1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
4 #include <linux/types.h>
5 #include <linux/module.h>
7 #include <linux/netdevice.h>
8 #include <linux/vmalloc.h>
9 #include <linux/string.h>
11 #include <linux/interrupt.h>
13 #include <linux/tcp.h>
14 #include <linux/sctp.h>
15 #include <linux/pkt_sched.h>
16 #include <linux/ipv6.h>
17 #include <linux/slab.h>
18 #include <net/checksum.h>
19 #include <net/ip6_checksum.h>
20 #include <linux/etherdevice.h>
21 #include <linux/ethtool.h>
23 #include <linux/if_vlan.h>
24 #include <linux/if_macvlan.h>
25 #include <linux/if_bridge.h>
26 #include <linux/prefetch.h>
27 #include <linux/bpf.h>
28 #include <linux/bpf_trace.h>
29 #include <linux/atomic.h>
30 #include <linux/numa.h>
31 #include <scsi/fc/fc_fcoe.h>
32 #include <net/udp_tunnel.h>
33 #include <net/pkt_cls.h>
34 #include <net/tc_act/tc_gact.h>
35 #include <net/tc_act/tc_mirred.h>
36 #include <net/vxlan.h>
38 #include <net/xdp_sock.h>
41 #include "ixgbe_common.h"
42 #include "ixgbe_dcb_82599.h"
43 #include "ixgbe_phy.h"
44 #include "ixgbe_sriov.h"
45 #include "ixgbe_model.h"
46 #include "ixgbe_txrx_common.h"
48 char ixgbe_driver_name[] = "ixgbe";
49 static const char ixgbe_driver_string[] =
50 "Intel(R) 10 Gigabit PCI Express Network Driver";
52 char ixgbe_default_device_descr[] =
53 "Intel(R) 10 Gigabit Network Connection";
55 static char ixgbe_default_device_descr[] =
56 "Intel(R) 10 Gigabit Network Connection";
58 #define DRV_VERSION "5.1.0-k"
59 const char ixgbe_driver_version[] = DRV_VERSION;
60 static const char ixgbe_copyright[] =
61 "Copyright (c) 1999-2016 Intel Corporation.";
63 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";
65 static const struct ixgbe_info *ixgbe_info_tbl[] = {
66 [board_82598] = &ixgbe_82598_info,
67 [board_82599] = &ixgbe_82599_info,
68 [board_X540] = &ixgbe_X540_info,
69 [board_X550] = &ixgbe_X550_info,
70 [board_X550EM_x] = &ixgbe_X550EM_x_info,
71 [board_x550em_x_fw] = &ixgbe_x550em_x_fw_info,
72 [board_x550em_a] = &ixgbe_x550em_a_info,
73 [board_x550em_a_fw] = &ixgbe_x550em_a_fw_info,
76 /* ixgbe_pci_tbl - PCI Device ID Table
78 * Wildcard entries (PCI_ANY_ID) should come last
79 * Last entry must be all 0s
81 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
82 * Class, Class Mask, private data (not used) }
84 static const struct pci_device_id ixgbe_pci_tbl[] = {
85 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T1), board_X550},
117 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
118 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_XFI), board_X550EM_x},
119 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
120 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x},
121 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x},
122 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_1G_T), board_x550em_x_fw},
123 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR), board_x550em_a },
124 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR_L), board_x550em_a },
125 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP_N), board_x550em_a },
126 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII), board_x550em_a },
127 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII_L), board_x550em_a },
128 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_10G_T), board_x550em_a},
129 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP), board_x550em_a },
130 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T), board_x550em_a_fw },
131 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T_L), board_x550em_a_fw },
132 /* required last entry */
135 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
137 #ifdef CONFIG_IXGBE_DCA
138 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
140 static struct notifier_block dca_notifier = {
141 .notifier_call = ixgbe_notify_dca,
147 #ifdef CONFIG_PCI_IOV
148 static unsigned int max_vfs;
149 module_param(max_vfs, uint, 0);
150 MODULE_PARM_DESC(max_vfs,
151 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
152 #endif /* CONFIG_PCI_IOV */
154 static unsigned int allow_unsupported_sfp;
155 module_param(allow_unsupported_sfp, uint, 0);
156 MODULE_PARM_DESC(allow_unsupported_sfp,
157 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
159 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
160 static int debug = -1;
161 module_param(debug, int, 0);
162 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
164 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
165 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
166 MODULE_LICENSE("GPL v2");
167 MODULE_VERSION(DRV_VERSION);
169 static struct workqueue_struct *ixgbe_wq;
171 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
172 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *);
174 static const struct net_device_ops ixgbe_netdev_ops;
176 static bool netif_is_ixgbe(struct net_device *dev)
178 return dev && (dev->netdev_ops == &ixgbe_netdev_ops);
181 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
184 struct pci_dev *parent_dev;
185 struct pci_bus *parent_bus;
187 parent_bus = adapter->pdev->bus->parent;
191 parent_dev = parent_bus->self;
195 if (!pci_is_pcie(parent_dev))
198 pcie_capability_read_word(parent_dev, reg, value);
199 if (*value == IXGBE_FAILED_READ_CFG_WORD &&
200 ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
205 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
207 struct ixgbe_hw *hw = &adapter->hw;
211 hw->bus.type = ixgbe_bus_type_pci_express;
213 /* Get the negotiated link width and speed from PCI config space of the
214 * parent, as this device is behind a switch
216 err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
218 /* assume caller will handle error case */
222 hw->bus.width = ixgbe_convert_bus_width(link_status);
223 hw->bus.speed = ixgbe_convert_bus_speed(link_status);
229 * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
230 * @hw: hw specific details
232 * This function is used by probe to determine whether a device's PCI-Express
233 * bandwidth details should be gathered from the parent bus instead of from the
234 * device. Used to ensure that various locations all have the correct device ID
237 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
239 switch (hw->device_id) {
240 case IXGBE_DEV_ID_82599_SFP_SF_QP:
241 case IXGBE_DEV_ID_82599_QSFP_SF_QP:
248 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
251 struct ixgbe_hw *hw = &adapter->hw;
252 struct pci_dev *pdev;
254 /* Some devices are not connected over PCIe and thus do not negotiate
255 * speed. These devices do not have valid bus info, and thus any report
256 * we generate may not be correct.
258 if (hw->bus.type == ixgbe_bus_type_internal)
261 /* determine whether to use the parent device */
262 if (ixgbe_pcie_from_parent(&adapter->hw))
263 pdev = adapter->pdev->bus->parent->self;
265 pdev = adapter->pdev;
267 pcie_print_link_status(pdev);
270 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
272 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
273 !test_bit(__IXGBE_REMOVING, &adapter->state) &&
274 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
275 queue_work(ixgbe_wq, &adapter->service_task);
278 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
280 struct ixgbe_adapter *adapter = hw->back;
285 e_dev_err("Adapter removed\n");
286 if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
287 ixgbe_service_event_schedule(adapter);
290 static u32 ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
292 u8 __iomem *reg_addr;
296 reg_addr = READ_ONCE(hw->hw_addr);
297 if (ixgbe_removed(reg_addr))
298 return IXGBE_FAILED_READ_REG;
300 /* Register read of 0xFFFFFFF can indicate the adapter has been removed,
301 * so perform several status register reads to determine if the adapter
304 for (i = 0; i < IXGBE_FAILED_READ_RETRIES; i++) {
305 value = readl(reg_addr + IXGBE_STATUS);
306 if (value != IXGBE_FAILED_READ_REG)
311 if (value == IXGBE_FAILED_READ_REG)
312 ixgbe_remove_adapter(hw);
314 value = readl(reg_addr + reg);
319 * ixgbe_read_reg - Read from device register
320 * @hw: hw specific details
321 * @reg: offset of register to read
323 * Returns : value read or IXGBE_FAILED_READ_REG if removed
325 * This function is used to read device registers. It checks for device
326 * removal by confirming any read that returns all ones by checking the
327 * status register value for all ones. This function avoids reading from
328 * the hardware if a removal was previously detected in which case it
329 * returns IXGBE_FAILED_READ_REG (all ones).
331 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
333 u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
336 if (ixgbe_removed(reg_addr))
337 return IXGBE_FAILED_READ_REG;
338 if (unlikely(hw->phy.nw_mng_if_sel &
339 IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE)) {
340 struct ixgbe_adapter *adapter;
343 for (i = 0; i < 200; ++i) {
344 value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY);
346 goto writes_completed;
347 if (value == IXGBE_FAILED_READ_REG) {
348 ixgbe_remove_adapter(hw);
349 return IXGBE_FAILED_READ_REG;
355 e_warn(hw, "register writes incomplete %08x\n", value);
359 value = readl(reg_addr + reg);
360 if (unlikely(value == IXGBE_FAILED_READ_REG))
361 value = ixgbe_check_remove(hw, reg);
365 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
369 pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
370 if (value == IXGBE_FAILED_READ_CFG_WORD) {
371 ixgbe_remove_adapter(hw);
377 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
379 struct ixgbe_adapter *adapter = hw->back;
382 if (ixgbe_removed(hw->hw_addr))
383 return IXGBE_FAILED_READ_CFG_WORD;
384 pci_read_config_word(adapter->pdev, reg, &value);
385 if (value == IXGBE_FAILED_READ_CFG_WORD &&
386 ixgbe_check_cfg_remove(hw, adapter->pdev))
387 return IXGBE_FAILED_READ_CFG_WORD;
391 #ifdef CONFIG_PCI_IOV
392 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
394 struct ixgbe_adapter *adapter = hw->back;
397 if (ixgbe_removed(hw->hw_addr))
398 return IXGBE_FAILED_READ_CFG_DWORD;
399 pci_read_config_dword(adapter->pdev, reg, &value);
400 if (value == IXGBE_FAILED_READ_CFG_DWORD &&
401 ixgbe_check_cfg_remove(hw, adapter->pdev))
402 return IXGBE_FAILED_READ_CFG_DWORD;
405 #endif /* CONFIG_PCI_IOV */
407 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
409 struct ixgbe_adapter *adapter = hw->back;
411 if (ixgbe_removed(hw->hw_addr))
413 pci_write_config_word(adapter->pdev, reg, value);
416 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
418 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
420 /* flush memory to make sure state is correct before next watchdog */
421 smp_mb__before_atomic();
422 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
425 struct ixgbe_reg_info {
430 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
432 /* General Registers */
433 {IXGBE_CTRL, "CTRL"},
434 {IXGBE_STATUS, "STATUS"},
435 {IXGBE_CTRL_EXT, "CTRL_EXT"},
437 /* Interrupt Registers */
438 {IXGBE_EICR, "EICR"},
441 {IXGBE_SRRCTL(0), "SRRCTL"},
442 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
443 {IXGBE_RDLEN(0), "RDLEN"},
444 {IXGBE_RDH(0), "RDH"},
445 {IXGBE_RDT(0), "RDT"},
446 {IXGBE_RXDCTL(0), "RXDCTL"},
447 {IXGBE_RDBAL(0), "RDBAL"},
448 {IXGBE_RDBAH(0), "RDBAH"},
451 {IXGBE_TDBAL(0), "TDBAL"},
452 {IXGBE_TDBAH(0), "TDBAH"},
453 {IXGBE_TDLEN(0), "TDLEN"},
454 {IXGBE_TDH(0), "TDH"},
455 {IXGBE_TDT(0), "TDT"},
456 {IXGBE_TXDCTL(0), "TXDCTL"},
458 /* List Terminator */
464 * ixgbe_regdump - register printout routine
466 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
472 switch (reginfo->ofs) {
473 case IXGBE_SRRCTL(0):
474 for (i = 0; i < 64; i++)
475 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
477 case IXGBE_DCA_RXCTRL(0):
478 for (i = 0; i < 64; i++)
479 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
482 for (i = 0; i < 64; i++)
483 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
486 for (i = 0; i < 64; i++)
487 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
490 for (i = 0; i < 64; i++)
491 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
493 case IXGBE_RXDCTL(0):
494 for (i = 0; i < 64; i++)
495 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
498 for (i = 0; i < 64; i++)
499 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
502 for (i = 0; i < 64; i++)
503 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
506 for (i = 0; i < 64; i++)
507 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
510 for (i = 0; i < 64; i++)
511 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
514 for (i = 0; i < 64; i++)
515 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
518 for (i = 0; i < 64; i++)
519 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
522 for (i = 0; i < 64; i++)
523 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
525 case IXGBE_TXDCTL(0):
526 for (i = 0; i < 64; i++)
527 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
530 pr_info("%-15s %08x\n",
531 reginfo->name, IXGBE_READ_REG(hw, reginfo->ofs));
541 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i, i + 7);
542 for (j = 0; j < 8; j++)
543 p += sprintf(p, " %08x", regs[i++]);
544 pr_err("%-15s%s\n", rname, buf);
549 static void ixgbe_print_buffer(struct ixgbe_ring *ring, int n)
551 struct ixgbe_tx_buffer *tx_buffer;
553 tx_buffer = &ring->tx_buffer_info[ring->next_to_clean];
554 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
555 n, ring->next_to_use, ring->next_to_clean,
556 (u64)dma_unmap_addr(tx_buffer, dma),
557 dma_unmap_len(tx_buffer, len),
558 tx_buffer->next_to_watch,
559 (u64)tx_buffer->time_stamp);
563 * ixgbe_dump - Print registers, tx-rings and rx-rings
565 static void ixgbe_dump(struct ixgbe_adapter *adapter)
567 struct net_device *netdev = adapter->netdev;
568 struct ixgbe_hw *hw = &adapter->hw;
569 struct ixgbe_reg_info *reginfo;
571 struct ixgbe_ring *ring;
572 struct ixgbe_tx_buffer *tx_buffer;
573 union ixgbe_adv_tx_desc *tx_desc;
574 struct my_u0 { u64 a; u64 b; } *u0;
575 struct ixgbe_ring *rx_ring;
576 union ixgbe_adv_rx_desc *rx_desc;
577 struct ixgbe_rx_buffer *rx_buffer_info;
580 if (!netif_msg_hw(adapter))
583 /* Print netdevice Info */
585 dev_info(&adapter->pdev->dev, "Net device Info\n");
586 pr_info("Device Name state "
588 pr_info("%-15s %016lX %016lX\n",
591 dev_trans_start(netdev));
594 /* Print Registers */
595 dev_info(&adapter->pdev->dev, "Register Dump\n");
596 pr_info(" Register Name Value\n");
597 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
598 reginfo->name; reginfo++) {
599 ixgbe_regdump(hw, reginfo);
602 /* Print TX Ring Summary */
603 if (!netdev || !netif_running(netdev))
606 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
607 pr_info(" %s %s %s %s\n",
608 "Queue [NTU] [NTC] [bi(ntc)->dma ]",
609 "leng", "ntw", "timestamp");
610 for (n = 0; n < adapter->num_tx_queues; n++) {
611 ring = adapter->tx_ring[n];
612 ixgbe_print_buffer(ring, n);
615 for (n = 0; n < adapter->num_xdp_queues; n++) {
616 ring = adapter->xdp_ring[n];
617 ixgbe_print_buffer(ring, n);
621 if (!netif_msg_tx_done(adapter))
622 goto rx_ring_summary;
624 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
626 /* Transmit Descriptor Formats
628 * 82598 Advanced Transmit Descriptor
629 * +--------------------------------------------------------------+
630 * 0 | Buffer Address [63:0] |
631 * +--------------------------------------------------------------+
632 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
633 * +--------------------------------------------------------------+
634 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
636 * 82598 Advanced Transmit Descriptor (Write-Back Format)
637 * +--------------------------------------------------------------+
639 * +--------------------------------------------------------------+
640 * 8 | RSV | STA | NXTSEQ |
641 * +--------------------------------------------------------------+
644 * 82599+ Advanced Transmit Descriptor
645 * +--------------------------------------------------------------+
646 * 0 | Buffer Address [63:0] |
647 * +--------------------------------------------------------------+
648 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN |
649 * +--------------------------------------------------------------+
650 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0
652 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
653 * +--------------------------------------------------------------+
655 * +--------------------------------------------------------------+
656 * 8 | RSV | STA | RSV |
