free_irq(entry->vector, q_vector); 3085 } 3086 3087 free_irq(adapter->msix_entries[vector++].vector, adapter); 3088 } 3089 3090 /** 3091 * ixgbe_irq_disable - Mask off interrupt generation on the NIC 3092 * @adapter: board private structure 3093 **/ 3094 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter) 3095 { 3096 switch (adapter->hw.mac.type) { 3097 case ixgbe_mac_82598EB: 3098 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0); 3099 break; 3100 case ixgbe_mac_82599EB: 3101 case ixgbe_mac_X540: 3102 case ixgbe_mac_X550: 3103 case ixgbe_mac_X550EM_x: 3104 case ixgbe_mac_x550em_a: 3105 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000); 3106 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0); 3107 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0); 3108 break; 3109 default: 3110 break; 3111 } 3112 IXGBE_WRITE_FLUSH(&adapter->hw); 3113 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) { 3114 int vector; 3115 3116 for (vector = 0; vector < adapter->num_q_vectors; vector++) 3117 synchronize_irq(adapter->msix_entries[vector].vector); 3118 3119 synchronize_irq(adapter->msix_entries[vector++].vector); 3120 } else { 3121 synchronize_irq(adapter->pdev->irq); 3122 } 3123 } 3124 3125 /** 3126 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts 3127 * 3128 **/ 3129 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter) 3130 { 3131 struct ixgbe_q_vector *q_vector = adapter->q_vector[0]; 3132 3133 ixgbe_write_eitr(q_vector); 3134 3135 ixgbe_set_ivar(adapter, 0, 0, 0); 3136 ixgbe_set_ivar(adapter, 1, 0, 0); 3137 3138 e_info(hw, "Legacy interrupt IVAR setup done\n"); 3139 } 3140 3141 /** 3142 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset 3143 * @adapter: board private structure 3144 * @ring: structure containing ring specific data 3145 * 3146 * Configure the Tx descriptor ring after a reset. 3147 **/ 3148 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter, 3149 struct ixgbe_ring *ring) 3150 { 3151 struct ixgbe_hw *hw = &adapter->hw; 3152 u64 tdba = ring->dma; 3153 int wait_loop = 10; 3154 u32 txdctl = IXGBE_TXDCTL_ENABLE; 3155 u8 reg_idx = ring->reg_idx; 3156 3157 /* disable queue to avoid issues while updating state */ 3158 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0); 3159 IXGBE_WRITE_FLUSH(hw); 3160 3161 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx), 3162 (tdba & DMA_BIT_MASK(32))); 3163 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32)); 3164 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx), 3165 ring->count * sizeof(union ixgbe_adv_tx_desc)); 3166 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0); 3167 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0); 3168 ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx); 3169 3170 /* 3171 * set WTHRESH to encourage burst writeback, it should not be set 3172 * higher than 1 when: 3173 * - ITR is 0 as it could cause false TX hangs 3174 * - ITR is set to > 100k int/sec and BQL is enabled 3175 * 3176 * In order to avoid issues WTHRESH + PTHRESH should always be equal 3177 * to or less than the number of on chip descriptors, which is 3178 * currently 40. 3179 */ 3180 if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR)) 3181 txdctl |= 1u << 16; /* WTHRESH = 1 */ 3182 else 3183 txdctl |= 8u << 16; /* WTHRESH = 8 */ 3184 3185 /* 3186 * Setting PTHRESH to 32 both improves performance 3187 * and avoids a TX hang with DFP enabled 3188 */ 3189 txdctl |= (1u << 8) | /* HTHRESH = 1 */ 3190 32; /* PTHRESH = 32 */ 3191 3192 /* reinitialize flowdirector state */ 3193 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) { 3194 ring->atr_sample_rate = adapter->atr_sample_rate; 3195 ring->atr_count = 0; 3196 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state); 3197 } else { 3198 ring->atr_sample_rate = 0; 3199 } 3200 3201 /* initialize XPS */ 3202 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) { 3203 struct ixgbe_q_vector *q_vector = ring->q_vector; 3204 3205 if (q_vector) 3206 netif_set_xps_queue(ring->netdev, 3207 &q_vector->affinity_mask, 3208 ring->queue_index); 3209 } 3210 3211 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state); 3212 3213 /* enable queue */ 3214 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl); 3215 3216 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */ 3217 if (hw->mac.type == ixgbe_mac_82598EB && 3218 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP)) 3219 return; 3220 3221 /* poll to verify queue is enabled */ 3222 do { 3223 usleep_range(1000, 2000); 3224 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx)); 3225 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE)); 3226 if (!wait_loop) 3227 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx); 3228 } 3229 3230 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter) 3231 { 3232 struct ixgbe_hw *hw = &adapter->hw; 3233 u32 rttdcs, mtqc; 3234 u8 tcs = netdev_get_num_tc(adapter->netdev); 3235 3236 if (hw->mac.type == ixgbe_mac_82598EB) 3237 return; 3238 3239 /* disable the arbiter while setting MTQC */ 3240 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS); 3241 rttdcs |= IXGBE_RTTDCS_ARBDIS; 3242 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs); 3243 3244 /* set transmit pool layout */ 3245 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) { 3246 mtqc = IXGBE_MTQC_VT_ENA; 3247 if (tcs > 4) 3248 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ; 3249 else if (tcs > 1) 3250 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ; 3251 else if (adapter->ring_feature[RING_F_RSS].indices == 4) 3252 mtqc |= IXGBE_MTQC_32VF; 3253 else 3254 mtqc |= IXGBE_MTQC_64VF; 3255 } else { 3256 if (tcs > 4) 3257 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ; 3258 else if (tcs > 1) 3259 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ; 3260 else 3261 mtqc = IXGBE_MTQC_64Q_1PB; 3262 } 3263 3264 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc); 3265 3266 /* Enable Security TX Buffer IFG for multiple pb */ 3267 if (tcs) { 3268 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG); 3269 sectx |= IXGBE_SECTX_DCB; 3270 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx); 3271 } 3272 3273 /* re-enable the arbiter */ 3274 rttdcs &= ~IXGBE_RTTDCS_ARBDIS; 3275 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs); 3276 } 3277 3278 /** 3279 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset 3280 * @adapter: board private structure 3281 * 3282 * Configure the Tx unit of the MAC after a reset. 3283 **/ 3284 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter) 3285 { 3286 struct ixgbe_hw *hw = &adapter->hw; 3287 u32 dmatxctl; 3288 u32 i; 3289 3290 ixgbe_setup_mtqc(adapter); 3291 3292 if (hw->mac.type != ixgbe_mac_82598EB) { 3293 /* DMATXCTL.EN must be before Tx queues are enabled */ 3294 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL); 3295 dmatxctl |= IXGBE_DMATXCTL_TE; 3296 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl); 3297 } 3298 3299 /* Setup the HW Tx Head and Tail descriptor pointers */ 3300 for (i = 0; i < adapter->num_tx_queues; i++) 3301 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]); 3302 } 3303 3304 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter, 3305 struct ixgbe_ring *ring) 3306 { 3307 struct ixgbe_hw *hw = &adapter->hw; 3308 u8 reg_idx = ring->reg_idx; 3309 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx)); 3310 3311 srrctl |= IXGBE_SRRCTL_DROP_EN; 3312 3313 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl); 3314 } 3315 3316 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter, 3317 struct ixgbe_ring *ring) 3318 { 3319 struct ixgbe_hw *hw = &adapter->hw; 3320 u8 reg_idx = ring->reg_idx; 3321 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx)); 3322 3323 srrctl &= ~IXGBE_SRRCTL_DROP_EN; 3324 3325 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl); 3326 } 3327 3328 #ifdef CONFIG_IXGBE_DCB 3329 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter) 3330 #else 3331 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter) 3332 #endif 3333 { 3334 int i; 3335 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable; 3336 3337 if (adapter->ixgbe_ieee_pfc) 3338 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en); 3339 3340 /* 3341 * We should set the drop enable bit if: 3342 * SR-IOV is enabled 3343 * or 3344 * Number of Rx queues > 1 and flow control is disabled 3345 * 3346 * This allows us to avoid head of line blocking for security 3347 * and performance reasons. 3348 */ 3349 if (adapter->num_vfs || (adapter->num_rx_queues > 1 && 3350 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) { 3351 for (i = 0; i < adapter->num_rx_queues; i++) 3352 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]); 3353 } else { 3354 for (i = 0; i < adapter->num_rx_queues; i++) 3355 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]); 3356 } 3357 } 3358 3359 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2 3360 3361 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter, 3362 struct ixgbe_ring *rx_ring) 3363 { 3364 struct ixgbe_hw *hw = &adapter->hw; 3365 u32 srrctl; 3366 u8 reg_idx = rx_ring->reg_idx; 3367 3368 if (hw->mac.type == ixgbe_mac_82598EB) { 3369 u16 mask = adapter->ring_feature[RING_F_RSS].mask; 3370 3371 /* 3372 * if VMDq is not active we must program one srrctl register 3373 * per RSS queue since we have enabled RDRXCTL.MVMEN 3374 */ 3375 reg_idx &= mask; 3376 } 3377 3378 /* configure header buffer length, needed for RSC */ 3379 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT; 3380 3381 /* configure the packet buffer length */ 3382 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT; 3383 3384 /* configure descriptor type */ 3385 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF; 3386 3387 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl); 3388 } 3389 3390 /** 3391 * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries 3392 * @adapter: device handle 3393 * 3394 * - 82598/82599/X540: 128 3395 * - X550(non-SRIOV mode): 512 3396 * - X550(SRIOV mode): 64 3397 */ 3398 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter) 3399 { 3400 if (adapter->hw.mac.type < ixgbe_mac_X550) 3401 return 128; 3402 else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) 3403 return 64; 3404 else 3405 return 512; 3406 } 3407 3408 /** 3409 * ixgbe_store_reta - Write the RETA table to HW 3410 * @adapter: device handle 3411 * 3412 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW. 3413 */ 3414 void ixgbe_store_reta(struct ixgbe_adapter *adapter) 3415 { 3416 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter); 3417 struct ixgbe_hw *hw = &adapter->hw; 3418 u32 reta = 0; 3419 u32 indices_multi; 3420 u8 *indir_tbl = adapter->rss_indir_tbl; 3421 3422 /* Fill out the redirection table as follows: 3423 * - 82598: 8 bit wide entries containing pair of 4 bit RSS 3424 * indices. 3425 * - 82599/X540: 8 bit wide entries containing 4 bit RSS index 3426 * - X550: 8 bit wide entries containing 6 bit RSS index 3427 */ 3428 if (adapter->hw.mac.type == ixgbe_mac_82598EB) 3429 indices_multi = 0x11; 3430 else 3431 indices_multi = 0x1; 3432 3433 /* Write redirection table to HW */ 3434 for (i = 0; i < reta_entries; i++) { 3435 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8; 3436 if ((i & 3) == 3) { 3437 if (i < 128) 3438 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta); 3439 else 3440 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32), 3441 reta); 3442 reta = 0; 3443 } 3444 } 3445 } 3446 3447 /** 3448 * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode) 3449 * @adapter: device handle 3450 * 3451 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW. 3452 */ 3453 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter) 3454 { 3455 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter); 3456 struct ixgbe_hw *hw = &adapter->hw; 3457 u32 vfreta = 0; 3458 unsigned int pf_pool = adapter->num_vfs; 3459 3460 /* Write redirection table to HW */ 3461 for (i = 0; i < reta_entries; i++) { 3462 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8; 3463 if ((i & 3) == 3) { 3464 IXGBE_WRITE_REG(hw, IXGBE_PFVFRETA(i >> 2, pf_pool), 3465 vfreta); 3466 vfreta = 0; 3467 } 3468 } 3469 } 3470 3471 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter) 3472 { 3473 struct ixgbe_hw *hw = &adapter->hw; 3474 u32 i, j; 3475 u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter); 3476 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices; 3477 3478 /* Program table for at least 2 queues w/ SR-IOV so that VFs can 3479 * make full use of any rings they may have. We will use the 3480 * PSRTYPE register to control how many rings we use within the PF. 3481 */ 3482 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2)) 3483 rss_i = 2; 3484 3485 /* Fill out hash function seeds */ 3486 for (i = 0; i < 10; i++) 3487 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]); 3488 3489 /* Fill out redirection table */ 3490 memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl)); 3491 3492 for (i = 0, j = 0; i < reta_entries; i++, j++) { 3493 if (j == rss_i) 3494 j = 0; 3495 3496 adapter->rss_indir_tbl[i] = j; 3497 } 3498 3499 ixgbe_store_reta(adapter); 3500 } 3501 3502 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter) 3503 { 3504 struct ixgbe_hw *hw = &adapter->hw; 3505 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices; 3506 unsigned int pf_pool = adapter->num_vfs; 3507 int i, j; 3508 3509 /* Fill out hash function seeds */ 3510 for (i = 0; i < 10; i++) 3511 IXGBE_WRITE_REG(hw, IXGBE_PFVFRSSRK(i, pf_pool), 3512 adapter->rss_key[i]); 3513 3514 /* Fill out the redirection table */ 3515 for (i = 0, j = 0; i < 64; i++, j++) { 3516 if (j == rss_i) 3517 j = 0; 3518 3519 adapter->rss_indir_tbl[i] = j; 3520 } 3521 3522 ixgbe_store_vfreta(adapter); 3523 } 3524 3525 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter) 3526 { 3527 struct ixgbe_hw *hw = &adapter->hw; 3528 u32 mrqc = 0, rss_field = 0, vfmrqc = 0; 3529 u32 rxcsum; 3530 3531 /* Disable indicating checksum in descriptor, enables RSS hash */ 3532 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM); 3533 rxcsum |= IXGBE_RXCSUM_PCSD; 3534 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum); 3535 3536 if (adapter->hw.mac.type == ixgbe_mac_82598EB) { 3537 if (adapter->ring_feature[RING_F_RSS].mask) 3538 mrqc = IXGBE_MRQC_RSSEN; 3539 } else { 3540 u8 tcs = netdev_get_num_tc(adapter->netdev); 3541 3542 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) { 3543 if (tcs > 4) 3544 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */ 3545 else if (tcs > 1) 3546 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */ 3547 else if (adapter->ring_feature[RING_F_RSS].indices == 4) 3548 mrqc = IXGBE_MRQC_VMDQRSS32EN; 3549 else 3550 mrqc = IXGBE_MRQC_VMDQRSS64EN; 3551 } else { 3552 if (tcs > 4) 3553 mrqc = IXGBE_MRQC_RTRSS8TCEN; 3554 else if (tcs > 1) 3555 mrqc = IXGBE_MRQC_RTRSS4TCEN; 3556 else 3557 mrqc = IXGBE_MRQC_RSSEN; 3558 } 3559 } 3560 3561 /* Perform hash on these packet types */ 3562 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 | 3563 IXGBE_MRQC_RSS_FIELD_IPV4_TCP | 3564 IXGBE_MRQC_RSS_FIELD_IPV6 | 3565 IXGBE_MRQC_RSS_FIELD_IPV6_TCP; 3566 3567 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP) 3568 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP; 3569 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP) 3570 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP; 3571 3572 netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key)); 3573 if ((hw->mac.type >= ixgbe_mac_X550) && 3574 (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) { 3575 unsigned int pf_pool = adapter->num_vfs; 3576 3577 /* Enable VF RSS mode */ 3578 mrqc |= IXGBE_MRQC_MULTIPLE_RSS; 3579 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc); 3580 3581 /* Setup RSS through the VF registers */ 3582 ixgbe_setup_vfreta(adapter); 3583 vfmrqc = IXGBE_MRQC_RSSEN; 3584 vfmrqc |= rss_field; 3585 IXGBE_WRITE_REG(hw, IXGBE_PFVFMRQC(pf_pool), vfmrqc); 3586 } else { 3587 ixgbe_setup_reta(adapter); 3588 mrqc |= rss_field; 3589 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc); 3590 } 3591 } 3592 3593 /** 3594 * ixgbe_configure_rscctl - enable RSC for the indicated ring 3595 * @adapter: address of board private structure 3596 * @index: index of ring to set 3597 **/ 3598 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter, 3599 struct ixgbe_ring *ring) 3600 { 3601 struct ixgbe_hw *hw = &adapter->hw; 3602 u32 rscctrl; 3603 u8 reg_idx = ring->reg_idx; 3604 3605 if (!ring_is_rsc_enabled(ring)) 3606 return; 3607 3608 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx)); 3609 rscctrl |= IXGBE_RSCCTL_RSCEN; 3610 /* 3611 * we must limit the number of descriptors so that the 3612 * total size of max desc * buf_len is not greater 3613 * than 65536 3614 */ 3615 rscctrl |= IXGBE_RSCCTL_MAXDESC_16; 3616 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl); 3617 } 3618 3619 #define IXGBE_MAX_RX_DESC_POLL 10 3620 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter, 3621 struct ixgbe_ring *ring) 3622 { 3623 struct ixgbe_hw *hw = &adapter->hw; 3624 int wait_loop = IXGBE_MAX_RX_DESC_POLL; 3625 u32 rxdctl; 3626 u8 reg_idx = ring->reg_idx; 3627 3628 if (ixgbe_removed(hw->hw_addr)) 3629 return; 3630 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */ 3631 if (hw->mac.type == ixgbe_mac_82598EB && 3632 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP)) 3633 return; 3634 3635 do { 3636 usleep_range(1000, 2000); 3637 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx)); 3638 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE)); 3639 3640 if (!wait_loop) { 3641 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within " 3642 "the polling period\n", reg_idx); 3643 } 3644 } 3645 3646 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter, 3647 struct ixgbe_ring *ring) 3648 { 3649 struct ixgbe_hw *hw = &adapter->hw; 3650 int wait_loop = IXGBE_MAX_RX_DESC_POLL; 3651 u32 rxdctl; 3652 u8 reg_idx = ring->reg_idx; 3653 3654 if (ixgbe_removed(hw->hw_addr)) 3655 return; 3656 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx)); 3657 rxdctl &= ~IXGBE_RXDCTL_ENABLE; 3658 3659 /* write value back with RXDCTL.ENABLE bit cleared */ 3660 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl); 3661 3662 if (hw->mac.type == ixgbe_mac_82598EB && 3663 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP)) 3664 return; 3665 3666 /* the hardware may take up to 100us to really disable the rx queue */ 3667 do { 3668 udelay(10); 3669 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx)); 3670 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE)); 3671 3672 if (!wait_loop) { 3673 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within " 3674 "the polling period\n", reg_idx); 3675 } 3676 } 3677 3678 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter, 3679 struct ixgbe_ring *ring) 3680 { 3681 struct ixgbe_hw *hw = &adapter->hw; 3682 u64 rdba = ring->dma; 3683 u32 rxdctl; 3684 u8 reg_idx = ring->reg_idx; 3685 3686 /* disable queue to avoid issues while updating state */ 3687 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx)); 3688 ixgbe_disable_rx_queue(adapter, ring); 3689 3690 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32))); 3691 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32)); 3692 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx), 3693 ring->count * sizeof(union ixgbe_adv_rx_desc)); 3694 /* Force flushing of IXGBE_RDLEN to prevent MDD */ 3695 IXGBE_WRITE_FLUSH(hw); 3696 3697 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0); 3698 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0); 3699 ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx); 3700 3701 ixgbe_configure_srrctl(adapter, ring); 3702 ixgbe_configure_rscctl(adapter, ring); 3703 3704 if (hw->mac.type == ixgbe_mac_82598EB) { 3705 /* 3706 * enable cache line friendly hardware writes: 3707 * PTHRESH=32 descriptors (half the internal cache), 3708 * this also removes ugly rx_no_buffer_count increment 3709 * HTHRESH=4 descriptors (to minimize latency on fetch) 3710 * WTHRESH=8 burst writeback up to two cache lines 3711 */ 3712 rxdctl &= ~0x3FFFFF; 3713 rxdctl |= 0x080420; 3714 } 3715 3716 /* enable receive descriptor ring */ 3717 rxdctl |= IXGBE_RXDCTL_ENABLE; 3718 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl); 3719 3720 ixgbe_rx_desc_queue_enable(adapter, ring); 3721 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring)); 3722 } 3723 3724 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter) 3725 { 3726 struct ixgbe_hw *hw = &adapter->hw; 3727 int rss_i = adapter->ring_feature[RING_F_RSS].indices; 3728 u16 pool; 3729 3730 /* PSRTYPE must be initialized in non 82598 adapters */ 3731 u32 psrtype = IXGBE_PSRTYPE_TCPHDR | 3732 IXGBE_PSRTYPE_UDPHDR | 3733 IXGBE_PSRTYPE_IPV4HDR | 3734 IXGBE_PSRTYPE_L2HDR | 3735 IXGBE_PSRTYPE_IPV6HDR; 3736 3737 if (hw->mac.type == ixgbe_mac_82598EB) 3738 return; 3739 3740 if (rss_i > 3) 3741 psrtype |= 2u << 29; 3742 else if (rss_i > 1) 3743 psrtype |= 1u << 29; 3744 3745 for_each_set_bit(pool, &adapter->fwd_bitmask, 32) 3746 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype); 3747 } 3748 3749 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter) 3750 { 3751 struct ixgbe_hw *hw = &adapter->hw; 3752 u32 reg_offset, vf_shift; 3753 u32 gcr_ext, vmdctl; 3754 int i; 3755 3756 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) 3757 return; 3758 3759 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL); 3760 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN; 3761 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK; 3762 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT; 3763 vmdctl |= IXGBE_VT_CTL_REPLEN; 3764 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl); 3765 3766 vf_shift = VMDQ_P(0) % 32; 3767 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0; 3768 3769 /* Enable only the PF's pool for Tx/Rx */ 3770 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift)); 3771 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1); 3772 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift)); 3773 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1); 3774 if (adapter->bridge_mode == BRIDGE_MODE_VEB) 3775 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN); 3776 3777 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */ 3778 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0)); 3779 3780 /* clear VLAN promisc flag so VFTA will be updated if necessary */ 3781 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC; 3782 3783 /* 3784 * Set up VF register offsets for selected VT Mode, 3785 * i.e. 32 or 64 VFs for SR-IOV 3786 */ 3787 switch (adapter->ring_feature[RING_F_VMDQ].mask) { 3788 case IXGBE_82599_VMDQ_8Q_MASK: 3789 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16; 3790 break; 3791 case IXGBE_82599_VMDQ_4Q_MASK: 3792 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32; 3793 break; 3794 default: 3795 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64; 3796 break; 3797 } 3798 3799 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext); 3800 3801 for (i = 0; i < adapter->num_vfs; i++) { 3802 /* configure spoof checking */ 3803 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, 3804 adapter->vfinfo[i].spoofchk_enabled); 3805 3806 /* Enable/Disable RSS query feature */ 3807 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i, 3808 adapter->vfinfo[i].rss_query_enabled); 3809 } 3810 } 3811 3812 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter) 3813 { 3814 struct ixgbe_hw *hw = &adapter->hw; 3815 struct net_device *netdev = adapter->netdev; 3816 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN; 3817 struct ixgbe_ring *rx_ring; 3818 int i; 3819 u32 mhadd, hlreg0; 3820 3821 #ifdef IXGBE_FCOE 3822 /* adjust max frame to be able to do baby jumbo for FCoE */ 3823 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) && 3824 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE)) 3825 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE; 3826 3827 #endif /* IXGBE_FCOE */ 3828 3829 /* adjust max frame to be at least the size of a standard frame */ 3830 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN)) 3831 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN); 3832 3833 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD); 3834 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) { 3835 mhadd &= ~IXGBE_MHADD_MFS_MASK; 3836 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT; 3837 3838 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd); 3839 } 3840 3841 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0); 3842 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */ 3843 hlreg0 |= IXGBE_HLREG0_JUMBOEN; 3844 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0); 3845 3846 /* 3847 * Setup the HW Rx Head and Tail Descriptor Pointers and 3848 * the Base and Length of the Rx Descriptor Ring 3849 */ 3850 for (i = 0; i < adapter->num_rx_queues; i++) { 3851 rx_ring = adapter->rx_ring[i]; 3852 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) 3853 set_ring_rsc_enabled(rx_ring); 3854 else 3855 clear_ring_rsc_enabled(rx_ring); 3856 } 3857 } 3858 3859 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter) 3860 { 3861 struct ixgbe_hw *hw = &adapter->hw; 3862 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL); 3863 3864 switch (hw->mac.type) { 3865 case ixgbe_mac_82598EB: 3866 /* 3867 * For VMDq support of different descriptor types or 3868 * buffer sizes through the use of multiple SRRCTL 3869 * registers, RDRXCTL.MVMEN must be set to 1 3870 * 3871 * also, the manual doesn't mention it clearly but DCA hints 3872 * will only use queue 0's tags unless this bit is set. Side 3873 * effects of setting this bit are only that SRRCTL must be 3874 * fully programmed [0..15] 3875 */ 3876 rdrxctl |= IXGBE_RDRXCTL_MVMEN; 3877 break; 3878 case ixgbe_mac_X550: 3879 case ixgbe_mac_X550EM_x: 3880 case ixgbe_mac_x550em_a: 3881 if (adapter->num_vfs) 3882 rdrxctl |= IXGBE_RDRXCTL_PSP; 3883 /* fall through for older HW */ 3884 case ixgbe_mac_82599EB: 3885 case ixgbe_mac_X540: 3886 /* Disable RSC for ACK packets */ 3887 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU, 3888 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU))); 3889 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE; 3890 /* hardware requires some bits to be set by default */ 3891 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX); 3892 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP; 3893 break; 3894 default: 3895 /* We should do nothing since we don't know this hardware */ 3896 return; 3897 } 3898 3899 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl); 3900 } 3901 3902 /** 3903 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset 3904 * @adapter: board private structure 3905 * 3906 * Configure the Rx unit of the MAC after a reset. 3907 **/ 3908 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter) 3909 { 3910 struct ixgbe_hw *hw = &adapter->hw; 3911 int i; 3912 u32 rxctrl, rfctl; 3913 3914 /* disable receives while setting up the descriptors */ 3915 hw->mac.ops.disable_rx(hw); 3916 3917 ixgbe_setup_psrtype(adapter); 3918 ixgbe_setup_rdrxctl(adapter); 3919 3920 /* RSC Setup */ 3921 rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL); 3922 rfctl &= ~IXGBE_RFCTL_RSC_DIS; 3923 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) 3924 rfctl |= IXGBE_RFCTL_RSC_DIS; 3925 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl); 3926 3927 /* Program registers for the distribution of queues */ 3928 ixgbe_setup_mrqc(adapter); 3929 3930 /* set_rx_buffer_len must be called before ring initialization */ 3931 ixgbe_set_rx_buffer_len(adapter); 3932 3933 /* 3934 * Setup the HW Rx Head and Tail Descriptor Pointers and 3935 * the Base and Length of the Rx Descriptor Ring 3936 */ 3937 for (i = 0; i < adapter->num_rx_queues; i++) 3938 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]); 3939 3940 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL); 3941 /* disable drop enable for 82598 parts */ 3942 if (hw->mac.type == ixgbe_mac_82598EB) 3943 rxctrl |= IXGBE_RXCTRL_DMBYPS; 3944 3945 /* enable all receives */ 3946 rxctrl |= IXGBE_RXCTRL_RXEN; 3947 hw->mac.ops.enable_rx_dma(hw, rxctrl); 3948 } 3949 3950 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev, 3951 __be16 proto, u16 vid) 3952 { 3953 struct ixgbe_adapter *adapter = netdev_priv(netdev); 3954 struct ixgbe_hw *hw = &adapter->hw; 3955 3956 /* add VID to filter table */ 3957 if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)) 3958 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid); 3959 3960 set_bit(vid, adapter->active_vlans); 3961 3962 return 0; 3963 } 3964 3965 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan) 3966 { 3967 u32 vlvf; 3968 int idx; 3969 3970 /* short cut the special case */ 3971 if (vlan == 0) 3972 return 0; 3973 3974 /* Search for the vlan id in the VLVF entries */ 3975 for (idx = IXGBE_VLVF_ENTRIES; --idx;) { 3976 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx)); 3977 if ((vlvf & VLAN_VID_MASK) == vlan) 3978 break; 3979 } 3980 3981 return idx; 3982 } 3983 3984 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid) 3985 { 3986 struct ixgbe_hw *hw = &adapter->hw; 3987 u32 bits, word; 3988 int idx; 3989 3990 idx = ixgbe_find_vlvf_entry(hw, vid); 3991 if (!idx) 3992 return; 3993 3994 /* See if any other pools are set for this VLAN filter 3995 * entry other than the PF. 3996 */ 3997 word = idx * 2 + (VMDQ_P(0) / 32); 3998 bits = ~BIT(VMDQ_P(0) % 32); 3999 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word)); 4000 4001 /* Disable the filter so this falls into the default pool. */ 4002 if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) { 4003 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)) 4004 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0); 4005 IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0); 4006 } 4007 } 4008 4009 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, 4010 __be16 proto, u16 vid) 4011 { 4012 struct ixgbe_adapter *adapter = netdev_priv(netdev); 4013 struct ixgbe_hw *hw = &adapter->hw; 4014 4015 /* remove VID from filter table */ 4016 if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)) 4017 hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true); 4018 4019 clear_bit(vid, adapter->active_vlans); 4020 4021 return 0; 4022 } 4023 4024 /** 4025 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping 4026 * @adapter: driver data 4027 */ 4028 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter) 4029 { 4030 struct ixgbe_hw *hw = &adapter->hw; 4031 u32 vlnctrl; 4032 int i, j; 4033 4034 switch (hw->mac.type) { 4035 case ixgbe_mac_82598EB: 4036 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL); 4037 vlnctrl &= ~IXGBE_VLNCTRL_VME; 4038 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl); 4039 break; 4040 case ixgbe_mac_82599EB: 4041 case ixgbe_mac_X540: 4042 case ixgbe_mac_X550: 4043 case ixgbe_mac_X550EM_x: 4044 case ixgbe_mac_x550em_a: 4045 for (i = 0; i < adapter->num_rx_queues; i++) { 4046 struct ixgbe_ring *ring = adapter->rx_ring[i]; 4047 4048 if (ring->l2_accel_priv) 4049 continue; 4050 j = ring->reg_idx; 4051 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j)); 4052 vlnctrl &= ~IXGBE_RXDCTL_VME; 4053 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl); 4054 } 4055 break; 4056 default: 4057 break; 4058 } 4059 } 4060 4061 /** 4062 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping 4063 * @adapter: driver data 4064 */ 4065 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter) 4066 { 4067 struct ixgbe_hw *hw = &adapter->hw; 4068 u32 vlnctrl; 4069 int i, j; 4070 4071 switch (hw->mac.type) { 4072 case ixgbe_mac_82598EB: 4073 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL); 4074 vlnctrl |= IXGBE_VLNCTRL_VME; 4075 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl); 4076 break; 4077 case ixgbe_mac_82599EB: 4078 case ixgbe_mac_X540: 4079 case ixgbe_mac_X550: 4080 case ixgbe_mac_X550EM_x: 4081 case ixgbe_mac_x550em_a: 4082 for (i = 0; i < adapter->num_rx_queues; i++) { 4083 struct ixgbe_ring *ring = adapter->rx_ring[i]; 4084 4085 if (ring->l2_accel_priv) 4086 continue; 4087 j = ring->reg_idx; 4088 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j)); 4089 vlnctrl |= IXGBE_RXDCTL_VME; 4090 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl); 4091 } 4092 break; 4093 default: 4094 break; 4095 } 4096 } 4097 4098 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter) 4099 { 4100 struct ixgbe_hw *hw = &adapter->hw; 4101 u32 vlnctrl, i; 4102 4103 switch (hw->mac.type) { 4104 case ixgbe_mac_82599EB: 4105 case ixgbe_mac_X540: 4106 case ixgbe_mac_X550: 4107 case ixgbe_mac_X550EM_x: 4108 case ixgbe_mac_x550em_a: 4109 default: 4110 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) 4111 break; 4112 /* fall through */ 4113 case ixgbe_mac_82598EB: 4114 /* legacy case, we can just disable VLAN filtering */ 4115 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL); 4116 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN); 4117 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl); 4118 return; 4119 } 4120 4121 /* We are already in VLAN promisc, nothing to do */ 4122 if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC) 4123 return; 4124 4125 /* Set flag so we don't redo unnecessary work */ 4126 adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC; 4127 4128 /* Add PF to all active pools */ 4129 for (i = IXGBE_VLVF_ENTRIES; --i;) { 4130 u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32); 4131 u32 vlvfb = IXGBE_READ_REG(hw, reg_offset); 4132 4133 vlvfb |= BIT(VMDQ_P(0) % 32); 4134 IXGBE_WRITE_REG(hw, reg_offset, vlvfb); 4135 } 4136 4137 /* Set all bits in the VLAN filter table array */ 4138 for (i = hw->mac.vft_size; i--;) 4139 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U); 4140 } 4141 4142 #define VFTA_BLOCK_SIZE 8 4143 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset) 4144 { 4145 struct ixgbe_hw *hw = &adapter->hw; 4146 u32 vfta[VFTA_BLOCK_SIZE] = { 0 }; 4147 u32 vid_start = vfta_offset * 32; 4148 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32); 4149 u32 i, vid, word, bits; 4150 4151 for (i = IXGBE_VLVF_ENTRIES; --i;) { 4152 u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i)); 4153 4154 /* pull VLAN ID from VLVF */ 4155 vid = vlvf & VLAN_VID_MASK; 4156 4157 /* only concern outselves with a certain range */ 4158 if (vid < vid_start || vid >= vid_end) 4159 continue; 4160 4161 if (vlvf) { 4162 /* record VLAN ID in VFTA */ 4163 vfta[(vid - vid_start) / 32] |= BIT(vid % 32); 4164 4165 /* if PF is part of this then continue */ 4166 if (test_bit(vid, adapter->active_vlans)) 4167 continue; 4168 } 4169 4170 /* remove PF from the pool */ 4171 word = i * 2 + VMDQ_P(0) / 32; 4172 bits = ~BIT(VMDQ_P(0) % 32); 4173 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word)); 4174 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits); 4175 } 4176 4177 /* extract values from active_vlans and write back to VFTA */ 4178 for (i = VFTA_BLOCK_SIZE; i--;) { 4179 vid = (vfta_offset + i) * 32; 4180 word = vid / BITS_PER_LONG; 4181 bits = vid % BITS_PER_LONG; 4182 4183 vfta[i] |= adapter->active_vlans[word] >> bits; 4184 4185 IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]); 4186 } 4187 } 4188 4189 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter) 4190 { 4191 struct ixgbe_hw *hw = &adapter->hw; 4192 u32 vlnctrl, i; 4193 4194 switch (hw->mac.type) { 4195 case ixgbe_mac_82599EB: 4196 case ixgbe_mac_X540: 4197 case ixgbe_mac_X550: 4198 case ixgbe_mac_X550EM_x: 4199 case ixgbe_mac_x550em_a: 4200 default: 4201 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) 4202 break; 4203 /* fall through */ 4204 case ixgbe_mac_82598EB: 4205 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL); 4206 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN; 4207 vlnctrl |= IXGBE_VLNCTRL_VFE; 4208 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl); 4209 return; 4210 } 4211 4212 /* We are not in VLAN promisc, nothing to do */ 4213 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)) 4214 return; 4215 4216 /* Set flag so we don't redo unnecessary work */ 4217 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC; 4218 4219 for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE) 4220 ixgbe_scrub_vfta(adapter, i); 4221 } 4222 4223 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter) 4224 { 4225 u16 vid = 1; 4226 4227 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0); 4228 4229 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID) 4230 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid); 4231 } 4232 4233 /** 4234 * ixgbe_write_mc_addr_list - write multicast addresses to MTA 4235 * @netdev: network interface device structure 4236 * 4237 * Writes multicast address list to the MTA hash table. 4238 * Returns: -ENOMEM on failure 4239 * 0 on no addresses written 4240 * X on writing X addresses to MTA 4241 **/ 4242 static int ixgbe_write_mc_addr_list(struct net_device *netdev) 4243 { 4244 struct ixgbe_adapter *adapter = netdev_priv(netdev); 4245 struct ixgbe_hw *hw = &adapter->hw; 4246 4247 if (!netif_running(netdev)) 4248 return 0; 4249 4250 if (hw->mac.ops.update_mc_addr_list) 4251 hw->mac.ops.update_mc_addr_list(hw, netdev); 4252 else 4253 return -ENOMEM; 4254 4255 #ifdef CONFIG_PCI_IOV 4256 ixgbe_restore_vf_multicasts(adapter); 4257 #endif 4258 4259 return netdev_mc_count(netdev); 4260 } 4261 4262 #ifdef CONFIG_PCI_IOV 4263 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter) 4264 { 4265 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0]; 4266 struct ixgbe_hw *hw = &adapter->hw; 4267 int i; 4268 4269 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) { 4270 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED; 4271 4272 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) 4273 hw->mac.ops.set_rar(hw, i, 4274 mac_table->addr, 4275 mac_table->pool, 4276 IXGBE_RAH_AV); 4277 else 4278 hw->mac.ops.clear_rar(hw, i); 4279 } 4280 } 4281 4282 #endif 4283 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter) 4284 { 4285 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0]; 4286 struct ixgbe_hw *hw = &adapter->hw; 4287 int i; 4288 4289 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) { 4290 if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED)) 4291 continue; 4292 4293 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED; 4294 4295 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) 4296 hw->mac.ops.set_rar(hw, i, 4297 mac_table->addr, 4298 mac_table->pool, 4299 IXGBE_RAH_AV); 4300 else 4301 hw->mac.ops.clear_rar(hw, i); 4302 } 4303 } 4304 4305 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter) 4306 { 4307 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0]; 4308 struct ixgbe_hw *hw = &adapter->hw; 4309 int i; 4310 4311 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) { 4312 mac_table->state |= IXGBE_MAC_STATE_MODIFIED; 4313 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE; 4314 } 4315 4316 ixgbe_sync_mac_table(adapter); 4317 } 4318 4319 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool) 4320 { 4321 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0]; 4322 struct ixgbe_hw *hw = &adapter->hw; 4323 int i, count = 0; 4324 4325 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) { 4326 /* do not count default RAR as available */ 4327 if (mac_table->state & IXGBE_MAC_STATE_DEFAULT) 4328 continue; 4329 4330 /* only count unused and addresses that belong to us */ 4331 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) { 4332 if (mac_table->pool != pool) 4333 continue; 4334 } 4335 4336 count++; 4337 } 4338 4339 return count; 4340 } 4341 4342 /* this function destroys the first RAR entry */ 4343 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter) 4344 { 4345 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0]; 4346 struct ixgbe_hw *hw = &adapter->hw; 4347 4348 memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN); 4349 mac_table->pool = VMDQ_P(0); 4350 4351 mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE; 4352 4353 hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool, 4354 IXGBE_RAH_AV); 4355 } 4356 4357 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter, 4358 const u8 *addr, u16 pool) 4359 { 4360 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0]; 4361 struct ixgbe_hw *hw = &adapter->hw; 4362 int i; 4363 4364 if (is_zero_ether_addr(addr)) 4365 return -EINVAL; 4366 4367 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) { 4368 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) 4369 continue; 4370 4371 ether_addr_copy(mac_table->addr, addr); 4372 mac_table->pool = pool; 4373 4374 mac_table->state |= IXGBE_MAC_STATE_MODIFIED | 4375 IXGBE_MAC_STATE_IN_USE; 4376 4377 ixgbe_sync_mac_table(adapter); 4378 4379 return i; 4380 } 4381 4382 return -ENOMEM; 4383 } 4384 4385 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter, 4386 const u8 *addr, u16 pool) 4387 { 4388 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0]; 4389 struct ixgbe_hw *hw = &adapter->hw; 4390 int i; 4391 4392 if (is_zero_ether_addr(addr)) 4393 return -EINVAL; 4394 4395 /* search table for addr, if found clear IN_USE flag and sync */ 4396 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) { 4397 /* we can only delete an entry if it is in use */ 4398 if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE)) 4399 continue; 4400 /* we only care about entries that belong to the given pool */ 4401 if (mac_table->pool != pool) 4402 continue; 4403 /* we only care about a specific MAC address */ 4404 if (!ether_addr_equal(addr, mac_table->addr)) 4405 continue; 4406 4407 mac_table->state |= IXGBE_MAC_STATE_MODIFIED; 4408 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE; 4409 4410 ixgbe_sync_mac_table(adapter); 4411 4412 return 0; 4413 } 4414 4415 return -ENOMEM; 4416 } 4417 /** 4418 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table 4419 * @netdev: network interface device structure 4420 * 4421 * Writes unicast address list to the RAR table. 4422 * Returns: -ENOMEM on failure/insufficient address space 4423 * 0 on no addresses written 4424 * X on writing X addresses to the RAR table 4425 **/ 4426 static int ixgbe_write_uc_addr_list(struct net_device *netdev, int vfn) 4427 { 4428 struct ixgbe_adapter *adapter = netdev_priv(netdev); 4429 int count = 0; 4430 4431 /* return ENOMEM indicating insufficient memory for addresses */ 4432 if (netdev_uc_count(netdev) > ixgbe_available_rars(adapter, vfn)) 4433 return -ENOMEM; 4434 4435 if (!netdev_uc_empty(netdev)) { 4436 struct netdev_hw_addr *ha; 4437 netdev_for_each_uc_addr(ha, netdev) { 4438 ixgbe_del_mac_filter(adapter, ha->addr, vfn); 4439 ixgbe_add_mac_filter(adapter, ha->addr, vfn); 4440 count++; 4441 } 4442 } 4443 return count; 4444 } 4445 4446 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr) 4447 { 4448 struct ixgbe_adapter *adapter = netdev_priv(netdev); 4449 int ret; 4450 4451 ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0)); 4452 4453 return min_t(int, ret, 0); 4454 } 4455 4456 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr) 4457 { 4458 struct ixgbe_adapter *adapter = netdev_priv(netdev); 4459 4460 ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0)); 4461 4462 return 0; 4463 } 4464 4465 /** 4466 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set 4467 * @netdev: network interface device structure 4468 * 4469 * The set_rx_method entry point is called whenever the unicast/multicast 4470 * address list or the network interface flags are updated. This routine is 4471 * responsible for configuring the hardware for proper unicast, multicast and 4472 * promiscuous mode. 4473 **/ 4474 void ixgbe_set_rx_mode(struct net_device *netdev) 4475 { 4476 struct ixgbe_adapter *adapter = netdev_priv(netdev); 4477 struct ixgbe_hw *hw = &adapter->hw; 4478 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE; 4479 netdev_features_t features = netdev->features; 4480 int count; 4481 4482 /* Check for Promiscuous and All Multicast modes */ 4483 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL); 4484 4485 /* set all bits that we expect to always be set */ 4486 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */ 4487 fctrl |= IXGBE_FCTRL_BAM; 4488 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */ 4489 fctrl |= IXGBE_FCTRL_PMCF; 4490 4491 /* clear the bits we are changing the status of */ 4492 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE); 4493 if (netdev->flags & IFF_PROMISC) { 4494 hw->addr_ctrl.user_set_promisc = true; 4495 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE); 4496 vmolr |= IXGBE_VMOLR_MPE; 4497 features &= ~NETIF_F_HW_VLAN_CTAG_FILTER; 4498 } else { 4499 if (netdev->flags & IFF_ALLMULTI) { 4500 fctrl |= IXGBE_FCTRL_MPE; 4501 vmolr |= IXGBE_VMOLR_MPE; 4502 } 4503 hw->addr_ctrl.user_set_promisc = false; 4504 } 4505 4506 /* 4507 * Write addresses to available RAR registers, if there is not 4508 * sufficient space to store all the addresses then enable 4509 * unicast promiscuous mode 4510 */ 4511 if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) { 4512 fctrl |= IXGBE_FCTRL_UPE; 4513 vmolr |= IXGBE_VMOLR_ROPE; 4514 } 4515 4516 /* Write addresses to the MTA, if the attempt fails 4517 * then we should just turn on promiscuous mode so 4518 * that we can at least receive multicast traffic 4519 */ 4520 count = ixgbe_write_mc_addr_list(netdev); 4521 if (count < 0) { 4522 fctrl |= IXGBE_FCTRL_MPE; 4523 vmolr |= IXGBE_VMOLR_MPE; 4524 } else if (count) { 4525 vmolr |= IXGBE_VMOLR_ROMPE; 4526 } 4527 4528 if (hw->mac.type != ixgbe_mac_82598EB) { 4529 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) & 4530 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE | 4531 IXGBE_VMOLR_ROPE); 4532 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr); 4533 } 4534 4535 /* This is useful for sniffing bad packets. */ 4536 if (features & NETIF_F_RXALL) { 4537 /* UPE and MPE will be handled by normal PROMISC logic 4538 * in e1000e_set_rx_mode */ 4539 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */ 4540 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */ 4541 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */ 4542 4543 fctrl &= ~(IXGBE_FCTRL_DPF); 4544 /* NOTE: VLAN filtering is disabled by setting PROMISC */ 4545 } 4546 4547 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl); 4548 4549 if (features & NETIF_F_HW_VLAN_CTAG_RX) 4550 ixgbe_vlan_strip_enable(adapter); 4551 else 4552 ixgbe_vlan_strip_disable(adapter); 4553 4554 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) 4555 ixgbe_vlan_promisc_disable(adapter); 4556 else 4557 ixgbe_vlan_promisc_enable(adapter); 4558 } 4559 4560 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter) 4561 { 4562 int q_idx; 4563 4564 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) { 4565 ixgbe_qv_init_lock(adapter->q_vector[q_idx]); 4566 napi_enable(&adapter->q_vector[q_idx]->napi); 4567 } 4568 } 4569 4570 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter) 4571 { 4572 int q_idx; 4573 4574 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) { 4575 napi_disable(&adapter->q_vector[q_idx]->napi); 4576 while (!ixgbe_qv_disable(adapter->q_vector[q_idx])) { 4577 pr_info("QV %d locked\n", q_idx); 4578 usleep_range(1000, 20000); 4579 } 4580 } 4581 } 4582 4583 static void ixgbe_clear_vxlan_port(struct ixgbe_adapter *adapter) 4584 { 4585 switch (adapter->hw.mac.type) { 4586 case ixgbe_mac_X550: 4587 case ixgbe_mac_X550EM_x: 4588 case ixgbe_mac_x550em_a: 4589 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VXLANCTRL, 0); 4590 adapter->vxlan_port = 0; 4591 break; 4592 default: 4593 break; 4594 } 4595 } 4596 4597 #ifdef CONFIG_IXGBE_DCB 4598 /** 4599 * ixgbe_configure_dcb - Configure DCB hardware 4600 * @adapter: ixgbe adapter struct 4601 * 4602 * This is called by the driver on open to configure the DCB hardware. 4603 * This is also called by the gennetlink interface when reconfiguring 4604 * the DCB state. 4605 */ 4606 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter) 4607 { 4608 struct ixgbe_hw *hw = &adapter->hw; 4609 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN; 4610 4611 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) { 4612 if (hw->mac.type == ixgbe_mac_82598EB) 4613 netif_set_gso_max_size(adapter->netdev, 65536); 4614 return; 4615 } 4616 4617 if (hw->mac.type == ixgbe_mac_82598EB) 4618 netif_set_gso_max_size(adapter->netdev, 32768); 4619 4620 #ifdef IXGBE_FCOE 4621 if (adapter->netdev->features & NETIF_F_FCOE_MTU) 4622 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE); 4623 #endif 4624 4625 /* reconfigure the hardware */ 4626 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) { 4627 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame, 4628 DCB_TX_CONFIG); 4629 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame, 4630 DCB_RX_CONFIG); 4631 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg); 4632 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) { 4633 ixgbe_dcb_hw_ets(&adapter->hw, 4634 adapter->ixgbe_ieee_ets, 4635 max_frame); 4636 ixgbe_dcb_hw_pfc_config(&adapter->hw, 4637 adapter->ixgbe_ieee_pfc->pfc_en, 4638 adapter->ixgbe_ieee_ets->prio_tc); 4639 } 4640 4641 /* Enable RSS Hash per TC */ 4642 if (hw->mac.type != ixgbe_mac_82598EB) { 4643 u32 msb = 0; 4644 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1; 4645 4646 while (rss_i) { 4647 msb++; 4648 rss_i >>= 1; 4649 } 4650 4651 /* write msb to all 8 TCs in one write */ 4652 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111); 4653 } 4654 } 4655 #endif 4656 4657 /* Additional bittime to account for IXGBE framing */ 4658 #define IXGBE_ETH_FRAMING 20 4659 4660 /** 4661 * ixgbe_hpbthresh - calculate high water mark for flow control 4662 * 4663 * @adapter: board private structure to calculate for 4664 * @pb: packet buffer to calculate 4665 */ 4666 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb) 4667 { 4668 struct ixgbe_hw *hw = &adapter->hw; 4669 struct net_device *dev = adapter->netdev; 4670 int link, tc, kb, marker; 4671 u32 dv_id, rx_pba; 4672 4673 /* Calculate max LAN frame size */ 4674 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING; 4675 4676 #ifdef IXGBE_FCOE 4677 /* FCoE traffic class uses FCOE jumbo frames */ 4678 if ((dev->features & NETIF_F_FCOE_MTU) && 4679 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) && 4680 (pb == ixgbe_fcoe_get_tc(adapter))) 4681 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE; 4682 #endif 4683 4684 /* Calculate delay value for device */ 4685 switch (hw->mac.type) { 4686 case ixgbe_mac_X540: 4687 case ixgbe_mac_X550: 4688 case ixgbe_mac_X550EM_x: 4689 case ixgbe_mac_x550em_a: 4690 dv_id = IXGBE_DV_X540(link, tc); 4691 break; 4692 default: 4693 dv_id = IXGBE_DV(link, tc); 4694 break; 4695 } 4696 4697 /* Loopback switch introduces additional latency */ 4698 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) 4699 dv_id += IXGBE_B2BT(tc); 4700 4701 /* Delay value is calculated in bit times convert to KB */ 4702 kb = IXGBE_BT2KB(dv_id); 4703 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10; 4704 4705 marker = rx_pba - kb; 4706 4707 /* It is possible that the packet buffer is not large enough 4708 * to provide required headroom. In this case throw an error 4709 * to user and a do the best we can. 4710 */ 4711 if (marker < 0) { 4712 e_warn(drv, "Packet Buffer(%i) can not provide enough" 4713 "headroom to support flow control." 4714 "Decrease MTU or number of traffic classes\n", pb); 4715 marker = tc + 1; 4716 } 4717 4718 return marker; 4719 } 4720 4721 /** 4722 * ixgbe_lpbthresh - calculate low water mark for for flow control 4723 * 4724 * @adapter: board private structure to calculate for 4725 * @pb: packet buffer to calculate 4726 */ 4727 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb) 4728 { 4729 struct ixgbe_hw *hw = &adapter->hw; 4730 struct net_device *dev = adapter->netdev; 4731 int tc; 4732 u32 dv_id; 4733 4734 /* Calculate max LAN frame size */ 4735 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN; 4736 4737 #ifdef IXGBE_FCOE 4738 /* FCoE traffic class uses FCOE jumbo frames */ 4739 if ((dev->features & NETIF_F_FCOE_MTU) && 4740 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) && 4741 (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up))) 4742 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE; 4743 #endif 4744 4745 /* Calculate delay value for device */ 4746 switch (hw->mac.type) { 4747 case ixgbe_mac_X540: 4748 case ixgbe_mac_X550: 4749 case ixgbe_mac_X550EM_x: 4750 case ixgbe_mac_x550em_a: 4751 dv_id = IXGBE_LOW_DV_X540(tc); 4752 break; 4753 default: 4754 dv_id = IXGBE_LOW_DV(tc); 4755 break; 4756 } 4757 4758 /* Delay value is calculated in bit times convert to KB */ 4759 return IXGBE_BT2KB(dv_id); 4760 } 4761 4762 /* 4763 * ixgbe_pbthresh_setup - calculate and setup high low water marks 4764 */ 4765 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter) 4766 { 4767 struct ixgbe_hw *hw = &adapter->hw; 4768 int num_tc = netdev_get_num_tc(adapter->netdev); 4769 int i; 4770 4771 if (!num_tc) 4772 num_tc = 1; 4773 4774 for (i = 0; i < num_tc; i++) { 4775 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i); 4776 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i); 4777 4778 /* Low water marks must not be larger than high water marks */ 4779 if (hw->fc.low_water[i] > hw->fc.high_water[i]) 4780 hw->fc.low_water[i] = 0; 4781 } 4782 4783 for (; i < MAX_TRAFFIC_CLASS; i++) 4784 hw->fc.high_water[i] = 0; 4785 } 4786 4787 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter) 4788 { 4789 struct ixgbe_hw *hw = &adapter->hw; 4790 int hdrm; 4791 u8 tc = netdev_get_num_tc(adapter->netdev); 4792 4793 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE || 4794 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) 4795 hdrm = 32 << adapter->fdir_pballoc; 4796 else 4797 hdrm = 0; 4798 4799 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL); 4800 ixgbe_pbthresh_setup(adapter); 4801 } 4802 4803 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter) 4804 { 4805 struct ixgbe_hw *hw = &adapter->hw; 4806 struct hlist_node *node2; 4807 struct ixgbe_fdir_filter *filter; 4808 4809 spin_lock(&adapter->fdir_perfect_lock); 4810 4811 if (!hlist_empty(&adapter->fdir_filter_list)) 4812 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask); 4813 4814 hlist_for_each_entry_safe(filter, node2, 4815 &adapter->fdir_filter_list, fdir_node) { 4816 ixgbe_fdir_write_perfect_filter_82599(hw, 4817 &filter->filter, 4818 filter->sw_idx, 4819 (filter->action == IXGBE_FDIR_DROP_QUEUE) ? 