657 * +--------------------------------------------------------------+
661 for (n = 0; n < adapter->num_tx_queues; n++) {
662 ring = adapter->tx_ring[n];
663 pr_info("------------------------------------\n");
664 pr_info("TX QUEUE INDEX = %d\n", ring->queue_index);
665 pr_info("------------------------------------\n");
666 pr_info("%s%s %s %s %s %s\n",
667 "T [desc] [address 63:0 ] ",
668 "[PlPOIdStDDt Ln] [bi->dma ] ",
669 "leng", "ntw", "timestamp", "bi->skb");
671 for (i = 0; ring->desc && (i < ring->count); i++) {
672 tx_desc = IXGBE_TX_DESC(ring, i);
673 tx_buffer = &ring->tx_buffer_info[i];
674 u0 = (struct my_u0 *)tx_desc;
675 if (dma_unmap_len(tx_buffer, len) > 0) {
676 const char *ring_desc;
678 if (i == ring->next_to_use &&
679 i == ring->next_to_clean)
680 ring_desc = " NTC/U";
681 else if (i == ring->next_to_use)
683 else if (i == ring->next_to_clean)
687 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p%s",
689 le64_to_cpu((__force __le64)u0->a),
690 le64_to_cpu((__force __le64)u0->b),
691 (u64)dma_unmap_addr(tx_buffer, dma),
692 dma_unmap_len(tx_buffer, len),
693 tx_buffer->next_to_watch,
694 (u64)tx_buffer->time_stamp,
698 if (netif_msg_pktdata(adapter) &&
700 print_hex_dump(KERN_INFO, "",
701 DUMP_PREFIX_ADDRESS, 16, 1,
702 tx_buffer->skb->data,
703 dma_unmap_len(tx_buffer, len),
709 /* Print RX Rings Summary */
711 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
712 pr_info("Queue [NTU] [NTC]\n");
713 for (n = 0; n < adapter->num_rx_queues; n++) {
714 rx_ring = adapter->rx_ring[n];
715 pr_info("%5d %5X %5X\n",
716 n, rx_ring->next_to_use, rx_ring->next_to_clean);
720 if (!netif_msg_rx_status(adapter))
723 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
725 /* Receive Descriptor Formats
727 * 82598 Advanced Receive Descriptor (Read) Format
729 * +-----------------------------------------------------+
730 * 0 | Packet Buffer Address [63:1] |A0/NSE|
731 * +----------------------------------------------+------+
732 * 8 | Header Buffer Address [63:1] | DD |
733 * +-----------------------------------------------------+
736 * 82598 Advanced Receive Descriptor (Write-Back) Format
738 * 63 48 47 32 31 30 21 20 16 15 4 3 0
739 * +------------------------------------------------------+
740 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS |
741 * | Packet | IP | | | | Type | Type |
742 * | Checksum | Ident | | | | | |
743 * +------------------------------------------------------+
744 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
745 * +------------------------------------------------------+
746 * 63 48 47 32 31 20 19 0
748 * 82599+ Advanced Receive Descriptor (Read) Format
750 * +-----------------------------------------------------+
751 * 0 | Packet Buffer Address [63:1] |A0/NSE|
752 * +----------------------------------------------+------+
753 * 8 | Header Buffer Address [63:1] | DD |
754 * +-----------------------------------------------------+
757 * 82599+ Advanced Receive Descriptor (Write-Back) Format
759 * 63 48 47 32 31 30 21 20 17 16 4 3 0
760 * +------------------------------------------------------+
761 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS |
762 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type |
763 * |/ Flow Dir Flt ID | | | | | |
764 * +------------------------------------------------------+
765 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
766 * +------------------------------------------------------+
767 * 63 48 47 32 31 20 19 0
770 for (n = 0; n < adapter->num_rx_queues; n++) {
771 rx_ring = adapter->rx_ring[n];
772 pr_info("------------------------------------\n");
773 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
774 pr_info("------------------------------------\n");
776 "R [desc] [ PktBuf A0] ",
777 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ",
778 "<-- Adv Rx Read format");
780 "RWB[desc] [PcsmIpSHl PtRs] ",
781 "[vl er S cks ln] ---------------- [bi->skb ] ",
782 "<-- Adv Rx Write-Back format");
784 for (i = 0; i < rx_ring->count; i++) {
785 const char *ring_desc;
787 if (i == rx_ring->next_to_use)
789 else if (i == rx_ring->next_to_clean)
794 rx_buffer_info = &rx_ring->rx_buffer_info[i];
795 rx_desc = IXGBE_RX_DESC(rx_ring, i);
796 u0 = (struct my_u0 *)rx_desc;
797 if (rx_desc->wb.upper.length) {
798 /* Descriptor Done */
799 pr_info("RWB[0x%03X] %016llX %016llX ---------------- %p%s\n",
801 le64_to_cpu((__force __le64)u0->a),
802 le64_to_cpu((__force __le64)u0->b),
806 pr_info("R [0x%03X] %016llX %016llX %016llX %p%s\n",
808 le64_to_cpu((__force __le64)u0->a),
809 le64_to_cpu((__force __le64)u0->b),
810 (u64)rx_buffer_info->dma,
814 if (netif_msg_pktdata(adapter) &&
815 rx_buffer_info->dma) {
816 print_hex_dump(KERN_INFO, "",
817 DUMP_PREFIX_ADDRESS, 16, 1,
818 page_address(rx_buffer_info->page) +
819 rx_buffer_info->page_offset,
820 ixgbe_rx_bufsz(rx_ring), true);
827 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
831 /* Let firmware take over control of h/w */
832 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
833 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
834 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
837 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
841 /* Let firmware know the driver has taken over */
842 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
843 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
844 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
848 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
849 * @adapter: pointer to adapter struct
850 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
851 * @queue: queue to map the corresponding interrupt to
852 * @msix_vector: the vector to map to the corresponding queue
855 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
856 u8 queue, u8 msix_vector)
859 struct ixgbe_hw *hw = &adapter->hw;
860 switch (hw->mac.type) {
861 case ixgbe_mac_82598EB:
862 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
865 index = (((direction * 64) + queue) >> 2) & 0x1F;
866 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
867 ivar &= ~(0xFF << (8 * (queue & 0x3)));
868 ivar |= (msix_vector << (8 * (queue & 0x3)));
869 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
871 case ixgbe_mac_82599EB:
874 case ixgbe_mac_X550EM_x:
875 case ixgbe_mac_x550em_a:
876 if (direction == -1) {
878 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
879 index = ((queue & 1) * 8);
880 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
881 ivar &= ~(0xFF << index);
882 ivar |= (msix_vector << index);
883 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
886 /* tx or rx causes */
887 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
888 index = ((16 * (queue & 1)) + (8 * direction));
889 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
890 ivar &= ~(0xFF << index);
891 ivar |= (msix_vector << index);
892 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
900 void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
905 switch (adapter->hw.mac.type) {
906 case ixgbe_mac_82598EB:
907 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
908 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
910 case ixgbe_mac_82599EB:
913 case ixgbe_mac_X550EM_x:
914 case ixgbe_mac_x550em_a:
915 mask = (qmask & 0xFFFFFFFF);
916 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
917 mask = (qmask >> 32);
918 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
925 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
927 struct ixgbe_hw *hw = &adapter->hw;
928 struct ixgbe_hw_stats *hwstats = &adapter->stats;
932 if ((hw->fc.current_mode != ixgbe_fc_full) &&
933 (hw->fc.current_mode != ixgbe_fc_rx_pause))
936 switch (hw->mac.type) {
937 case ixgbe_mac_82598EB:
938 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
941 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
943 hwstats->lxoffrxc += data;
945 /* refill credits (no tx hang) if we received xoff */
949 for (i = 0; i < adapter->num_tx_queues; i++)
950 clear_bit(__IXGBE_HANG_CHECK_ARMED,
951 &adapter->tx_ring[i]->state);
953 for (i = 0; i < adapter->num_xdp_queues; i++)
954 clear_bit(__IXGBE_HANG_CHECK_ARMED,
955 &adapter->xdp_ring[i]->state);
958 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
960 struct ixgbe_hw *hw = &adapter->hw;
961 struct ixgbe_hw_stats *hwstats = &adapter->stats;
965 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
967 if (adapter->ixgbe_ieee_pfc)
968 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
970 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
971 ixgbe_update_xoff_rx_lfc(adapter);
975 /* update stats for each tc, only valid with PFC enabled */
976 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
979 switch (hw->mac.type) {
980 case ixgbe_mac_82598EB:
981 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
984 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
986 hwstats->pxoffrxc[i] += pxoffrxc;
987 /* Get the TC for given UP */
988 tc = netdev_get_prio_tc_map(adapter->netdev, i);
989 xoff[tc] += pxoffrxc;
992 /* disarm tx queues that have received xoff frames */
993 for (i = 0; i < adapter->num_tx_queues; i++) {
994 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
996 tc = tx_ring->dcb_tc;
998 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1001 for (i = 0; i < adapter->num_xdp_queues; i++) {
1002 struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
1004 tc = xdp_ring->dcb_tc;
1006 clear_bit(__IXGBE_HANG_CHECK_ARMED, &xdp_ring->state);
1010 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1012 return ring->stats.packets;
1015 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1017 unsigned int head, tail;
1019 head = ring->next_to_clean;
1020 tail = ring->next_to_use;
1022 return ((head <= tail) ? tail : tail + ring->count) - head;
1025 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1027 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1028 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1029 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1031 clear_check_for_tx_hang(tx_ring);
1034 * Check for a hung queue, but be thorough. This verifies
1035 * that a transmit has been completed since the previous
1036 * check AND there is at least one packet pending. The
1037 * ARMED bit is set to indicate a potential hang. The
1038 * bit is cleared if a pause frame is received to remove
1039 * false hang detection due to PFC or 802.3x frames. By
1040 * requiring this to fail twice we avoid races with
1041 * pfc clearing the ARMED bit and conditions where we
1042 * run the check_tx_hang logic with a transmit completion
1043 * pending but without time to complete it yet.
1045 if (tx_done_old == tx_done && tx_pending)
1046 /* make sure it is true for two checks in a row */
1047 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1049 /* update completed stats and continue */
1050 tx_ring->tx_stats.tx_done_old = tx_done;
1051 /* reset the countdown */
1052 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1058 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1059 * @adapter: driver private struct
1061 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1064 /* Do the reset outside of interrupt context */
1065 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1066 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
1067 e_warn(drv, "initiating reset due to tx timeout\n");
1068 ixgbe_service_event_schedule(adapter);
1073 * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1074 * @netdev: network interface device structure
1075 * @queue_index: Tx queue to set
1076 * @maxrate: desired maximum transmit bitrate
1078 static int ixgbe_tx_maxrate(struct net_device *netdev,
1079 int queue_index, u32 maxrate)
1081 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1082 struct ixgbe_hw *hw = &adapter->hw;
1083 u32 bcnrc_val = ixgbe_link_mbps(adapter);
1088 /* Calculate the rate factor values to set */
1089 bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT;
1090 bcnrc_val /= maxrate;
1092 /* clear everything but the rate factor */
1093 bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK |
1094 IXGBE_RTTBCNRC_RF_DEC_MASK;
1096 /* enable the rate scheduler */
1097 bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA;
1099 IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index);
1100 IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val);
1106 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1107 * @q_vector: structure containing interrupt and ring information
1108 * @tx_ring: tx ring to clean
1109 * @napi_budget: Used to determine if we are in netpoll
1111 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1112 struct ixgbe_ring *tx_ring, int napi_budget)
1114 struct ixgbe_adapter *adapter = q_vector->adapter;
1115 struct ixgbe_tx_buffer *tx_buffer;
1116 union ixgbe_adv_tx_desc *tx_desc;
1117 unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
1118 unsigned int budget = q_vector->tx.work_limit;
1119 unsigned int i = tx_ring->next_to_clean;
1121 if (test_bit(__IXGBE_DOWN, &adapter->state))
1124 tx_buffer = &tx_ring->tx_buffer_info[i];
1125 tx_desc = IXGBE_TX_DESC(tx_ring, i);
1126 i -= tx_ring->count;
1129 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1131 /* if next_to_watch is not set then there is no work pending */
1135 /* prevent any other reads prior to eop_desc */
1138 /* if DD is not set pending work has not been completed */
1139 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1142 /* clear next_to_watch to prevent false hangs */
1143 tx_buffer->next_to_watch = NULL;
1145 /* update the statistics for this packet */
1146 total_bytes += tx_buffer->bytecount;
1147 total_packets += tx_buffer->gso_segs;
1148 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
1152 if (ring_is_xdp(tx_ring))
1153 xdp_return_frame(tx_buffer->xdpf);
1155 napi_consume_skb(tx_buffer->skb, napi_budget);
1157 /* unmap skb header data */
1158 dma_unmap_single(tx_ring->dev,
1159 dma_unmap_addr(tx_buffer, dma),
1160 dma_unmap_len(tx_buffer, len),
1163 /* clear tx_buffer data */
1164 dma_unmap_len_set(tx_buffer, len, 0);
1166 /* unmap remaining buffers */
1167 while (tx_desc != eop_desc) {
1172 i -= tx_ring->count;
1173 tx_buffer = tx_ring->tx_buffer_info;
1174 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1177 /* unmap any remaining paged data */
1178 if (dma_unmap_len(tx_buffer, len)) {
1179 dma_unmap_page(tx_ring->dev,
1180 dma_unmap_addr(tx_buffer, dma),
1181 dma_unmap_len(tx_buffer, len),
1183 dma_unmap_len_set(tx_buffer, len, 0);
1187 /* move us one more past the eop_desc for start of next pkt */
1192 i -= tx_ring->count;
1193 tx_buffer = tx_ring->tx_buffer_info;
1194 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1197 /* issue prefetch for next Tx descriptor */
1200 /* update budget accounting */
1202 } while (likely(budget));
1204 i += tx_ring->count;
1205 tx_ring->next_to_clean = i;
1206 u64_stats_update_begin(&tx_ring->syncp);
1207 tx_ring->stats.bytes += total_bytes;
1208 tx_ring->stats.packets += total_packets;
1209 u64_stats_update_end(&tx_ring->syncp);
1210 q_vector->tx.total_bytes += total_bytes;
1211 q_vector->tx.total_packets += total_packets;
1212 adapter->tx_ipsec += total_ipsec;
1214 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1215 /* schedule immediate reset if we believe we hung */
1216 struct ixgbe_hw *hw = &adapter->hw;
1217 e_err(drv, "Detected Tx Unit Hang %s\n"
1219 " TDH, TDT <%x>, <%x>\n"
1220 " next_to_use <%x>\n"
1221 " next_to_clean <%x>\n"
1222 "tx_buffer_info[next_to_clean]\n"
1223 " time_stamp <%lx>\n"
1225 ring_is_xdp(tx_ring) ? "(XDP)" : "",
1226 tx_ring->queue_index,
1227 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1228 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1229 tx_ring->next_to_use, i,
1230 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1232 if (!ring_is_xdp(tx_ring))
1233 netif_stop_subqueue(tx_ring->netdev,
1234 tx_ring->queue_index);
1237 "tx hang %d detected on queue %d, resetting adapter\n",
1238 adapter->tx_timeout_count + 1, tx_ring->queue_index);
1240 /* schedule immediate reset if we believe we hung */
1241 ixgbe_tx_timeout_reset(adapter);
1243 /* the adapter is about to reset, no point in enabling stuff */
1247 if (ring_is_xdp(tx_ring))
1250 netdev_tx_completed_queue(txring_txq(tx_ring),
1251 total_packets, total_bytes);
1253 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1254 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1255 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1256 /* Make sure that anybody stopping the queue after this
1257 * sees the new next_to_clean.
1260 if (__netif_subqueue_stopped(tx_ring->netdev,
1261 tx_ring->queue_index)
1262 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1263 netif_wake_subqueue(tx_ring->netdev,
1264 tx_ring->queue_index);
1265 ++tx_ring->tx_stats.restart_queue;
1272 #ifdef CONFIG_IXGBE_DCA
1273 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1274 struct ixgbe_ring *tx_ring,
1277 struct ixgbe_hw *hw = &adapter->hw;
1281 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1282 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1284 switch (hw->mac.type) {
1285 case ixgbe_mac_82598EB:
1286 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1288 case ixgbe_mac_82599EB:
1289 case ixgbe_mac_X540:
1290 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1291 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1294 /* for unknown hardware do not write register */
1299 * We can enable relaxed ordering for reads, but not writes when
1300 * DCA is enabled. This is due to a known issue in some chipsets
1301 * which will cause the DCA tag to be cleared.
1303 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1304 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1305 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1307 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1310 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1311 struct ixgbe_ring *rx_ring,
1314 struct ixgbe_hw *hw = &adapter->hw;
1316 u8 reg_idx = rx_ring->reg_idx;
1318 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1319 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1321 switch (hw->mac.type) {
1322 case ixgbe_mac_82599EB:
1323 case ixgbe_mac_X540:
1324 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1331 * We can enable relaxed ordering for reads, but not writes when
1332 * DCA is enabled. This is due to a known issue in some chipsets
1333 * which will cause the DCA tag to be cleared.
1335 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1336 IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1337 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1339 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1342 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1344 struct ixgbe_adapter *adapter = q_vector->adapter;
1345 struct ixgbe_ring *ring;
1346 int cpu = get_cpu();
1348 if (q_vector->cpu == cpu)
1351 ixgbe_for_each_ring(ring, q_vector->tx)
1352 ixgbe_update_tx_dca(adapter, ring, cpu);
1354 ixgbe_for_each_ring(ring, q_vector->rx)
1355 ixgbe_update_rx_dca(adapter, ring, cpu);
1357 q_vector->cpu = cpu;
1362 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1366 /* always use CB2 mode, difference is masked in the CB driver */
1367 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1368 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1369 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1371 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1372 IXGBE_DCA_CTRL_DCA_DISABLE);
1374 for (i = 0; i < adapter->num_q_vectors; i++) {
1375 adapter->q_vector[i]->cpu = -1;
1376 ixgbe_update_dca(adapter->q_vector[i]);
1380 static int __ixgbe_notify_dca(struct device *dev, void *data)
1382 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1383 unsigned long event = *(unsigned long *)data;
1385 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1389 case DCA_PROVIDER_ADD:
1390 /* if we're already enabled, don't do it again */
1391 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1393 if (dca_add_requester(dev) == 0) {
1394 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1395 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1396 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1399 /* fall through - DCA is disabled. */
1400 case DCA_PROVIDER_REMOVE:
1401 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1402 dca_remove_requester(dev);
1403 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1404 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1405 IXGBE_DCA_CTRL_DCA_DISABLE);
1413 #endif /* CONFIG_IXGBE_DCA */
1415 #define IXGBE_RSS_L4_TYPES_MASK \
1416 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1417 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1418 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1419 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1421 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1422 union ixgbe_adv_rx_desc *rx_desc,
1423 struct sk_buff *skb)
1427 if (!(ring->netdev->features & NETIF_F_RXHASH))
1430 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1431 IXGBE_RXDADV_RSSTYPE_MASK;
1436 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1437 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1438 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1443 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1444 * @ring: structure containing ring specific data
1445 * @rx_desc: advanced rx descriptor
1447 * Returns : true if it is FCoE pkt
1449 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1450 union ixgbe_adv_rx_desc *rx_desc)
1452 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1454 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1455 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1456 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1457 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1460 #endif /* IXGBE_FCOE */
1462 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1463 * @ring: structure containing ring specific data
1464 * @rx_desc: current Rx descriptor being processed
1465 * @skb: skb currently being received and modified
1467 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1468 union ixgbe_adv_rx_desc *rx_desc,
1469 struct sk_buff *skb)
1471 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1472 bool encap_pkt = false;
1474 skb_checksum_none_assert(skb);
1476 /* Rx csum disabled */
1477 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1480 /* check for VXLAN and Geneve packets */
1481 if (pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) {
1483 skb->encapsulation = 1;
1486 /* if IP and error */
1487 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1488 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1489 ring->rx_stats.csum_err++;
1493 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1496 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1498 * 82599 errata, UDP frames with a 0 checksum can be marked as
1501 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1502 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1505 ring->rx_stats.csum_err++;
1509 /* It must be a TCP or UDP packet with a valid checksum */
1510 skb->ip_summed = CHECKSUM_UNNECESSARY;
1512 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1515 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1516 skb->ip_summed = CHECKSUM_NONE;
1519 /* If we checked the outer header let the stack know */
1520 skb->csum_level = 1;
1524 static inline unsigned int ixgbe_rx_offset(struct ixgbe_ring *rx_ring)
1526 return ring_uses_build_skb(rx_ring) ? IXGBE_SKB_PAD : 0;
1529 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1530 struct ixgbe_rx_buffer *bi)
1532 struct page *page = bi->page;
1535 /* since we are recycling buffers we should seldom need to alloc */
1539 /* alloc new page for storage */
1540 page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1541 if (unlikely(!page)) {
1542 rx_ring->rx_stats.alloc_rx_page_failed++;
1546 /* map page for use */
1547 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1548 ixgbe_rx_pg_size(rx_ring),
1553 * if mapping failed free memory back to system since
1554 * there isn't much point in holding memory we can't use
1556 if (dma_mapping_error(rx_ring->dev, dma)) {
1557 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1559 rx_ring->rx_stats.alloc_rx_page_failed++;
1565 bi->page_offset = ixgbe_rx_offset(rx_ring);
1566 page_ref_add(page, USHRT_MAX - 1);
1567 bi->pagecnt_bias = USHRT_MAX;
1568 rx_ring->rx_stats.alloc_rx_page++;
1574 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1575 * @rx_ring: ring to place buffers on
1576 * @cleaned_count: number of buffers to replace
1578 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1580 union ixgbe_adv_rx_desc *rx_desc;
1581 struct ixgbe_rx_buffer *bi;
1582 u16 i = rx_ring->next_to_use;
1589 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1590 bi = &rx_ring->rx_buffer_info[i];
1591 i -= rx_ring->count;
1593 bufsz = ixgbe_rx_bufsz(rx_ring);
1596 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1599 /* sync the buffer for use by the device */
1600 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1601 bi->page_offset, bufsz,
1605 * Refresh the desc even if buffer_addrs didn't change
1606 * because each write-back erases this info.
1608 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1614 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1615 bi = rx_ring->rx_buffer_info;
1616 i -= rx_ring->count;
1619 /* clear the length for the next_to_use descriptor */
1620 rx_desc->wb.upper.length = 0;
1623 } while (cleaned_count);
1625 i += rx_ring->count;
1627 if (rx_ring->next_to_use != i) {
1628 rx_ring->next_to_use = i;
1630 /* update next to alloc since we have filled the ring */
1631 rx_ring->next_to_alloc = i;
1633 /* Force memory writes to complete before letting h/w
1634 * know there are new descriptors to fetch. (Only
1635 * applicable for weak-ordered memory model archs,
1639 writel(i, rx_ring->tail);
1643 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1644 struct sk_buff *skb)
1646 u16 hdr_len = skb_headlen(skb);
1648 /* set gso_size to avoid messing up TCP MSS */
1649 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1650 IXGBE_CB(skb)->append_cnt);
1651 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1654 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1655 struct sk_buff *skb)
1657 /* if append_cnt is 0 then frame is not RSC */
1658 if (!IXGBE_CB(skb)->append_cnt)
1661 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1662 rx_ring->rx_stats.rsc_flush++;
1664 ixgbe_set_rsc_gso_size(rx_ring, skb);
1666 /* gso_size is computed using append_cnt so always clear it last */
1667 IXGBE_CB(skb)->append_cnt = 0;
1671 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1672 * @rx_ring: rx descriptor ring packet is being transacted on
1673 * @rx_desc: pointer to the EOP Rx descriptor
1674 * @skb: pointer to current skb being populated
1676 * This function checks the ring, descriptor, and packet information in
1677 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1678 * other fields within the skb.
1680 void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1681 union ixgbe_adv_rx_desc *rx_desc,
1682 struct sk_buff *skb)
1684 struct net_device *dev = rx_ring->netdev;
1685 u32 flags = rx_ring->q_vector->adapter->flags;
1687 ixgbe_update_rsc_stats(rx_ring, skb);
1689 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1691 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1693 if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1694 ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1696 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1697 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1698 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1699 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1702 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
1703 ixgbe_ipsec_rx(rx_ring, rx_desc, skb);
1705 /* record Rx queue, or update MACVLAN statistics */
1706 if (netif_is_ixgbe(dev))
1707 skb_record_rx_queue(skb, rx_ring->queue_index);
1709 macvlan_count_rx(netdev_priv(dev), skb->len + ETH_HLEN, true,
1712 skb->protocol = eth_type_trans(skb, dev);
1715 void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1716 struct sk_buff *skb)
1718 napi_gro_receive(&q_vector->napi, skb);
1722 * ixgbe_is_non_eop - process handling of non-EOP buffers
1723 * @rx_ring: Rx ring being processed
1724 * @rx_desc: Rx descriptor for current buffer
1725 * @skb: Current socket buffer containing buffer in progress
1727 * This function updates next to clean. If the buffer is an EOP buffer
1728 * this function exits returning false, otherwise it will place the
1729 * sk_buff in the next buffer to be chained and return true indicating
1730 * that this is in fact a non-EOP buffer.
1732 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1733 union ixgbe_adv_rx_desc *rx_desc,
1734 struct sk_buff *skb)
1736 u32 ntc = rx_ring->next_to_clean + 1;
1738 /* fetch, update, and store next to clean */
1739 ntc = (ntc < rx_ring->count) ? ntc : 0;
1740 rx_ring->next_to_clean = ntc;
1742 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1744 /* update RSC append count if present */
1745 if (ring_is_rsc_enabled(rx_ring)) {
1746 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1747 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1749 if (unlikely(rsc_enabled)) {
1750 u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1752 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1753 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1755 /* update ntc based on RSC value */
1756 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1757 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1758 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1762 /* if we are the last buffer then there is nothing else to do */
1763 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1766 /* place skb in next buffer to be received */
1767 rx_ring->rx_buffer_info[ntc].skb = skb;
1768 rx_ring->rx_stats.non_eop_descs++;
1774 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1775 * @rx_ring: rx descriptor ring packet is being transacted on
1776 * @skb: pointer to current skb being adjusted
1778 * This function is an ixgbe specific version of __pskb_pull_tail. The
1779 * main difference between this version and the original function is that
1780 * this function can make several assumptions about the state of things
1781 * that allow for significant optimizations versus the standard function.
1782 * As a result we can do things like drop a frag and maintain an accurate
1783 * truesize for the skb.
1785 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1786 struct sk_buff *skb)
1788 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1790 unsigned int pull_len;
1793 * it is valid to use page_address instead of kmap since we are
1794 * working with pages allocated out of the lomem pool per
1795 * alloc_page(GFP_ATOMIC)
1797 va = skb_frag_address(frag);
1800 * we need the header to contain the greater of either ETH_HLEN or
1801 * 60 bytes if the skb->len is less than 60 for skb_pad.
1803 pull_len = eth_get_headlen(va, IXGBE_RX_HDR_SIZE);
1805 /* align pull length to size of long to optimize memcpy performance */
1806 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1808 /* update all of the pointers */
1809 skb_frag_size_sub(frag, pull_len);
1810 frag->page_offset += pull_len;
1811 skb->data_len -= pull_len;
1812 skb->tail += pull_len;
1816 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1817 * @rx_ring: rx descriptor ring packet is being transacted on
1818 * @skb: pointer to current skb being updated
1820 * This function provides a basic DMA sync up for the first fragment of an
1821 * skb. The reason for doing this is that the first fragment cannot be
1822 * unmapped until we have reached the end of packet descriptor for a buffer
1825 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1826 struct sk_buff *skb)
1828 /* if the page was released unmap it, else just sync our portion */
1829 if (unlikely(IXGBE_CB(skb)->page_released)) {
1830 dma_unmap_page_attrs(rx_ring->dev, IXGBE_CB(skb)->dma,
1831 ixgbe_rx_pg_size(rx_ring),
1834 } else if (ring_uses_build_skb(rx_ring)) {
1835 unsigned long offset = (unsigned long)(skb->data) & ~PAGE_MASK;
1837 dma_sync_single_range_for_cpu(rx_ring->dev,
1843 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1845 dma_sync_single_range_for_cpu(rx_ring->dev,
1848 skb_frag_size(frag),
1854 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1855 * @rx_ring: rx descriptor ring packet is being transacted on
1856 * @rx_desc: pointer to the EOP Rx descriptor
1857 * @skb: pointer to current skb being fixed
1859 * Check if the skb is valid in the XDP case it will be an error pointer.
1860 * Return true in this case to abort processing and advance to next
1863 * Check for corrupted packet headers caused by senders on the local L2
1864 * embedded NIC switch not setting up their Tx Descriptors right. These
1865 * should be very rare.
1867 * Also address the case where we are pulling data in on pages only
1868 * and as such no data is present in the skb header.
1870 * In addition if skb is not at least 60 bytes we need to pad it so that
1871 * it is large enough to qualify as a valid Ethernet frame.
1873 * Returns true if an error was encountered and skb was freed.
1875 bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1876 union ixgbe_adv_rx_desc *rx_desc,
1877 struct sk_buff *skb)
1879 struct net_device *netdev = rx_ring->netdev;
1881 /* XDP packets use error pointer so abort at this point */
1885 /* Verify netdev is present, and that packet does not have any
1886 * errors that would be unacceptable to the netdev.
1889 (unlikely(ixgbe_test_staterr(rx_desc,
1890 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1891 !(netdev->features & NETIF_F_RXALL)))) {
1892 dev_kfree_skb_any(skb);
1896 /* place header in linear portion of buffer */
1897 if (!skb_headlen(skb))
1898 ixgbe_pull_tail(rx_ring, skb);
1901 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1902 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1906 /* if eth_skb_pad returns an error the skb was freed */
1907 if (eth_skb_pad(skb))
1914 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1915 * @rx_ring: rx descriptor ring to store buffers on
1916 * @old_buff: donor buffer to have page reused
1918 * Synchronizes page for reuse by the adapter
1920 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1921 struct ixgbe_rx_buffer *old_buff)
1923 struct ixgbe_rx_buffer *new_buff;
1924 u16 nta = rx_ring->next_to_alloc;
1926 new_buff = &rx_ring->rx_buffer_info[nta];
1928 /* update, and store next to alloc */
1930 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1932 /* Transfer page from old buffer to new buffer.
1933 * Move each member individually to avoid possible store
1934 * forwarding stalls and unnecessary copy of skb.