4820 IXGBE_FDIR_DROP_QUEUE : 4821 adapter->rx_ring[filter->action]->reg_idx); 4822 } 4823 4824 spin_unlock(&adapter->fdir_perfect_lock); 4825 } 4826 4827 static void ixgbe_macvlan_set_rx_mode(struct net_device *dev, unsigned int pool, 4828 struct ixgbe_adapter *adapter) 4829 { 4830 struct ixgbe_hw *hw = &adapter->hw; 4831 u32 vmolr; 4832 4833 /* No unicast promiscuous support for VMDQ devices. */ 4834 vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(pool)); 4835 vmolr |= (IXGBE_VMOLR_ROMPE | IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE); 4836 4837 /* clear the affected bit */ 4838 vmolr &= ~IXGBE_VMOLR_MPE; 4839 4840 if (dev->flags & IFF_ALLMULTI) { 4841 vmolr |= IXGBE_VMOLR_MPE; 4842 } else { 4843 vmolr |= IXGBE_VMOLR_ROMPE; 4844 hw->mac.ops.update_mc_addr_list(hw, dev); 4845 } 4846 ixgbe_write_uc_addr_list(adapter->netdev, pool); 4847 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(pool), vmolr); 4848 } 4849 4850 static void ixgbe_fwd_psrtype(struct ixgbe_fwd_adapter *vadapter) 4851 { 4852 struct ixgbe_adapter *adapter = vadapter->real_adapter; 4853 int rss_i = adapter->num_rx_queues_per_pool; 4854 struct ixgbe_hw *hw = &adapter->hw; 4855 u16 pool = vadapter->pool; 4856 u32 psrtype = IXGBE_PSRTYPE_TCPHDR | 4857 IXGBE_PSRTYPE_UDPHDR | 4858 IXGBE_PSRTYPE_IPV4HDR | 4859 IXGBE_PSRTYPE_L2HDR | 4860 IXGBE_PSRTYPE_IPV6HDR; 4861 4862 if (hw->mac.type == ixgbe_mac_82598EB) 4863 return; 4864 4865 if (rss_i > 3) 4866 psrtype |= 2u << 29; 4867 else if (rss_i > 1) 4868 psrtype |= 1u << 29; 4869 4870 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype); 4871 } 4872 4873 /** 4874 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue 4875 * @rx_ring: ring to free buffers from 4876 **/ 4877 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring) 4878 { 4879 struct device *dev = rx_ring->dev; 4880 unsigned long size; 4881 u16 i; 4882 4883 /* ring already cleared, nothing to do */ 4884 if (!rx_ring->rx_buffer_info) 4885 return; 4886 4887 /* Free all the Rx ring sk_buffs */ 4888 for (i = 0; i < rx_ring->count; i++) { 4889 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i]; 4890 4891 if (rx_buffer->skb) { 4892 struct sk_buff *skb = rx_buffer->skb; 4893 if (IXGBE_CB(skb)->page_released) 4894 dma_unmap_page(dev, 4895 IXGBE_CB(skb)->dma, 4896 ixgbe_rx_bufsz(rx_ring), 4897 DMA_FROM_DEVICE); 4898 dev_kfree_skb(skb); 4899 rx_buffer->skb = NULL; 4900 } 4901 4902 if (!rx_buffer->page) 4903 continue; 4904 4905 dma_unmap_page(dev, rx_buffer->dma, 4906 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE); 4907 __free_pages(rx_buffer->page, ixgbe_rx_pg_order(rx_ring)); 4908 4909 rx_buffer->page = NULL; 4910 } 4911 4912 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count; 4913 memset(rx_ring->rx_buffer_info, 0, size); 4914 4915 /* Zero out the descriptor ring */ 4916 memset(rx_ring->desc, 0, rx_ring->size); 4917 4918 rx_ring->next_to_alloc = 0; 4919 rx_ring->next_to_clean = 0; 4920 rx_ring->next_to_use = 0; 4921 } 4922 4923 static void ixgbe_disable_fwd_ring(struct ixgbe_fwd_adapter *vadapter, 4924 struct ixgbe_ring *rx_ring) 4925 { 4926 struct ixgbe_adapter *adapter = vadapter->real_adapter; 4927 int index = rx_ring->queue_index + vadapter->rx_base_queue; 4928 4929 /* shutdown specific queue receive and wait for dma to settle */ 4930 ixgbe_disable_rx_queue(adapter, rx_ring); 4931 usleep_range(10000, 20000); 4932 ixgbe_irq_disable_queues(adapter, BIT_ULL(index)); 4933 ixgbe_clean_rx_ring(rx_ring); 4934 rx_ring->l2_accel_priv = NULL; 4935 } 4936 4937 static int ixgbe_fwd_ring_down(struct net_device *vdev, 4938 struct ixgbe_fwd_adapter *accel) 4939 { 4940 struct ixgbe_adapter *adapter = accel->real_adapter; 4941 unsigned int rxbase = accel->rx_base_queue; 4942 unsigned int txbase = accel->tx_base_queue; 4943 int i; 4944 4945 netif_tx_stop_all_queues(vdev); 4946 4947 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) { 4948 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]); 4949 adapter->rx_ring[rxbase + i]->netdev = adapter->netdev; 4950 } 4951 4952 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) { 4953 adapter->tx_ring[txbase + i]->l2_accel_priv = NULL; 4954 adapter->tx_ring[txbase + i]->netdev = adapter->netdev; 4955 } 4956 4957 4958 return 0; 4959 } 4960 4961 static int ixgbe_fwd_ring_up(struct net_device *vdev, 4962 struct ixgbe_fwd_adapter *accel) 4963 { 4964 struct ixgbe_adapter *adapter = accel->real_adapter; 4965 unsigned int rxbase, txbase, queues; 4966 int i, baseq, err = 0; 4967 4968 if (!test_bit(accel->pool, &adapter->fwd_bitmask)) 4969 return 0; 4970 4971 baseq = accel->pool * adapter->num_rx_queues_per_pool; 4972 netdev_dbg(vdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n", 4973 accel->pool, adapter->num_rx_pools, 4974 baseq, baseq + adapter->num_rx_queues_per_pool, 4975 adapter->fwd_bitmask); 4976 4977 accel->netdev = vdev; 4978 accel->rx_base_queue = rxbase = baseq; 4979 accel->tx_base_queue = txbase = baseq; 4980 4981 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) 4982 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]); 4983 4984 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) { 4985 adapter->rx_ring[rxbase + i]->netdev = vdev; 4986 adapter->rx_ring[rxbase + i]->l2_accel_priv = accel; 4987 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[rxbase + i]); 4988 } 4989 4990 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) { 4991 adapter->tx_ring[txbase + i]->netdev = vdev; 4992 adapter->tx_ring[txbase + i]->l2_accel_priv = accel; 4993 } 4994 4995 queues = min_t(unsigned int, 4996 adapter->num_rx_queues_per_pool, vdev->num_tx_queues); 4997 err = netif_set_real_num_tx_queues(vdev, queues); 4998 if (err) 4999 goto fwd_queue_err; 5000 5001 err = netif_set_real_num_rx_queues(vdev, queues); 5002 if (err) 5003 goto fwd_queue_err; 5004 5005 if (is_valid_ether_addr(vdev->dev_addr)) 5006 ixgbe_add_mac_filter(adapter, vdev->dev_addr, accel->pool); 5007 5008 ixgbe_fwd_psrtype(accel); 5009 ixgbe_macvlan_set_rx_mode(vdev, accel->pool, adapter); 5010 return err; 5011 fwd_queue_err: 5012 ixgbe_fwd_ring_down(vdev, accel); 5013 return err; 5014 } 5015 5016 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter) 5017 { 5018 struct net_device *upper; 5019 struct list_head *iter; 5020 int err; 5021 5022 netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) { 5023 if (netif_is_macvlan(upper)) { 5024 struct macvlan_dev *dfwd = netdev_priv(upper); 5025 struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv; 5026 5027 if (dfwd->fwd_priv) { 5028 err = ixgbe_fwd_ring_up(upper, vadapter); 5029 if (err) 5030 continue; 5031 } 5032 } 5033 } 5034 } 5035 5036 static void ixgbe_configure(struct ixgbe_adapter *adapter) 5037 { 5038 struct ixgbe_hw *hw = &adapter->hw; 5039 5040 ixgbe_configure_pb(adapter); 5041 #ifdef CONFIG_IXGBE_DCB 5042 ixgbe_configure_dcb(adapter); 5043 #endif 5044 /* 5045 * We must restore virtualization before VLANs or else 5046 * the VLVF registers will not be populated 5047 */ 5048 ixgbe_configure_virtualization(adapter); 5049 5050 ixgbe_set_rx_mode(adapter->netdev); 5051 ixgbe_restore_vlan(adapter); 5052 5053 switch (hw->mac.type) { 5054 case ixgbe_mac_82599EB: 5055 case ixgbe_mac_X540: 5056 hw->mac.ops.disable_rx_buff(hw); 5057 break; 5058 default: 5059 break; 5060 } 5061 5062 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) { 5063 ixgbe_init_fdir_signature_82599(&adapter->hw, 5064 adapter->fdir_pballoc); 5065 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) { 5066 ixgbe_init_fdir_perfect_82599(&adapter->hw, 5067 adapter->fdir_pballoc); 5068 ixgbe_fdir_filter_restore(adapter); 5069 } 5070 5071 switch (hw->mac.type) { 5072 case ixgbe_mac_82599EB: 5073 case ixgbe_mac_X540: 5074 hw->mac.ops.enable_rx_buff(hw); 5075 break; 5076 default: 5077 break; 5078 } 5079 5080 #ifdef CONFIG_IXGBE_DCA 5081 /* configure DCA */ 5082 if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE) 5083 ixgbe_setup_dca(adapter); 5084 #endif /* CONFIG_IXGBE_DCA */ 5085 5086 #ifdef IXGBE_FCOE 5087 /* configure FCoE L2 filters, redirection table, and Rx control */ 5088 ixgbe_configure_fcoe(adapter); 5089 5090 #endif /* IXGBE_FCOE */ 5091 ixgbe_configure_tx(adapter); 5092 ixgbe_configure_rx(adapter); 5093 ixgbe_configure_dfwd(adapter); 5094 } 5095 5096 /** 5097 * ixgbe_sfp_link_config - set up SFP+ link 5098 * @adapter: pointer to private adapter struct 5099 **/ 5100 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter) 5101 { 5102 /* 5103 * We are assuming the worst case scenario here, and that 5104 * is that an SFP was inserted/removed after the reset 5105 * but before SFP detection was enabled. As such the best 5106 * solution is to just start searching as soon as we start 5107 */ 5108 if (adapter->hw.mac.type == ixgbe_mac_82598EB) 5109 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP; 5110 5111 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET; 5112 adapter->sfp_poll_time = 0; 5113 } 5114 5115 /** 5116 * ixgbe_non_sfp_link_config - set up non-SFP+ link 5117 * @hw: pointer to private hardware struct 5118 * 5119 * Returns 0 on success, negative on failure 5120 **/ 5121 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw) 5122 { 5123 u32 speed; 5124 bool autoneg, link_up = false; 5125 int ret = IXGBE_ERR_LINK_SETUP; 5126 5127 if (hw->mac.ops.check_link) 5128 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false); 5129 5130 if (ret) 5131 return ret; 5132 5133 speed = hw->phy.autoneg_advertised; 5134 if ((!speed) && (hw->mac.ops.get_link_capabilities)) 5135 ret = hw->mac.ops.get_link_capabilities(hw, &speed, 5136 &autoneg); 5137 if (ret) 5138 return ret; 5139 5140 if (hw->mac.ops.setup_link) 5141 ret = hw->mac.ops.setup_link(hw, speed, link_up); 5142 5143 return ret; 5144 } 5145 5146 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter) 5147 { 5148 struct ixgbe_hw *hw = &adapter->hw; 5149 u32 gpie = 0; 5150 5151 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) { 5152 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT | 5153 IXGBE_GPIE_OCD; 5154 gpie |= IXGBE_GPIE_EIAME; 5155 /* 5156 * use EIAM to auto-mask when MSI-X interrupt is asserted 5157 * this saves a register write for every interrupt 5158 */ 5159 switch (hw->mac.type) { 5160 case ixgbe_mac_82598EB: 5161 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE); 5162 break; 5163 case ixgbe_mac_82599EB: 5164 case ixgbe_mac_X540: 5165 case ixgbe_mac_X550: 5166 case ixgbe_mac_X550EM_x: 5167 case ixgbe_mac_x550em_a: 5168 default: 5169 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF); 5170 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF); 5171 break; 5172 } 5173 } else { 5174 /* legacy interrupts, use EIAM to auto-mask when reading EICR, 5175 * specifically only auto mask tx and rx interrupts */ 5176 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE); 5177 } 5178 5179 /* XXX: to interrupt immediately for EICS writes, enable this */ 5180 /* gpie |= IXGBE_GPIE_EIMEN; */ 5181 5182 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) { 5183 gpie &= ~IXGBE_GPIE_VTMODE_MASK; 5184 5185 switch (adapter->ring_feature[RING_F_VMDQ].mask) { 5186 case IXGBE_82599_VMDQ_8Q_MASK: 5187 gpie |= IXGBE_GPIE_VTMODE_16; 5188 break; 5189 case IXGBE_82599_VMDQ_4Q_MASK: 5190 gpie |= IXGBE_GPIE_VTMODE_32; 5191 break; 5192 default: 5193 gpie |= IXGBE_GPIE_VTMODE_64; 5194 break; 5195 } 5196 } 5197 5198 /* Enable Thermal over heat sensor interrupt */ 5199 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) { 5200 switch (adapter->hw.mac.type) { 5201 case ixgbe_mac_82599EB: 5202 gpie |= IXGBE_SDP0_GPIEN_8259X; 5203 break; 5204 default: 5205 break; 5206 } 5207 } 5208 5209 /* Enable fan failure interrupt */ 5210 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) 5211 gpie |= IXGBE_SDP1_GPIEN(hw); 5212 5213 switch (hw->mac.type) { 5214 case ixgbe_mac_82599EB: 5215 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X; 5216 break; 5217 case ixgbe_mac_X550EM_x: 5218 case ixgbe_mac_x550em_a: 5219 gpie |= IXGBE_SDP0_GPIEN_X540; 5220 break; 5221 default: 5222 break; 5223 } 5224 5225 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie); 5226 } 5227 5228 static void ixgbe_up_complete(struct ixgbe_adapter *adapter) 5229 { 5230 struct ixgbe_hw *hw = &adapter->hw; 5231 int err; 5232 u32 ctrl_ext; 5233 5234 ixgbe_get_hw_control(adapter); 5235 ixgbe_setup_gpie(adapter); 5236 5237 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) 5238 ixgbe_configure_msix(adapter); 5239 else 5240 ixgbe_configure_msi_and_legacy(adapter); 5241 5242 /* enable the optics for 82599 SFP+ fiber */ 5243 if (hw->mac.ops.enable_tx_laser) 5244 hw->mac.ops.enable_tx_laser(hw); 5245 5246 if (hw->phy.ops.set_phy_power) 5247 hw->phy.ops.set_phy_power(hw, true); 5248 5249 smp_mb__before_atomic(); 5250 clear_bit(__IXGBE_DOWN, &adapter->state); 5251 ixgbe_napi_enable_all(adapter); 5252 5253 if (ixgbe_is_sfp(hw)) { 5254 ixgbe_sfp_link_config(adapter); 5255 } else { 5256 err = ixgbe_non_sfp_link_config(hw); 5257 if (err) 5258 e_err(probe, "link_config FAILED %d\n", err); 5259 } 5260 5261 /* clear any pending interrupts, may auto mask */ 5262 IXGBE_READ_REG(hw, IXGBE_EICR); 5263 ixgbe_irq_enable(adapter, true, true); 5264 5265 /* 5266 * If this adapter has a fan, check to see if we had a failure 5267 * before we enabled the interrupt. 5268 */ 5269 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) { 5270 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP); 5271 if (esdp & IXGBE_ESDP_SDP1) 5272 e_crit(drv, "Fan has stopped, replace the adapter\n"); 5273 } 5274 5275 /* bring the link up in the watchdog, this could race with our first 5276 * link up interrupt but shouldn't be a problem */ 5277 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE; 5278 adapter->link_check_timeout = jiffies; 5279 mod_timer(&adapter->service_timer, jiffies); 5280 5281 /* Set PF Reset Done bit so PF/VF Mail Ops can work */ 5282 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT); 5283 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD; 5284 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext); 5285 } 5286 5287 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter) 5288 { 5289 WARN_ON(in_interrupt()); 5290 /* put off any impending NetWatchDogTimeout */ 5291 netif_trans_update(adapter->netdev); 5292 5293 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state)) 5294 usleep_range(1000, 2000); 5295 ixgbe_down(adapter); 5296 /* 5297 * If SR-IOV enabled then wait a bit before bringing the adapter 5298 * back up to give the VFs time to respond to the reset. The 5299 * two second wait is based upon the watchdog timer cycle in 5300 * the VF driver. 5301 */ 5302 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) 5303 msleep(2000); 5304 ixgbe_up(adapter); 5305 clear_bit(__IXGBE_RESETTING, &adapter->state); 5306 } 5307 5308 void ixgbe_up(struct ixgbe_adapter *adapter) 5309 { 5310 /* hardware has been reset, we need to reload some things */ 5311 ixgbe_configure(adapter); 5312 5313 ixgbe_up_complete(adapter); 5314 } 5315 5316 void ixgbe_reset(struct ixgbe_adapter *adapter) 5317 { 5318 struct ixgbe_hw *hw = &adapter->hw; 5319 struct net_device *netdev = adapter->netdev; 5320 int err; 5321 5322 if (ixgbe_removed(hw->hw_addr)) 5323 return; 5324 /* lock SFP init bit to prevent race conditions with the watchdog */ 5325 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state)) 5326 usleep_range(1000, 2000); 5327 5328 /* clear all SFP and link config related flags while holding SFP_INIT */ 5329 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP | 5330 IXGBE_FLAG2_SFP_NEEDS_RESET); 5331 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG; 5332 5333 err = hw->mac.ops.init_hw(hw); 5334 switch (err) { 5335 case 0: 5336 case IXGBE_ERR_SFP_NOT_PRESENT: 5337 case IXGBE_ERR_SFP_NOT_SUPPORTED: 5338 break; 5339 case IXGBE_ERR_MASTER_REQUESTS_PENDING: 5340 e_dev_err("master disable timed out\n"); 5341 break; 5342 case IXGBE_ERR_EEPROM_VERSION: 5343 /* We are running on a pre-production device, log a warning */ 5344 e_dev_warn("This device is a pre-production adapter/LOM. " 5345 "Please be aware there may be issues associated with " 5346 "your hardware. If you are experiencing problems " 5347 "please contact your Intel or hardware " 5348 "representative who provided you with this " 5349 "hardware.\n"); 5350 break; 5351 default: 5352 e_dev_err("Hardware Error: %d\n", err); 5353 } 5354 5355 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state); 5356 5357 /* flush entries out of MAC table */ 5358 ixgbe_flush_sw_mac_table(adapter); 5359 __dev_uc_unsync(netdev, NULL); 5360 5361 /* do not flush user set addresses */ 5362 ixgbe_mac_set_default_filter(adapter); 5363 5364 /* update SAN MAC vmdq pool selection */ 5365 if (hw->mac.san_mac_rar_index) 5366 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0)); 5367 5368 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) 5369 ixgbe_ptp_reset(adapter); 5370 5371 if (hw->phy.ops.set_phy_power) { 5372 if (!netif_running(adapter->netdev) && !adapter->wol) 5373 hw->phy.ops.set_phy_power(hw, false); 5374 else 5375 hw->phy.ops.set_phy_power(hw, true); 5376 } 5377 } 5378 5379 /** 5380 * ixgbe_clean_tx_ring - Free Tx Buffers 5381 * @tx_ring: ring to be cleaned 5382 **/ 5383 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring) 5384 { 5385 struct ixgbe_tx_buffer *tx_buffer_info; 5386 unsigned long size; 5387 u16 i; 5388 5389 /* ring already cleared, nothing to do */ 5390 if (!tx_ring->tx_buffer_info) 5391 return; 5392 5393 /* Free all the Tx ring sk_buffs */ 5394 for (i = 0; i < tx_ring->count; i++) { 5395 tx_buffer_info = &tx_ring->tx_buffer_info[i]; 5396 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info); 5397 } 5398 5399 netdev_tx_reset_queue(txring_txq(tx_ring)); 5400 5401 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count; 5402 memset(tx_ring->tx_buffer_info, 0, size); 5403 5404 /* Zero out the descriptor ring */ 5405 memset(tx_ring->desc, 0, tx_ring->size); 5406 5407 tx_ring->next_to_use = 0; 5408 tx_ring->next_to_clean = 0; 5409 } 5410 5411 /** 5412 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues 5413 * @adapter: board private structure 5414 **/ 5415 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter) 5416 { 5417 int i; 5418 5419 for (i = 0; i < adapter->num_rx_queues; i++) 5420 ixgbe_clean_rx_ring(adapter->rx_ring[i]); 5421 } 5422 5423 /** 5424 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues 5425 * @adapter: board private structure 5426 **/ 5427 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter) 5428 { 5429 int i; 5430 5431 for (i = 0; i < adapter->num_tx_queues; i++) 5432 ixgbe_clean_tx_ring(adapter->tx_ring[i]); 5433 } 5434 5435 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter) 5436 { 5437 struct hlist_node *node2; 5438 struct ixgbe_fdir_filter *filter; 5439 5440 spin_lock(&adapter->fdir_perfect_lock); 5441 5442 hlist_for_each_entry_safe(filter, node2, 5443 &adapter->fdir_filter_list, fdir_node) { 5444 hlist_del(&filter->fdir_node); 5445 kfree(filter); 5446 } 5447 adapter->fdir_filter_count = 0; 5448 5449 spin_unlock(&adapter->fdir_perfect_lock); 5450 } 5451 5452 void ixgbe_down(struct ixgbe_adapter *adapter) 5453 { 5454 struct net_device *netdev = adapter->netdev; 5455 struct ixgbe_hw *hw = &adapter->hw; 5456 struct net_device *upper; 5457 struct list_head *iter; 5458 int i; 5459 5460 /* signal that we are down to the interrupt handler */ 5461 if (test_and_set_bit(__IXGBE_DOWN, &adapter->state)) 5462 return; /* do nothing if already down */ 5463 5464 /* disable receives */ 5465 hw->mac.ops.disable_rx(hw); 5466 5467 /* disable all enabled rx queues */ 5468 for (i = 0; i < adapter->num_rx_queues; i++) 5469 /* this call also flushes the previous write */ 5470 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]); 5471 5472 usleep_range(10000, 20000); 5473 5474 netif_tx_stop_all_queues(netdev); 5475 5476 /* call carrier off first to avoid false dev_watchdog timeouts */ 5477 netif_carrier_off(netdev); 5478 netif_tx_disable(netdev); 5479 5480 /* disable any upper devices */ 5481 netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) { 5482 if (netif_is_macvlan(upper)) { 5483 struct macvlan_dev *vlan = netdev_priv(upper); 5484 5485 if (vlan->fwd_priv) { 5486 netif_tx_stop_all_queues(upper); 5487 netif_carrier_off(upper); 5488 netif_tx_disable(upper); 5489 } 5490 } 5491 } 5492 5493 ixgbe_irq_disable(adapter); 5494 5495 ixgbe_napi_disable_all(adapter); 5496 5497 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT | 5498 IXGBE_FLAG2_RESET_REQUESTED); 5499 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE; 5500 5501 del_timer_sync(&adapter->service_timer); 5502 5503 if (adapter->num_vfs) { 5504 /* Clear EITR Select mapping */ 5505 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0); 5506 5507 /* Mark all the VFs as inactive */ 5508 for (i = 0 ; i < adapter->num_vfs; i++) 5509 adapter->vfinfo[i].clear_to_send = false; 5510 5511 /* ping all the active vfs to let them know we are going down */ 5512 ixgbe_ping_all_vfs(adapter); 5513 5514 /* Disable all VFTE/VFRE TX/RX */ 5515 ixgbe_disable_tx_rx(adapter); 5516 } 5517 5518 /* disable transmits in the hardware now that interrupts are off */ 5519 for (i = 0; i < adapter->num_tx_queues; i++) { 5520 u8 reg_idx = adapter->tx_ring[i]->reg_idx; 5521 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH); 5522 } 5523 5524 /* Disable the Tx DMA engine on 82599 and later MAC */ 5525 switch (hw->mac.type) { 5526 case ixgbe_mac_82599EB: 5527 case ixgbe_mac_X540: 5528 case ixgbe_mac_X550: 5529 case ixgbe_mac_X550EM_x: 5530 case ixgbe_mac_x550em_a: 5531 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, 5532 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) & 5533 ~IXGBE_DMATXCTL_TE)); 5534 break; 5535 default: 5536 break; 5537 } 5538 5539 if (!pci_channel_offline(adapter->pdev)) 5540 ixgbe_reset(adapter); 5541 5542 /* power down the optics for 82599 SFP+ fiber */ 5543 if (hw->mac.ops.disable_tx_laser) 5544 hw->mac.ops.disable_tx_laser(hw); 5545 5546 ixgbe_clean_all_tx_rings(adapter); 5547 ixgbe_clean_all_rx_rings(adapter); 5548 } 5549 5550 /** 5551 * ixgbe_tx_timeout - Respond to a Tx Hang 5552 * @netdev: network interface device structure 5553 **/ 5554 static void ixgbe_tx_timeout(struct net_device *netdev) 5555 { 5556 struct ixgbe_adapter *adapter = netdev_priv(netdev); 5557 5558 /* Do the reset outside of interrupt context */ 5559 ixgbe_tx_timeout_reset(adapter); 5560 } 5561 5562 #ifdef CONFIG_IXGBE_DCB 5563 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter) 5564 { 5565 struct ixgbe_hw *hw = &adapter->hw; 5566 struct tc_configuration *tc; 5567 int j; 5568 5569 switch (hw->mac.type) { 5570 case ixgbe_mac_82598EB: 5571 case ixgbe_mac_82599EB: 5572 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS; 5573 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS; 5574 break; 5575 case ixgbe_mac_X540: 5576 case ixgbe_mac_X550: 5577 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS; 5578 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS; 5579 break; 5580 case ixgbe_mac_X550EM_x: 5581 case ixgbe_mac_x550em_a: 5582 default: 5583 adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS; 5584 adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS; 5585 break; 5586 } 5587 5588 /* Configure DCB traffic classes */ 5589 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) { 5590 tc = &adapter->dcb_cfg.tc_config[j]; 5591 tc->path[DCB_TX_CONFIG].bwg_id = 0; 5592 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1); 5593 tc->path[DCB_RX_CONFIG].bwg_id = 0; 5594 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1); 5595 tc->dcb_pfc = pfc_disabled; 5596 } 5597 5598 /* Initialize default user to priority mapping, UPx->TC0 */ 5599 tc = &adapter->dcb_cfg.tc_config[0]; 5600 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF; 5601 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF; 5602 5603 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100; 5604 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100; 5605 adapter->dcb_cfg.pfc_mode_enable = false; 5606 adapter->dcb_set_bitmap = 0x00; 5607 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE) 5608 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE; 5609 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg, 5610 sizeof(adapter->temp_dcb_cfg)); 5611 } 5612 #endif 5613 5614 /** 5615 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter) 5616 * @adapter: board private structure to initialize 5617 * 5618 * ixgbe_sw_init initializes the Adapter private data structure. 5619 * Fields are initialized based on PCI device information and 5620 * OS network device settings (MTU size). 