1936 new_buff->dma = old_buff->dma;
1937 new_buff->page = old_buff->page;
1938 new_buff->page_offset = old_buff->page_offset;
1939 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
1942 static inline bool ixgbe_page_is_reserved(struct page *page)
1944 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1947 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer)
1949 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1950 struct page *page = rx_buffer->page;
1952 /* avoid re-using remote pages */
1953 if (unlikely(ixgbe_page_is_reserved(page)))
1956 #if (PAGE_SIZE < 8192)
1957 /* if we are only owner of page we can reuse it */
1958 if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
1961 /* The last offset is a bit aggressive in that we assume the
1962 * worst case of FCoE being enabled and using a 3K buffer.
1963 * However this should have minimal impact as the 1K extra is
1964 * still less than one buffer in size.
1966 #define IXGBE_LAST_OFFSET \
1967 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1968 if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1972 /* If we have drained the page fragment pool we need to update
1973 * the pagecnt_bias and page count so that we fully restock the
1974 * number of references the driver holds.
1976 if (unlikely(pagecnt_bias == 1)) {
1977 page_ref_add(page, USHRT_MAX - 1);
1978 rx_buffer->pagecnt_bias = USHRT_MAX;
1985 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1986 * @rx_ring: rx descriptor ring to transact packets on
1987 * @rx_buffer: buffer containing page to add
1988 * @skb: sk_buff to place the data into
1989 * @size: size of data in rx_buffer
1991 * This function will add the data contained in rx_buffer->page to the skb.
1992 * This is done either through a direct copy if the data in the buffer is
1993 * less than the skb header size, otherwise it will just attach the page as
1994 * a frag to the skb.
1996 * The function will then update the page offset if necessary and return
1997 * true if the buffer can be reused by the adapter.
1999 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
2000 struct ixgbe_rx_buffer *rx_buffer,
2001 struct sk_buff *skb,
2004 #if (PAGE_SIZE < 8192)
2005 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2007 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2008 SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2009 SKB_DATA_ALIGN(size);
2011 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2012 rx_buffer->page_offset, size, truesize);
2013 #if (PAGE_SIZE < 8192)
2014 rx_buffer->page_offset ^= truesize;
2016 rx_buffer->page_offset += truesize;
2020 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2021 union ixgbe_adv_rx_desc *rx_desc,
2022 struct sk_buff **skb,
2023 const unsigned int size)
2025 struct ixgbe_rx_buffer *rx_buffer;
2027 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2028 prefetchw(rx_buffer->page);
2029 *skb = rx_buffer->skb;
2031 /* Delay unmapping of the first packet. It carries the header
2032 * information, HW may still access the header after the writeback.
2033 * Only unmap it when EOP is reached
2035 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2040 ixgbe_dma_sync_frag(rx_ring, *skb);
2043 /* we are reusing so sync this buffer for CPU use */
2044 dma_sync_single_range_for_cpu(rx_ring->dev,
2046 rx_buffer->page_offset,
2050 rx_buffer->pagecnt_bias--;
2055 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2056 struct ixgbe_rx_buffer *rx_buffer,
2057 struct sk_buff *skb)
2059 if (ixgbe_can_reuse_rx_page(rx_buffer)) {
2060 /* hand second half of page back to the ring */
2061 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2063 if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2064 /* the page has been released from the ring */
2065 IXGBE_CB(skb)->page_released = true;
2067 /* we are not reusing the buffer so unmap it */
2068 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2069 ixgbe_rx_pg_size(rx_ring),
2073 __page_frag_cache_drain(rx_buffer->page,
2074 rx_buffer->pagecnt_bias);
2077 /* clear contents of rx_buffer */
2078 rx_buffer->page = NULL;
2079 rx_buffer->skb = NULL;
2082 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2083 struct ixgbe_rx_buffer *rx_buffer,
2084 struct xdp_buff *xdp,
2085 union ixgbe_adv_rx_desc *rx_desc)
2087 unsigned int size = xdp->data_end - xdp->data;
2088 #if (PAGE_SIZE < 8192)
2089 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2091 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2092 xdp->data_hard_start);
2094 struct sk_buff *skb;
2096 /* prefetch first cache line of first page */
2097 prefetch(xdp->data);
2098 #if L1_CACHE_BYTES < 128
2099 prefetch(xdp->data + L1_CACHE_BYTES);
2101 /* Note, we get here by enabling legacy-rx via:
2103 * ethtool --set-priv-flags <dev> legacy-rx on
2105 * In this mode, we currently get 0 extra XDP headroom as
2106 * opposed to having legacy-rx off, where we process XDP
2107 * packets going to stack via ixgbe_build_skb(). The latter
2108 * provides us currently with 192 bytes of headroom.
2110 * For ixgbe_construct_skb() mode it means that the
2111 * xdp->data_meta will always point to xdp->data, since
2112 * the helper cannot expand the head. Should this ever
2113 * change in future for legacy-rx mode on, then lets also
2114 * add xdp->data_meta handling here.
2117 /* allocate a skb to store the frags */
2118 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2122 if (size > IXGBE_RX_HDR_SIZE) {
2123 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2124 IXGBE_CB(skb)->dma = rx_buffer->dma;
2126 skb_add_rx_frag(skb, 0, rx_buffer->page,
2127 xdp->data - page_address(rx_buffer->page),
2129 #if (PAGE_SIZE < 8192)
2130 rx_buffer->page_offset ^= truesize;
2132 rx_buffer->page_offset += truesize;
2135 memcpy(__skb_put(skb, size),
2136 xdp->data, ALIGN(size, sizeof(long)));
2137 rx_buffer->pagecnt_bias++;
2143 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2144 struct ixgbe_rx_buffer *rx_buffer,
2145 struct xdp_buff *xdp,
2146 union ixgbe_adv_rx_desc *rx_desc)
2148 unsigned int metasize = xdp->data - xdp->data_meta;
2149 #if (PAGE_SIZE < 8192)
2150 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2152 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2153 SKB_DATA_ALIGN(xdp->data_end -
2154 xdp->data_hard_start);
2156 struct sk_buff *skb;
2158 /* Prefetch first cache line of first page. If xdp->data_meta
2159 * is unused, this points extactly as xdp->data, otherwise we
2160 * likely have a consumer accessing first few bytes of meta
2161 * data, and then actual data.
2163 prefetch(xdp->data_meta);
2164 #if L1_CACHE_BYTES < 128
2165 prefetch(xdp->data_meta + L1_CACHE_BYTES);
2168 /* build an skb to around the page buffer */
2169 skb = build_skb(xdp->data_hard_start, truesize);
2173 /* update pointers within the skb to store the data */
2174 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2175 __skb_put(skb, xdp->data_end - xdp->data);
2177 skb_metadata_set(skb, metasize);
2179 /* record DMA address if this is the start of a chain of buffers */
2180 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2181 IXGBE_CB(skb)->dma = rx_buffer->dma;
2183 /* update buffer offset */
2184 #if (PAGE_SIZE < 8192)
2185 rx_buffer->page_offset ^= truesize;
2187 rx_buffer->page_offset += truesize;
2193 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2194 struct ixgbe_ring *rx_ring,
2195 struct xdp_buff *xdp)
2197 int err, result = IXGBE_XDP_PASS;
2198 struct bpf_prog *xdp_prog;
2199 struct xdp_frame *xdpf;
2203 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2208 prefetchw(xdp->data_hard_start); /* xdp_frame write */
2210 act = bpf_prog_run_xdp(xdp_prog, xdp);
2215 xdpf = convert_to_xdp_frame(xdp);
2216 if (unlikely(!xdpf)) {
2217 result = IXGBE_XDP_CONSUMED;
2220 result = ixgbe_xmit_xdp_ring(adapter, xdpf);
2223 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2225 result = IXGBE_XDP_REDIR;
2227 result = IXGBE_XDP_CONSUMED;
2230 bpf_warn_invalid_xdp_action(act);
2233 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2234 /* fallthrough -- handle aborts by dropping packet */
2236 result = IXGBE_XDP_CONSUMED;
2241 return ERR_PTR(-result);
2244 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2245 struct ixgbe_rx_buffer *rx_buffer,
2248 #if (PAGE_SIZE < 8192)
2249 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2251 rx_buffer->page_offset ^= truesize;
2253 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2254 SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2255 SKB_DATA_ALIGN(size);
2257 rx_buffer->page_offset += truesize;
2262 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2263 * @q_vector: structure containing interrupt and ring information
2264 * @rx_ring: rx descriptor ring to transact packets on
2265 * @budget: Total limit on number of packets to process
2267 * This function provides a "bounce buffer" approach to Rx interrupt
2268 * processing. The advantage to this is that on systems that have
2269 * expensive overhead for IOMMU access this provides a means of avoiding
2270 * it by maintaining the mapping of the page to the syste.
2272 * Returns amount of work completed
2274 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2275 struct ixgbe_ring *rx_ring,
2278 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2279 struct ixgbe_adapter *adapter = q_vector->adapter;
2282 unsigned int mss = 0;
2283 #endif /* IXGBE_FCOE */
2284 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2285 unsigned int xdp_xmit = 0;
2286 struct xdp_buff xdp;
2288 xdp.rxq = &rx_ring->xdp_rxq;
2290 while (likely(total_rx_packets < budget)) {
2291 union ixgbe_adv_rx_desc *rx_desc;
2292 struct ixgbe_rx_buffer *rx_buffer;
2293 struct sk_buff *skb;
2296 /* return some buffers to hardware, one at a time is too slow */
2297 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2298 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2302 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2303 size = le16_to_cpu(rx_desc->wb.upper.length);
2307 /* This memory barrier is needed to keep us from reading
2308 * any other fields out of the rx_desc until we know the
2309 * descriptor has been written back
2313 rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size);
2315 /* retrieve a buffer from the ring */
2317 xdp.data = page_address(rx_buffer->page) +
2318 rx_buffer->page_offset;
2319 xdp.data_meta = xdp.data;
2320 xdp.data_hard_start = xdp.data -
2321 ixgbe_rx_offset(rx_ring);
2322 xdp.data_end = xdp.data + size;
2324 skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2328 unsigned int xdp_res = -PTR_ERR(skb);
2330 if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2331 xdp_xmit |= xdp_res;
2332 ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2334 rx_buffer->pagecnt_bias++;
2337 total_rx_bytes += size;
2339 ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2340 } else if (ring_uses_build_skb(rx_ring)) {
2341 skb = ixgbe_build_skb(rx_ring, rx_buffer,
2344 skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2348 /* exit if we failed to retrieve a buffer */
2350 rx_ring->rx_stats.alloc_rx_buff_failed++;
2351 rx_buffer->pagecnt_bias++;
2355 ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb);
2358 /* place incomplete frames back on ring for completion */
2359 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2362 /* verify the packet layout is correct */
2363 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2366 /* probably a little skewed due to removing CRC */
2367 total_rx_bytes += skb->len;
2369 /* populate checksum, timestamp, VLAN, and protocol */
2370 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2373 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2374 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2375 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2376 /* include DDPed FCoE data */
2377 if (ddp_bytes > 0) {
2379 mss = rx_ring->netdev->mtu -
2380 sizeof(struct fcoe_hdr) -
2381 sizeof(struct fc_frame_header) -
2382 sizeof(struct fcoe_crc_eof);
2386 total_rx_bytes += ddp_bytes;
2387 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2391 dev_kfree_skb_any(skb);
2396 #endif /* IXGBE_FCOE */
2397 ixgbe_rx_skb(q_vector, skb);
2399 /* update budget accounting */
2403 if (xdp_xmit & IXGBE_XDP_REDIR)
2406 if (xdp_xmit & IXGBE_XDP_TX) {
2407 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2409 /* Force memory writes to complete before letting h/w
2410 * know there are new descriptors to fetch.
2413 writel(ring->next_to_use, ring->tail);
2416 u64_stats_update_begin(&rx_ring->syncp);
2417 rx_ring->stats.packets += total_rx_packets;
2418 rx_ring->stats.bytes += total_rx_bytes;
2419 u64_stats_update_end(&rx_ring->syncp);
2420 q_vector->rx.total_packets += total_rx_packets;
2421 q_vector->rx.total_bytes += total_rx_bytes;
2423 return total_rx_packets;
2427 * ixgbe_configure_msix - Configure MSI-X hardware
2428 * @adapter: board private structure
2430 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2433 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2435 struct ixgbe_q_vector *q_vector;
2439 /* Populate MSIX to EITR Select */
2440 if (adapter->num_vfs > 32) {
2441 u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2442 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2446 * Populate the IVAR table and set the ITR values to the
2447 * corresponding register.
2449 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2450 struct ixgbe_ring *ring;
2451 q_vector = adapter->q_vector[v_idx];
2453 ixgbe_for_each_ring(ring, q_vector->rx)
2454 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2456 ixgbe_for_each_ring(ring, q_vector->tx)
2457 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2459 ixgbe_write_eitr(q_vector);
2462 switch (adapter->hw.mac.type) {
2463 case ixgbe_mac_82598EB:
2464 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2467 case ixgbe_mac_82599EB:
2468 case ixgbe_mac_X540:
2469 case ixgbe_mac_X550:
2470 case ixgbe_mac_X550EM_x:
2471 case ixgbe_mac_x550em_a:
2472 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2477 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2479 /* set up to autoclear timer, and the vectors */
2480 mask = IXGBE_EIMS_ENABLE_MASK;
2481 mask &= ~(IXGBE_EIMS_OTHER |
2482 IXGBE_EIMS_MAILBOX |
2485 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2489 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2490 * @q_vector: structure containing interrupt and ring information
2491 * @ring_container: structure containing ring performance data
2493 * Stores a new ITR value based on packets and byte
2494 * counts during the last interrupt. The advantage of per interrupt
2495 * computation is faster updates and more accurate ITR for the current
2496 * traffic pattern. Constants in this function were computed
2497 * based on theoretical maximum wire speed and thresholds were set based
2498 * on testing data as well as attempting to minimize response time
2499 * while increasing bulk throughput.
2501 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2502 struct ixgbe_ring_container *ring_container)
2504 unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2505 IXGBE_ITR_ADAPTIVE_LATENCY;
2506 unsigned int avg_wire_size, packets, bytes;
2507 unsigned long next_update = jiffies;
2509 /* If we don't have any rings just leave ourselves set for maximum
2510 * possible latency so we take ourselves out of the equation.
2512 if (!ring_container->ring)
2515 /* If we didn't update within up to 1 - 2 jiffies we can assume
2516 * that either packets are coming in so slow there hasn't been
2517 * any work, or that there is so much work that NAPI is dealing
2518 * with interrupt moderation and we don't need to do anything.
2520 if (time_after(next_update, ring_container->next_update))
2523 packets = ring_container->total_packets;
2525 /* We have no packets to actually measure against. This means
2526 * either one of the other queues on this vector is active or
2527 * we are a Tx queue doing TSO with too high of an interrupt rate.
2529 * When this occurs just tick up our delay by the minimum value
2530 * and hope that this extra delay will prevent us from being called
2531 * without any work on our queue.
2534 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2535 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2536 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2537 itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2541 bytes = ring_container->total_bytes;
2543 /* If packets are less than 4 or bytes are less than 9000 assume
2544 * insufficient data to use bulk rate limiting approach. We are
2545 * likely latency driven.
2547 if (packets < 4 && bytes < 9000) {
2548 itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2549 goto adjust_by_size;
2552 /* Between 4 and 48 we can assume that our current interrupt delay
2553 * is only slightly too low. As such we should increase it by a small
2557 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2558 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2559 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2563 /* Between 48 and 96 is our "goldilocks" zone where we are working
2564 * out "just right". Just report that our current ITR is good for us.
2567 itr = q_vector->itr >> 2;
2571 /* If packet count is 96 or greater we are likely looking at a slight
2572 * overrun of the delay we want. Try halving our delay to see if that
2573 * will cut the number of packets in half per interrupt.