5621 **/ 5622 static int ixgbe_sw_init(struct ixgbe_adapter *adapter) 5623 { 5624 struct ixgbe_hw *hw = &adapter->hw; 5625 struct pci_dev *pdev = adapter->pdev; 5626 unsigned int rss, fdir; 5627 u32 fwsm; 5628 u16 device_caps; 5629 int i; 5630 5631 /* PCI config space info */ 5632 5633 hw->vendor_id = pdev->vendor; 5634 hw->device_id = pdev->device; 5635 hw->revision_id = pdev->revision; 5636 hw->subsystem_vendor_id = pdev->subsystem_vendor; 5637 hw->subsystem_device_id = pdev->subsystem_device; 5638 5639 /* Set common capability flags and settings */ 5640 rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus()); 5641 adapter->ring_feature[RING_F_RSS].limit = rss; 5642 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE; 5643 adapter->max_q_vectors = MAX_Q_VECTORS_82599; 5644 adapter->atr_sample_rate = 20; 5645 fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus()); 5646 adapter->ring_feature[RING_F_FDIR].limit = fdir; 5647 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K; 5648 #ifdef CONFIG_IXGBE_DCA 5649 adapter->flags |= IXGBE_FLAG_DCA_CAPABLE; 5650 #endif 5651 #ifdef CONFIG_IXGBE_DCB 5652 adapter->flags |= IXGBE_FLAG_DCB_CAPABLE; 5653 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED; 5654 #endif 5655 #ifdef IXGBE_FCOE 5656 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE; 5657 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED; 5658 #ifdef CONFIG_IXGBE_DCB 5659 /* Default traffic class to use for FCoE */ 5660 adapter->fcoe.up = IXGBE_FCOE_DEFTC; 5661 #endif /* CONFIG_IXGBE_DCB */ 5662 #endif /* IXGBE_FCOE */ 5663 5664 /* initialize static ixgbe jump table entries */ 5665 adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]), 5666 GFP_KERNEL); 5667 if (!adapter->jump_tables[0]) 5668 return -ENOMEM; 5669 adapter->jump_tables[0]->mat = ixgbe_ipv4_fields; 5670 5671 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) 5672 adapter->jump_tables[i] = NULL; 5673 5674 adapter->mac_table = kzalloc(sizeof(struct ixgbe_mac_addr) * 5675 hw->mac.num_rar_entries, 5676 GFP_ATOMIC); 5677 if (!adapter->mac_table) 5678 return -ENOMEM; 5679 5680 /* Set MAC specific capability flags and exceptions */ 5681 switch (hw->mac.type) { 5682 case ixgbe_mac_82598EB: 5683 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE; 5684 5685 if (hw->device_id == IXGBE_DEV_ID_82598AT) 5686 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE; 5687 5688 adapter->max_q_vectors = MAX_Q_VECTORS_82598; 5689 adapter->ring_feature[RING_F_FDIR].limit = 0; 5690 adapter->atr_sample_rate = 0; 5691 adapter->fdir_pballoc = 0; 5692 #ifdef IXGBE_FCOE 5693 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE; 5694 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED; 5695 #ifdef CONFIG_IXGBE_DCB 5696 adapter->fcoe.up = 0; 5697 #endif /* IXGBE_DCB */ 5698 #endif /* IXGBE_FCOE */ 5699 break; 5700 case ixgbe_mac_82599EB: 5701 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM) 5702 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE; 5703 break; 5704 case ixgbe_mac_X540: 5705 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw)); 5706 if (fwsm & IXGBE_FWSM_TS_ENABLED) 5707 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE; 5708 break; 5709 case ixgbe_mac_X550EM_x: 5710 case ixgbe_mac_x550em_a: 5711 #ifdef CONFIG_IXGBE_DCB 5712 adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE; 5713 #endif 5714 #ifdef IXGBE_FCOE 5715 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE; 5716 #ifdef CONFIG_IXGBE_DCB 5717 adapter->fcoe.up = 0; 5718 #endif /* IXGBE_DCB */ 5719 #endif /* IXGBE_FCOE */ 5720 /* Fall Through */ 5721 case ixgbe_mac_X550: 5722 #ifdef CONFIG_IXGBE_DCA 5723 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE; 5724 #endif 5725 #ifdef CONFIG_IXGBE_VXLAN 5726 adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE; 5727 #endif 5728 break; 5729 default: 5730 break; 5731 } 5732 5733 #ifdef IXGBE_FCOE 5734 /* FCoE support exists, always init the FCoE lock */ 5735 spin_lock_init(&adapter->fcoe.lock); 5736 5737 #endif 5738 /* n-tuple support exists, always init our spinlock */ 5739 spin_lock_init(&adapter->fdir_perfect_lock); 5740 5741 #ifdef CONFIG_IXGBE_DCB 5742 ixgbe_init_dcb(adapter); 5743 #endif 5744 5745 /* default flow control settings */ 5746 hw->fc.requested_mode = ixgbe_fc_full; 5747 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */ 5748 ixgbe_pbthresh_setup(adapter); 5749 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE; 5750 hw->fc.send_xon = true; 5751 hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw); 5752 5753 #ifdef CONFIG_PCI_IOV 5754 if (max_vfs > 0) 5755 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n"); 5756 5757 /* assign number of SR-IOV VFs */ 5758 if (hw->mac.type != ixgbe_mac_82598EB) { 5759 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) { 5760 adapter->num_vfs = 0; 5761 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n"); 5762 } else { 5763 adapter->num_vfs = max_vfs; 5764 } 5765 } 5766 #endif /* CONFIG_PCI_IOV */ 5767 5768 /* enable itr by default in dynamic mode */ 5769 adapter->rx_itr_setting = 1; 5770 adapter->tx_itr_setting = 1; 5771 5772 /* set default ring sizes */ 5773 adapter->tx_ring_count = IXGBE_DEFAULT_TXD; 5774 adapter->rx_ring_count = IXGBE_DEFAULT_RXD; 5775 5776 /* Cache bit indicating need for crosstalk fix */ 5777 switch (hw->mac.type) { 5778 case ixgbe_mac_82599EB: 5779 case ixgbe_mac_X550EM_x: 5780 case ixgbe_mac_x550em_a: 5781 hw->mac.ops.get_device_caps(hw, &device_caps); 5782 if (device_caps & IXGBE_DEVICE_CAPS_NO_CROSSTALK_WR) 5783 adapter->need_crosstalk_fix = false; 5784 else 5785 adapter->need_crosstalk_fix = true; 5786 break; 5787 default: 5788 adapter->need_crosstalk_fix = false; 5789 break; 5790 } 5791 5792 /* set default work limits */ 5793 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK; 5794 5795 /* initialize eeprom parameters */ 5796 if (ixgbe_init_eeprom_params_generic(hw)) { 5797 e_dev_err("EEPROM initialization failed\n"); 5798 return -EIO; 5799 } 5800 5801 /* PF holds first pool slot */ 5802 set_bit(0, &adapter->fwd_bitmask); 5803 set_bit(__IXGBE_DOWN, &adapter->state); 5804 5805 return 0; 5806 } 5807 5808 /** 5809 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors) 5810 * @tx_ring: tx descriptor ring (for a specific queue) to setup 5811 * 5812 * Return 0 on success, negative on failure 5813 **/ 5814 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring) 5815 { 5816 struct device *dev = tx_ring->dev; 5817 int orig_node = dev_to_node(dev); 5818 int ring_node = -1; 5819 int size; 5820 5821 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count; 5822 5823 if (tx_ring->q_vector) 5824 ring_node = tx_ring->q_vector->numa_node; 5825 5826 tx_ring->tx_buffer_info = vzalloc_node(size, ring_node); 5827 if (!tx_ring->tx_buffer_info) 5828 tx_ring->tx_buffer_info = vzalloc(size); 5829 if (!tx_ring->tx_buffer_info) 5830 goto err; 5831 5832 u64_stats_init(&tx_ring->syncp); 5833 5834 /* round up to nearest 4K */ 5835 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc); 5836 tx_ring->size = ALIGN(tx_ring->size, 4096); 5837 5838 set_dev_node(dev, ring_node); 5839 tx_ring->desc = dma_alloc_coherent(dev, 5840 tx_ring->size, 5841 &tx_ring->dma, 5842 GFP_KERNEL); 5843 set_dev_node(dev, orig_node); 5844 if (!tx_ring->desc) 5845 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size, 5846 &tx_ring->dma, GFP_KERNEL); 5847 if (!tx_ring->desc) 5848 goto err; 5849 5850 tx_ring->next_to_use = 0; 5851 tx_ring->next_to_clean = 0; 5852 return 0; 5853 5854 err: 5855 vfree(tx_ring->tx_buffer_info); 5856 tx_ring->tx_buffer_info = NULL; 5857 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n"); 5858 return -ENOMEM; 5859 } 5860 5861 /** 5862 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources 5863 * @adapter: board private structure 5864 * 5865 * If this function returns with an error, then it's possible one or 5866 * more of the rings is populated (while the rest are not). It is the 5867 * callers duty to clean those orphaned rings. 5868 * 5869 * Return 0 on success, negative on failure 5870 **/ 5871 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter) 5872 { 5873 int i, err = 0; 5874 5875 for (i = 0; i < adapter->num_tx_queues; i++) { 5876 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]); 5877 if (!err) 5878 continue; 5879 5880 e_err(probe, "Allocation for Tx Queue %u failed\n", i); 5881 goto err_setup_tx; 5882 } 5883 5884 return 0; 5885 err_setup_tx: 5886 /* rewind the index freeing the rings as we go */ 5887 while (i--) 5888 ixgbe_free_tx_resources(adapter->tx_ring[i]); 5889 return err; 5890 } 5891 5892 /** 5893 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors) 5894 * @rx_ring: rx descriptor ring (for a specific queue) to setup 5895 * 5896 * Returns 0 on success, negative on failure 5897 **/ 5898 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring) 5899 { 5900 struct device *dev = rx_ring->dev; 5901 int orig_node = dev_to_node(dev); 5902 int ring_node = -1; 5903 int size; 5904 5905 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count; 5906 5907 if (rx_ring->q_vector) 5908 ring_node = rx_ring->q_vector->numa_node; 5909 5910 rx_ring->rx_buffer_info = vzalloc_node(size, ring_node); 5911 if (!rx_ring->rx_buffer_info) 5912 rx_ring->rx_buffer_info = vzalloc(size); 5913 if (!rx_ring->rx_buffer_info) 5914 goto err; 5915 5916 u64_stats_init(&rx_ring->syncp); 5917 5918 /* Round up to nearest 4K */ 5919 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc); 5920 rx_ring->size = ALIGN(rx_ring->size, 4096); 5921 5922 set_dev_node(dev, ring_node); 5923 rx_ring->desc = dma_alloc_coherent(dev, 5924 rx_ring->size, 5925 &rx_ring->dma, 5926 GFP_KERNEL); 5927 set_dev_node(dev, orig_node); 5928 if (!rx_ring->desc) 5929 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size, 5930 &rx_ring->dma, GFP_KERNEL); 5931 if (!rx_ring->desc) 5932 goto err; 5933 5934 rx_ring->next_to_clean = 0; 5935 rx_ring->next_to_use = 0; 5936 5937 return 0; 5938 err: 5939 vfree(rx_ring->rx_buffer_info); 5940 rx_ring->rx_buffer_info = NULL; 5941 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n"); 5942 return -ENOMEM; 5943 } 5944 5945 /** 5946 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources 5947 * @adapter: board private structure 5948 * 5949 * If this function returns with an error, then it's possible one or 5950 * more of the rings is populated (while the rest are not). It is the 5951 * callers duty to clean those orphaned rings. 5952 * 5953 * Return 0 on success, negative on failure 5954 **/ 5955 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter) 5956 { 5957 int i, err = 0; 5958 5959 for (i = 0; i < adapter->num_rx_queues; i++) { 5960 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]); 5961 if (!err) 5962 continue; 5963 5964 e_err(probe, "Allocation for Rx Queue %u failed\n", i); 5965 goto err_setup_rx; 5966 } 5967 5968 #ifdef IXGBE_FCOE 5969 err = ixgbe_setup_fcoe_ddp_resources(adapter); 5970 if (!err) 5971 #endif 5972 return 0; 5973 err_setup_rx: 5974 /* rewind the index freeing the rings as we go */ 5975 while (i--) 5976 ixgbe_free_rx_resources(adapter->rx_ring[i]); 5977 return err; 5978 } 5979 5980 /** 5981 * ixgbe_free_tx_resources - Free Tx Resources per Queue 5982 * @tx_ring: Tx descriptor ring for a specific queue 5983 * 5984 * Free all transmit software resources 5985 **/ 5986 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring) 5987 { 5988 ixgbe_clean_tx_ring(tx_ring); 5989 5990 vfree(tx_ring->tx_buffer_info); 5991 tx_ring->tx_buffer_info = NULL; 5992 5993 /* if not set, then don't free */ 5994 if (!tx_ring->desc) 5995 return; 5996 5997 dma_free_coherent(tx_ring->dev, tx_ring->size, 5998 tx_ring->desc, tx_ring->dma); 5999 6000 tx_ring->desc = NULL; 6001 } 6002 6003 /** 6004 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues 6005 * @adapter: board private structure 6006 * 6007 * Free all transmit software resources 6008 **/ 6009 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter) 6010 { 6011 int i; 6012 6013 for (i = 0; i < adapter->num_tx_queues; i++) 6014 if (adapter->tx_ring[i]->desc) 6015 ixgbe_free_tx_resources(adapter->tx_ring[i]); 6016 } 6017 6018 /** 6019 * ixgbe_free_rx_resources - Free Rx Resources 6020 * @rx_ring: ring to clean the resources from 6021 * 6022 * Free all receive software resources 6023 **/ 6024 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring) 6025 { 6026 ixgbe_clean_rx_ring(rx_ring); 6027 6028 vfree(rx_ring->rx_buffer_info); 6029 rx_ring->rx_buffer_info = NULL; 6030 6031 /* if not set, then don't free */ 6032 if (!rx_ring->desc) 6033 return; 6034 6035 dma_free_coherent(rx_ring->dev, rx_ring->size, 6036 rx_ring->desc, rx_ring->dma); 6037 6038 rx_ring->desc = NULL; 6039 } 6040 6041 /** 6042 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues 6043 * @adapter: board private structure 6044 * 6045 * Free all receive software resources 6046 **/ 6047 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter) 6048 { 6049 int i; 6050 6051 #ifdef IXGBE_FCOE 6052 ixgbe_free_fcoe_ddp_resources(adapter); 6053 6054 #endif 6055 for (i = 0; i < adapter->num_rx_queues; i++) 6056 if (adapter->rx_ring[i]->desc) 6057 ixgbe_free_rx_resources(adapter->rx_ring[i]); 6058 } 6059 6060 /** 6061 * ixgbe_change_mtu - Change the Maximum Transfer Unit 6062 * @netdev: network interface device structure 6063 * @new_mtu: new value for maximum frame size 6064 * 6065 * Returns 0 on success, negative on failure 6066 **/ 6067 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu) 6068 { 6069 struct ixgbe_adapter *adapter = netdev_priv(netdev); 6070 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN; 6071 6072 /* MTU < 68 is an error and causes problems on some kernels */ 6073 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE)) 6074 return -EINVAL; 6075 6076 /* 6077 * For 82599EB we cannot allow legacy VFs to enable their receive 6078 * paths when MTU greater than 1500 is configured. So display a 6079 * warning that legacy VFs will be disabled. 6080 */ 6081 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && 6082 (adapter->hw.mac.type == ixgbe_mac_82599EB) && 6083 (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN))) 6084 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n"); 6085 6086 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu); 6087 6088 /* must set new MTU before calling down or up */ 6089 netdev->mtu = new_mtu; 6090 6091 if (netif_running(netdev)) 6092 ixgbe_reinit_locked(adapter); 6093 6094 return 0; 6095 } 6096 6097 /** 6098 * ixgbe_open - Called when a network interface is made active 6099 * @netdev: network interface device structure 6100 * 6101 * Returns 0 on success, negative value on failure 6102 * 6103 * The open entry point is called when a network interface is made 6104 * active by the system (IFF_UP). At this point all resources needed 6105 * for transmit and receive operations are allocated, the interrupt 6106 * handler is registered with the OS, the watchdog timer is started, 6107 * and the stack is notified that the interface is ready. 6108 **/ 6109 int ixgbe_open(struct net_device *netdev) 6110 { 6111 struct ixgbe_adapter *adapter = netdev_priv(netdev); 6112 struct ixgbe_hw *hw = &adapter->hw; 6113 int err, queues; 6114 6115 /* disallow open during test */ 6116 if (test_bit(__IXGBE_TESTING, &adapter->state)) 6117 return -EBUSY; 6118 6119 netif_carrier_off(netdev); 6120 6121 /* allocate transmit descriptors */ 6122 err = ixgbe_setup_all_tx_resources(adapter); 6123 if (err) 6124 goto err_setup_tx; 6125 6126 /* allocate receive descriptors */ 6127 err = ixgbe_setup_all_rx_resources(adapter); 6128 if (err) 6129 goto err_setup_rx; 6130 6131 ixgbe_configure(adapter); 6132 6133 err = ixgbe_request_irq(adapter); 6134 if (err) 6135 goto err_req_irq; 6136 6137 /* Notify the stack of the actual queue counts. */ 6138 if (adapter->num_rx_pools > 1) 6139 queues = adapter->num_rx_queues_per_pool; 6140 else 6141 queues = adapter->num_tx_queues; 6142 6143 err = netif_set_real_num_tx_queues(netdev, queues); 6144 if (err) 6145 goto err_set_queues; 6146 6147 if (adapter->num_rx_pools > 1 && 6148 adapter->num_rx_queues > IXGBE_MAX_L2A_QUEUES) 6149 queues = IXGBE_MAX_L2A_QUEUES; 6150 else 6151 queues = adapter->num_rx_queues; 6152 err = netif_set_real_num_rx_queues(netdev, queues); 6153 if (err) 6154 goto err_set_queues; 6155 6156 ixgbe_ptp_init(adapter); 6157 6158 ixgbe_up_complete(adapter); 6159 6160 ixgbe_clear_vxlan_port(adapter); 6161 #ifdef CONFIG_IXGBE_VXLAN 6162 vxlan_get_rx_port(netdev); 6163 #endif 6164 6165 return 0; 6166 6167 err_set_queues: 6168 ixgbe_free_irq(adapter); 6169 err_req_irq: 6170 ixgbe_free_all_rx_resources(adapter); 6171 if (hw->phy.ops.set_phy_power && !adapter->wol) 6172 hw->phy.ops.set_phy_power(&adapter->hw, false); 6173 err_setup_rx: 6174 ixgbe_free_all_tx_resources(adapter); 6175 err_setup_tx: 6176 ixgbe_reset(adapter); 6177 6178 return err; 6179 } 6180 6181 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter) 6182 { 6183 ixgbe_ptp_suspend(adapter); 6184 6185 if (adapter->hw.phy.ops.enter_lplu) { 6186 adapter->hw.phy.reset_disable = true; 6187 ixgbe_down(adapter); 6188 adapter->hw.phy.ops.enter_lplu(&adapter->hw); 6189 adapter->hw.phy.reset_disable = false; 6190 } else { 6191 ixgbe_down(adapter); 6192 } 6193 6194 ixgbe_free_irq(adapter); 6195 6196 ixgbe_free_all_tx_resources(adapter); 6197 ixgbe_free_all_rx_resources(adapter); 6198 } 6199 6200 /** 6201 * ixgbe_close - Disables a network interface 6202 * @netdev: network interface device structure 6203 * 6204 * Returns 0, this is not allowed to fail 6205 * 6206 * The close entry point is called when an interface is de-activated 6207 * by the OS. The hardware is still under the drivers control, but 6208 * needs to be disabled. A global MAC reset is issued to stop the 6209 * hardware, and all transmit and receive resources are freed. 6210 **/ 6211 int ixgbe_close(struct net_device *netdev) 6212 { 6213 struct ixgbe_adapter *adapter = netdev_priv(netdev); 6214 6215 ixgbe_ptp_stop(adapter); 6216 6217 ixgbe_close_suspend(adapter); 6218 6219 ixgbe_fdir_filter_exit(adapter); 6220 6221 ixgbe_release_hw_control(adapter); 6222 6223 return 0; 6224 } 6225 6226 #ifdef CONFIG_PM 6227 static int ixgbe_resume(struct pci_dev *pdev) 6228 { 6229 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev); 6230 struct net_device *netdev = adapter->netdev; 6231 u32 err; 6232 6233 adapter->hw.hw_addr = adapter->io_addr; 6234 pci_set_power_state(pdev, PCI_D0); 6235 pci_restore_state(pdev); 6236 /* 6237 * pci_restore_state clears dev->state_saved so call 6238 * pci_save_state to restore it. 6239 */ 6240 pci_save_state(pdev); 6241 6242 err = pci_enable_device_mem(pdev); 6243 if (err) { 6244 e_dev_err("Cannot enable PCI device from suspend\n"); 6245 return err; 6246 } 6247 smp_mb__before_atomic(); 6248 clear_bit(__IXGBE_DISABLED, &adapter->state); 6249 pci_set_master(pdev); 6250 6251 pci_wake_from_d3(pdev, false); 6252 6253 ixgbe_reset(adapter); 6254 6255 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0); 6256 6257 rtnl_lock(); 6258 err = ixgbe_init_interrupt_scheme(adapter); 6259 if (!err && netif_running(netdev)) 6260 err = ixgbe_open(netdev); 6261 6262 rtnl_unlock(); 6263 6264 if (err) 6265 return err; 6266 6267 netif_device_attach(netdev); 6268 6269 return 0; 6270 } 6271 #endif /* CONFIG_PM */ 6272 6273 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake) 6274 { 6275 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev); 6276 struct net_device *netdev = adapter->netdev; 6277 struct ixgbe_hw *hw = &adapter->hw; 6278 u32 ctrl, fctrl; 6279 u32 wufc = adapter->wol; 6280 #ifdef CONFIG_PM 6281 int retval = 0; 6282 #endif 6283 6284 netif_device_detach(netdev); 6285 6286 rtnl_lock(); 6287 if (netif_running(netdev)) 6288 ixgbe_close_suspend(adapter); 6289 rtnl_unlock(); 6290 6291 ixgbe_clear_interrupt_scheme(adapter); 6292 6293 #ifdef CONFIG_PM 6294 retval = pci_save_state(pdev); 6295 if (retval) 6296 return retval; 6297 6298 #endif 6299 if (hw->mac.ops.stop_link_on_d3) 6300 hw->mac.ops.stop_link_on_d3(hw); 6301 6302 if (wufc) { 6303 ixgbe_set_rx_mode(netdev); 6304 6305 /* enable the optics for 82599 SFP+ fiber as we can WoL */ 6306 if (hw->mac.ops.enable_tx_laser) 6307 hw->mac.ops.enable_tx_laser(hw); 6308 6309 /* turn on all-multi mode if wake on multicast is enabled */ 6310 if (wufc & IXGBE_WUFC_MC) { 6311 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL); 6312 fctrl |= IXGBE_FCTRL_MPE; 6313 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl); 6314 } 6315 6316 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL); 6317 ctrl |= IXGBE_CTRL_GIO_DIS; 6318 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl); 6319 6320 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc); 6321 } else { 6322 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0); 6323 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0); 6324 } 6325 6326 switch (hw->mac.type) { 6327 case ixgbe_mac_82598EB: 6328 pci_wake_from_d3(pdev, false); 6329 break; 6330 case ixgbe_mac_82599EB: 6331 case ixgbe_mac_X540: 6332 case ixgbe_mac_X550: 6333 case ixgbe_mac_X550EM_x: 6334 case ixgbe_mac_x550em_a: 6335 pci_wake_from_d3(pdev, !!wufc); 6336 break; 6337 default: 6338 break; 6339 } 6340 6341 *enable_wake = !!wufc; 6342 if (hw->phy.ops.set_phy_power && !*enable_wake) 6343 hw->phy.ops.set_phy_power(hw, false); 6344 6345 ixgbe_release_hw_control(adapter); 6346 6347 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state)) 6348 pci_disable_device(pdev); 6349 6350 return 0; 6351 } 6352 6353 #ifdef CONFIG_PM 6354 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state) 6355 { 6356 int retval; 6357 bool wake; 6358 6359 retval = __ixgbe_shutdown(pdev, &wake); 6360 if (retval) 6361 return retval; 6362 6363 if (wake) { 6364 pci_prepare_to_sleep(pdev); 6365 } else { 6366 pci_wake_from_d3(pdev, false); 6367 pci_set_power_state(pdev, PCI_D3hot); 6368 } 6369 6370 return 0; 6371 } 6372 #endif /* CONFIG_PM */ 6373 6374 static void ixgbe_shutdown(struct pci_dev *pdev) 6375 { 6376 bool wake; 6377 6378 __ixgbe_shutdown(pdev, &wake); 6379 6380 if (system_state == SYSTEM_POWER_OFF) { 6381 pci_wake_from_d3(pdev, wake); 6382 pci_set_power_state(pdev, PCI_D3hot); 6383 } 6384 } 6385 6386 /** 6387 * ixgbe_update_stats - Update the board statistics counters. 6388 * @adapter: board private structure 6389 **/ 6390 void ixgbe_update_stats(struct ixgbe_adapter *adapter) 6391 { 6392 struct net_device *netdev = adapter->netdev; 6393 struct ixgbe_hw *hw = &adapter->hw; 6394 struct ixgbe_hw_stats *hwstats = &adapter->stats; 6395 u64 total_mpc = 0; 6396 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot; 6397 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0; 6398 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0; 6399 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0; 6400 6401 if (test_bit(__IXGBE_DOWN, &adapter->state) || 6402 test_bit(__IXGBE_RESETTING, &adapter->state)) 6403 return; 6404 6405 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) { 6406 u64 rsc_count = 0; 6407 u64 rsc_flush = 0; 6408 for (i = 0; i < adapter->num_rx_queues; i++) { 6409 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count; 6410 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush; 6411 } 6412 adapter->rsc_total_count = rsc_count; 6413 adapter->rsc_total_flush = rsc_flush; 6414 } 6415 6416 for (i = 0; i < adapter->num_rx_queues; i++) { 6417 struct ixgbe_ring *rx_ring = adapter->rx_ring[i]; 6418 non_eop_descs += rx_ring->rx_stats.non_eop_descs; 6419 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed; 6420 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed; 6421 hw_csum_rx_error += rx_ring->rx_stats.csum_err; 6422 bytes += rx_ring->stats.bytes; 6423 packets += rx_ring->stats.packets; 6424 } 6425 adapter->non_eop_descs = non_eop_descs; 6426 adapter->alloc_rx_page_failed = alloc_rx_page_failed; 6427 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed; 6428 adapter->hw_csum_rx_error = hw_csum_rx_error; 6429 netdev->stats.rx_bytes = bytes; 6430 netdev->stats.rx_packets = packets; 6431 6432 bytes = 0; 6433 packets = 0; 6434 /* gather some stats to the adapter struct that are per queue */ 6435 for (i = 0; i < adapter->num_tx_queues; i++) { 6436 struct ixgbe_ring *tx_ring = adapter->tx_ring[i]; 6437 restart_queue += tx_ring->tx_stats.restart_queue; 6438 tx_busy += tx_ring->tx_stats.tx_busy; 6439 bytes += tx_ring->stats.bytes; 6440 packets += tx_ring->stats.packets; 6441 } 6442 adapter->restart_queue = restart_queue; 6443 adapter->tx_busy = tx_busy; 6444 netdev->stats.tx_bytes = bytes; 6445 netdev->stats.tx_packets = packets; 6446 6447 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS); 6448 6449 /* 8 register reads */ 6450 for (i = 0; i < 8; i++) { 6451 /* for packet buffers not used, the register should read 0 */ 6452 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i)); 6453 missed_rx += mpc; 6454 hwstats->mpc[i] += mpc; 6455 total_mpc += hwstats->mpc[i]; 6456 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i)); 6457 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i)); 6458 switch (hw->mac.type) { 6459 case ixgbe_mac_82598EB: 6460 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i)); 6461 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i)); 6462 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i)); 6463 hwstats->pxonrxc[i] += 6464 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i)); 6465 break; 6466 case ixgbe_mac_82599EB: 6467 case ixgbe_mac_X540: 6468 case ixgbe_mac_X550: 6469 case ixgbe_mac_X550EM_x: 6470 case ixgbe_mac_x550em_a: 6471 hwstats->pxonrxc[i] += 6472 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i)); 6473 break; 6474 default: 6475 break; 6476 } 6477 } 6478 6479 /*16 register reads */ 6480 for (i = 0; i < 16; i++) { 6481 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i)); 6482 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i)); 6483 if ((hw->mac.type == ixgbe_mac_82599EB) || 6484 (hw->mac.type == ixgbe_mac_X540) || 6485 (hw->mac.type == ixgbe_mac_X550) || 6486 (hw->mac.type == ixgbe_mac_X550EM_x) || 6487 (hw->mac.type == ixgbe_mac_x550em_a)) { 6488 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i)); 6489 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */ 6490 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i)); 6491 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */ 6492 } 6493 } 6494 6495 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC); 6496 /* work around hardware counting issue */ 6497 hwstats->gprc -= missed_rx; 6498 6499 ixgbe_update_xoff_received(adapter); 6500 6501 /* 82598 hardware only has a 32 bit counter in the high register */ 6502 switch (hw->mac.type) { 6503 case ixgbe_mac_82598EB: 6504 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC); 6505 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH); 6506 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH); 6507 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH); 6508 break; 6509 case ixgbe_mac_X540: 6510 case ixgbe_mac_X550: 6511 case ixgbe_mac_X550EM_x: 6512 case ixgbe_mac_x550em_a: 6513 /* OS2BMC stats are X540 and later */ 6514 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC); 6515 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC); 6516 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC); 6517 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC); 6518 case ixgbe_mac_82599EB: 6519 for (i = 0; i < 16; i++) 6520 adapter->hw_rx_no_dma_resources += 6521 IXGBE_READ_REG(hw, IXGBE_QPRDC(i)); 6522 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL); 6523 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */ 6524 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL); 6525 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */ 6526 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL); 6527 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */ 6528 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT); 6529 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH); 6530 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS); 6531 #ifdef IXGBE_FCOE 6532 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC); 6533 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC); 6534 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC); 6535 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC); 6536 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC); 6537 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC); 6538 /* Add up per cpu counters for total ddp aloc fail */ 6539 if (adapter->fcoe.ddp_pool) { 6540 struct ixgbe_fcoe *fcoe = &adapter->fcoe; 6541 struct ixgbe_fcoe_ddp_pool *ddp_pool; 6542 unsigned int cpu; 6543 u64 noddp = 0, noddp_ext_buff = 0; 6544 for_each_possible_cpu(cpu) { 6545 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu); 6546 noddp += ddp_pool->noddp; 6547 noddp_ext_buff += ddp_pool->noddp_ext_buff; 6548 } 6549 hwstats->fcoe_noddp = noddp; 6550 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff; 6551 } 6552 #endif /* IXGBE_FCOE */ 6553 break; 6554 default: 6555 break; 6556 } 6557 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC); 6558 hwstats->bprc += bprc; 6559 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC); 6560 if (hw->mac.type == ixgbe_mac_82598EB) 6561 hwstats->mprc -= bprc; 6562 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC); 6563 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64); 6564 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127); 6565 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255); 6566 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511); 6567 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023); 6568 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522); 6569 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC); 6570 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC); 6571 hwstats->lxontxc += lxon; 6572 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC); 6573 hwstats->lxofftxc += lxoff; 6574 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC); 6575 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC); 6576 /* 6577 * 82598 errata - tx of flow control packets is included in tx counters 6578 */ 6579 xon_off_tot = lxon + lxoff; 6580 hwstats->gptc -= xon_off_tot; 6581 hwstats->mptc -= xon_off_tot; 6582 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN)); 6583 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC); 6584 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC); 6585 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC); 6586 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR); 6587 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64); 6588 hwstats->ptc64 -= xon_off_tot; 6589 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127); 6590 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255); 6591 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511); 6592 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023); 6593 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522); 6594 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC); 6595 6596 /* Fill out the OS statistics structure */ 6597 netdev->stats.multicast = hwstats->mprc; 6598 6599 /* Rx Errors */ 6600 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec; 6601 netdev->stats.rx_dropped = 0; 6602 netdev->stats.rx_length_errors = hwstats->rlec; 6603 netdev->stats.rx_crc_errors = hwstats->crcerrs; 6604 netdev->stats.rx_missed_errors = total_mpc; 6605 } 6606 6607 /** 6608 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table 6609 * @adapter: pointer to the device adapter structure 6610 **/ 6611 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter) 6612 { 6613 struct ixgbe_hw *hw = &adapter->hw; 6614 int i; 6615 6616 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT)) 6617 return; 6618 6619 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT; 6620 6621 /* if interface is down do nothing */ 6622 if (test_bit(__IXGBE_DOWN, &adapter->state)) 6623 return; 6624 6625 /* do nothing if we are not using signature filters */ 6626 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)) 6627 return; 6628 6629 adapter->fdir_overflow++; 6630 6631 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) { 6632 for (i = 0; i < adapter->num_tx_queues; i++) 6633 set_bit(__IXGBE_TX_FDIR_INIT_DONE, 6634 &(adapter->tx_ring[i]->state)); 6635 /* re-enable flow director interrupts */ 6636 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR); 6637 } else { 6638 e_err(probe, "failed to finish FDIR re-initialization, " 6639 "ignored adding FDIR ATR filters\n"); 6640 } 6641 } 6642 6643 /** 6644 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts 6645 * @adapter: pointer to the device adapter structure 6646 * 6647 * This function serves two purposes. First it strobes the interrupt lines 6648 * in order to make certain interrupts are occurring. Secondly it sets the 6649 * bits needed to check for TX hangs. As a result we should immediately 6650 * determine if a hang has occurred. 6651 */ 6652 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter) 6653 { 6654 struct ixgbe_hw *hw = &adapter->hw; 6655 u64 eics = 0; 6656 int i; 6657 6658 /* If we're down, removing or resetting, just bail */ 6659 if (test_bit(__IXGBE_DOWN, &adapter->state) || 6660 test_bit(__IXGBE_REMOVING, &adapter->state) || 6661 test_bit(__IXGBE_RESETTING, &adapter->state)) 6662 return; 6663 6664 /* Force detection of hung controller */ 6665 if (netif_carrier_ok(adapter->netdev)) { 6666 for (i = 0; i < adapter->num_tx_queues; i++) 6667 set_check_for_tx_hang(adapter->tx_ring[i]); 6668 } 6669 6670 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) { 6671 /* 6672 * for legacy and MSI interrupts don't set any bits 6673 * that are enabled for EIAM, because this operation 6674 * would set *both* EIMS and EICS for any bit in EIAM 6675 */ 6676 IXGBE_WRITE_REG(hw, IXGBE_EICS, 6677 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER)); 6678 } else { 6679 /* get one bit for every active tx/rx interrupt vector */ 6680 for (i = 0; i < adapter->num_q_vectors; i++) { 6681 struct ixgbe_q_vector *qv = adapter->q_vector[i]; 6682 if (qv->rx.ring || qv->tx.ring) 6683 eics |= BIT_ULL(i); 6684 } 6685 } 6686 6687 /* Cause software interrupt to ensure rings are cleaned */ 6688 ixgbe_irq_rearm_queues(adapter, eics); 6689 } 6690 6691 /** 6692 * ixgbe_watchdog_update_link - update the link status 6693 * @adapter: pointer to the device adapter structure 6694 * @link_speed: pointer to a u32 to store the link_speed 6695 **/ 6696 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter) 6697 { 6698 struct ixgbe_hw *hw = &adapter->hw; 6699 u32 link_speed = adapter->link_speed; 6700 bool link_up = adapter->link_up; 6701 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable; 6702 6703 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)) 6704 return; 6705 6706 if (hw->mac.ops.check_link) { 6707 hw->mac.ops.check_link(hw, &link_speed, &link_up, false); 6708 } else { 6709 /* always assume link is up, if no check link function */ 6710 link_speed = IXGBE_LINK_SPEED_10GB_FULL; 6711 link_up = true; 6712 } 6713 6714 /* If Crosstalk fix enabled do the sanity check of making sure 6715 * the SFP+ cage is empty. 6716 */ 6717 if (adapter->need_crosstalk_fix) { 6718 u32 sfp_cage_full; 6719 6720 sfp_cage_full = IXGBE_READ_REG(hw, IXGBE_ESDP) & 6721 IXGBE_ESDP_SDP2; 6722 if (ixgbe_is_sfp(hw) && link_up && !sfp_cage_full) 6723 link_up = false; 6724 } 6725 6726 if (adapter->ixgbe_ieee_pfc) 6727 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en); 6728 6729 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) { 6730 hw->mac.ops.fc_enable(hw); 6731 ixgbe_set_rx_drop_en(adapter); 6732 } 6733 6734 if (link_up || 6735 time_after(jiffies, (adapter->link_check_timeout + 6736 IXGBE_TRY_LINK_TIMEOUT))) { 6737 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE; 6738 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC); 6739 IXGBE_WRITE_FLUSH(hw); 6740 } 6741 6742 adapter->link_up = link_up; 6743 adapter->link_speed = link_speed; 6744 } 6745 6746 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter) 6747 { 6748 #ifdef CONFIG_IXGBE_DCB 6749 struct net_device *netdev = adapter->netdev; 6750 struct dcb_app app = { 6751 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE, 6752 .protocol = 0, 6753 }; 6754 u8 up = 0; 6755 6756 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE) 6757 up = dcb_ieee_getapp_mask(netdev, &app); 6758 6759 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0; 6760 #endif 6761 } 6762 6763 /** 6764 * ixgbe_watchdog_link_is_up - update netif_carrier status and 6765 * print link up message 6766 * @adapter: pointer to the device adapter structure 6767 **/ 6768 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter) 6769 { 6770 struct net_device *netdev = adapter->netdev; 6771 struct ixgbe_hw *hw = &adapter->hw; 6772 struct net_device *upper; 6773 struct list_head *iter; 6774 u32 link_speed = adapter->link_speed; 6775 const char *speed_str; 6776 bool flow_rx, flow_tx; 6777 6778 /* only continue if link was previously down */ 6779 if (netif_carrier_ok(netdev)) 6780 return; 6781 6782 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP; 6783 6784 switch (hw->mac.type) { 6785 case ixgbe_mac_82598EB: { 6786 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL); 6787 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS); 6788 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE); 6789 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X); 6790 } 6791 break; 6792 case ixgbe_mac_X540: 6793 case ixgbe_mac_X550: 6794 case ixgbe_mac_X550EM_x: 6795 case ixgbe_mac_x550em_a: 6796 case ixgbe_mac_82599EB: { 6797 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN); 6798 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG); 6799 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE); 6800 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X); 6801 } 6802 break; 6803 default: 6804 flow_tx = false; 6805 flow_rx = false; 6806 break; 6807 } 6808 6809 adapter->last_rx_ptp_check = jiffies; 6810 6811 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) 6812 ixgbe_ptp_start_cyclecounter(adapter); 6813 6814 switch (link_speed) { 6815 case IXGBE_LINK_SPEED_10GB_FULL: 6816 speed_str = "10 Gbps"; 6817 break; 6818 case IXGBE_LINK_SPEED_2_5GB_FULL: 6819 speed_str = "2.5 Gbps"; 6820 break; 6821 case IXGBE_LINK_SPEED_1GB_FULL: 6822 speed_str = "1 Gbps"; 6823 break; 6824 case IXGBE_LINK_SPEED_100_FULL: 6825 speed_str = "100 Mbps"; 6826 break; 6827 default: 6828 speed_str = "unknown speed"; 6829 break; 6830 } 6831 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str, 6832 ((flow_rx && flow_tx) ? "RX/TX" : 6833 (flow_rx ? "RX" : 6834 (flow_tx ? "TX" : "None")))); 6835 6836 netif_carrier_on(netdev); 6837 ixgbe_check_vf_rate_limit(adapter); 6838 6839 /* enable transmits */ 6840 netif_tx_wake_all_queues(adapter->netdev); 6841 6842 /* enable any upper devices */ 6843 rtnl_lock(); 6844 netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) { 6845 if (netif_is_macvlan(upper)) { 6846 struct macvlan_dev *vlan = netdev_priv(upper); 6847 6848 if (vlan->fwd_priv) 6849 netif_tx_wake_all_queues(upper); 6850 } 6851 } 6852 rtnl_unlock(); 6853 6854 /* update the default user priority for VFs */ 6855 ixgbe_update_default_up(adapter); 6856 6857 /* ping all the active vfs to let them know link has changed */ 6858 ixgbe_ping_all_vfs(adapter); 6859 } 6860 6861 /** 6862 * ixgbe_watchdog_link_is_down - update netif_carrier status and 6863 * print link down message 6864 * @adapter: pointer to the adapter structure 6865 **/ 6866 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter) 6867 { 6868 struct net_device *netdev = adapter->netdev; 6869 struct ixgbe_hw *hw = &adapter->hw; 6870 6871 adapter->link_up = false; 6872 adapter->link_speed = 0; 6873 6874 /* only continue if link was up previously */ 6875 if (!netif_carrier_ok(netdev)) 6876 return; 6877 6878 /* poll for SFP+ cable when link is down */ 6879 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB) 6880 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP; 6881 6882 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) 6883 ixgbe_ptp_start_cyclecounter(adapter); 6884 6885 e_info(drv, "NIC Link is Down\n"); 6886 netif_carrier_off(netdev); 6887 6888 /* ping all the active vfs to let them know link has changed */ 6889 ixgbe_ping_all_vfs(adapter); 6890 } 6891 6892 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter) 6893 { 6894 int i; 6895 6896 for (i = 0; i < adapter->num_tx_queues; i++) { 6897 struct ixgbe_ring *tx_ring = adapter->tx_ring[i]; 6898 6899 if (tx_ring->next_to_use != tx_ring->next_to_clean) 6900 return true; 6901 } 6902 6903 return false; 6904 } 6905 6906 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter) 6907 { 6908 struct ixgbe_hw *hw = &adapter->hw; 6909 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ]; 6910 u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask); 6911 6912 int i, j; 6913 6914 if (!adapter->num_vfs) 6915 return false; 6916 6917 /* resetting the PF is only needed for MAC before X550 */ 6918 if (hw->mac.type >= ixgbe_mac_X550) 6919 return false; 6920 6921 for (i = 0; i < adapter->num_vfs; i++) { 6922 for (j = 0; j < q_per_pool; j++) { 6923 u32 h, t; 6924 6925 h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j)); 6926 t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j)); 6927 6928 if (h != t) 6929 return true; 6930 } 6931 } 6932 6933 return false; 6934 } 6935 6936 /** 6937 * ixgbe_watchdog_flush_tx - flush queues on link down 6938 * @adapter: pointer to the device adapter structure 6939 **/ 6940 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter) 6941 { 6942 if (!netif_carrier_ok(adapter->netdev)) { 6943 if (ixgbe_ring_tx_pending(adapter) || 6944 ixgbe_vf_tx_pending(adapter)) { 6945 /* We've lost link, so the controller stops DMA, 6946 * but we've got queued Tx work that's never going 6947 * to get done, so reset controller to flush Tx. 6948 * (Do the reset outside of interrupt context). 6949 */ 6950 e_warn(drv, "initiating reset to clear Tx work after link loss\n"); 6951 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED; 6952 } 6953 } 6954 } 6955 6956 #ifdef CONFIG_PCI_IOV 6957 static inline void ixgbe_issue_vf_flr(struct ixgbe_adapter *adapter, 6958 struct pci_dev *vfdev) 6959 { 6960 if (!pci_wait_for_pending_transaction(vfdev)) 6961 e_dev_warn("Issuing VFLR with pending transactions\n"); 6962 6963 e_dev_err("Issuing VFLR for VF %s\n", pci_name(vfdev)); 6964 pcie_capability_set_word(vfdev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR); 6965 6966 msleep(100); 6967 } 6968 6969 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter) 6970 { 6971 struct ixgbe_hw *hw = &adapter->hw; 6972 struct pci_dev *pdev = adapter->pdev; 6973 unsigned int vf; 6974 u32 gpc; 6975 6976 if (!(netif_carrier_ok(adapter->netdev))) 6977 return; 6978 6979 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC); 6980 if (gpc) /* If incrementing then no need for the check below */ 6981 return; 6982 /* Check to see if a bad DMA write target from an errant or 6983 * malicious VF has caused a PCIe error. If so then we can 6984 * issue a VFLR to the offending VF(s) and then resume without 6985 * requesting a full slot reset. 6986 */ 6987 6988 if (!pdev) 6989 return; 6990 6991 /* check status reg for all VFs owned by this PF */ 6992 for (vf = 0; vf < adapter->num_vfs; ++vf) { 6993 struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev; 6994 u16 status_reg; 6995 6996 if (!vfdev) 6997 continue; 6998 pci_read_config_word(vfdev, PCI_STATUS, &status_reg); 6999 if (status_reg != IXGBE_FAILED_READ_CFG_WORD && 7000 status_reg & PCI_STATUS_REC_MASTER_ABORT) 7001 ixgbe_issue_vf_flr(adapter, vfdev); 7002 } 7003 } 7004 7005 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter) 7006 { 7007 u32 ssvpc; 7008 7009 /* Do not perform spoof check for 82598 or if not in IOV mode */ 7010 if (adapter->hw.mac.type == ixgbe_mac_82598EB || 7011 adapter->num_vfs == 0) 7012 return; 7013 7014 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC); 7015 7016 /* 7017 * ssvpc register is cleared on read, if zero then no 7018 * spoofed packets in the last interval. 7019 */ 7020 if (!ssvpc) 7021 return; 7022 7023 e_warn(drv, "%u Spoofed packets detected\n", ssvpc); 7024 } 7025 #else 7026 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter) 7027 { 7028 } 7029 7030 static void 7031 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter) 7032 { 7033 } 7034 #endif /* CONFIG_PCI_IOV */ 7035 7036 7037 /** 7038 * ixgbe_watchdog_subtask - check and bring link up 7039 * @adapter: pointer to the device adapter structure 7040 **/ 7041 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter) 7042 { 7043 /* if interface is down, removing or resetting, do nothing */ 7044 if (test_bit(__IXGBE_DOWN, &adapter->state) || 7045 test_bit(__IXGBE_REMOVING, &adapter->state) || 7046 test_bit(__IXGBE_RESETTING, &adapter->state)) 7047 return; 7048 7049 ixgbe_watchdog_update_link(adapter); 7050 7051 if (adapter->link_up) 7052 ixgbe_watchdog_link_is_up(adapter); 7053 else 7054 ixgbe_watchdog_link_is_down(adapter); 7055 7056 ixgbe_check_for_bad_vf(adapter); 7057 ixgbe_spoof_check(adapter); 7058 ixgbe_update_stats(adapter); 7059 7060 ixgbe_watchdog_flush_tx(adapter); 7061 } 7062 7063 /** 7064 * ixgbe_sfp_detection_subtask - poll for SFP+ cable 7065 * @adapter: the ixgbe adapter structure 7066 **/ 7067 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter) 7068 { 7069 struct ixgbe_hw *hw = &adapter->hw; 7070 s32 err; 7071 7072 /* If crosstalk fix enabled verify the SFP+ cage is full */ 7073 if (adapter->need_crosstalk_fix) { 7074 u32 sfp_cage_full; 7075 7076 sfp_cage_full = IXGBE_READ_REG(hw, IXGBE_ESDP) & 7077 IXGBE_ESDP_SDP2; 7078 if (!sfp_cage_full) 7079 return; 7080 } 7081 7082 /* not searching for SFP so there is nothing to do here */ 7083 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) && 7084 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET)) 7085 return; 7086 7087 if (adapter->sfp_poll_time && 7088 time_after(adapter->sfp_poll_time, jiffies)) 7089 return; /* If not yet time to poll for SFP */ 7090 7091 /* someone else is in init, wait until next service event */ 7092 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state)) 7093 return; 7094 7095 adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1; 7096 7097 err = hw->phy.ops.identify_sfp(hw); 7098 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) 7099 goto sfp_out; 7100 7101 if (err == IXGBE_ERR_SFP_NOT_PRESENT) { 7102 /* If no cable is present, then we need to reset 7103 * the next time we find a good cable. */ 7104 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET; 7105 } 7106 7107 /* exit on error */ 7108 if (err) 7109 goto sfp_out; 7110 7111 /* exit if reset not needed */ 7112 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET)) 7113 goto sfp_out; 7114 7115 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET; 7116 7117 /* 7118 * A module may be identified correctly, but the EEPROM may not have 7119 * support for that module. setup_sfp() will fail in that case, so 7120 * we should not allow that module to load. 7121 */ 7122 if (hw->mac.type == ixgbe_mac_82598EB) 7123 err = hw->phy.ops.reset(hw); 7124 else 7125 err = hw->mac.ops.setup_sfp(hw); 7126 7127 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) 7128 goto sfp_out; 7129 7130 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG; 7131 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type); 7132 7133 sfp_out: 7134 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state); 7135 7136 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) && 7137 (adapter->netdev->reg_state == NETREG_REGISTERED)) { 7138 e_dev_err("failed to initialize because an unsupported " 7139 "SFP+ module type was detected.\n"); 7140 e_dev_err("Reload the driver after installing a " 7141 "supported module.\n"); 7142 unregister_netdev(adapter->netdev); 7143 } 7144 } 7145 7146 /** 7147 * ixgbe_sfp_link_config_subtask - set up link SFP after module install 7148 * @adapter: the ixgbe adapter structure 7149 **/ 7150 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter) 7151 { 7152 struct ixgbe_hw *hw = &adapter->hw; 7153 u32 speed; 7154 bool autoneg = false; 7155 7156 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG)) 7157 return; 7158 7159 /* someone else is in init, wait until next service event */ 7160 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state)) 7161 return; 7162 7163 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG; 7164 7165 speed = hw->phy.autoneg_advertised; 7166 if ((!speed) && (hw->mac.ops.get_link_capabilities)) { 7167 hw->mac.ops.get_link_capabilities(hw, &speed, &autoneg); 7168 7169 /* setup the highest link when no autoneg */ 7170 if (!autoneg) { 7171 if (speed & IXGBE_LINK_SPEED_10GB_FULL) 7172 speed = IXGBE_LINK_SPEED_10GB_FULL; 7173 } 7174 } 7175 7176 if (hw->mac.ops.setup_link) 7177 hw->mac.ops.setup_link(hw, speed, true); 7178 7179 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE; 7180 adapter->link_check_timeout = jiffies; 7181 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state); 7182 } 7183 7184 /** 7185 * ixgbe_service_timer - Timer Call-back 7186 * @data: pointer to adapter cast into an unsigned long 7187 **/ 7188 static void ixgbe_service_timer(unsigned long data) 7189 { 7190 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data; 7191 unsigned long next_event_offset; 7192 7193 /* poll faster when waiting for link */ 7194 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) 7195 next_event_offset = HZ / 10; 7196 else 7197 next_event_offset = HZ * 2; 7198 7199 /* Reset the timer */ 7200 mod_timer(&adapter->service_timer, next_event_offset + jiffies); 7201 7202 ixgbe_service_event_schedule(adapter); 7203 } 7204 7205 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter) 7206 { 7207 struct ixgbe_hw *hw = &adapter->hw; 7208 u32 status; 7209 7210 if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT)) 7211 return; 7212 7213 adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT; 7214 7215 if (!hw->phy.ops.handle_lasi) 7216 return; 7217 7218 status = hw->phy.ops.handle_lasi(&adapter->hw); 7219 if (status != IXGBE_ERR_OVERTEMP) 7220 return; 7221 7222 e_crit(drv, "%s\n", ixgbe_overheat_msg); 7223 } 7224 7225 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter) 7226 { 7227 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED)) 7228 return; 7229 7230 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED; 7231 7232 /* If we're already down, removing or resetting, just bail */ 7233 if (test_bit(__IXGBE_DOWN, &adapter->state) || 7234 test_bit(__IXGBE_REMOVING, &adapter->state) || 7235 test_bit(__IXGBE_RESETTING, &adapter->state)) 7236 return; 7237 7238 ixgbe_dump(adapter); 7239 netdev_err(adapter->netdev, "Reset adapter\n"); 7240 adapter->tx_timeout_count++; 7241 7242 rtnl_lock(); 7243 ixgbe_reinit_locked(adapter); 7244 rtnl_unlock(); 7245 } 7246 7247 /** 7248 * ixgbe_service_task - manages and runs subtasks 7249 * @work: pointer to work_struct containing our data 7250 **/ 7251 static void ixgbe_service_task(struct work_struct *work) 7252 { 7253 struct ixgbe_adapter *adapter = container_of(work, 7254 struct ixgbe_adapter, 7255 service_task); 7256 if (ixgbe_removed(adapter->hw.hw_addr)) { 7257 if (!test_bit(__IXGBE_DOWN, &adapter->state)) { 7258 rtnl_lock(); 7259 ixgbe_down(adapter); 7260 rtnl_unlock(); 7261 } 7262 ixgbe_service_event_complete(adapter); 7263 return; 7264 } 7265 #ifdef CONFIG_IXGBE_VXLAN 7266 rtnl_lock(); 7267 if (adapter->flags2 & IXGBE_FLAG2_VXLAN_REREG_NEEDED) { 7268 adapter->flags2 &= ~IXGBE_FLAG2_VXLAN_REREG_NEEDED; 7269 vxlan_get_rx_port(adapter->netdev); 7270 } 7271 rtnl_unlock(); 7272 #endif /* CONFIG_IXGBE_VXLAN */ 7273 ixgbe_reset_subtask(adapter); 7274 ixgbe_phy_interrupt_subtask(adapter); 7275 ixgbe_sfp_detection_subtask(adapter); 7276 ixgbe_sfp_link_config_subtask(adapter); 7277 ixgbe_check_overtemp_subtask(adapter); 7278 ixgbe_watchdog_subtask(adapter); 7279 ixgbe_fdir_reinit_subtask(adapter); 7280 ixgbe_check_hang_subtask(adapter); 7281 7282 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) { 7283 ixgbe_ptp_overflow_check(adapter); 7284 ixgbe_ptp_rx_hang(adapter); 7285 } 7286 7287 ixgbe_service_event_complete(adapter); 7288 } 7289 7290 static int ixgbe_tso(struct ixgbe_ring *tx_ring, 7291 struct ixgbe_tx_buffer *first, 7292 u8 *hdr_len) 7293 { 7294 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx; 7295 struct sk_buff *skb = first->skb; 7296 union { 7297 struct iphdr *v4; 7298 struct ipv6hdr *v6; 7299 unsigned char *hdr; 7300 } ip; 7301 union { 7302 struct tcphdr *tcp; 7303 unsigned char *hdr; 7304 } l4; 7305 u32 paylen, l4_offset; 7306 int err; 7307 7308 if (skb->ip_summed != CHECKSUM_PARTIAL) 7309 return 0; 7310 7311 if (!skb_is_gso(skb)) 7312 return 0; 7313 7314 err = skb_cow_head(skb, 0); 7315 if (err < 0) 7316 return err; 7317 7318 ip.hdr = skb_network_header(skb); 7319 l4.hdr = skb_checksum_start(skb); 7320 7321 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */ 7322 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP; 7323 7324 /* initialize outer IP header fields */ 7325 if (ip.v4->version == 4) { 7326 /* IP header will have to cancel out any data that 7327 * is not a part of the outer IP header 7328 */ 7329 ip.v4->check = csum_fold(csum_add(lco_csum(skb), 7330 csum_unfold(l4.tcp->check))); 7331 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4; 7332 7333 ip.v4->tot_len = 0; 7334 first->tx_flags |= IXGBE_TX_FLAGS_TSO | 7335 IXGBE_TX_FLAGS_CSUM | 7336 IXGBE_TX_FLAGS_IPV4; 7337 } else { 7338 ip.v6->payload_len = 0; 7339 first->tx_flags |= IXGBE_TX_FLAGS_TSO | 7340 IXGBE_TX_FLAGS_CSUM; 7341 } 7342 7343 /* determine offset of inner transport header */ 7344 l4_offset = l4.hdr - skb->data; 7345 7346 /* compute length of segmentation header */ 7347 *hdr_len = (l4.tcp->doff * 4) + l4_offset; 7348 7349 /* remove payload length from inner checksum */ 7350 paylen = skb->len - l4_offset; 7351 csum_replace_by_diff(&l4.tcp->check, htonl(paylen)); 7352 7353 /* update gso size and bytecount with header size */ 7354 first->gso_segs = skb_shinfo(skb)->gso_segs; 7355 first->bytecount += (first->gso_segs - 1) * *hdr_len; 7356 7357 /* mss_l4len_id: use 0 as index for TSO */ 7358 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT; 7359 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT; 7360 7361 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */ 7362 vlan_macip_lens = l4.hdr - ip.hdr; 7363 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT; 7364 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK; 7365 7366 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd, 7367 mss_l4len_idx); 7368 7369 return 1; 7370 } 7371 7372 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb) 7373 { 7374 unsigned int offset = 0; 7375 7376 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL); 7377 7378 return offset == skb_checksum_start_offset(skb); 7379 } 7380 7381 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring, 7382 struct ixgbe_tx_buffer *first) 7383 { 7384 struct sk_buff *skb = first->skb; 7385 u32 vlan_macip_lens = 0; 7386 u32 type_tucmd = 0; 7387 7388 if (skb->ip_summed != CHECKSUM_PARTIAL) { 7389 csum_failed: 7390 if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | 7391 IXGBE_TX_FLAGS_CC))) 7392 return; 7393 goto no_csum; 7394 } 7395 7396 switch (skb->csum_offset) { 7397 case offsetof(struct tcphdr, check): 7398 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP; 7399 /* fall through */ 7400 case offsetof(struct udphdr, check): 7401 break; 7402 case offsetof(struct sctphdr, checksum): 7403 /* validate that this is actually an SCTP request */ 7404 if (((first->protocol == htons(ETH_P_IP)) && 7405 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) || 7406 ((first->protocol == htons(ETH_P_IPV6)) && 7407 ixgbe_ipv6_csum_is_sctp(skb))) { 7408 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP; 7409 break; 7410 } 7411 /* fall through */ 7412 default: 7413 skb_checksum_help(skb); 7414 goto csum_failed; 7415 } 7416 7417 /* update TX checksum flag */ 7418 first->tx_flags |= IXGBE_TX_FLAGS_CSUM; 7419 vlan_macip_lens = skb_checksum_start_offset(skb) - 7420 skb_network_offset(skb); 7421 no_csum: 7422 /* vlan_macip_lens: MACLEN, VLAN tag */ 7423 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT; 7424 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK; 7425 7426 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd, 0); 7427 } 7428 7429 #define IXGBE_SET_FLAG(_input, _flag, _result) \ 7430 ((_flag <= _result) ? \ 7431 ((u32)(_input & _flag) * (_result / _flag)) : \ 7432 ((u32)(_input & _flag) / (_flag / _result))) 7433 7434 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags) 7435 { 7436 /* set type for advanced descriptor with frame checksum insertion */ 7437 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA | 7438 IXGBE_ADVTXD_DCMD_DEXT | 7439 IXGBE_ADVTXD_DCMD_IFCS; 7440 7441 /* set HW vlan bit if vlan is present */ 7442 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN, 7443 IXGBE_ADVTXD_DCMD_VLE); 7444 7445 /* set segmentation enable bits for TSO/FSO */ 7446 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO, 7447 IXGBE_ADVTXD_DCMD_TSE); 7448 7449 /* set timestamp bit if present */ 7450 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP, 7451 IXGBE_ADVTXD_MAC_TSTAMP); 7452 7453 /* insert frame checksum */ 7454 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS); 7455 7456 return cmd_type; 7457 } 7458 7459 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc, 7460 u32 tx_flags, unsigned int paylen) 7461 { 7462 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT; 7463 7464 /* enable L4 checksum for TSO and TX checksum offload */ 7465 olinfo_status |= IXGBE_SET_FLAG(tx_flags, 7466 IXGBE_TX_FLAGS_CSUM, 7467 IXGBE_ADVTXD_POPTS_TXSM); 7468 7469 /* enble IPv4 checksum for TSO */ 7470 olinfo_status |= IXGBE_SET_FLAG(tx_flags, 7471 IXGBE_TX_FLAGS_IPV4, 7472 IXGBE_ADVTXD_POPTS_IXSM); 7473 7474 /* 7475 * Check Context must be set if Tx switch is enabled, which it 7476 * always is for case where virtual functions are running 7477 */ 7478 olinfo_status |= IXGBE_SET_FLAG(tx_flags, 7479 IXGBE_TX_FLAGS_CC, 7480 IXGBE_ADVTXD_CC); 7481 7482 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status); 7483 } 7484 7485 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size) 7486 { 7487 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index); 7488 7489 /* Herbert's original patch had: 7490 * smp_mb__after_netif_stop_queue(); 7491 * but since that doesn't exist yet, just open code it. 7492 */ 7493 smp_mb(); 7494 7495 /* We need to check again in a case another CPU has just 7496 * made room available. 