2575 if (packets < 256) {
2576 itr = q_vector->itr >> 3;
2577 if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2578 itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2582 /* The paths below assume we are dealing with a bulk ITR since number
2583 * of packets is 256 or greater. We are just going to have to compute
2584 * a value and try to bring the count under control, though for smaller
2585 * packet sizes there isn't much we can do as NAPI polling will likely
2586 * be kicking in sooner rather than later.
2588 itr = IXGBE_ITR_ADAPTIVE_BULK;
2591 /* If packet counts are 256 or greater we can assume we have a gross
2592 * overestimation of what the rate should be. Instead of trying to fine
2593 * tune it just use the formula below to try and dial in an exact value
2594 * give the current packet size of the frame.
2596 avg_wire_size = bytes / packets;
2598 /* The following is a crude approximation of:
2599 * wmem_default / (size + overhead) = desired_pkts_per_int
2600 * rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2601 * (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2603 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2604 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2607 * (170 * (size + 24)) / (size + 640) = ITR
2609 * We first do some math on the packet size and then finally bitshift
2610 * by 8 after rounding up. We also have to account for PCIe link speed
2611 * difference as ITR scales based on this.
2613 if (avg_wire_size <= 60) {
2614 /* Start at 50k ints/sec */
2615 avg_wire_size = 5120;
2616 } else if (avg_wire_size <= 316) {
2617 /* 50K ints/sec to 16K ints/sec */
2618 avg_wire_size *= 40;
2619 avg_wire_size += 2720;
2620 } else if (avg_wire_size <= 1084) {
2621 /* 16K ints/sec to 9.2K ints/sec */
2622 avg_wire_size *= 15;
2623 avg_wire_size += 11452;
2624 } else if (avg_wire_size <= 1980) {
2625 /* 9.2K ints/sec to 8K ints/sec */
2627 avg_wire_size += 22420;
2629 /* plateau at a limit of 8K ints/sec */
2630 avg_wire_size = 32256;
2633 /* If we are in low latency mode half our delay which doubles the rate
2634 * to somewhere between 100K to 16K ints/sec
2636 if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2637 avg_wire_size >>= 1;
2639 /* Resultant value is 256 times larger than it needs to be. This
2640 * gives us room to adjust the value as needed to either increase
2641 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2643 * Use addition as we have already recorded the new latency flag
2644 * for the ITR value.
2646 switch (q_vector->adapter->link_speed) {
2647 case IXGBE_LINK_SPEED_10GB_FULL:
2648 case IXGBE_LINK_SPEED_100_FULL:
2650 itr += DIV_ROUND_UP(avg_wire_size,
2651 IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2652 IXGBE_ITR_ADAPTIVE_MIN_INC;
2654 case IXGBE_LINK_SPEED_2_5GB_FULL:
2655 case IXGBE_LINK_SPEED_1GB_FULL:
2656 case IXGBE_LINK_SPEED_10_FULL:
2657 itr += DIV_ROUND_UP(avg_wire_size,
2658 IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2659 IXGBE_ITR_ADAPTIVE_MIN_INC;
2664 /* write back value */
2665 ring_container->itr = itr;
2667 /* next update should occur within next jiffy */
2668 ring_container->next_update = next_update + 1;
2670 ring_container->total_bytes = 0;
2671 ring_container->total_packets = 0;
2675 * ixgbe_write_eitr - write EITR register in hardware specific way
2676 * @q_vector: structure containing interrupt and ring information
2678 * This function is made to be called by ethtool and by the driver
2679 * when it needs to update EITR registers at runtime. Hardware
2680 * specific quirks/differences are taken care of here.
2682 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2684 struct ixgbe_adapter *adapter = q_vector->adapter;
2685 struct ixgbe_hw *hw = &adapter->hw;
2686 int v_idx = q_vector->v_idx;
2687 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2689 switch (adapter->hw.mac.type) {
2690 case ixgbe_mac_82598EB:
2691 /* must write high and low 16 bits to reset counter */
2692 itr_reg |= (itr_reg << 16);
2694 case ixgbe_mac_82599EB:
2695 case ixgbe_mac_X540:
2696 case ixgbe_mac_X550:
2697 case ixgbe_mac_X550EM_x:
2698 case ixgbe_mac_x550em_a:
2700 * set the WDIS bit to not clear the timer bits and cause an
2701 * immediate assertion of the interrupt
2703 itr_reg |= IXGBE_EITR_CNT_WDIS;
2708 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2711 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2715 ixgbe_update_itr(q_vector, &q_vector->tx);
2716 ixgbe_update_itr(q_vector, &q_vector->rx);
2718 /* use the smallest value of new ITR delay calculations */
2719 new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2721 /* Clear latency flag if set, shift into correct position */
2722 new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2725 if (new_itr != q_vector->itr) {
2726 /* save the algorithm value here */
2727 q_vector->itr = new_itr;
2729 ixgbe_write_eitr(q_vector);
2734 * ixgbe_check_overtemp_subtask - check for over temperature
2735 * @adapter: pointer to adapter
2737 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2739 struct ixgbe_hw *hw = &adapter->hw;
2740 u32 eicr = adapter->interrupt_event;
2743 if (test_bit(__IXGBE_DOWN, &adapter->state))
2746 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2749 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2751 switch (hw->device_id) {
2752 case IXGBE_DEV_ID_82599_T3_LOM:
2754 * Since the warning interrupt is for both ports
2755 * we don't have to check if:
2756 * - This interrupt wasn't for our port.
2757 * - We may have missed the interrupt so always have to
2758 * check if we got a LSC
2760 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2761 !(eicr & IXGBE_EICR_LSC))
2764 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2766 bool link_up = false;
2768 hw->mac.ops.check_link(hw, &speed, &link_up, false);
2774 /* Check if this is not due to overtemp */
2775 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2779 case IXGBE_DEV_ID_X550EM_A_1G_T:
2780 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2781 rc = hw->phy.ops.check_overtemp(hw);
2782 if (rc != IXGBE_ERR_OVERTEMP)
2786 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2788 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2792 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2794 adapter->interrupt_event = 0;
2797 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2799 struct ixgbe_hw *hw = &adapter->hw;
2801 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2802 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2803 e_crit(probe, "Fan has stopped, replace the adapter\n");
2804 /* write to clear the interrupt */
2805 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2809 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2811 struct ixgbe_hw *hw = &adapter->hw;
2813 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2816 switch (adapter->hw.mac.type) {
2817 case ixgbe_mac_82599EB:
2819 * Need to check link state so complete overtemp check
2822 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2823 (eicr & IXGBE_EICR_LSC)) &&
2824 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2825 adapter->interrupt_event = eicr;
2826 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2827 ixgbe_service_event_schedule(adapter);
2831 case ixgbe_mac_x550em_a:
2832 if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2833 adapter->interrupt_event = eicr;
2834 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2835 ixgbe_service_event_schedule(adapter);
2836 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2837 IXGBE_EICR_GPI_SDP0_X550EM_a);
2838 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2839 IXGBE_EICR_GPI_SDP0_X550EM_a);
2842 case ixgbe_mac_X550:
2843 case ixgbe_mac_X540:
2844 if (!(eicr & IXGBE_EICR_TS))
2851 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2854 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2856 switch (hw->mac.type) {
2857 case ixgbe_mac_82598EB:
2858 if (hw->phy.type == ixgbe_phy_nl)
2861 case ixgbe_mac_82599EB:
2862 case ixgbe_mac_X550EM_x:
2863 case ixgbe_mac_x550em_a:
2864 switch (hw->mac.ops.get_media_type(hw)) {
2865 case ixgbe_media_type_fiber:
2866 case ixgbe_media_type_fiber_qsfp:
2876 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2878 struct ixgbe_hw *hw = &adapter->hw;
2879 u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2881 if (!ixgbe_is_sfp(hw))
2884 /* Later MAC's use different SDP */
2885 if (hw->mac.type >= ixgbe_mac_X540)
2886 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2888 if (eicr & eicr_mask) {
2889 /* Clear the interrupt */
2890 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2891 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2892 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2893 adapter->sfp_poll_time = 0;
2894 ixgbe_service_event_schedule(adapter);
2898 if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2899 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2900 /* Clear the interrupt */
2901 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2902 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2903 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2904 ixgbe_service_event_schedule(adapter);
2909 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2911 struct ixgbe_hw *hw = &adapter->hw;
2914 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2915 adapter->link_check_timeout = jiffies;
2916 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2917 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2918 IXGBE_WRITE_FLUSH(hw);
2919 ixgbe_service_event_schedule(adapter);
2923 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2927 struct ixgbe_hw *hw = &adapter->hw;
2929 switch (hw->mac.type) {
2930 case ixgbe_mac_82598EB:
2931 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2932 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2934 case ixgbe_mac_82599EB:
2935 case ixgbe_mac_X540:
2936 case ixgbe_mac_X550:
2937 case ixgbe_mac_X550EM_x:
2938 case ixgbe_mac_x550em_a:
2939 mask = (qmask & 0xFFFFFFFF);
2941 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2942 mask = (qmask >> 32);
2944 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2949 /* skip the flush */
2952 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2956 struct ixgbe_hw *hw = &adapter->hw;
2958 switch (hw->mac.type) {
2959 case ixgbe_mac_82598EB:
2960 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2961 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2963 case ixgbe_mac_82599EB:
2964 case ixgbe_mac_X540:
2965 case ixgbe_mac_X550:
2966 case ixgbe_mac_X550EM_x:
2967 case ixgbe_mac_x550em_a:
2968 mask = (qmask & 0xFFFFFFFF);
2970 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2971 mask = (qmask >> 32);
2973 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2978 /* skip the flush */
2982 * ixgbe_irq_enable - Enable default interrupt generation settings
2983 * @adapter: board private structure
2984 * @queues: enable irqs for queues
2985 * @flush: flush register write
2987 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2990 struct ixgbe_hw *hw = &adapter->hw;
2991 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2993 /* don't reenable LSC while waiting for link */
2994 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2995 mask &= ~IXGBE_EIMS_LSC;
2997 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2998 switch (adapter->hw.mac.type) {
2999 case ixgbe_mac_82599EB:
3000 mask |= IXGBE_EIMS_GPI_SDP0(hw);
3002 case ixgbe_mac_X540:
3003 case ixgbe_mac_X550:
3004 case ixgbe_mac_X550EM_x:
3005 case ixgbe_mac_x550em_a:
3006 mask |= IXGBE_EIMS_TS;
3011 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3012 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3013 switch (adapter->hw.mac.type) {
3014 case ixgbe_mac_82599EB:
3015 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3016 mask |= IXGBE_EIMS_GPI_SDP2(hw);
3018 case ixgbe_mac_X540:
3019 case ixgbe_mac_X550:
3020 case ixgbe_mac_X550EM_x:
3021 case ixgbe_mac_x550em_a:
3022 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3023 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3024 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3025 mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3026 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3027 mask |= IXGBE_EICR_GPI_SDP0_X540;
3028 mask |= IXGBE_EIMS_ECC;
3029 mask |= IXGBE_EIMS_MAILBOX;
3035 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3036 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3037 mask |= IXGBE_EIMS_FLOW_DIR;
3039 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3041 ixgbe_irq_enable_queues(adapter, ~0);
3043 IXGBE_WRITE_FLUSH(&adapter->hw);
3046 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3048 struct ixgbe_adapter *adapter = data;
3049 struct ixgbe_hw *hw = &adapter->hw;
3053 * Workaround for Silicon errata. Use clear-by-write instead
3054 * of clear-by-read. Reading with EICS will return the
3055 * interrupt causes without clearing, which later be done
3056 * with the write to EICR.
3058 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3060 /* The lower 16bits of the EICR register are for the queue interrupts
3061 * which should be masked here in order to not accidentally clear them if
3062 * the bits are high when ixgbe_msix_other is called. There is a race
3063 * condition otherwise which results in possible performance loss
3064 * especially if the ixgbe_msix_other interrupt is triggering
3065 * consistently (as it would when PPS is turned on for the X540 device)
3069 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3071 if (eicr & IXGBE_EICR_LSC)
3072 ixgbe_check_lsc(adapter);
3074 if (eicr & IXGBE_EICR_MAILBOX)
3075 ixgbe_msg_task(adapter);
3077 switch (hw->mac.type) {
3078 case ixgbe_mac_82599EB:
3079 case ixgbe_mac_X540:
3080 case ixgbe_mac_X550:
3081 case ixgbe_mac_X550EM_x:
3082 case ixgbe_mac_x550em_a:
3083 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3084 (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3085 adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3086 ixgbe_service_event_schedule(adapter);
3087 IXGBE_WRITE_REG(hw, IXGBE_EICR,
3088 IXGBE_EICR_GPI_SDP0_X540);
3090 if (eicr & IXGBE_EICR_ECC) {
3091 e_info(link, "Received ECC Err, initiating reset\n");
3092 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3093 ixgbe_service_event_schedule(adapter);
3094 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3096 /* Handle Flow Director Full threshold interrupt */
3097 if (eicr & IXGBE_EICR_FLOW_DIR) {
3098 int reinit_count = 0;
3100 for (i = 0; i < adapter->num_tx_queues; i++) {
3101 struct ixgbe_ring *ring = adapter->tx_ring[i];
3102 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3107 /* no more flow director interrupts until after init */
3108 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3109 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3110 ixgbe_service_event_schedule(adapter);
3113 ixgbe_check_sfp_event(adapter, eicr);
3114 ixgbe_check_overtemp_event(adapter, eicr);
3120 ixgbe_check_fan_failure(adapter, eicr);
3122 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3123 ixgbe_ptp_check_pps_event(adapter);
3125 /* re-enable the original interrupt state, no lsc, no queues */
3126 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3127 ixgbe_irq_enable(adapter, false, false);
3132 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3134 struct ixgbe_q_vector *q_vector = data;
3136 /* EIAM disabled interrupts (on this vector) for us */
3138 if (q_vector->rx.ring || q_vector->tx.ring)
3139 napi_schedule_irqoff(&q_vector->napi);
3145 * ixgbe_poll - NAPI Rx polling callback
3146 * @napi: structure for representing this polling device
3147 * @budget: how many packets driver is allowed to clean
3149 * This function is used for legacy and MSI, NAPI mode
3151 int ixgbe_poll(struct napi_struct *napi, int budget)
3153 struct ixgbe_q_vector *q_vector =
3154 container_of(napi, struct ixgbe_q_vector, napi);
3155 struct ixgbe_adapter *adapter = q_vector->adapter;
3156 struct ixgbe_ring *ring;
3157 int per_ring_budget, work_done = 0;
3158 bool clean_complete = true;
3160 #ifdef CONFIG_IXGBE_DCA
3161 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3162 ixgbe_update_dca(q_vector);
3165 ixgbe_for_each_ring(ring, q_vector->tx) {
3166 bool wd = ring->xsk_umem ?
3167 ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3168 ixgbe_clean_tx_irq(q_vector, ring, budget);
3171 clean_complete = false;
3174 /* Exit if we are called by netpoll */
3178 /* attempt to distribute budget to each queue fairly, but don't allow
3179 * the budget to go below 1 because we'll exit polling */
3180 if (q_vector->rx.count > 1)
3181 per_ring_budget = max(budget/q_vector->rx.count, 1);
3183 per_ring_budget = budget;
3185 ixgbe_for_each_ring(ring, q_vector->rx) {
3186 int cleaned = ring->xsk_umem ?