7497 */ 7498 if (likely(ixgbe_desc_unused(tx_ring) < size)) 7499 return -EBUSY; 7500 7501 /* A reprieve! - use start_queue because it doesn't call schedule */ 7502 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index); 7503 ++tx_ring->tx_stats.restart_queue; 7504 return 0; 7505 } 7506 7507 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size) 7508 { 7509 if (likely(ixgbe_desc_unused(tx_ring) >= size)) 7510 return 0; 7511 7512 return __ixgbe_maybe_stop_tx(tx_ring, size); 7513 } 7514 7515 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \ 7516 IXGBE_TXD_CMD_RS) 7517 7518 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring, 7519 struct ixgbe_tx_buffer *first, 7520 const u8 hdr_len) 7521 { 7522 struct sk_buff *skb = first->skb; 7523 struct ixgbe_tx_buffer *tx_buffer; 7524 union ixgbe_adv_tx_desc *tx_desc; 7525 struct skb_frag_struct *frag; 7526 dma_addr_t dma; 7527 unsigned int data_len, size; 7528 u32 tx_flags = first->tx_flags; 7529 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags); 7530 u16 i = tx_ring->next_to_use; 7531 7532 tx_desc = IXGBE_TX_DESC(tx_ring, i); 7533 7534 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len); 7535 7536 size = skb_headlen(skb); 7537 data_len = skb->data_len; 7538 7539 #ifdef IXGBE_FCOE 7540 if (tx_flags & IXGBE_TX_FLAGS_FCOE) { 7541 if (data_len < sizeof(struct fcoe_crc_eof)) { 7542 size -= sizeof(struct fcoe_crc_eof) - data_len; 7543 data_len = 0; 7544 } else { 7545 data_len -= sizeof(struct fcoe_crc_eof); 7546 } 7547 } 7548 7549 #endif 7550 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE); 7551 7552 tx_buffer = first; 7553 7554 for (frag = &skb_shinfo(skb)->frags[0];; frag++) { 7555 if (dma_mapping_error(tx_ring->dev, dma)) 7556 goto dma_error; 7557 7558 /* record length, and DMA address */ 7559 dma_unmap_len_set(tx_buffer, len, size); 7560 dma_unmap_addr_set(tx_buffer, dma, dma); 7561 7562 tx_desc->read.buffer_addr = cpu_to_le64(dma); 7563 7564 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) { 7565 tx_desc->read.cmd_type_len = 7566 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD); 7567 7568 i++; 7569 tx_desc++; 7570 if (i == tx_ring->count) { 7571 tx_desc = IXGBE_TX_DESC(tx_ring, 0); 7572 i = 0; 7573 } 7574 tx_desc->read.olinfo_status = 0; 7575 7576 dma += IXGBE_MAX_DATA_PER_TXD; 7577 size -= IXGBE_MAX_DATA_PER_TXD; 7578 7579 tx_desc->read.buffer_addr = cpu_to_le64(dma); 7580 } 7581 7582 if (likely(!data_len)) 7583 break; 7584 7585 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size); 7586 7587 i++; 7588 tx_desc++; 7589 if (i == tx_ring->count) { 7590 tx_desc = IXGBE_TX_DESC(tx_ring, 0); 7591 i = 0; 7592 } 7593 tx_desc->read.olinfo_status = 0; 7594 7595 #ifdef IXGBE_FCOE 7596 size = min_t(unsigned int, data_len, skb_frag_size(frag)); 7597 #else 7598 size = skb_frag_size(frag); 7599 #endif 7600 data_len -= size; 7601 7602 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size, 7603 DMA_TO_DEVICE); 7604 7605 tx_buffer = &tx_ring->tx_buffer_info[i]; 7606 } 7607 7608 /* write last descriptor with RS and EOP bits */ 7609 cmd_type |= size | IXGBE_TXD_CMD; 7610 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type); 7611 7612 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount); 7613 7614 /* set the timestamp */ 7615 first->time_stamp = jiffies; 7616 7617 /* 7618 * Force memory writes to complete before letting h/w know there 7619 * are new descriptors to fetch. (Only applicable for weak-ordered 7620 * memory model archs, such as IA-64). 7621 * 7622 * We also need this memory barrier to make certain all of the 7623 * status bits have been updated before next_to_watch is written. 7624 */ 7625 wmb(); 7626 7627 /* set next_to_watch value indicating a packet is present */ 7628 first->next_to_watch = tx_desc; 7629 7630 i++; 7631 if (i == tx_ring->count) 7632 i = 0; 7633 7634 tx_ring->next_to_use = i; 7635 7636 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED); 7637 7638 if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) { 7639 writel(i, tx_ring->tail); 7640 7641 /* we need this if more than one processor can write to our tail 7642 * at a time, it synchronizes IO on IA64/Altix systems 7643 */ 7644 mmiowb(); 7645 } 7646 7647 return; 7648 dma_error: 7649 dev_err(tx_ring->dev, "TX DMA map failed\n"); 7650 7651 /* clear dma mappings for failed tx_buffer_info map */ 7652 for (;;) { 7653 tx_buffer = &tx_ring->tx_buffer_info[i]; 7654 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer); 7655 if (tx_buffer == first) 7656 break; 7657 if (i == 0) 7658 i = tx_ring->count; 7659 i--; 7660 } 7661 7662 tx_ring->next_to_use = i; 7663 } 7664 7665 static void ixgbe_atr(struct ixgbe_ring *ring, 7666 struct ixgbe_tx_buffer *first) 7667 { 7668 struct ixgbe_q_vector *q_vector = ring->q_vector; 7669 union ixgbe_atr_hash_dword input = { .dword = 0 }; 7670 union ixgbe_atr_hash_dword common = { .dword = 0 }; 7671 union { 7672 unsigned char *network; 7673 struct iphdr *ipv4; 7674 struct ipv6hdr *ipv6; 7675 } hdr; 7676 struct tcphdr *th; 7677 unsigned int hlen; 7678 struct sk_buff *skb; 7679 __be16 vlan_id; 7680 int l4_proto; 7681 7682 /* if ring doesn't have a interrupt vector, cannot perform ATR */ 7683 if (!q_vector) 7684 return; 7685 7686 /* do nothing if sampling is disabled */ 7687 if (!ring->atr_sample_rate) 7688 return; 7689 7690 ring->atr_count++; 7691 7692 /* currently only IPv4/IPv6 with TCP is supported */ 7693 if ((first->protocol != htons(ETH_P_IP)) && 7694 (first->protocol != htons(ETH_P_IPV6))) 7695 return; 7696 7697 /* snag network header to get L4 type and address */ 7698 skb = first->skb; 7699 hdr.network = skb_network_header(skb); 7700 #ifdef CONFIG_IXGBE_VXLAN 7701 if (skb->encapsulation && 7702 first->protocol == htons(ETH_P_IP) && 7703 hdr.ipv4->protocol != IPPROTO_UDP) { 7704 struct ixgbe_adapter *adapter = q_vector->adapter; 7705 7706 /* verify the port is recognized as VXLAN */ 7707 if (adapter->vxlan_port && 7708 udp_hdr(skb)->dest == adapter->vxlan_port) 7709 hdr.network = skb_inner_network_header(skb); 7710 } 7711 #endif /* CONFIG_IXGBE_VXLAN */ 7712 7713 /* Currently only IPv4/IPv6 with TCP is supported */ 7714 switch (hdr.ipv4->version) { 7715 case IPVERSION: 7716 /* access ihl as u8 to avoid unaligned access on ia64 */ 7717 hlen = (hdr.network[0] & 0x0F) << 2; 7718 l4_proto = hdr.ipv4->protocol; 7719 break; 7720 case 6: 7721 hlen = hdr.network - skb->data; 7722 l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL); 7723 hlen -= hdr.network - skb->data; 7724 break; 7725 default: 7726 return; 7727 } 7728 7729 if (l4_proto != IPPROTO_TCP) 7730 return; 7731 7732 th = (struct tcphdr *)(hdr.network + hlen); 7733 7734 /* skip this packet since the socket is closing */ 7735 if (th->fin) 7736 return; 7737 7738 /* sample on all syn packets or once every atr sample count */ 7739 if (!th->syn && (ring->atr_count < ring->atr_sample_rate)) 7740 return; 7741 7742 /* reset sample count */ 7743 ring->atr_count = 0; 7744 7745 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT); 7746 7747 /* 7748 * src and dst are inverted, think how the receiver sees them 7749 * 7750 * The input is broken into two sections, a non-compressed section 7751 * containing vm_pool, vlan_id, and flow_type. The rest of the data 7752 * is XORed together and stored in the compressed dword. 7753 */ 7754 input.formatted.vlan_id = vlan_id; 7755 7756 /* 7757 * since src port and flex bytes occupy the same word XOR them together 7758 * and write the value to source port portion of compressed dword 7759 */ 7760 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN)) 7761 common.port.src ^= th->dest ^ htons(ETH_P_8021Q); 7762 else 7763 common.port.src ^= th->dest ^ first->protocol; 7764 common.port.dst ^= th->source; 7765 7766 switch (hdr.ipv4->version) { 7767 case IPVERSION: 7768 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4; 7769 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr; 7770 break; 7771 case 6: 7772 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6; 7773 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^ 7774 hdr.ipv6->saddr.s6_addr32[1] ^ 7775 hdr.ipv6->saddr.s6_addr32[2] ^ 7776 hdr.ipv6->saddr.s6_addr32[3] ^ 7777 hdr.ipv6->daddr.s6_addr32[0] ^ 7778 hdr.ipv6->daddr.s6_addr32[1] ^ 7779 hdr.ipv6->daddr.s6_addr32[2] ^ 7780 hdr.ipv6->daddr.s6_addr32[3]; 7781 break; 7782 default: 7783 break; 7784 } 7785 7786 if (hdr.network != skb_network_header(skb)) 7787 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK; 7788 7789 /* This assumes the Rx queue and Tx queue are bound to the same CPU */ 7790 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw, 7791 input, common, ring->queue_index); 7792 } 7793 7794 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb, 7795 void *accel_priv, select_queue_fallback_t fallback) 7796 { 7797 struct ixgbe_fwd_adapter *fwd_adapter = accel_priv; 7798 #ifdef IXGBE_FCOE 7799 struct ixgbe_adapter *adapter; 7800 struct ixgbe_ring_feature *f; 7801 int txq; 7802 #endif 7803 7804 if (fwd_adapter) 7805 return skb->queue_mapping + fwd_adapter->tx_base_queue; 7806 7807 #ifdef IXGBE_FCOE 7808 7809 /* 7810 * only execute the code below if protocol is FCoE 7811 * or FIP and we have FCoE enabled on the adapter 7812 */ 7813 switch (vlan_get_protocol(skb)) { 7814 case htons(ETH_P_FCOE): 7815 case htons(ETH_P_FIP): 7816 adapter = netdev_priv(dev); 7817 7818 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) 7819 break; 7820 default: 7821 return fallback(dev, skb); 7822 } 7823 7824 f = &adapter->ring_feature[RING_F_FCOE]; 7825 7826 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 7827 smp_processor_id(); 7828 7829 while (txq >= f->indices) 7830 txq -= f->indices; 7831 7832 return txq + f->offset; 7833 #else 7834 return fallback(dev, skb); 7835 #endif 7836 } 7837 7838 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb, 7839 struct ixgbe_adapter *adapter, 7840 struct ixgbe_ring *tx_ring) 7841 { 7842 struct ixgbe_tx_buffer *first; 7843 int tso; 7844 u32 tx_flags = 0; 7845 unsigned short f; 7846 u16 count = TXD_USE_COUNT(skb_headlen(skb)); 7847 __be16 protocol = skb->protocol; 7848 u8 hdr_len = 0; 7849 7850 /* 7851 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD, 7852 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD, 7853 * + 2 desc gap to keep tail from touching head, 7854 * + 1 desc for context descriptor, 7855 * otherwise try next time 7856 */ 7857 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) 7858 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size); 7859 7860 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) { 7861 tx_ring->tx_stats.tx_busy++; 7862 return NETDEV_TX_BUSY; 7863 } 7864 7865 /* record the location of the first descriptor for this packet */ 7866 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use]; 7867 first->skb = skb; 7868 first->bytecount = skb->len; 7869 first->gso_segs = 1; 7870 7871 /* if we have a HW VLAN tag being added default to the HW one */ 7872 if (skb_vlan_tag_present(skb)) { 7873 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT; 7874 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN; 7875 /* else if it is a SW VLAN check the next protocol and store the tag */ 7876 } else if (protocol == htons(ETH_P_8021Q)) { 7877 struct vlan_hdr *vhdr, _vhdr; 7878 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr); 7879 if (!vhdr) 7880 goto out_drop; 7881 7882 tx_flags |= ntohs(vhdr->h_vlan_TCI) << 7883 IXGBE_TX_FLAGS_VLAN_SHIFT; 7884 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN; 7885 } 7886 protocol = vlan_get_protocol(skb); 7887 7888 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) && 7889 adapter->ptp_clock && 7890 !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS, 7891 &adapter->state)) { 7892 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS; 7893 tx_flags |= IXGBE_TX_FLAGS_TSTAMP; 7894 7895 /* schedule check for Tx timestamp */ 7896 adapter->ptp_tx_skb = skb_get(skb); 7897 adapter->ptp_tx_start = jiffies; 7898 schedule_work(&adapter->ptp_tx_work); 7899 } 7900 7901 skb_tx_timestamp(skb); 7902 7903 #ifdef CONFIG_PCI_IOV 7904 /* 7905 * Use the l2switch_enable flag - would be false if the DMA 7906 * Tx switch had been disabled. 7907 */ 7908 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) 7909 tx_flags |= IXGBE_TX_FLAGS_CC; 7910 7911 #endif 7912 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */ 7913 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && 7914 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) || 7915 (skb->priority != TC_PRIO_CONTROL))) { 7916 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK; 7917 tx_flags |= (skb->priority & 0x7) << 7918 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT; 7919 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) { 7920 struct vlan_ethhdr *vhdr; 7921 7922 if (skb_cow_head(skb, 0)) 7923 goto out_drop; 7924 vhdr = (struct vlan_ethhdr *)skb->data; 7925 vhdr->h_vlan_TCI = htons(tx_flags >> 7926 IXGBE_TX_FLAGS_VLAN_SHIFT); 7927 } else { 7928 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN; 7929 } 7930 } 7931 7932 /* record initial flags and protocol */ 7933 first->tx_flags = tx_flags; 7934 first->protocol = protocol; 7935 7936 #ifdef IXGBE_FCOE 7937 /* setup tx offload for FCoE */ 7938 if ((protocol == htons(ETH_P_FCOE)) && 7939 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) { 7940 tso = ixgbe_fso(tx_ring, first, &hdr_len); 7941 if (tso < 0) 7942 goto out_drop; 7943 7944 goto xmit_fcoe; 7945 } 7946 7947 #endif /* IXGBE_FCOE */ 7948 tso = ixgbe_tso(tx_ring, first, &hdr_len); 7949 if (tso < 0) 7950 goto out_drop; 7951 else if (!tso) 7952 ixgbe_tx_csum(tx_ring, first); 7953 7954 /* add the ATR filter if ATR is on */ 7955 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state)) 7956 ixgbe_atr(tx_ring, first); 7957 7958 #ifdef IXGBE_FCOE 7959 xmit_fcoe: 7960 #endif /* IXGBE_FCOE */ 7961 ixgbe_tx_map(tx_ring, first, hdr_len); 7962 7963 return NETDEV_TX_OK; 7964 7965 out_drop: 7966 dev_kfree_skb_any(first->skb); 7967 first->skb = NULL; 7968 7969 return NETDEV_TX_OK; 7970 } 7971 7972 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb, 7973 struct net_device *netdev, 7974 struct ixgbe_ring *ring) 7975 { 7976 struct ixgbe_adapter *adapter = netdev_priv(netdev); 7977 struct ixgbe_ring *tx_ring; 7978 7979 /* 7980 * The minimum packet size for olinfo paylen is 17 so pad the skb 7981 * in order to meet this minimum size requirement. 7982 */ 7983 if (skb_put_padto(skb, 17)) 7984 return NETDEV_TX_OK; 7985 7986 tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping]; 7987 7988 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring); 7989 } 7990 7991 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, 7992 struct net_device *netdev) 7993 { 7994 return __ixgbe_xmit_frame(skb, netdev, NULL); 7995 } 7996 7997 /** 7998 * ixgbe_set_mac - Change the Ethernet Address of the NIC 7999 * @netdev: network interface device structure 8000 * @p: pointer to an address structure 8001 * 8002 * Returns 0 on success, negative on failure 8003 **/ 8004 static int ixgbe_set_mac(struct net_device *netdev, void *p) 8005 { 8006 struct ixgbe_adapter *adapter = netdev_priv(netdev); 8007 struct ixgbe_hw *hw = &adapter->hw; 8008 struct sockaddr *addr = p; 8009 8010 if (!is_valid_ether_addr(addr->sa_data)) 8011 return -EADDRNOTAVAIL; 8012 8013 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len); 8014 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len); 8015 8016 ixgbe_mac_set_default_filter(adapter); 8017 8018 return 0; 8019 } 8020 8021 static int 8022 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr) 8023 { 8024 struct ixgbe_adapter *adapter = netdev_priv(netdev); 8025 struct ixgbe_hw *hw = &adapter->hw; 8026 u16 value; 8027 int rc; 8028 8029 if (prtad != hw->phy.mdio.prtad) 8030 return -EINVAL; 8031 rc = hw->phy.ops.read_reg(hw, addr, devad, &value); 8032 if (!rc) 8033 rc = value; 8034 return rc; 8035 } 8036 8037 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad, 8038 u16 addr, u16 value) 8039 { 8040 struct ixgbe_adapter *adapter = netdev_priv(netdev); 8041 struct ixgbe_hw *hw = &adapter->hw; 8042 8043 if (prtad != hw->phy.mdio.prtad) 8044 return -EINVAL; 8045 return hw->phy.ops.write_reg(hw, addr, devad, value); 8046 } 8047 8048 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd) 8049 { 8050 struct ixgbe_adapter *adapter = netdev_priv(netdev); 8051 8052 switch (cmd) { 8053 case SIOCSHWTSTAMP: 8054 return ixgbe_ptp_set_ts_config(adapter, req); 8055 case SIOCGHWTSTAMP: 8056 return ixgbe_ptp_get_ts_config(adapter, req); 8057 default: 8058 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd); 8059 } 8060 } 8061 8062 /** 8063 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding 8064 * netdev->dev_addrs 8065 * @netdev: network interface device structure 8066 * 8067 * Returns non-zero on failure 8068 **/ 8069 static int ixgbe_add_sanmac_netdev(struct net_device *dev) 8070 { 8071 int err = 0; 8072 struct ixgbe_adapter *adapter = netdev_priv(dev); 8073 struct ixgbe_hw *hw = &adapter->hw; 8074 8075 if (is_valid_ether_addr(hw->mac.san_addr)) { 8076 rtnl_lock(); 8077 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN); 8078 rtnl_unlock(); 8079 8080 /* update SAN MAC vmdq pool selection */ 8081 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0)); 8082 } 8083 return err; 8084 } 8085 8086 /** 8087 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding 8088 * netdev->dev_addrs 8089 * @netdev: network interface device structure 8090 * 8091 * Returns non-zero on failure 8092 **/ 8093 static int ixgbe_del_sanmac_netdev(struct net_device *dev) 8094 { 8095 int err = 0; 8096 struct ixgbe_adapter *adapter = netdev_priv(dev); 8097 struct ixgbe_mac_info *mac = &adapter->hw.mac; 8098 8099 if (is_valid_ether_addr(mac->san_addr)) { 8100 rtnl_lock(); 8101 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN); 8102 rtnl_unlock(); 8103 } 8104 return err; 8105 } 8106 8107 #ifdef CONFIG_NET_POLL_CONTROLLER 8108 /* 8109 * Polling 'interrupt' - used by things like netconsole to send skbs 8110 * without having to re-enable interrupts. It's not called while 8111 * the interrupt routine is executing. 8112 */ 8113 static void ixgbe_netpoll(struct net_device *netdev) 8114 { 8115 struct ixgbe_adapter *adapter = netdev_priv(netdev); 8116 int i; 8117 8118 /* if interface is down do nothing */ 8119 if (test_bit(__IXGBE_DOWN, &adapter->state)) 8120 return; 8121 8122 /* loop through and schedule all active queues */ 8123 for (i = 0; i < adapter->num_q_vectors; i++) 8124 ixgbe_msix_clean_rings(0, adapter->q_vector[i]); 8125 } 8126 8127 #endif 8128 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev, 8129 struct rtnl_link_stats64 *stats) 8130 { 8131 struct ixgbe_adapter *adapter = netdev_priv(netdev); 8132 int i; 8133 8134 rcu_read_lock(); 8135 for (i = 0; i < adapter->num_rx_queues; i++) { 8136 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]); 8137 u64 bytes, packets; 8138 unsigned int start; 8139 8140 if (ring) { 8141 do { 8142 start = u64_stats_fetch_begin_irq(&ring->syncp); 8143 packets = ring->stats.packets; 8144 bytes = ring->stats.bytes; 8145 } while (u64_stats_fetch_retry_irq(&ring->syncp, start)); 8146 stats->rx_packets += packets; 8147 stats->rx_bytes += bytes; 8148 } 8149 } 8150 8151 for (i = 0; i < adapter->num_tx_queues; i++) { 8152 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]); 8153 u64 bytes, packets; 8154 unsigned int start; 8155 8156 if (ring) { 8157 do { 8158 start = u64_stats_fetch_begin_irq(&ring->syncp); 8159 packets = ring->stats.packets; 8160 bytes = ring->stats.bytes; 8161 } while (u64_stats_fetch_retry_irq(&ring->syncp, start)); 8162 stats->tx_packets += packets; 8163 stats->tx_bytes += bytes; 8164 } 8165 } 8166 rcu_read_unlock(); 8167 /* following stats updated by ixgbe_watchdog_task() */ 8168 stats->multicast = netdev->stats.multicast; 8169 stats->rx_errors = netdev->stats.rx_errors; 8170 stats->rx_length_errors = netdev->stats.rx_length_errors; 8171 stats->rx_crc_errors = netdev->stats.rx_crc_errors; 8172 stats->rx_missed_errors = netdev->stats.rx_missed_errors; 8173 return stats; 8174 } 8175 8176 #ifdef CONFIG_IXGBE_DCB 8177 /** 8178 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid. 8179 * @adapter: pointer to ixgbe_adapter 8180 * @tc: number of traffic classes currently enabled 8181 * 8182 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm 8183 * 802.1Q priority maps to a packet buffer that exists. 8184 */ 8185 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc) 8186 { 8187 struct ixgbe_hw *hw = &adapter->hw; 8188 u32 reg, rsave; 8189 int i; 8190 8191 /* 82598 have a static priority to TC mapping that can not 8192 * be changed so no validation is needed. 8193 */ 8194 if (hw->mac.type == ixgbe_mac_82598EB) 8195 return; 8196 8197 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC); 8198 rsave = reg; 8199 8200 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) { 8201 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT); 8202 8203 /* If up2tc is out of bounds default to zero */ 8204 if (up2tc > tc) 8205 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT); 8206 } 8207 8208 if (reg != rsave) 8209 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg); 8210 8211 return; 8212 } 8213 8214 /** 8215 * ixgbe_set_prio_tc_map - Configure netdev prio tc map 8216 * @adapter: Pointer to adapter struct 8217 * 8218 * Populate the netdev user priority to tc map 8219 */ 8220 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter) 8221 { 8222 struct net_device *dev = adapter->netdev; 8223 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg; 8224 struct ieee_ets *ets = adapter->ixgbe_ieee_ets; 8225 u8 prio; 8226 8227 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) { 8228 u8 tc = 0; 8229 8230 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) 8231 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio); 8232 else if (ets) 8233 tc = ets->prio_tc[prio]; 8234 8235 netdev_set_prio_tc_map(dev, prio, tc); 8236 } 8237 } 8238 8239 #endif /* CONFIG_IXGBE_DCB */ 8240 /** 8241 * ixgbe_setup_tc - configure net_device for multiple traffic classes 8242 * 8243 * @netdev: net device to configure 8244 * @tc: number of traffic classes to enable 8245 */ 8246 int ixgbe_setup_tc(struct net_device *dev, u8 tc) 8247 { 8248 struct ixgbe_adapter *adapter = netdev_priv(dev); 8249 struct ixgbe_hw *hw = &adapter->hw; 8250 bool pools; 8251 8252 /* Hardware supports up to 8 traffic classes */ 8253 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs) 8254 return -EINVAL; 8255 8256 if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS) 8257 return -EINVAL; 8258 8259 pools = (find_first_zero_bit(&adapter->fwd_bitmask, 32) > 1); 8260 if (tc && pools && adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS) 8261 return -EBUSY; 8262 8263 /* Hardware has to reinitialize queues and interrupts to 8264 * match packet buffer alignment. Unfortunately, the 8265 * hardware is not flexible enough to do this dynamically. 8266 */ 8267 if (netif_running(dev)) 8268 ixgbe_close(dev); 8269 else 8270 ixgbe_reset(adapter); 8271 8272 ixgbe_clear_interrupt_scheme(adapter); 8273 8274 #ifdef CONFIG_IXGBE_DCB 8275 if (tc) { 8276 netdev_set_num_tc(dev, tc); 8277 ixgbe_set_prio_tc_map(adapter); 8278 8279 adapter->flags |= IXGBE_FLAG_DCB_ENABLED; 8280 8281 if (adapter->hw.mac.type == ixgbe_mac_82598EB) { 8282 adapter->last_lfc_mode = adapter->hw.fc.requested_mode; 8283 adapter->hw.fc.requested_mode = ixgbe_fc_none; 8284 } 8285 } else { 8286 netdev_reset_tc(dev); 8287 8288 if (adapter->hw.mac.type == ixgbe_mac_82598EB) 8289 adapter->hw.fc.requested_mode = adapter->last_lfc_mode; 8290 8291 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED; 8292 8293 adapter->temp_dcb_cfg.pfc_mode_enable = false; 8294 adapter->dcb_cfg.pfc_mode_enable = false; 8295 } 8296 8297 ixgbe_validate_rtr(adapter, tc); 8298 8299 #endif /* CONFIG_IXGBE_DCB */ 8300 ixgbe_init_interrupt_scheme(adapter); 8301 8302 if (netif_running(dev)) 8303 return ixgbe_open(dev); 8304 8305 return 0; 8306 } 8307 8308 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter, 8309 struct tc_cls_u32_offload *cls) 8310 { 8311 u32 uhtid = TC_U32_USERHTID(cls->knode.handle); 8312 u32 loc; 8313 int err; 8314 8315 if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE)) 8316 return -EINVAL; 8317 8318 loc = cls->knode.handle & 0xfffff; 8319 8320 spin_lock(&adapter->fdir_perfect_lock); 8321 err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc); 8322 spin_unlock(&adapter->fdir_perfect_lock); 8323 return err; 8324 } 8325 8326 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter, 8327 __be16 protocol, 8328 struct tc_cls_u32_offload *cls) 8329 { 8330 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle); 8331 8332 if (uhtid >= IXGBE_MAX_LINK_HANDLE) 8333 return -EINVAL; 8334 8335 /* This ixgbe devices do not support hash tables at the moment 8336 * so abort when given hash tables. 8337 */ 8338 if (cls->hnode.divisor > 0) 8339 return -EINVAL; 8340 8341 set_bit(uhtid - 1, &adapter->tables); 8342 return 0; 8343 } 8344 8345 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter, 8346 struct tc_cls_u32_offload *cls) 8347 { 8348 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle); 8349 8350 if (uhtid >= IXGBE_MAX_LINK_HANDLE) 8351 return -EINVAL; 8352 8353 clear_bit(uhtid - 1, &adapter->tables); 8354 return 0; 8355 } 8356 8357 #ifdef CONFIG_NET_CLS_ACT 8358 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex, 8359 u8 *queue, u64 *action) 8360 { 8361 unsigned int num_vfs = adapter->num_vfs, vf; 8362 struct net_device *upper; 8363 struct list_head *iter; 8364 8365 /* redirect to a SRIOV VF */ 8366 for (vf = 0; vf < num_vfs; ++vf) { 8367 upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev); 8368 if (upper->ifindex == ifindex) { 8369 if (adapter->num_rx_pools > 1) 8370 *queue = vf * 2; 8371 else 8372 *queue = vf * adapter->num_rx_queues_per_pool; 8373 8374 *action = vf + 1; 8375 *action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF; 8376 return 0; 8377 } 8378 } 8379 8380 /* redirect to a offloaded macvlan netdev */ 8381 netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) { 8382 if (netif_is_macvlan(upper)) { 8383 struct macvlan_dev *dfwd = netdev_priv(upper); 8384 struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv; 8385 8386 if (vadapter && vadapter->netdev->ifindex == ifindex) { 8387 *queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx; 8388 *action = *queue; 8389 return 0; 8390 } 8391 } 8392 } 8393 8394 return -EINVAL; 8395 } 8396 8397 static int parse_tc_actions(struct ixgbe_adapter *adapter, 8398 struct tcf_exts *exts, u64 *action, u8 *queue) 8399 { 8400 const struct tc_action *a; 8401 int err; 8402 8403 if (tc_no_actions(exts)) 8404 return -EINVAL; 8405 8406 tc_for_each_action(a, exts) { 8407 8408 /* Drop action */ 8409 if (is_tcf_gact_shot(a)) { 8410 *action = IXGBE_FDIR_DROP_QUEUE; 8411 *queue = IXGBE_FDIR_DROP_QUEUE; 8412 return 0; 8413 } 8414 8415 /* Redirect to a VF or a offloaded macvlan */ 8416 if (is_tcf_mirred_redirect(a)) { 8417 int ifindex = tcf_mirred_ifindex(a); 8418 8419 err = handle_redirect_action(adapter, ifindex, queue, 8420 action); 8421 if (err == 0) 8422 return err; 8423 } 8424 } 8425 8426 return -EINVAL; 8427 } 8428 #else 8429 static int parse_tc_actions(struct ixgbe_adapter *adapter, 8430 struct tcf_exts *exts, u64 *action, u8 *queue) 8431 { 8432 return -EINVAL; 8433 } 8434 #endif /* CONFIG_NET_CLS_ACT */ 8435 8436 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input, 8437 union ixgbe_atr_input *mask, 8438 struct tc_cls_u32_offload *cls, 8439 struct ixgbe_mat_field *field_ptr, 8440 struct ixgbe_nexthdr *nexthdr) 8441 { 8442 int i, j, off; 8443 __be32 val, m; 8444 bool found_entry = false, found_jump_field = false; 8445 8446 for (i = 0; i < cls->knode.sel->nkeys; i++) { 8447 off = cls->knode.sel->keys[i].off; 8448 val = cls->knode.sel->keys[i].val; 8449 m = cls->knode.sel->keys[i].mask; 8450 8451 for (j = 0; field_ptr[j].val; j++) { 8452 if (field_ptr[j].off == off) { 8453 field_ptr[j].val(input, mask, val, m); 8454 input->filter.formatted.flow_type |= 8455 field_ptr[j].type; 8456 found_entry = true; 8457 break; 8458 } 8459 } 8460 if (nexthdr) { 8461 if (nexthdr->off == cls->knode.sel->keys[i].off && 8462 nexthdr->val == cls->knode.sel->keys[i].val && 8463 nexthdr->mask == cls->knode.sel->keys[i].mask) 8464 found_jump_field = true; 8465 else 8466 continue; 8467 } 8468 } 8469 8470 if (nexthdr && !found_jump_field) 8471 return -EINVAL; 8472 8473 if (!found_entry) 8474 return 0; 8475 8476 mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK | 8477 IXGBE_ATR_L4TYPE_MASK; 8478 8479 if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4) 8480 mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK; 8481 8482 return 0; 8483 } 8484 8485 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter, 8486 __be16 protocol, 8487 struct tc_cls_u32_offload *cls) 8488 { 8489 u32 loc = cls->knode.handle & 0xfffff; 8490 struct ixgbe_hw *hw = &adapter->hw; 8491 struct ixgbe_mat_field *field_ptr; 8492 struct ixgbe_fdir_filter *input = NULL; 8493 union ixgbe_atr_input *mask = NULL; 8494 struct ixgbe_jump_table *jump = NULL; 8495 int i, err = -EINVAL; 8496 u8 queue; 8497 u32 uhtid, link_uhtid; 8498 8499 uhtid = TC_U32_USERHTID(cls->knode.handle); 8500 link_uhtid = TC_U32_USERHTID(cls->knode.link_handle); 8501 8502 /* At the moment cls_u32 jumps to network layer and skips past 8503 * L2 headers. The canonical method to match L2 frames is to use 8504 * negative values. However this is error prone at best but really 8505 * just broken because there is no way to "know" what sort of hdr 8506 * is in front of the network layer. Fix cls_u32 to support L2 8507 * headers when needed. 8508 */ 8509 if (protocol != htons(ETH_P_IP)) 8510 return err; 8511 8512 if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) { 8513 e_err(drv, "Location out of range\n"); 8514 return err; 8515 } 8516 8517 /* cls u32 is a graph starting at root node 0x800. The driver tracks 8518 * links and also the fields used to advance the parser across each 8519 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map 8520 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h 8521 * To add support for new nodes update ixgbe_model.h parse structures 8522 * this function _should_ be generic try not to hardcode values here. 8523 */ 8524 if (uhtid == 0x800) { 8525 field_ptr = (adapter->jump_tables[0])->mat; 8526 } else { 8527 if (uhtid >= IXGBE_MAX_LINK_HANDLE) 8528 return err; 8529 if (!adapter->jump_tables[uhtid]) 8530 return err; 8531 field_ptr = (adapter->jump_tables[uhtid])->mat; 8532 } 8533 8534 if (!field_ptr) 8535 return err; 8536 8537 /* At this point we know the field_ptr is valid and need to either 8538 * build cls_u32 link or attach filter. Because adding a link to 8539 * a handle that does not exist is invalid and the same for adding 8540 * rules to handles that don't exist. 8541 */ 8542 8543 if (link_uhtid) { 8544 struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps; 8545 8546 if (link_uhtid >= IXGBE_MAX_LINK_HANDLE) 8547 return err; 8548 8549 if (!test_bit(link_uhtid - 1, &adapter->tables)) 8550 return err; 8551 8552 for (i = 0; nexthdr[i].jump; i++) { 8553 if (nexthdr[i].o != cls->knode.sel->offoff || 8554 nexthdr[i].s != cls->knode.sel->offshift || 8555 nexthdr[i].m != cls->knode.sel->offmask) 8556 return err; 8557 8558 jump = kzalloc(sizeof(*jump), GFP_KERNEL); 8559 if (!jump) 8560 return -ENOMEM; 8561 input = kzalloc(sizeof(*input), GFP_KERNEL); 8562 if (!input) { 8563 err = -ENOMEM; 8564 goto free_jump; 8565 } 8566 mask = kzalloc(sizeof(*mask), GFP_KERNEL); 8567 if (!mask) { 8568 err = -ENOMEM; 8569 goto free_input; 8570 } 8571 jump->input = input; 8572 jump->mask = mask; 8573 err = ixgbe_clsu32_build_input(input, mask, cls, 8574 field_ptr, &nexthdr[i]); 8575 if (!err) { 8576 jump->mat = nexthdr[i].jump; 8577 adapter->jump_tables[link_uhtid] = jump; 8578 break; 8579 } 8580 } 8581 return 0; 8582 } 8583 8584 input = kzalloc(sizeof(*input), GFP_KERNEL); 8585 if (!input) 8586 return -ENOMEM; 8587 mask = kzalloc(sizeof(*mask), GFP_KERNEL); 8588 if (!mask) { 8589 err = -ENOMEM; 8590 goto free_input; 8591 } 8592 8593 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) { 8594 if ((adapter->jump_tables[uhtid])->input) 8595 memcpy(input, (adapter->jump_tables[uhtid])->input, 8596 sizeof(*input)); 8597 if ((adapter->jump_tables[uhtid])->mask) 8598 memcpy(mask, (adapter->jump_tables[uhtid])->mask, 8599 sizeof(*mask)); 8600 } 8601 err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL); 8602 if (err) 8603 goto err_out; 8604 8605 err = parse_tc_actions(adapter, cls->knode.exts, &input->action, 8606 &queue); 8607 if (err < 0) 8608 goto err_out; 8609 8610 input->sw_idx = loc; 8611 8612 spin_lock(&adapter->fdir_perfect_lock); 8613 8614 if (hlist_empty(&adapter->fdir_filter_list)) { 8615 memcpy(&adapter->fdir_mask, mask, sizeof(*mask)); 8616 err = ixgbe_fdir_set_input_mask_82599(hw, mask); 8617 if (err) 8618 goto err_out_w_lock; 8619 } else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) { 8620 err = -EINVAL; 8621 goto err_out_w_lock; 8622 } 8623 8624 ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask); 8625 err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter, 8626 input->sw_idx, queue); 8627 if (!err) 8628 ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx); 8629 spin_unlock(&adapter->fdir_perfect_lock); 8630 8631 kfree(mask); 8632 return err; 8633 err_out_w_lock: 8634 spin_unlock(&adapter->fdir_perfect_lock); 8635 err_out: 8636 kfree(mask); 8637 free_input: 8638 kfree(input); 8639 free_jump: 8640 kfree(jump); 8641 return err; 8642 } 8643 8644 static int __ixgbe_setup_tc(struct net_device *dev, u32 handle, __be16 proto, 8645 struct tc_to_netdev *tc) 8646 { 8647 struct ixgbe_adapter *adapter = netdev_priv(dev); 8648 8649 if (TC_H_MAJ(handle) == TC_H_MAJ(TC_H_INGRESS) && 8650 tc->type == TC_SETUP_CLSU32) { 8651 switch (tc->cls_u32->command) { 8652 case TC_CLSU32_NEW_KNODE: 8653 case TC_CLSU32_REPLACE_KNODE: 8654 return ixgbe_configure_clsu32(adapter, 8655 proto, tc->cls_u32); 8656 case TC_CLSU32_DELETE_KNODE: 8657 return ixgbe_delete_clsu32(adapter, tc->cls_u32); 8658 case TC_CLSU32_NEW_HNODE: 8659 case TC_CLSU32_REPLACE_HNODE: 8660 return ixgbe_configure_clsu32_add_hnode(adapter, proto, 8661 tc->cls_u32); 8662 case TC_CLSU32_DELETE_HNODE: 8663 return ixgbe_configure_clsu32_del_hnode(adapter, 8664 tc->cls_u32); 8665 default: 8666 return -EINVAL; 8667 } 8668 } 8669 8670 if (tc->type != TC_SETUP_MQPRIO) 8671 return -EINVAL; 8672 8673 return ixgbe_setup_tc(dev, tc->tc); 8674 } 8675 8676 #ifdef CONFIG_PCI_IOV 8677 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter) 8678 { 8679 struct net_device *netdev = adapter->netdev; 8680 8681 rtnl_lock(); 8682 ixgbe_setup_tc(netdev, netdev_get_num_tc(netdev)); 8683 rtnl_unlock(); 8684 } 8685 8686 #endif 8687 void ixgbe_do_reset(struct net_device *netdev) 8688 { 8689 struct ixgbe_adapter *adapter = netdev_priv(netdev); 8690 8691 if (netif_running(netdev)) 8692 ixgbe_reinit_locked(adapter); 8693 else 8694 ixgbe_reset(adapter); 8695 } 8696 8697 static netdev_features_t ixgbe_fix_features(struct net_device *netdev, 8698 netdev_features_t features) 8699 { 8700 struct ixgbe_adapter *adapter = netdev_priv(netdev); 8701 8702 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */ 8703 if (!(features & NETIF_F_RXCSUM)) 8704 features &= ~NETIF_F_LRO; 8705 8706 /* Turn off LRO if not RSC capable */ 8707 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)) 8708 features &= ~NETIF_F_LRO; 8709 8710 return features; 8711 } 8712 8713 static int ixgbe_set_features(struct net_device *netdev, 8714 netdev_features_t features) 8715 { 8716 struct ixgbe_adapter *adapter = netdev_priv(netdev); 8717 netdev_features_t changed = netdev->features ^ features; 8718 bool need_reset = false; 8719 8720 /* Make sure RSC matches LRO, reset if change */ 8721 if (!(features & NETIF_F_LRO)) { 8722 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) 8723 need_reset = true; 8724 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED; 8725 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) && 8726 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) { 8727 if (adapter->rx_itr_setting == 1 || 8728 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) { 8729 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED; 8730 need_reset = true; 8731 } else if ((changed ^ features) & NETIF_F_LRO) { 8732 e_info(probe, "rx-usecs set too low, " 8733 "disabling RSC\n"); 8734 } 8735 } 8736 8737 /* 8738 * Check if Flow Director n-tuple support or hw_tc support was 8739 * enabled or disabled. If the state changed, we need to reset. 8740 */ 8741 if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) { 8742 /* turn off ATR, enable perfect filters and reset */ 8743 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) 8744 need_reset = true; 8745 8746 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE; 8747 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE; 8748 } else { 8749 /* turn off perfect filters, enable ATR and reset */ 8750 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) 8751 need_reset = true; 8752 8753 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE; 8754 8755 /* We cannot enable ATR if SR-IOV is enabled */ 8756 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED || 8757 /* We cannot enable ATR if we have 2 or more tcs */ 8758 (netdev_get_num_tc(netdev) > 1) || 8759 /* We cannot enable ATR if RSS is disabled */ 8760 (adapter->ring_feature[RING_F_RSS].limit <= 1) || 8761 /* A sample rate of 0 indicates ATR disabled */ 8762 (!adapter->atr_sample_rate)) 8763 ; /* do nothing not supported */ 8764 else /* otherwise supported and set the flag */ 8765 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE; 8766 } 8767 8768 if (changed & NETIF_F_RXALL) 8769 need_reset = true; 8770 8771 netdev->features = features; 8772 8773 #ifdef CONFIG_IXGBE_VXLAN 8774 if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) { 8775 if (features & NETIF_F_RXCSUM) 8776 adapter->flags2 |= IXGBE_FLAG2_VXLAN_REREG_NEEDED; 8777 else 8778 ixgbe_clear_vxlan_port(adapter); 8779 } 8780 #endif /* CONFIG_IXGBE_VXLAN */ 8781 8782 if (need_reset) 8783 ixgbe_do_reset(netdev); 8784 else if (changed & (NETIF_F_HW_VLAN_CTAG_RX | 8785 NETIF_F_HW_VLAN_CTAG_FILTER)) 8786 ixgbe_set_rx_mode(netdev); 8787 8788 return 0; 8789 } 8790 8791 #ifdef CONFIG_IXGBE_VXLAN 8792 /** 8793 * ixgbe_add_vxlan_port - Get notifications about VXLAN ports that come up 8794 * @dev: The port's netdev 8795 * @sa_family: Socket Family that VXLAN is notifiying us about 8796 * @port: New UDP port number that VXLAN started listening to 8797 **/ 8798 static void ixgbe_add_vxlan_port(struct net_device *dev, sa_family_t sa_family, 8799 __be16 port) 8800 { 8801 struct ixgbe_adapter *adapter = netdev_priv(dev); 8802 struct ixgbe_hw *hw = &adapter->hw; 8803 8804 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) 8805 return; 8806 8807 if (sa_family == AF_INET6) 8808 return; 8809 8810 if (adapter->vxlan_port == port) 8811 return; 8812 8813 if (adapter->vxlan_port) { 8814 netdev_info(dev, 8815 "Hit Max num of VXLAN ports, not adding port %d\n", 8816 ntohs(port)); 8817 return; 8818 } 8819 8820 adapter->vxlan_port = port; 8821 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, ntohs(port)); 8822 } 8823 8824 /** 8825 * ixgbe_del_vxlan_port - Get notifications about VXLAN ports that go away 8826 * @dev: The port's netdev 8827 * @sa_family: Socket Family that VXLAN is notifying us about 8828 * @port: UDP port number that VXLAN stopped listening to 8829 **/ 8830 static void ixgbe_del_vxlan_port(struct net_device *dev, sa_family_t sa_family, 8831 __be16 port) 8832 { 8833 struct ixgbe_adapter *adapter = netdev_priv(dev); 8834 8835 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) 8836 return; 8837 8838 if (sa_family == AF_INET6) 8839 return; 8840 8841 if (adapter->vxlan_port != port) { 8842 netdev_info(dev, "Port %d was not found, not deleting\n", 8843 ntohs(port)); 8844 return; 8845 } 8846 8847 ixgbe_clear_vxlan_port(adapter); 8848 adapter->flags2 |= IXGBE_FLAG2_VXLAN_REREG_NEEDED; 8849 } 8850 #endif /* CONFIG_IXGBE_VXLAN */ 8851 8852 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[], 8853 struct net_device *dev, 8854 const unsigned char *addr, u16 vid, 8855 u16 flags) 8856 { 8857 /* guarantee we can provide a unique filter for the unicast address */ 8858 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) { 8859 struct ixgbe_adapter *adapter = netdev_priv(dev); 8860 u16 pool = VMDQ_P(0); 8861 8862 if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool)) 8863 return -ENOMEM; 8864 } 8865 8866 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags); 8867 } 8868 8869 /** 8870 * ixgbe_configure_bridge_mode - set various bridge modes 8871 * @adapter - the private structure 8872 * @mode - requested bridge mode 8873 * 8874 * Configure some settings require for various bridge modes. 8875 **/ 8876 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter, 8877 __u16 mode) 8878 { 8879 struct ixgbe_hw *hw = &adapter->hw; 8880 unsigned int p, num_pools; 8881 u32 vmdctl; 8882 8883 switch (mode) { 8884 case BRIDGE_MODE_VEPA: 8885 /* disable Tx loopback, rely on switch hairpin mode */ 8886 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0); 8887 8888 /* must enable Rx switching replication to allow multicast 8889 * packet reception on all VFs, and to enable source address 8890 * pruning. 8891 */ 8892 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL); 8893 vmdctl |= IXGBE_VT_CTL_REPLEN; 8894 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl); 8895 8896 /* enable Rx source address pruning. Note, this requires 8897 * replication to be enabled or else it does nothing. 8898 */ 8899 num_pools = adapter->num_vfs + adapter->num_rx_pools; 8900 for (p = 0; p < num_pools; p++) { 8901 if (hw->mac.ops.set_source_address_pruning) 8902 hw->mac.ops.set_source_address_pruning(hw, 8903 true, 8904 p); 8905 } 8906 break; 8907 case BRIDGE_MODE_VEB: 8908 /* enable Tx loopback for internal VF/PF communication */ 8909 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 8910 IXGBE_PFDTXGSWC_VT_LBEN); 8911 8912 /* disable Rx switching replication unless we have SR-IOV 8913 * virtual functions 8914 */ 8915 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL); 8916 if (!adapter->num_vfs) 8917 vmdctl &= ~IXGBE_VT_CTL_REPLEN; 8918 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl); 8919 8920 /* disable Rx source address pruning, since we don't expect to 8921 * be receiving external loopback of our transmitted frames. 8922 */ 8923 num_pools = adapter->num_vfs + adapter->num_rx_pools; 8924 for (p = 0; p < num_pools; p++) { 8925 if (hw->mac.ops.set_source_address_pruning) 8926 hw->mac.ops.set_source_address_pruning(hw, 8927 false, 8928 p); 8929 } 8930 break; 8931 default: 8932 return -EINVAL; 8933 } 8934 8935 adapter->bridge_mode = mode; 8936 8937 e_info(drv, "enabling bridge mode: %s\n", 8938 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB"); 8939 8940 return 0; 8941 } 8942 8943 static int ixgbe_ndo_bridge_setlink(struct net_device *dev, 8944 struct nlmsghdr *nlh, u16 flags) 8945 { 8946 struct ixgbe_adapter *adapter = netdev_priv(dev); 8947 struct nlattr *attr, *br_spec; 8948 int rem; 8949 8950 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) 8951 return -EOPNOTSUPP; 8952 8953 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC); 8954 if (!br_spec) 8955 return -EINVAL; 8956 8957 nla_for_each_nested(attr, br_spec, rem) { 8958 int status; 8959 __u16 mode; 8960 8961 if (nla_type(attr) != IFLA_BRIDGE_MODE) 8962 continue; 8963 8964 if (nla_len(attr) < sizeof(mode)) 8965 return -EINVAL; 8966 8967 mode = nla_get_u16(attr); 8968 status = ixgbe_configure_bridge_mode(adapter, mode); 8969 if (status) 8970 return status; 8971 8972 break; 8973 } 8974 8975 return 0; 8976 } 8977 8978 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq, 8979 struct net_device *dev, 8980 u32 filter_mask, int nlflags) 8981 { 8982 struct ixgbe_adapter *adapter = netdev_priv(dev); 8983 8984 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) 8985 return 0; 8986 8987 return ndo_dflt_bridge_getlink(skb, pid, seq, dev, 8988 adapter->bridge_mode, 0, 0, nlflags, 8989 filter_mask, NULL); 8990 } 8991 8992 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev) 8993 { 8994 struct ixgbe_fwd_adapter *fwd_adapter = NULL; 8995 struct ixgbe_adapter *adapter = netdev_priv(pdev); 8996 int used_pools = adapter->num_vfs + adapter->num_rx_pools; 8997 unsigned int limit; 8998 int pool, err; 8999 9000 /* Hardware has a limited number of available pools. Each VF, and the 9001 * PF require a pool. Check to ensure we don't attempt to use more 9002 * then the available number of pools. 9003 */ 9004 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS) 9005 return ERR_PTR(-EINVAL); 9006 9007 #ifdef CONFIG_RPS 9008 if (vdev->num_rx_queues != vdev->num_tx_queues) { 9009 netdev_info(pdev, "%s: Only supports a single queue count for TX and RX\n", 9010 vdev->name); 9011 return ERR_PTR(-EINVAL); 9012 } 9013 #endif 9014 /* Check for hardware restriction on number of rx/tx queues */ 9015 if (vdev->num_tx_queues > IXGBE_MAX_L2A_QUEUES || 9016 vdev->num_tx_queues == IXGBE_BAD_L2A_QUEUE) { 9017 netdev_info(pdev, 9018 "%s: Supports RX/TX Queue counts 1,2, and 4\n", 9019 pdev->name); 9020 return ERR_PTR(-EINVAL); 9021 } 9022 9023 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && 9024 adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS - 1) || 9025 (adapter->num_rx_pools > IXGBE_MAX_MACVLANS)) 9026 return ERR_PTR(-EBUSY); 9027 9028 fwd_adapter = kzalloc(sizeof(*fwd_adapter), GFP_KERNEL); 9029 if (!fwd_adapter) 9030 return ERR_PTR(-ENOMEM); 9031 9032 pool = find_first_zero_bit(&adapter->fwd_bitmask, 32); 9033 adapter->num_rx_pools++; 9034 set_bit(pool, &adapter->fwd_bitmask); 9035 limit = find_last_bit(&adapter->fwd_bitmask, 32); 9036 9037 /* Enable VMDq flag so device will be set in VM mode */ 9038 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED | IXGBE_FLAG_SRIOV_ENABLED; 9039 adapter->ring_feature[RING_F_VMDQ].limit = limit + 1; 9040 adapter->ring_feature[RING_F_RSS].limit = vdev->num_tx_queues; 9041 9042 /* Force reinit of ring allocation with VMDQ enabled */ 9043 err = ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev)); 9044 if (err) 9045 goto fwd_add_err; 9046 fwd_adapter->pool = pool; 9047 fwd_adapter->real_adapter = adapter; 9048 err = ixgbe_fwd_ring_up(vdev, fwd_adapter); 9049 if (err) 9050 goto fwd_add_err; 9051 netif_tx_start_all_queues(vdev); 9052 return fwd_adapter; 9053 fwd_add_err: 9054 /* unwind counter and free adapter struct */ 9055 netdev_info(pdev, 9056 "%s: dfwd hardware acceleration failed\n", vdev->name); 9057 clear_bit(pool, &adapter->fwd_bitmask); 9058 adapter->num_rx_pools--; 9059 kfree(fwd_adapter); 9060 return ERR_PTR(err); 9061 } 9062 9063 static void ixgbe_fwd_del(struct net_device *pdev, void *priv) 9064 { 9065 struct ixgbe_fwd_adapter *fwd_adapter = priv; 9066 struct ixgbe_adapter *adapter = fwd_adapter->real_adapter; 9067 unsigned int limit; 9068 9069 clear_bit(fwd_adapter->pool, &adapter->fwd_bitmask); 9070 adapter->num_rx_pools--; 9071 9072 limit = find_last_bit(&adapter->fwd_bitmask, 32); 9073 adapter->ring_feature[RING_F_VMDQ].limit = limit + 1; 9074 ixgbe_fwd_ring_down(fwd_adapter->netdev, fwd_adapter); 9075 ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev)); 9076 netdev_dbg(pdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n", 9077 fwd_adapter->pool, adapter->num_rx_pools, 9078 fwd_adapter->rx_base_queue, 9079 fwd_adapter->rx_base_queue + adapter->num_rx_queues_per_pool, 9080 adapter->fwd_bitmask); 9081 kfree(fwd_adapter); 9082 } 9083 9084 #define IXGBE_MAX_MAC_HDR_LEN 127 9085 #define IXGBE_MAX_NETWORK_HDR_LEN 511 9086 9087 static netdev_features_t 9088 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev, 9089 netdev_features_t features) 9090 { 9091 unsigned int network_hdr_len, mac_hdr_len; 9092 9093 /* Make certain the headers can be described by a context descriptor */ 9094 mac_hdr_len = skb_network_header(skb) - skb->data; 9095 if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN)) 9096 return features & ~(NETIF_F_HW_CSUM | 9097 NETIF_F_SCTP_CRC | 9098 NETIF_F_HW_VLAN_CTAG_TX | 9099 NETIF_F_TSO | 9100 NETIF_F_TSO6); 9101 9102 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb); 9103 if (unlikely(network_hdr_len > IXGBE_MAX_NETWORK_HDR_LEN)) 9104 return features & ~(NETIF_F_HW_CSUM | 9105 NETIF_F_SCTP_CRC | 9106 NETIF_F_TSO | 9107 NETIF_F_TSO6); 9108 9109 /* We can only support IPV4 TSO in tunnels if we can mangle the 9110 * inner IP ID field, so strip TSO if MANGLEID is not supported. 9111 */ 9112 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) 9113 features &= ~NETIF_F_TSO; 9114 9115 return features; 9116 } 9117 9118 static const struct net_device_ops ixgbe_netdev_ops = { 9119 .ndo_open = ixgbe_open, 9120 .ndo_stop = ixgbe_close, 9121 .ndo_start_xmit = ixgbe_xmit_frame, 9122 .ndo_select_queue = ixgbe_select_queue, 9123 .ndo_set_rx_mode = ixgbe_set_rx_mode, 9124 .ndo_validate_addr = eth_validate_addr, 9125 .ndo_set_mac_address = ixgbe_set_mac, 9126 .ndo_change_mtu = ixgbe_change_mtu, 9127 .ndo_tx_timeout = ixgbe_tx_timeout, 9128 .ndo_set_tx_maxrate = ixgbe_tx_maxrate, 9129 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid, 9130 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid, 9131 .ndo_do_ioctl = ixgbe_ioctl, 9132 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac, 9133 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan, 9134 .ndo_set_vf_rate = ixgbe_ndo_set_vf_bw, 9135 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk, 9136 .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en, 9137 .ndo_set_vf_trust = ixgbe_ndo_set_vf_trust, 9138 .ndo_get_vf_config = ixgbe_ndo_get_vf_config, 9139 .ndo_get_stats64 = ixgbe_get_stats64, 9140 .ndo_setup_tc = __ixgbe_setup_tc, 9141 #ifdef CONFIG_NET_POLL_CONTROLLER 9142 .ndo_poll_controller = ixgbe_netpoll, 9143 #endif 9144 #ifdef CONFIG_NET_RX_BUSY_POLL 9145 .ndo_busy_poll = ixgbe_low_latency_recv, 9146 #endif 9147 #ifdef IXGBE_FCOE 9148 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get, 9149 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target, 9150 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put, 9151 .ndo_fcoe_enable = ixgbe_fcoe_enable, 9152 .ndo_fcoe_disable = ixgbe_fcoe_disable, 9153 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn, 9154 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo, 9155 #endif /* IXGBE_FCOE */ 9156 .ndo_set_features = ixgbe_set_features, 9157 .ndo_fix_features = ixgbe_fix_features, 9158 .ndo_fdb_add = ixgbe_ndo_fdb_add, 9159 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink, 9160 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink, 9161 .ndo_dfwd_add_station = ixgbe_fwd_add, 9162 .ndo_dfwd_del_station = ixgbe_fwd_del, 9163 #ifdef CONFIG_IXGBE_VXLAN 9164 .ndo_add_vxlan_port = ixgbe_add_vxlan_port, 9165 .ndo_del_vxlan_port = ixgbe_del_vxlan_port, 9166 #endif /* CONFIG_IXGBE_VXLAN */ 9167 .ndo_features_check = ixgbe_features_check, 9168 }; 9169 9170 /** 9171 * ixgbe_enumerate_functions - Get the number of ports this device has 9172 * @adapter: adapter structure 9173 * 9174 * This function enumerates the phsyical functions co-located on a single slot, 9175 * in order to determine how many ports a device has. This is most useful in 9176 * determining the required GT/s of PCIe bandwidth necessary for optimal 9177 * performance. 9178 **/ 9179 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter) 9180 { 9181 struct pci_dev *entry, *pdev = adapter->pdev; 9182 int physfns = 0; 9183 9184 /* Some cards can not use the generic count PCIe functions method, 9185 * because they are behind a parent switch, so we hardcode these with 9186 * the correct number of functions. 