3187 ixgbe_clean_rx_irq_zc(q_vector, ring,
3189 ixgbe_clean_rx_irq(q_vector, ring,
3192 work_done += cleaned;
3193 if (cleaned >= per_ring_budget)
3194 clean_complete = false;
3197 /* If all work not completed, return budget and keep polling */
3198 if (!clean_complete)
3201 /* all work done, exit the polling mode */
3202 if (likely(napi_complete_done(napi, work_done))) {
3203 if (adapter->rx_itr_setting & 1)
3204 ixgbe_set_itr(q_vector);
3205 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3206 ixgbe_irq_enable_queues(adapter,
3207 BIT_ULL(q_vector->v_idx));
3210 return min(work_done, budget - 1);
3214 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3215 * @adapter: board private structure
3217 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3218 * interrupts from the kernel.
3220 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3222 struct net_device *netdev = adapter->netdev;
3223 unsigned int ri = 0, ti = 0;
3226 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3227 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3228 struct msix_entry *entry = &adapter->msix_entries[vector];
3230 if (q_vector->tx.ring && q_vector->rx.ring) {
3231 snprintf(q_vector->name, sizeof(q_vector->name),
3232 "%s-TxRx-%u", netdev->name, ri++);
3234 } else if (q_vector->rx.ring) {
3235 snprintf(q_vector->name, sizeof(q_vector->name),
3236 "%s-rx-%u", netdev->name, ri++);
3237 } else if (q_vector->tx.ring) {
3238 snprintf(q_vector->name, sizeof(q_vector->name),
3239 "%s-tx-%u", netdev->name, ti++);
3241 /* skip this unused q_vector */
3244 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3245 q_vector->name, q_vector);
3247 e_err(probe, "request_irq failed for MSIX interrupt "
3248 "Error: %d\n", err);
3249 goto free_queue_irqs;
3251 /* If Flow Director is enabled, set interrupt affinity */
3252 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3253 /* assign the mask for this irq */
3254 irq_set_affinity_hint(entry->vector,
3255 &q_vector->affinity_mask);
3259 err = request_irq(adapter->msix_entries[vector].vector,
3260 ixgbe_msix_other, 0, netdev->name, adapter);
3262 e_err(probe, "request_irq for msix_other failed: %d\n", err);
3263 goto free_queue_irqs;
3271 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3273 free_irq(adapter->msix_entries[vector].vector,
3274 adapter->q_vector[vector]);
3276 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3277 pci_disable_msix(adapter->pdev);
3278 kfree(adapter->msix_entries);
3279 adapter->msix_entries = NULL;
3284 * ixgbe_intr - legacy mode Interrupt Handler
3285 * @irq: interrupt number
3286 * @data: pointer to a network interface device structure
3288 static irqreturn_t ixgbe_intr(int irq, void *data)
3290 struct ixgbe_adapter *adapter = data;
3291 struct ixgbe_hw *hw = &adapter->hw;
3292 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3296 * Workaround for silicon errata #26 on 82598. Mask the interrupt
3297 * before the read of EICR.
3299 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3301 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3302 * therefore no explicit interrupt disable is necessary */
3303 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3306 * shared interrupt alert!
3307 * make sure interrupts are enabled because the read will
3308 * have disabled interrupts due to EIAM
3309 * finish the workaround of silicon errata on 82598. Unmask
3310 * the interrupt that we masked before the EICR read.
3312 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3313 ixgbe_irq_enable(adapter, true, true);
3314 return IRQ_NONE; /* Not our interrupt */
3317 if (eicr & IXGBE_EICR_LSC)
3318 ixgbe_check_lsc(adapter);
3320 switch (hw->mac.type) {
3321 case ixgbe_mac_82599EB:
3322 ixgbe_check_sfp_event(adapter, eicr);
3324 case ixgbe_mac_X540:
3325 case ixgbe_mac_X550:
3326 case ixgbe_mac_X550EM_x:
3327 case ixgbe_mac_x550em_a:
3328 if (eicr & IXGBE_EICR_ECC) {
3329 e_info(link, "Received ECC Err, initiating reset\n");
3330 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3331 ixgbe_service_event_schedule(adapter);
3332 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3334 ixgbe_check_overtemp_event(adapter, eicr);
3340 ixgbe_check_fan_failure(adapter, eicr);
3341 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3342 ixgbe_ptp_check_pps_event(adapter);
3344 /* would disable interrupts here but EIAM disabled it */
3345 napi_schedule_irqoff(&q_vector->napi);
3348 * re-enable link(maybe) and non-queue interrupts, no flush.
3349 * ixgbe_poll will re-enable the queue interrupts
3351 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3352 ixgbe_irq_enable(adapter, false, false);
3358 * ixgbe_request_irq - initialize interrupts
3359 * @adapter: board private structure
3361 * Attempts to configure interrupts using the best available
3362 * capabilities of the hardware and kernel.
3364 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3366 struct net_device *netdev = adapter->netdev;
3369 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3370 err = ixgbe_request_msix_irqs(adapter);
3371 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3372 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3373 netdev->name, adapter);
3375 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3376 netdev->name, adapter);
3379 e_err(probe, "request_irq failed, Error %d\n", err);
3384 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3388 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3389 free_irq(adapter->pdev->irq, adapter);
3393 if (!adapter->msix_entries)
3396 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3397 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3398 struct msix_entry *entry = &adapter->msix_entries[vector];
3400 /* free only the irqs that were actually requested */
3401 if (!q_vector->rx.ring && !q_vector->tx.ring)
3404 /* clear the affinity_mask in the IRQ descriptor */
3405 irq_set_affinity_hint(entry->vector, NULL);
3407 free_irq(entry->vector, q_vector);
3410 free_irq(adapter->msix_entries[vector].vector, adapter);
3414 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3415 * @adapter: board private structure
3417 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3419 switch (adapter->hw.mac.type) {
3420 case ixgbe_mac_82598EB:
3421 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3423 case ixgbe_mac_82599EB:
3424 case ixgbe_mac_X540:
3425 case ixgbe_mac_X550:
3426 case ixgbe_mac_X550EM_x:
3427 case ixgbe_mac_x550em_a:
3428 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3429 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3430 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3435 IXGBE_WRITE_FLUSH(&adapter->hw);
3436 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3439 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3440 synchronize_irq(adapter->msix_entries[vector].vector);
3442 synchronize_irq(adapter->msix_entries[vector++].vector);
3444 synchronize_irq(adapter->pdev->irq);
3449 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3450 * @adapter: board private structure
3453 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3455 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3457 ixgbe_write_eitr(q_vector);
3459 ixgbe_set_ivar(adapter, 0, 0, 0);
3460 ixgbe_set_ivar(adapter, 1, 0, 0);
3462 e_info(hw, "Legacy interrupt IVAR setup done\n");
3466 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3467 * @adapter: board private structure
3468 * @ring: structure containing ring specific data
3470 * Configure the Tx descriptor ring after a reset.
3472 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3473 struct ixgbe_ring *ring)
3475 struct ixgbe_hw *hw = &adapter->hw;
3476 u64 tdba = ring->dma;
3478 u32 txdctl = IXGBE_TXDCTL_ENABLE;
3479 u8 reg_idx = ring->reg_idx;
3481 ring->xsk_umem = NULL;
3482 if (ring_is_xdp(ring))
3483 ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
3485 /* disable queue to avoid issues while updating state */
3486 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3487 IXGBE_WRITE_FLUSH(hw);
3489 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3490 (tdba & DMA_BIT_MASK(32)));
3491 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3492 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3493 ring->count * sizeof(union ixgbe_adv_tx_desc));
3494 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3495 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3496 ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3499 * set WTHRESH to encourage burst writeback, it should not be set
3500 * higher than 1 when:
3501 * - ITR is 0 as it could cause false TX hangs
3502 * - ITR is set to > 100k int/sec and BQL is enabled
3504 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3505 * to or less than the number of on chip descriptors, which is
3508 if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3509 txdctl |= 1u << 16; /* WTHRESH = 1 */
3511 txdctl |= 8u << 16; /* WTHRESH = 8 */
3514 * Setting PTHRESH to 32 both improves performance
3515 * and avoids a TX hang with DFP enabled
3517 txdctl |= (1u << 8) | /* HTHRESH = 1 */
3518 32; /* PTHRESH = 32 */
3520 /* reinitialize flowdirector state */
3521 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3522 ring->atr_sample_rate = adapter->atr_sample_rate;
3523 ring->atr_count = 0;
3524 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3526 ring->atr_sample_rate = 0;
3529 /* initialize XPS */
3530 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3531 struct ixgbe_q_vector *q_vector = ring->q_vector;
3534 netif_set_xps_queue(ring->netdev,
3535 &q_vector->affinity_mask,
3539 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3541 /* reinitialize tx_buffer_info */
3542 memset(ring->tx_buffer_info, 0,
3543 sizeof(struct ixgbe_tx_buffer) * ring->count);
3546 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3548 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3549 if (hw->mac.type == ixgbe_mac_82598EB &&
3550 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3553 /* poll to verify queue is enabled */
3555 usleep_range(1000, 2000);
3556 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3557 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3559 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3562 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3564 struct ixgbe_hw *hw = &adapter->hw;
3566 u8 tcs = adapter->hw_tcs;
3568 if (hw->mac.type == ixgbe_mac_82598EB)
3571 /* disable the arbiter while setting MTQC */
3572 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3573 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3574 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3576 /* set transmit pool layout */
3577 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3578 mtqc = IXGBE_MTQC_VT_ENA;
3580 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3582 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3583 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3584 IXGBE_82599_VMDQ_4Q_MASK)
3585 mtqc |= IXGBE_MTQC_32VF;
3587 mtqc |= IXGBE_MTQC_64VF;
3590 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3591 } else if (tcs > 1) {
3592 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3594 u8 max_txq = adapter->num_tx_queues +
3595 adapter->num_xdp_queues;
3597 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3599 mtqc = IXGBE_MTQC_64Q_1PB;
3603 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3605 /* Enable Security TX Buffer IFG for multiple pb */
3607 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3608 sectx |= IXGBE_SECTX_DCB;
3609 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3612 /* re-enable the arbiter */
3613 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3614 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3618 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3619 * @adapter: board private structure
3621 * Configure the Tx unit of the MAC after a reset.
3623 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3625 struct ixgbe_hw *hw = &adapter->hw;
3629 ixgbe_setup_mtqc(adapter);
3631 if (hw->mac.type != ixgbe_mac_82598EB) {
3632 /* DMATXCTL.EN must be before Tx queues are enabled */
3633 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3634 dmatxctl |= IXGBE_DMATXCTL_TE;
3635 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3638 /* Setup the HW Tx Head and Tail descriptor pointers */
3639 for (i = 0; i < adapter->num_tx_queues; i++)
3640 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3641 for (i = 0; i < adapter->num_xdp_queues; i++)
3642 ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3645 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3646 struct ixgbe_ring *ring)
3648 struct ixgbe_hw *hw = &adapter->hw;
3649 u8 reg_idx = ring->reg_idx;
3650 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3652 srrctl |= IXGBE_SRRCTL_DROP_EN;
3654 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3657 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3658 struct ixgbe_ring *ring)
3660 struct ixgbe_hw *hw = &adapter->hw;
3661 u8 reg_idx = ring->reg_idx;
3662 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3664 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3666 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3669 #ifdef CONFIG_IXGBE_DCB
3670 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3672 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3676 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3678 if (adapter->ixgbe_ieee_pfc)
3679 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3682 * We should set the drop enable bit if:
3685 * Number of Rx queues > 1 and flow control is disabled
3687 * This allows us to avoid head of line blocking for security
3688 * and performance reasons.
3690 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3691 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3692 for (i = 0; i < adapter->num_rx_queues; i++)
3693 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3695 for (i = 0; i < adapter->num_rx_queues; i++)
3696 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3700 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3702 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3703 struct ixgbe_ring *rx_ring)
3705 struct ixgbe_hw *hw = &adapter->hw;
3707 u8 reg_idx = rx_ring->reg_idx;
3709 if (hw->mac.type == ixgbe_mac_82598EB) {
3710 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3713 * if VMDq is not active we must program one srrctl register
3714 * per RSS queue since we have enabled RDRXCTL.MVMEN
3719 /* configure header buffer length, needed for RSC */
3720 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3722 /* configure the packet buffer length */
3723 if (rx_ring->xsk_umem) {
3724 u32 xsk_buf_len = rx_ring->xsk_umem->chunk_size_nohr -
3725 XDP_PACKET_HEADROOM;
3727 /* If the MAC support setting RXDCTL.RLPML, the
3728 * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3729 * RXDCTL.RLPML is set to the actual UMEM buffer
3730 * size. If not, then we are stuck with a 1k buffer
3731 * size resolution. In this case frames larger than
3732 * the UMEM buffer size viewed in a 1k resolution will
3735 if (hw->mac.type != ixgbe_mac_82599EB)
3736 srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3738 srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3739 } else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3740 srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3742 srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3745 /* configure descriptor type */
3746 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3748 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3752 * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3753 * @adapter: device handle
3755 * - 82598/82599/X540: 128
3756 * - X550(non-SRIOV mode): 512
3757 * - X550(SRIOV mode): 64
3759 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3761 if (adapter->hw.mac.type < ixgbe_mac_X550)
3763 else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3770 * ixgbe_store_key - Write the RSS key to HW
3771 * @adapter: device handle
3773 * Write the RSS key stored in adapter.rss_key to HW.
3775 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3777 struct ixgbe_hw *hw = &adapter->hw;
3780 for (i = 0; i < 10; i++)
3781 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3785 * ixgbe_init_rss_key - Initialize adapter RSS key
3786 * @adapter: device handle
3788 * Allocates and initializes the RSS key if it is not allocated.
3790 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3794 if (!adapter->rss_key) {
3795 rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3796 if (unlikely(!rss_key))
3799 netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3800 adapter->rss_key = rss_key;
3807 * ixgbe_store_reta - Write the RETA table to HW
3808 * @adapter: device handle
3810 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3812 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3814 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3815 struct ixgbe_hw *hw = &adapter->hw;
3818 u8 *indir_tbl = adapter->rss_indir_tbl;
3820 /* Fill out the redirection table as follows:
3821 * - 82598: 8 bit wide entries containing pair of 4 bit RSS
3823 * - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3824 * - X550: 8 bit wide entries containing 6 bit RSS index
3826 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3827 indices_multi = 0x11;
3829 indices_multi = 0x1;
3831 /* Write redirection table to HW */
3832 for (i = 0; i < reta_entries; i++) {
3833 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3836 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3838 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3846 * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3847 * @adapter: device handle
3849 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3851 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3853 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3854 struct ixgbe_hw *hw = &adapter->hw;
3857 /* Write redirection table to HW */
3858 for (i = 0; i < reta_entries; i++) {
3859 u16 pool = adapter->num_rx_pools;
3861 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3867 IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3873 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3876 u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3877 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3879 /* Program table for at least 4 queues w/ SR-IOV so that VFs can
3880 * make full use of any rings they may have. We will use the
3881 * PSRTYPE register to control how many rings we use within the PF.