9187 */ 9188 if (ixgbe_pcie_from_parent(&adapter->hw)) 9189 physfns = 4; 9190 9191 list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) { 9192 /* don't count virtual functions */ 9193 if (entry->is_virtfn) 9194 continue; 9195 9196 /* When the devices on the bus don't all match our device ID, 9197 * we can't reliably determine the correct number of 9198 * functions. This can occur if a function has been direct 9199 * attached to a virtual machine using VT-d, for example. In 9200 * this case, simply return -1 to indicate this. 9201 */ 9202 if ((entry->vendor != pdev->vendor) || 9203 (entry->device != pdev->device)) 9204 return -1; 9205 9206 physfns++; 9207 } 9208 9209 return physfns; 9210 } 9211 9212 /** 9213 * ixgbe_wol_supported - Check whether device supports WoL 9214 * @adapter: the adapter private structure 9215 * @device_id: the device ID 9216 * @subdev_id: the subsystem device ID 9217 * 9218 * This function is used by probe and ethtool to determine 9219 * which devices have WoL support 9220 * 9221 **/ 9222 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id, 9223 u16 subdevice_id) 9224 { 9225 struct ixgbe_hw *hw = &adapter->hw; 9226 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK; 9227 9228 /* WOL not supported on 82598 */ 9229 if (hw->mac.type == ixgbe_mac_82598EB) 9230 return false; 9231 9232 /* check eeprom to see if WOL is enabled for X540 and newer */ 9233 if (hw->mac.type >= ixgbe_mac_X540) { 9234 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) || 9235 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) && 9236 (hw->bus.func == 0))) 9237 return true; 9238 } 9239 9240 /* WOL is determined based on device IDs for 82599 MACs */ 9241 switch (device_id) { 9242 case IXGBE_DEV_ID_82599_SFP: 9243 /* Only these subdevices could supports WOL */ 9244 switch (subdevice_id) { 9245 case IXGBE_SUBDEV_ID_82599_560FLR: 9246 case IXGBE_SUBDEV_ID_82599_LOM_SNAP6: 9247 case IXGBE_SUBDEV_ID_82599_SFP_WOL0: 9248 case IXGBE_SUBDEV_ID_82599_SFP_2OCP: 9249 /* only support first port */ 9250 if (hw->bus.func != 0) 9251 break; 9252 case IXGBE_SUBDEV_ID_82599_SP_560FLR: 9253 case IXGBE_SUBDEV_ID_82599_SFP: 9254 case IXGBE_SUBDEV_ID_82599_RNDC: 9255 case IXGBE_SUBDEV_ID_82599_ECNA_DP: 9256 case IXGBE_SUBDEV_ID_82599_SFP_1OCP: 9257 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1: 9258 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2: 9259 return true; 9260 } 9261 break; 9262 case IXGBE_DEV_ID_82599EN_SFP: 9263 /* Only these subdevices support WOL */ 9264 switch (subdevice_id) { 9265 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1: 9266 return true; 9267 } 9268 break; 9269 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE: 9270 /* All except this subdevice support WOL */ 9271 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ) 9272 return true; 9273 break; 9274 case IXGBE_DEV_ID_82599_KX4: 9275 return true; 9276 default: 9277 break; 9278 } 9279 9280 return false; 9281 } 9282 9283 /** 9284 * ixgbe_probe - Device Initialization Routine 9285 * @pdev: PCI device information struct 9286 * @ent: entry in ixgbe_pci_tbl 9287 * 9288 * Returns 0 on success, negative on failure 9289 * 9290 * ixgbe_probe initializes an adapter identified by a pci_dev structure. 9291 * The OS initialization, configuring of the adapter private structure, 9292 * and a hardware reset occur. 9293 **/ 9294 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 9295 { 9296 struct net_device *netdev; 9297 struct ixgbe_adapter *adapter = NULL; 9298 struct ixgbe_hw *hw; 9299 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data]; 9300 int i, err, pci_using_dac, expected_gts; 9301 unsigned int indices = MAX_TX_QUEUES; 9302 u8 part_str[IXGBE_PBANUM_LENGTH]; 9303 bool disable_dev = false; 9304 #ifdef IXGBE_FCOE 9305 u16 device_caps; 9306 #endif 9307 u32 eec; 9308 9309 /* Catch broken hardware that put the wrong VF device ID in 9310 * the PCIe SR-IOV capability. 9311 */ 9312 if (pdev->is_virtfn) { 9313 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n", 9314 pci_name(pdev), pdev->vendor, pdev->device); 9315 return -EINVAL; 9316 } 9317 9318 err = pci_enable_device_mem(pdev); 9319 if (err) 9320 return err; 9321 9322 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) { 9323 pci_using_dac = 1; 9324 } else { 9325 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); 9326 if (err) { 9327 dev_err(&pdev->dev, 9328 "No usable DMA configuration, aborting\n"); 9329 goto err_dma; 9330 } 9331 pci_using_dac = 0; 9332 } 9333 9334 err = pci_request_selected_regions(pdev, pci_select_bars(pdev, 9335 IORESOURCE_MEM), ixgbe_driver_name); 9336 if (err) { 9337 dev_err(&pdev->dev, 9338 "pci_request_selected_regions failed 0x%x\n", err); 9339 goto err_pci_reg; 9340 } 9341 9342 pci_enable_pcie_error_reporting(pdev); 9343 9344 pci_set_master(pdev); 9345 pci_save_state(pdev); 9346 9347 if (ii->mac == ixgbe_mac_82598EB) { 9348 #ifdef CONFIG_IXGBE_DCB 9349 /* 8 TC w/ 4 queues per TC */ 9350 indices = 4 * MAX_TRAFFIC_CLASS; 9351 #else 9352 indices = IXGBE_MAX_RSS_INDICES; 9353 #endif 9354 } 9355 9356 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices); 9357 if (!netdev) { 9358 err = -ENOMEM; 9359 goto err_alloc_etherdev; 9360 } 9361 9362 SET_NETDEV_DEV(netdev, &pdev->dev); 9363 9364 adapter = netdev_priv(netdev); 9365 9366 adapter->netdev = netdev; 9367 adapter->pdev = pdev; 9368 hw = &adapter->hw; 9369 hw->back = adapter; 9370 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE); 9371 9372 hw->hw_addr = ioremap(pci_resource_start(pdev, 0), 9373 pci_resource_len(pdev, 0)); 9374 adapter->io_addr = hw->hw_addr; 9375 if (!hw->hw_addr) { 9376 err = -EIO; 9377 goto err_ioremap; 9378 } 9379 9380 netdev->netdev_ops = &ixgbe_netdev_ops; 9381 ixgbe_set_ethtool_ops(netdev); 9382 netdev->watchdog_timeo = 5 * HZ; 9383 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name)); 9384 9385 /* Setup hw api */ 9386 hw->mac.ops = *ii->mac_ops; 9387 hw->mac.type = ii->mac; 9388 hw->mvals = ii->mvals; 9389 9390 /* EEPROM */ 9391 hw->eeprom.ops = *ii->eeprom_ops; 9392 eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw)); 9393 if (ixgbe_removed(hw->hw_addr)) { 9394 err = -EIO; 9395 goto err_ioremap; 9396 } 9397 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */ 9398 if (!(eec & BIT(8))) 9399 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic; 9400 9401 /* PHY */ 9402 hw->phy.ops = *ii->phy_ops; 9403 hw->phy.sfp_type = ixgbe_sfp_type_unknown; 9404 /* ixgbe_identify_phy_generic will set prtad and mmds properly */ 9405 hw->phy.mdio.prtad = MDIO_PRTAD_NONE; 9406 hw->phy.mdio.mmds = 0; 9407 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22; 9408 hw->phy.mdio.dev = netdev; 9409 hw->phy.mdio.mdio_read = ixgbe_mdio_read; 9410 hw->phy.mdio.mdio_write = ixgbe_mdio_write; 9411 9412 ii->get_invariants(hw); 9413 9414 /* setup the private structure */ 9415 err = ixgbe_sw_init(adapter); 9416 if (err) 9417 goto err_sw_init; 9418 9419 /* Make sure the SWFW semaphore is in a valid state */ 9420 if (hw->mac.ops.init_swfw_sync) 9421 hw->mac.ops.init_swfw_sync(hw); 9422 9423 /* Make it possible the adapter to be woken up via WOL */ 9424 switch (adapter->hw.mac.type) { 9425 case ixgbe_mac_82599EB: 9426 case ixgbe_mac_X540: 9427 case ixgbe_mac_X550: 9428 case ixgbe_mac_X550EM_x: 9429 case ixgbe_mac_x550em_a: 9430 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0); 9431 break; 9432 default: 9433 break; 9434 } 9435 9436 /* 9437 * If there is a fan on this device and it has failed log the 9438 * failure. 9439 */ 9440 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) { 9441 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP); 9442 if (esdp & IXGBE_ESDP_SDP1) 9443 e_crit(probe, "Fan has stopped, replace the adapter\n"); 9444 } 9445 9446 if (allow_unsupported_sfp) 9447 hw->allow_unsupported_sfp = allow_unsupported_sfp; 9448 9449 /* reset_hw fills in the perm_addr as well */ 9450 hw->phy.reset_if_overtemp = true; 9451 err = hw->mac.ops.reset_hw(hw); 9452 hw->phy.reset_if_overtemp = false; 9453 if (err == IXGBE_ERR_SFP_NOT_PRESENT) { 9454 err = 0; 9455 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) { 9456 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n"); 9457 e_dev_err("Reload the driver after installing a supported module.\n"); 9458 goto err_sw_init; 9459 } else if (err) { 9460 e_dev_err("HW Init failed: %d\n", err); 9461 goto err_sw_init; 9462 } 9463 9464 #ifdef CONFIG_PCI_IOV 9465 /* SR-IOV not supported on the 82598 */ 9466 if (adapter->hw.mac.type == ixgbe_mac_82598EB) 9467 goto skip_sriov; 9468 /* Mailbox */ 9469 ixgbe_init_mbx_params_pf(hw); 9470 hw->mbx.ops = ii->mbx_ops; 9471 pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT); 9472 ixgbe_enable_sriov(adapter); 9473 skip_sriov: 9474 9475 #endif 9476 netdev->features = NETIF_F_SG | 9477 NETIF_F_TSO | 9478 NETIF_F_TSO6 | 9479 NETIF_F_RXHASH | 9480 NETIF_F_RXCSUM | 9481 NETIF_F_HW_CSUM; 9482 9483 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \ 9484 NETIF_F_GSO_GRE_CSUM | \ 9485 NETIF_F_GSO_IPXIP4 | \ 9486 NETIF_F_GSO_IPXIP6 | \ 9487 NETIF_F_GSO_UDP_TUNNEL | \ 9488 NETIF_F_GSO_UDP_TUNNEL_CSUM) 9489 9490 netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES; 9491 netdev->features |= NETIF_F_GSO_PARTIAL | 9492 IXGBE_GSO_PARTIAL_FEATURES; 9493 9494 if (hw->mac.type >= ixgbe_mac_82599EB) 9495 netdev->features |= NETIF_F_SCTP_CRC; 9496 9497 /* copy netdev features into list of user selectable features */ 9498 netdev->hw_features |= netdev->features | 9499 NETIF_F_HW_VLAN_CTAG_RX | 9500 NETIF_F_HW_VLAN_CTAG_TX | 9501 NETIF_F_RXALL | 9502 NETIF_F_HW_L2FW_DOFFLOAD; 9503 9504 if (hw->mac.type >= ixgbe_mac_82599EB) 9505 netdev->hw_features |= NETIF_F_NTUPLE | 9506 NETIF_F_HW_TC; 9507 9508 if (pci_using_dac) 9509 netdev->features |= NETIF_F_HIGHDMA; 9510 9511 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID; 9512 netdev->hw_enc_features |= netdev->vlan_features; 9513 netdev->mpls_features |= NETIF_F_HW_CSUM; 9514 9515 /* set this bit last since it cannot be part of vlan_features */ 9516 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER | 9517 NETIF_F_HW_VLAN_CTAG_RX | 9518 NETIF_F_HW_VLAN_CTAG_TX; 9519 9520 netdev->priv_flags |= IFF_UNICAST_FLT; 9521 netdev->priv_flags |= IFF_SUPP_NOFCS; 9522 9523 #ifdef CONFIG_IXGBE_DCB 9524 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE) 9525 netdev->dcbnl_ops = &dcbnl_ops; 9526 #endif 9527 9528 #ifdef IXGBE_FCOE 9529 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) { 9530 unsigned int fcoe_l; 9531 9532 if (hw->mac.ops.get_device_caps) { 9533 hw->mac.ops.get_device_caps(hw, &device_caps); 9534 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS) 9535 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE; 9536 } 9537 9538 9539 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus()); 9540 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l; 9541 9542 netdev->features |= NETIF_F_FSO | 9543 NETIF_F_FCOE_CRC; 9544 9545 netdev->vlan_features |= NETIF_F_FSO | 9546 NETIF_F_FCOE_CRC | 9547 NETIF_F_FCOE_MTU; 9548 } 9549 #endif /* IXGBE_FCOE */ 9550 9551 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) 9552 netdev->hw_features |= NETIF_F_LRO; 9553 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) 9554 netdev->features |= NETIF_F_LRO; 9555 9556 /* make sure the EEPROM is good */ 9557 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) { 9558 e_dev_err("The EEPROM Checksum Is Not Valid\n"); 9559 err = -EIO; 9560 goto err_sw_init; 9561 } 9562 9563 eth_platform_get_mac_address(&adapter->pdev->dev, 9564 adapter->hw.mac.perm_addr); 9565 9566 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len); 9567 9568 if (!is_valid_ether_addr(netdev->dev_addr)) { 9569 e_dev_err("invalid MAC address\n"); 9570 err = -EIO; 9571 goto err_sw_init; 9572 } 9573 9574 /* Set hw->mac.addr to permanent MAC address */ 9575 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr); 9576 ixgbe_mac_set_default_filter(adapter); 9577 9578 setup_timer(&adapter->service_timer, &ixgbe_service_timer, 9579 (unsigned long) adapter); 9580 9581 if (ixgbe_removed(hw->hw_addr)) { 9582 err = -EIO; 9583 goto err_sw_init; 9584 } 9585 INIT_WORK(&adapter->service_task, ixgbe_service_task); 9586 set_bit(__IXGBE_SERVICE_INITED, &adapter->state); 9587 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state); 9588 9589 err = ixgbe_init_interrupt_scheme(adapter); 9590 if (err) 9591 goto err_sw_init; 9592 9593 /* WOL not supported for all devices */ 9594 adapter->wol = 0; 9595 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap); 9596 hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device, 9597 pdev->subsystem_device); 9598 if (hw->wol_enabled) 9599 adapter->wol = IXGBE_WUFC_MAG; 9600 9601 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol); 9602 9603 /* save off EEPROM version number */ 9604 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh); 9605 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl); 9606 9607 /* pick up the PCI bus settings for reporting later */ 9608 if (ixgbe_pcie_from_parent(hw)) 9609 ixgbe_get_parent_bus_info(adapter); 9610 else 9611 hw->mac.ops.get_bus_info(hw); 9612 9613 /* calculate the expected PCIe bandwidth required for optimal 9614 * performance. Note that some older parts will never have enough 9615 * bandwidth due to being older generation PCIe parts. We clamp these 9616 * parts to ensure no warning is displayed if it can't be fixed. 9617 */ 9618 switch (hw->mac.type) { 9619 case ixgbe_mac_82598EB: 9620 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16); 9621 break; 9622 default: 9623 expected_gts = ixgbe_enumerate_functions(adapter) * 10; 9624 break; 9625 } 9626 9627 /* don't check link if we failed to enumerate functions */ 9628 if (expected_gts > 0) 9629 ixgbe_check_minimum_link(adapter, expected_gts); 9630 9631 err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str)); 9632 if (err) 9633 strlcpy(part_str, "Unknown", sizeof(part_str)); 9634 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present) 9635 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n", 9636 hw->mac.type, hw->phy.type, hw->phy.sfp_type, 9637 part_str); 9638 else 9639 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n", 9640 hw->mac.type, hw->phy.type, part_str); 9641 9642 e_dev_info("%pM\n", netdev->dev_addr); 9643 9644 /* reset the hardware with the new settings */ 9645 err = hw->mac.ops.start_hw(hw); 9646 if (err == IXGBE_ERR_EEPROM_VERSION) { 9647 /* We are running on a pre-production device, log a warning */ 9648 e_dev_warn("This device is a pre-production adapter/LOM. " 9649 "Please be aware there may be issues associated " 9650 "with your hardware. If you are experiencing " 9651 "problems please contact your Intel or hardware " 9652 "representative who provided you with this " 9653 "hardware.\n"); 9654 } 9655 strcpy(netdev->name, "eth%d"); 9656 err = register_netdev(netdev); 9657 if (err) 9658 goto err_register; 9659 9660 pci_set_drvdata(pdev, adapter); 9661 9662 /* power down the optics for 82599 SFP+ fiber */ 9663 if (hw->mac.ops.disable_tx_laser) 9664 hw->mac.ops.disable_tx_laser(hw); 9665 9666 /* carrier off reporting is important to ethtool even BEFORE open */ 9667 netif_carrier_off(netdev); 9668 9669 #ifdef CONFIG_IXGBE_DCA 9670 if (dca_add_requester(&pdev->dev) == 0) { 9671 adapter->flags |= IXGBE_FLAG_DCA_ENABLED; 9672 ixgbe_setup_dca(adapter); 9673 } 9674 #endif 9675 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) { 9676 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs); 9677 for (i = 0; i < adapter->num_vfs; i++) 9678 ixgbe_vf_configuration(pdev, (i | 0x10000000)); 9679 } 9680 9681 /* firmware requires driver version to be 0xFFFFFFFF 9682 * since os does not support feature 9683 */ 9684 if (hw->mac.ops.set_fw_drv_ver) 9685 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 9686 0xFF); 9687 9688 /* add san mac addr to netdev */ 9689 ixgbe_add_sanmac_netdev(netdev); 9690 9691 e_dev_info("%s\n", ixgbe_default_device_descr); 9692 9693 #ifdef CONFIG_IXGBE_HWMON 9694 if (ixgbe_sysfs_init(adapter)) 9695 e_err(probe, "failed to allocate sysfs resources\n"); 9696 #endif /* CONFIG_IXGBE_HWMON */ 9697 9698 ixgbe_dbg_adapter_init(adapter); 9699 9700 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */ 9701 if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link) 9702 hw->mac.ops.setup_link(hw, 9703 IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL, 9704 true); 9705 9706 return 0; 9707 9708 err_register: 9709 ixgbe_release_hw_control(adapter); 9710 ixgbe_clear_interrupt_scheme(adapter); 9711 err_sw_init: 9712 ixgbe_disable_sriov(adapter); 9713 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP; 9714 iounmap(adapter->io_addr); 9715 kfree(adapter->jump_tables[0]); 9716 kfree(adapter->mac_table); 9717 err_ioremap: 9718 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state); 9719 free_netdev(netdev); 9720 err_alloc_etherdev: 9721 pci_release_selected_regions(pdev, 9722 pci_select_bars(pdev, IORESOURCE_MEM)); 9723 err_pci_reg: 9724 err_dma: 9725 if (!adapter || disable_dev) 9726 pci_disable_device(pdev); 9727 return err; 9728 } 9729 9730 /** 9731 * ixgbe_remove - Device Removal Routine 9732 * @pdev: PCI device information struct 9733 * 9734 * ixgbe_remove is called by the PCI subsystem to alert the driver 9735 * that it should release a PCI device. The could be caused by a 9736 * Hot-Plug event, or because the driver is going to be removed from 9737 * memory. 9738 **/ 9739 static void ixgbe_remove(struct pci_dev *pdev) 9740 { 9741 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev); 9742 struct net_device *netdev; 9743 bool disable_dev; 9744 int i; 9745 9746 /* if !adapter then we already cleaned up in probe */ 9747 if (!adapter) 9748 return; 9749 9750 netdev = adapter->netdev; 9751 ixgbe_dbg_adapter_exit(adapter); 9752 9753 set_bit(__IXGBE_REMOVING, &adapter->state); 9754 cancel_work_sync(&adapter->service_task); 9755 9756 9757 #ifdef CONFIG_IXGBE_DCA 9758 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) { 9759 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED; 9760 dca_remove_requester(&pdev->dev); 9761 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 9762 IXGBE_DCA_CTRL_DCA_DISABLE); 9763 } 9764 9765 #endif 9766 #ifdef CONFIG_IXGBE_HWMON 9767 ixgbe_sysfs_exit(adapter); 9768 #endif /* CONFIG_IXGBE_HWMON */ 9769 9770 /* remove the added san mac */ 9771 ixgbe_del_sanmac_netdev(netdev); 9772 9773 #ifdef CONFIG_PCI_IOV 9774 ixgbe_disable_sriov(adapter); 9775 #endif 9776 if (netdev->reg_state == NETREG_REGISTERED) 9777 unregister_netdev(netdev); 9778 9779 ixgbe_clear_interrupt_scheme(adapter); 9780 9781 ixgbe_release_hw_control(adapter); 9782 9783 #ifdef CONFIG_DCB 9784 kfree(adapter->ixgbe_ieee_pfc); 9785 kfree(adapter->ixgbe_ieee_ets); 9786 9787 #endif 9788 iounmap(adapter->io_addr); 9789 pci_release_selected_regions(pdev, pci_select_bars(pdev, 9790 IORESOURCE_MEM)); 9791 9792 e_dev_info("complete\n"); 9793 9794 for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) { 9795 if (adapter->jump_tables[i]) { 9796 kfree(adapter->jump_tables[i]->input); 9797 kfree(adapter->jump_tables[i]->mask); 9798 } 9799 kfree(adapter->jump_tables[i]); 9800 } 9801 9802 kfree(adapter->mac_table); 9803 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state); 9804 free_netdev(netdev); 9805 9806 pci_disable_pcie_error_reporting(pdev); 9807 9808 if (disable_dev) 9809 pci_disable_device(pdev); 9810 } 9811 9812 /** 9813 * ixgbe_io_error_detected - called when PCI error is detected 9814 * @pdev: Pointer to PCI device 9815 * @state: The current pci connection state 9816 * 9817 * This function is called after a PCI bus error affecting 9818 * this device has been detected. 9819 */ 9820 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev, 9821 pci_channel_state_t state) 9822 { 9823 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev); 9824 struct net_device *netdev = adapter->netdev; 9825 9826 #ifdef CONFIG_PCI_IOV 9827 struct ixgbe_hw *hw = &adapter->hw; 9828 struct pci_dev *bdev, *vfdev; 9829 u32 dw0, dw1, dw2, dw3; 9830 int vf, pos; 9831 u16 req_id, pf_func; 9832 9833 if (adapter->hw.mac.type == ixgbe_mac_82598EB || 9834 adapter->num_vfs == 0) 9835 goto skip_bad_vf_detection; 9836 9837 bdev = pdev->bus->self; 9838 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT)) 9839 bdev = bdev->bus->self; 9840 9841 if (!bdev) 9842 goto skip_bad_vf_detection; 9843 9844 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR); 9845 if (!pos) 9846 goto skip_bad_vf_detection; 9847 9848 dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG); 9849 dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4); 9850 dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8); 9851 dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12); 9852 if (ixgbe_removed(hw->hw_addr)) 9853 goto skip_bad_vf_detection; 9854 9855 req_id = dw1 >> 16; 9856 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */ 9857 if (!(req_id & 0x0080)) 9858 goto skip_bad_vf_detection; 9859 9860 pf_func = req_id & 0x01; 9861 if ((pf_func & 1) == (pdev->devfn & 1)) { 9862 unsigned int device_id; 9863 9864 vf = (req_id & 0x7F) >> 1; 9865 e_dev_err("VF %d has caused a PCIe error\n", vf); 9866 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: " 9867 "%8.8x\tdw3: %8.8x\n", 9868 dw0, dw1, dw2, dw3); 9869 switch (adapter->hw.mac.type) { 9870 case ixgbe_mac_82599EB: 9871 device_id = IXGBE_82599_VF_DEVICE_ID; 9872 break; 9873 case ixgbe_mac_X540: 9874 device_id = IXGBE_X540_VF_DEVICE_ID; 9875 break; 9876 case ixgbe_mac_X550: 9877 device_id = IXGBE_DEV_ID_X550_VF; 9878 break; 9879 case ixgbe_mac_X550EM_x: 9880 device_id = IXGBE_DEV_ID_X550EM_X_VF; 9881 break; 9882 case ixgbe_mac_x550em_a: 9883 device_id = IXGBE_DEV_ID_X550EM_A_VF; 9884 break; 9885 default: 9886 device_id = 0; 9887 break; 9888 } 9889 9890 /* Find the pci device of the offending VF */ 9891 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL); 9892 while (vfdev) { 9893 if (vfdev->devfn == (req_id & 0xFF)) 9894 break; 9895 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, 9896 device_id, vfdev); 9897 } 9898 /* 9899 * There's a slim chance the VF could have been hot plugged, 9900 * so if it is no longer present we don't need to issue the 9901 * VFLR. Just clean up the AER in that case. 9902 */ 9903 if (vfdev) { 9904 ixgbe_issue_vf_flr(adapter, vfdev); 9905 /* Free device reference count */ 9906 pci_dev_put(vfdev); 9907 } 9908 9909 pci_cleanup_aer_uncorrect_error_status(pdev); 9910 } 9911 9912 /* 9913 * Even though the error may have occurred on the other port 9914 * we still need to increment the vf error reference count for 9915 * both ports because the I/O resume function will be called 9916 * for both of them. 9917 */ 9918 adapter->vferr_refcount++; 9919 9920 return PCI_ERS_RESULT_RECOVERED; 9921 9922 skip_bad_vf_detection: 9923 #endif /* CONFIG_PCI_IOV */ 9924 if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state)) 9925 return PCI_ERS_RESULT_DISCONNECT; 9926 9927 rtnl_lock(); 9928 netif_device_detach(netdev); 9929 9930 if (state == pci_channel_io_perm_failure) { 9931 rtnl_unlock(); 9932 return PCI_ERS_RESULT_DISCONNECT; 9933 } 9934 9935 if (netif_running(netdev)) 9936 ixgbe_down(adapter); 9937 9938 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state)) 9939 pci_disable_device(pdev); 9940 rtnl_unlock(); 9941 9942 /* Request a slot reset. */ 9943 return PCI_ERS_RESULT_NEED_RESET; 9944 } 9945 9946 /** 9947 * ixgbe_io_slot_reset - called after the pci bus has been reset. 9948 * @pdev: Pointer to PCI device 9949 * 9950 * Restart the card from scratch, as if from a cold-boot. 9951 */ 9952 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev) 9953 { 9954 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev); 9955 pci_ers_result_t result; 9956 int err; 9957 9958 if (pci_enable_device_mem(pdev)) { 9959 e_err(probe, "Cannot re-enable PCI device after reset.\n"); 9960 result = PCI_ERS_RESULT_DISCONNECT; 9961 } else { 9962 smp_mb__before_atomic(); 9963 clear_bit(__IXGBE_DISABLED, &adapter->state); 9964 adapter->hw.hw_addr = adapter->io_addr; 9965 pci_set_master(pdev); 9966 pci_restore_state(pdev); 9967 pci_save_state(pdev); 9968 9969 pci_wake_from_d3(pdev, false); 9970 9971 ixgbe_reset(adapter); 9972 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0); 9973 result = PCI_ERS_RESULT_RECOVERED; 9974 } 9975 9976 err = pci_cleanup_aer_uncorrect_error_status(pdev); 9977 if (err) { 9978 e_dev_err("pci_cleanup_aer_uncorrect_error_status " 9979 "failed 0x%0x\n", err); 9980 /* non-fatal, continue */ 9981 } 9982 9983 return result; 9984 } 9985 9986 /** 9987 * ixgbe_io_resume - called when traffic can start flowing again. 9988 * @pdev: Pointer to PCI device 9989 * 9990 * This callback is called when the error recovery driver tells us that 9991 * its OK to resume normal operation. 9992 */ 9993 static void ixgbe_io_resume(struct pci_dev *pdev) 9994 { 9995 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev); 9996 struct net_device *netdev = adapter->netdev; 9997 9998 #ifdef CONFIG_PCI_IOV 9999 if (adapter->vferr_refcount) { 10000 e_info(drv, "Resuming after VF err\n"); 10001 adapter->vferr_refcount--; 10002 return; 10003 } 10004 10005 #endif 10006 if (netif_running(netdev)) 10007 ixgbe_up(adapter); 10008 10009 netif_device_attach(netdev); 10010 } 10011 10012 static const struct pci_error_handlers ixgbe_err_handler = { 10013 .error_detected = ixgbe_io_error_detected, 10014 .slot_reset = ixgbe_io_slot_reset, 10015 .resume = ixgbe_io_resume, 10016 }; 10017 10018 static struct pci_driver ixgbe_driver = { 10019 .name = ixgbe_driver_name, 10020 .id_table = ixgbe_pci_tbl, 10021 .probe = ixgbe_probe, 10022 .remove = ixgbe_remove, 10023 #ifdef CONFIG_PM 10024 .suspend = ixgbe_suspend, 10025 .resume = ixgbe_resume, 10026 #endif 10027 .shutdown = ixgbe_shutdown, 10028 .sriov_configure = ixgbe_pci_sriov_configure, 10029 .err_handler = &ixgbe_err_handler 10030 }; 10031 10032 /** 10033 * ixgbe_init_module - Driver Registration Routine 10034 * 10035 * ixgbe_init_module is the first routine called when the driver is 10036 * loaded. All it does is register with the PCI subsystem. 10037 **/ 10038 static int __init ixgbe_init_module(void) 10039 { 10040 int ret; 10041 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version); 10042 pr_info("%s\n", ixgbe_copyright); 10043 10044 ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name); 10045 if (!ixgbe_wq) { 10046 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name); 10047 return -ENOMEM; 10048 } 10049 10050 ixgbe_dbg_init(); 10051 10052 ret = pci_register_driver(&ixgbe_driver); 10053 if (ret) { 10054 ixgbe_dbg_exit(); 10055 return ret; 10056 } 10057 10058 #ifdef CONFIG_IXGBE_DCA 10059 dca_register_notify(&dca_notifier); 10060 #endif 10061 10062 return 0; 10063 } 10064 10065 module_init(ixgbe_init_module); 10066 10067 /** 10068 * ixgbe_exit_module - Driver Exit Cleanup Routine 10069 * 10070 * ixgbe_exit_module is called just before the driver is removed 10071 * from memory. 10072 **/ 10073 static void __exit ixgbe_exit_module(void) 10074 { 10075 #ifdef CONFIG_IXGBE_DCA 10076 dca_unregister_notify(&dca_notifier); 10077 #endif 10078 pci_unregister_driver(&ixgbe_driver); 10079 10080 ixgbe_dbg_exit(); 10081 if (ixgbe_wq) { 10082 destroy_workqueue(ixgbe_wq); 10083 ixgbe_wq = NULL; 10084 } 10085 } 10086 10087 #ifdef CONFIG_IXGBE_DCA 10088 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event, 10089 void *p) 10090 { 10091 int ret_val; 10092 10093 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event, 10094 __ixgbe_notify_dca); 10095 10096 return ret_val ? NOTIFY_BAD : NOTIFY_DONE; 10097 } 10098 10099 #endif /* CONFIG_IXGBE_DCA */ 10100 10101 module_exit(ixgbe_exit_module); 10102 10103 /* ixgbe_main.c */ 10104