3883 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3886 /* Fill out hash function seeds */
3887 ixgbe_store_key(adapter);
3889 /* Fill out redirection table */
3890 memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3892 for (i = 0, j = 0; i < reta_entries; i++, j++) {
3896 adapter->rss_indir_tbl[i] = j;
3899 ixgbe_store_reta(adapter);
3902 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3904 struct ixgbe_hw *hw = &adapter->hw;
3905 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3908 /* Fill out hash function seeds */
3909 for (i = 0; i < 10; i++) {
3910 u16 pool = adapter->num_rx_pools;
3914 IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3915 *(adapter->rss_key + i));
3918 /* Fill out the redirection table */
3919 for (i = 0, j = 0; i < 64; i++, j++) {
3923 adapter->rss_indir_tbl[i] = j;
3926 ixgbe_store_vfreta(adapter);
3929 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3931 struct ixgbe_hw *hw = &adapter->hw;
3932 u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3935 /* Disable indicating checksum in descriptor, enables RSS hash */
3936 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3937 rxcsum |= IXGBE_RXCSUM_PCSD;
3938 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3940 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3941 if (adapter->ring_feature[RING_F_RSS].mask)
3942 mrqc = IXGBE_MRQC_RSSEN;
3944 u8 tcs = adapter->hw_tcs;
3946 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3948 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3950 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3951 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3952 IXGBE_82599_VMDQ_4Q_MASK)
3953 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3955 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3957 /* Enable L3/L4 for Tx Switched packets only for X550,
3958 * older devices do not support this feature
3960 if (hw->mac.type >= ixgbe_mac_X550)
3961 mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3964 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3966 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3968 mrqc = IXGBE_MRQC_RSSEN;
3972 /* Perform hash on these packet types */
3973 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3974 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3975 IXGBE_MRQC_RSS_FIELD_IPV6 |
3976 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3978 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3979 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3980 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3981 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3983 if ((hw->mac.type >= ixgbe_mac_X550) &&
3984 (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3985 u16 pool = adapter->num_rx_pools;
3987 /* Enable VF RSS mode */
3988 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3989 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3991 /* Setup RSS through the VF registers */
3992 ixgbe_setup_vfreta(adapter);
3993 vfmrqc = IXGBE_MRQC_RSSEN;
3994 vfmrqc |= rss_field;
3998 IXGBE_PFVFMRQC(VMDQ_P(pool)),
4001 ixgbe_setup_reta(adapter);
4003 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4008 * ixgbe_configure_rscctl - enable RSC for the indicated ring
4009 * @adapter: address of board private structure
4010 * @ring: structure containing ring specific data
4012 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4013 struct ixgbe_ring *ring)
4015 struct ixgbe_hw *hw = &adapter->hw;
4017 u8 reg_idx = ring->reg_idx;
4019 if (!ring_is_rsc_enabled(ring))
4022 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4023 rscctrl |= IXGBE_RSCCTL_RSCEN;
4025 * we must limit the number of descriptors so that the
4026 * total size of max desc * buf_len is not greater
4029 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4030 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4033 #define IXGBE_MAX_RX_DESC_POLL 10
4034 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4035 struct ixgbe_ring *ring)
4037 struct ixgbe_hw *hw = &adapter->hw;
4038 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4040 u8 reg_idx = ring->reg_idx;
4042 if (ixgbe_removed(hw->hw_addr))
4044 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4045 if (hw->mac.type == ixgbe_mac_82598EB &&
4046 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4050 usleep_range(1000, 2000);
4051 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4052 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4055 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4056 "the polling period\n", reg_idx);
4060 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4061 struct ixgbe_ring *ring)
4063 struct ixgbe_hw *hw = &adapter->hw;
4064 union ixgbe_adv_rx_desc *rx_desc;
4065 u64 rdba = ring->dma;
4067 u8 reg_idx = ring->reg_idx;
4069 xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4070 ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
4071 if (ring->xsk_umem) {
4072 ring->zca.free = ixgbe_zca_free;
4073 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4078 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4079 MEM_TYPE_PAGE_SHARED, NULL));
4082 /* disable queue to avoid use of these values while updating state */
4083 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4084 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4086 /* write value back with RXDCTL.ENABLE bit cleared */
4087 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4088 IXGBE_WRITE_FLUSH(hw);
4090 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4091 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4092 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4093 ring->count * sizeof(union ixgbe_adv_rx_desc));
4094 /* Force flushing of IXGBE_RDLEN to prevent MDD */
4095 IXGBE_WRITE_FLUSH(hw);
4097 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4098 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4099 ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4101 ixgbe_configure_srrctl(adapter, ring);
4102 ixgbe_configure_rscctl(adapter, ring);
4104 if (hw->mac.type == ixgbe_mac_82598EB) {
4106 * enable cache line friendly hardware writes:
4107 * PTHRESH=32 descriptors (half the internal cache),
4108 * this also removes ugly rx_no_buffer_count increment
4109 * HTHRESH=4 descriptors (to minimize latency on fetch)
4110 * WTHRESH=8 burst writeback up to two cache lines
4112 rxdctl &= ~0x3FFFFF;
4114 #if (PAGE_SIZE < 8192)
4115 /* RXDCTL.RLPML does not work on 82599 */
4116 } else if (hw->mac.type != ixgbe_mac_82599EB) {
4117 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4118 IXGBE_RXDCTL_RLPML_EN);
4120 /* Limit the maximum frame size so we don't overrun the skb.
4121 * This can happen in SRIOV mode when the MTU of the VF is
4122 * higher than the MTU of the PF.
4124 if (ring_uses_build_skb(ring) &&
4125 !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4126 rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4127 IXGBE_RXDCTL_RLPML_EN;
4131 if (ring->xsk_umem && hw->mac.type != ixgbe_mac_82599EB) {
4132 u32 xsk_buf_len = ring->xsk_umem->chunk_size_nohr -
4133 XDP_PACKET_HEADROOM;
4135 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4136 IXGBE_RXDCTL_RLPML_EN);
4137 rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4139 ring->rx_buf_len = xsk_buf_len;
4142 /* initialize rx_buffer_info */
4143 memset(ring->rx_buffer_info, 0,
4144 sizeof(struct ixgbe_rx_buffer) * ring->count);
4146 /* initialize Rx descriptor 0 */
4147 rx_desc = IXGBE_RX_DESC(ring, 0);
4148 rx_desc->wb.upper.length = 0;
4150 /* enable receive descriptor ring */
4151 rxdctl |= IXGBE_RXDCTL_ENABLE;
4152 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4154 ixgbe_rx_desc_queue_enable(adapter, ring);
4156 ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4158 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4161 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4163 struct ixgbe_hw *hw = &adapter->hw;
4164 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4165 u16 pool = adapter->num_rx_pools;
4167 /* PSRTYPE must be initialized in non 82598 adapters */
4168 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4169 IXGBE_PSRTYPE_UDPHDR |
4170 IXGBE_PSRTYPE_IPV4HDR |
4171 IXGBE_PSRTYPE_L2HDR |
4172 IXGBE_PSRTYPE_IPV6HDR;
4174 if (hw->mac.type == ixgbe_mac_82598EB)
4178 psrtype |= 2u << 29;
4180 psrtype |= 1u << 29;
4183 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4186 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4188 struct ixgbe_hw *hw = &adapter->hw;
4189 u16 pool = adapter->num_rx_pools;
4190 u32 reg_offset, vf_shift, vmolr;
4191 u32 gcr_ext, vmdctl;
4194 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4197 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4198 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4199 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4200 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4201 vmdctl |= IXGBE_VT_CTL_REPLEN;
4202 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4204 /* accept untagged packets until a vlan tag is
4205 * specifically set for the VMDQ queue/pool
4207 vmolr = IXGBE_VMOLR_AUPE;
4209 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4211 vf_shift = VMDQ_P(0) % 32;
4212 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4214 /* Enable only the PF's pool for Tx/Rx */
4215 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4216 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4217 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4218 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4219 if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4220 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4222 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4223 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4225 /* clear VLAN promisc flag so VFTA will be updated if necessary */
4226 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4229 * Set up VF register offsets for selected VT Mode,
4230 * i.e. 32 or 64 VFs for SR-IOV
4232 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4233 case IXGBE_82599_VMDQ_8Q_MASK:
4234 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4236 case IXGBE_82599_VMDQ_4Q_MASK:
4237 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4240 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4244 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4246 for (i = 0; i < adapter->num_vfs; i++) {
4247 /* configure spoof checking */
4248 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4249 adapter->vfinfo[i].spoofchk_enabled);
4251 /* Enable/Disable RSS query feature */
4252 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4253 adapter->vfinfo[i].rss_query_enabled);
4257 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4259 struct ixgbe_hw *hw = &adapter->hw;
4260 struct net_device *netdev = adapter->netdev;
4261 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4262 struct ixgbe_ring *rx_ring;
4267 /* adjust max frame to be able to do baby jumbo for FCoE */
4268 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4269 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4270 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4272 #endif /* IXGBE_FCOE */
4274 /* adjust max frame to be at least the size of a standard frame */
4275 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4276 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4278 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4279 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4280 mhadd &= ~IXGBE_MHADD_MFS_MASK;
4281 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4283 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4286 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4287 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4288 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4289 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4292 * Setup the HW Rx Head and Tail Descriptor Pointers and
4293 * the Base and Length of the Rx Descriptor Ring
4295 for (i = 0; i < adapter->num_rx_queues; i++) {
4296 rx_ring = adapter->rx_ring[i];
4298 clear_ring_rsc_enabled(rx_ring);
4299 clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4300 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4302 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4303 set_ring_rsc_enabled(rx_ring);
4305 if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4306 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4308 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4309 if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4312 set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4314 #if (PAGE_SIZE < 8192)
4315 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4316 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4318 if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4319 (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4320 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4325 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4327 struct ixgbe_hw *hw = &adapter->hw;
4328 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4330 switch (hw->mac.type) {
4331 case ixgbe_mac_82598EB:
4333 * For VMDq support of different descriptor types or
4334 * buffer sizes through the use of multiple SRRCTL
4335 * registers, RDRXCTL.MVMEN must be set to 1
4337 * also, the manual doesn't mention it clearly but DCA hints
4338 * will only use queue 0's tags unless this bit is set. Side
4339 * effects of setting this bit are only that SRRCTL must be
4340 * fully programmed [0..15]
4342 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4344 case ixgbe_mac_X550:
4345 case ixgbe_mac_X550EM_x:
4346 case ixgbe_mac_x550em_a:
4347 if (adapter->num_vfs)
4348 rdrxctl |= IXGBE_RDRXCTL_PSP;
4350 case ixgbe_mac_82599EB:
4351 case ixgbe_mac_X540:
4352 /* Disable RSC for ACK packets */
4353 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4354 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4355 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4356 /* hardware requires some bits to be set by default */
4357 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4358 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4361 /* We should do nothing since we don't know this hardware */
4365 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4369 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4370 * @adapter: board private structure
4372 * Configure the Rx unit of the MAC after a reset.
4374 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4376 struct ixgbe_hw *hw = &adapter->hw;
4380 /* disable receives while setting up the descriptors */
4381 hw->mac.ops.disable_rx(hw);
4383 ixgbe_setup_psrtype(adapter);
4384 ixgbe_setup_rdrxctl(adapter);
4387 rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4388 rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4389 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4390 rfctl |= IXGBE_RFCTL_RSC_DIS;
4392 /* disable NFS filtering */
4393 rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4394 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4396 /* Program registers for the distribution of queues */
4397 ixgbe_setup_mrqc(adapter);
4399 /* set_rx_buffer_len must be called before ring initialization */
4400 ixgbe_set_rx_buffer_len(adapter);
4403 * Setup the HW Rx Head and Tail Descriptor Pointers and
4404 * the Base and Length of the Rx Descriptor Ring
4406 for (i = 0; i < adapter->num_rx_queues; i++)
4407 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4409 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4410 /* disable drop enable for 82598 parts */
4411 if (hw->mac.type == ixgbe_mac_82598EB)
4412 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4414 /* enable all receives */
4415 rxctrl |= IXGBE_RXCTRL_RXEN;
4416 hw->mac.ops.enable_rx_dma(hw, rxctrl);
4419 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4420 __be16 proto, u16 vid)
4422 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4423 struct ixgbe_hw *hw = &adapter->hw;
4425 /* add VID to filter table */
4426 if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4427 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4429 set_bit(vid, adapter->active_vlans);
4434 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4439 /* short cut the special case */
4443 /* Search for the vlan id in the VLVF entries */
4444 for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4445 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4446 if ((vlvf & VLAN_VID_MASK) == vlan)
4453 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4455 struct ixgbe_hw *hw = &adapter->hw;
4459 idx = ixgbe_find_vlvf_entry(hw, vid);
4463 /* See if any other pools are set for this VLAN filter
4464 * entry other than the PF.
4466 word = idx * 2 + (VMDQ_P(0) / 32);
4467 bits = ~BIT(VMDQ_P(0) % 32);
4468 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4470 /* Disable the filter so this falls into the default pool. */
4471 if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4472 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4473 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4474 IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4478 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4479 __be16 proto, u16 vid)
4481 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4482 struct ixgbe_hw *hw = &adapter->hw;
4484 /* remove VID from filter table */
4485 if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4486 hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4488 clear_bit(vid, adapter->active_vlans);
4494 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4495 * @adapter: driver data
4497 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4499 struct ixgbe_hw *hw = &adapter->hw;
4503 switch (hw->mac.type) {
4504 case ixgbe_mac_82598EB:
4505 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4506 vlnctrl &= ~IXGBE_VLNCTRL_VME;
4507 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4509 case ixgbe_mac_82599EB:
4510 case ixgbe_mac_X540:
4511 case ixgbe_mac_X550:
4512 case ixgbe_mac_X550EM_x:
4513 case ixgbe_mac_x550em_a:
4514 for (i = 0; i < adapter->num_rx_queues; i++) {
4515 struct ixgbe_ring *ring = adapter->rx_ring[i];
4517 if (!netif_is_ixgbe(ring->netdev))
4521 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4522 vlnctrl &= ~IXGBE_RXDCTL_VME;
4523 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4532 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4533 * @adapter: driver data
4535 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4537 struct ixgbe_hw *hw = &adapter->hw;
4541 switch (hw->mac.type) {
4542 case ixgbe_mac_82598EB:
4543 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4544 vlnctrl |= IXGBE_VLNCTRL_VME;
4545 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4547 case ixgbe_mac_82599EB:
4548 case ixgbe_mac_X540:
4549 case ixgbe_mac_X550:
4550 case ixgbe_mac_X550EM_x:
4551 case ixgbe_mac_x550em_a:
4552 for (i = 0; i < adapter->num_rx_queues; i++) {
4553 struct ixgbe_ring *ring = adapter->rx_ring[i];
4555 if (!netif_is_ixgbe(ring->netdev))
4559 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4560 vlnctrl |= IXGBE_RXDCTL_VME;
4561 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4569 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4571 struct ixgbe_hw *hw = &adapter->hw;
4574 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4576 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4577 /* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4578 vlnctrl |= IXGBE_VLNCTRL_VFE;
4579 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4581 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4582 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4586 /* Nothing to do for 82598 */
4587 if (hw->mac.type == ixgbe_mac_82598EB)
4590 /* We are already in VLAN promisc, nothing to do */
4591 if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4594 /* Set flag so we don't redo unnecessary work */
4595 adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4597 /* Add PF to all active pools */
4598 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4599 u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4600 u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4602 vlvfb |= BIT(VMDQ_P(0) % 32);
4603 IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4606 /* Set all bits in the VLAN filter table array */
4607 for (i = hw->mac.vft_size; i--;)
4608 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4611 #define VFTA_BLOCK_SIZE 8
4612 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4614 struct ixgbe_hw *hw = &adapter->hw;
4615 u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4616 u32 vid_start = vfta_offset * 32;
4617 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4618 u32 i, vid, word, bits;
4620 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4621 u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4623 /* pull VLAN ID from VLVF */
4624 vid = vlvf & VLAN_VID_MASK;
4626 /* only concern outselves with a certain range */
4627 if (vid < vid_start || vid >= vid_end)
4631 /* record VLAN ID in VFTA */
4632 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4634 /* if PF is part of this then continue */
4635 if (test_bit(vid, adapter->active_vlans))
4639 /* remove PF from the pool */
4640 word = i * 2 + VMDQ_P(0) / 32;
4641 bits = ~BIT(VMDQ_P(0) % 32);
4642 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4643 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4646 /* extract values from active_vlans and write back to VFTA */
4647 for (i = VFTA_BLOCK_SIZE; i--;) {
4648 vid = (vfta_offset + i) * 32;
4649 word = vid / BITS_PER_LONG;
4650 bits = vid % BITS_PER_LONG;
4652 vfta[i] |= adapter->active_vlans[word] >> bits;
4654 IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4658 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4660 struct ixgbe_hw *hw = &adapter->hw;
4663 /* Set VLAN filtering to enabled */
4664 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4665 vlnctrl |= IXGBE_VLNCTRL_VFE;
4666 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4668 if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4669 hw->mac.type == ixgbe_mac_82598EB)
4672 /* We are not in VLAN promisc, nothing to do */
4673 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4676 /* Set flag so we don't redo unnecessary work */
4677 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4679 for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4680 ixgbe_scrub_vfta(adapter, i);
4683 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4687 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4689 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4690 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4694 * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4695 * @netdev: network interface device structure
4697 * Writes multicast address list to the MTA hash table.
4698 * Returns: -ENOMEM on failure
4699 * 0 on no addresses written
4700 * X on writing X addresses to MTA
4702 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4704 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4705 struct ixgbe_hw *hw = &adapter->hw;
4707 if (!netif_running(netdev))
4710 if (hw->mac.ops.update_mc_addr_list)
4711 hw->mac.ops.update_mc_addr_list(hw, netdev);
4715 #ifdef CONFIG_PCI_IOV
4716 ixgbe_restore_vf_multicasts(adapter);
4719 return netdev_mc_count(netdev);
4722 #ifdef CONFIG_PCI_IOV
4723 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4725 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4726 struct ixgbe_hw *hw = &adapter->hw;
4729 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4730 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4732 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4733 hw->mac.ops.set_rar(hw, i,
4738 hw->mac.ops.clear_rar(hw, i);
4743 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4745 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4746 struct ixgbe_hw *hw = &adapter->hw;
4749 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4750 if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4753 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4755 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4756 hw->mac.ops.set_rar(hw, i,
4761 hw->mac.ops.clear_rar(hw, i);
4765 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4767 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4768 struct ixgbe_hw *hw = &adapter->hw;
4771 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4772 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4773 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4776 ixgbe_sync_mac_table(adapter);
4779 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4781 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4782 struct ixgbe_hw *hw = &adapter->hw;
4785 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4786 /* do not count default RAR as available */
4787 if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4790 /* only count unused and addresses that belong to us */
4791 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4792 if (mac_table->pool != pool)
4802 /* this function destroys the first RAR entry */
4803 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4805 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4806 struct ixgbe_hw *hw = &adapter->hw;
4808 memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4809 mac_table->pool = VMDQ_P(0);
4811 mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4813 hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4817 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4818 const u8 *addr, u16 pool)
4820 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4821 struct ixgbe_hw *hw = &adapter->hw;
4824 if (is_zero_ether_addr(addr))
4827 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4828 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4831 ether_addr_copy(mac_table->addr, addr);
4832 mac_table->pool = pool;
4834 mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4835 IXGBE_MAC_STATE_IN_USE;
4837 ixgbe_sync_mac_table(adapter);
4845 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4846 const u8 *addr, u16 pool)
4848 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4849 struct ixgbe_hw *hw = &adapter->hw;
4852 if (is_zero_ether_addr(addr))
4855 /* search table for addr, if found clear IN_USE flag and sync */
4856 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4857 /* we can only delete an entry if it is in use */
4858 if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4860 /* we only care about entries that belong to the given pool */
4861 if (mac_table->pool != pool)
4863 /* we only care about a specific MAC address */
4864 if (!ether_addr_equal(addr, mac_table->addr))
4867 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4868 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4870 ixgbe_sync_mac_table(adapter);
4878 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4880 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4883 ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4885 return min_t(int, ret, 0);
4888 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4890 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4892 ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4898 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4899 * @netdev: network interface device structure
4901 * The set_rx_method entry point is called whenever the unicast/multicast
4902 * address list or the network interface flags are updated. This routine is
4903 * responsible for configuring the hardware for proper unicast, multicast and
4906 void ixgbe_set_rx_mode(struct net_device *netdev)
4908 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4909 struct ixgbe_hw *hw = &adapter->hw;
4910 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4911 netdev_features_t features = netdev->features;
4914 /* Check for Promiscuous and All Multicast modes */
4915 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4917 /* set all bits that we expect to always be set */
4918 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4919 fctrl |= IXGBE_FCTRL_BAM;
4920 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4921 fctrl |= IXGBE_FCTRL_PMCF;
4923 /* clear the bits we are changing the status of */
4924 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4925 if (netdev->flags & IFF_PROMISC) {
4926 hw->addr_ctrl.user_set_promisc = true;
4927 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4928 vmolr |= IXGBE_VMOLR_MPE;
4929 features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4931 if (netdev->flags & IFF_ALLMULTI) {
4932 fctrl |= IXGBE_FCTRL_MPE;
4933 vmolr |= IXGBE_VMOLR_MPE;
4935 hw->addr_ctrl.user_set_promisc = false;
4939 * Write addresses to available RAR registers, if there is not
4940 * sufficient space to store all the addresses then enable
4941 * unicast promiscuous mode
4943 if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4944 fctrl |= IXGBE_FCTRL_UPE;
4945 vmolr |= IXGBE_VMOLR_ROPE;
4948 /* Write addresses to the MTA, if the attempt fails
4949 * then we should just turn on promiscuous mode so
4950 * that we can at least receive multicast traffic
4952 count = ixgbe_write_mc_addr_list(netdev);
4954 fctrl |= IXGBE_FCTRL_MPE;
4955 vmolr |= IXGBE_VMOLR_MPE;
4957 vmolr |= IXGBE_VMOLR_ROMPE;
4960 if (hw->mac.type != ixgbe_mac_82598EB) {
4961 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4962 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4964 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4967 /* This is useful for sniffing bad packets. */
4968 if (features & NETIF_F_RXALL) {
4969 /* UPE and MPE will be handled by normal PROMISC logic
4970 * in e1000e_set_rx_mode */
4971 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4972 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4973 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4975 fctrl &= ~(IXGBE_FCTRL_DPF);
4976 /* NOTE: VLAN filtering is disabled by setting PROMISC */
4979 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4981 if (features & NETIF_F_HW_VLAN_CTAG_RX)
4982 ixgbe_vlan_strip_enable(adapter);
4984 ixgbe_vlan_strip_disable(adapter);
4986 if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4987 ixgbe_vlan_promisc_disable(adapter);
4989 ixgbe_vlan_promisc_enable(adapter);
4992 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4996 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4997 napi_enable(&adapter->q_vector[q_idx]->napi);
5000 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
5004 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
5005 napi_disable(&adapter->q_vector[q_idx]->napi);
5008 static void ixgbe_clear_udp_tunnel_port(struct ixgbe_adapter *adapter, u32 mask)
5010 struct ixgbe_hw *hw = &adapter->hw;
5013 if (!(adapter->flags & (IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE |
5014 IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)))
5017 vxlanctrl = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) & ~mask;
5018 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, vxlanctrl);
5020 if (mask & IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK)
5021 adapter->vxlan_port = 0;
5023 if (mask & IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK)
5024 adapter->geneve_port = 0;
5027 #ifdef CONFIG_IXGBE_DCB
5029 * ixgbe_configure_dcb - Configure DCB hardware
5030 * @adapter: ixgbe adapter struct
5032 * This is called by the driver on open to configure the DCB hardware.
5033 * This is also called by the gennetlink interface when reconfiguring
5036 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5038 struct ixgbe_hw *hw = &adapter->hw;
5039 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5041 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5042 if (hw->mac.type == ixgbe_mac_82598EB)
5043 netif_set_gso_max_size(adapter->netdev, 65536);
5047 if (hw->mac.type == ixgbe_mac_82598EB)
5048 netif_set_gso_max_size(adapter->netdev, 32768);
5051 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5052 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5055 /* reconfigure the hardware */
5056 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5057 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5059 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5061 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5062 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5063 ixgbe_dcb_hw_ets(&adapter->hw,
5064 adapter->ixgbe_ieee_ets,
5066 ixgbe_dcb_hw_pfc_config(&adapter->hw,
5067 adapter->ixgbe_ieee_pfc->pfc_en,
5068 adapter->ixgbe_ieee_ets->prio_tc);
5071 /* Enable RSS Hash per TC */
5072 if (hw->mac.type != ixgbe_mac_82598EB) {
5074 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5081 /* write msb to all 8 TCs in one write */
5082 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5087 /* Additional bittime to account for IXGBE framing */
5088 #define IXGBE_ETH_FRAMING 20
5091 * ixgbe_hpbthresh - calculate high water mark for flow control
5093 * @adapter: board private structure to calculate for
5094 * @pb: packet buffer to calculate
5096 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5098 struct ixgbe_hw *hw = &adapter->hw;
5099 struct net_device *dev = adapter->netdev;
5100 int link, tc, kb, marker;
5103 /* Calculate max LAN frame size */
5104 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5107 /* FCoE traffic class uses FCOE jumbo frames */
5108 if ((dev->features & NETIF_F_FCOE_MTU) &&
5109 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5110 (pb == ixgbe_fcoe_get_tc(adapter)))
5111 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5114 /* Calculate delay value for device */
5115 switch (hw->mac.type) {
5116 case ixgbe_mac_X540:
5117 case ixgbe_mac_X550:
5118 case ixgbe_mac_X550EM_x:
5119 case ixgbe_mac_x550em_a:
5120 dv_id = IXGBE_DV_X540(link, tc);
5123 dv_id = IXGBE_DV(link, tc);
5127 /* Loopback switch introduces additional latency */
5128 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5129 dv_id += IXGBE_B2BT(tc);
5131 /* Delay value is calculated in bit times convert to KB */
5132 kb = IXGBE_BT2KB(dv_id);
5133 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5135 marker = rx_pba - kb;
5137 /* It is possible that the packet buffer is not large enough
5138 * to provide required headroom. In this case throw an error
5139 * to user and a do the best we can.
5142 e_warn(drv, "Packet Buffer(%i) can not provide enough"
5143 "headroom to support flow control."
5144 "Decrease MTU or number of traffic classes\n", pb);
5152 * ixgbe_lpbthresh - calculate low water mark for for flow control
5154 * @adapter: board private structure to calculate for
5155 * @pb: packet buffer to calculate
5157 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5159 struct ixgbe_hw *hw = &adapter->hw;
5160 struct net_device *dev = adapter->netdev;
5164 /* Calculate max LAN frame size */
5165 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5168 /* FCoE traffic class uses FCOE jumbo frames */
5169 if ((dev->features & NETIF_F_FCOE_MTU) &&
5170 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5171 (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5172 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5175 /* Calculate delay value for device */
5176 switch (hw->mac.type) {
5177 case ixgbe_mac_X540:
5178 case ixgbe_mac_X550:
5179 case ixgbe_mac_X550EM_x:
5180 case ixgbe_mac_x550em_a:
5181 dv_id = IXGBE_LOW_DV_X540(tc);
5184 dv_id = IXGBE_LOW_DV(tc);
5188 /* Delay value is calculated in bit times convert to KB */
5189 return IXGBE_BT2KB(dv_id);
5193 * ixgbe_pbthresh_setup - calculate and setup high low water marks
5195 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5197 struct ixgbe_hw *hw = &adapter->hw;
5198 int num_tc = adapter->hw_tcs;
5204 for (i = 0; i < num_tc; i++) {
5205 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5206 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5208 /* Low water marks must not be larger than high water marks */
5209 if (hw->fc.low_water[i] > hw->fc.high_water[i])
5210 hw->fc.low_water[i] = 0;
5213 for (; i < MAX_TRAFFIC_CLASS; i++)
5214 hw->fc.high_water[i] = 0;
5217 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5219 struct ixgbe_hw *hw = &adapter->hw;
5221 u8 tc = adapter->hw_tcs;
5223 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5224 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5225 hdrm = 32 << adapter->fdir_pballoc;
5229 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5230 ixgbe_pbthresh_setup(adapter);
5233 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5235 struct ixgbe_hw *hw = &adapter->hw;
5236 struct hlist_node *node2;
5237 struct ixgbe_fdir_filter *filter;
5240 spin_lock(&adapter->fdir_perfect_lock);
5242 if (!hlist_empty(&adapter->fdir_filter_list))
5243 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5245 hlist_for_each_entry_safe(filter, node2,
5246 &adapter->fdir_filter_list, fdir_node) {
5247 action = filter->action;
5248 if (action != IXGBE_FDIR_DROP_QUEUE && action != 0)
5250 (action >> ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF) - 1;
5252 ixgbe_fdir_write_perfect_filter_82599(hw,
5255 (action == IXGBE_FDIR_DROP_QUEUE) ?
5256 IXGBE_FDIR_DROP_QUEUE :
5257 adapter->rx_ring[action]->reg_idx);
5260 spin_unlock(&adapter->fdir_perfect_lock);
5264 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5265 * @rx_ring: ring to free buffers from
5267 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5269 u16 i = rx_ring->next_to_clean;
5270 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5272 if (rx_ring->xsk_umem) {
5273 ixgbe_xsk_clean_rx_ring(rx_ring);
5277 /* Free all the Rx ring sk_buffs */
5278 while (i != rx_ring->next_to_alloc) {
5279 if (rx_buffer->skb) {
5280 struct sk_buff *skb = rx_buffer->skb;
5281 if (IXGBE_CB(skb)->page_released)
5282 dma_unmap_page_attrs(rx_ring->dev,
5284 ixgbe_rx_pg_size(rx_ring),
5290 /* Invalidate cache lines that may have been written to by
5291 * device so that we avoid corrupting memory.
5293 dma_sync_single_range_for_cpu(rx_ring->dev,
5295 rx_buffer->page_offset,
5296 ixgbe_rx_bufsz(rx_ring),
5299 /* free resources associated with mapping */
5300 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5301 ixgbe_rx_pg_size(rx_ring),
5304 __page_frag_cache_drain(rx_buffer->page,
5305 rx_buffer->pagecnt_bias);
5309 if (i == rx_ring->count) {
5311 rx_buffer = rx_ring->rx_buffer_info;
5316 rx_ring->next_to_alloc = 0;
5317 rx_ring->next_to_clean = 0;
5318 rx_ring->next_to_use = 0;
5321 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5322 struct ixgbe_fwd_adapter *accel)
5324 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5325 int num_tc = netdev_get_num_tc(adapter->netdev);
5326 struct net_device *vdev = accel->netdev;
5329 baseq = accel->pool * adapter->num_rx_queues_per_pool;
5330 netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5331 accel->pool, adapter->num_rx_pools,
5332 baseq, baseq + adapter->num_rx_queues_per_pool);
5334 accel->rx_base_queue = baseq;
5335 accel->tx_base_queue = baseq;
5337 /* record configuration for macvlan interface in vdev */
5338 for (i = 0; i < num_tc; i++)
5339 netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5340 i, rss_i, baseq + (rss_i * i));
5342 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5343 adapter->rx_ring[baseq + i]->netdev = vdev;
5345 /* Guarantee all rings are updated before we update the
5346 * MAC address filter.
5350 /* ixgbe_add_mac_filter will return an index if it succeeds, so we
5351 * need to only treat it as an error value if it is negative.
5353 err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5354 VMDQ_P(accel->pool));
5358 /* if we cannot add the MAC rule then disable the offload */
5359 macvlan_release_l2fw_offload(vdev);
5361 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5362 adapter->rx_ring[baseq + i]->netdev = NULL;
5364 netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5366 /* unbind the queues and drop the subordinate channel config */
5367 netdev_unbind_sb_channel(adapter->netdev, vdev);
5368 netdev_set_sb_channel(vdev, 0);
5370 clear_bit(accel->pool, adapter->fwd_bitmask);
5376 static int ixgbe_macvlan_up(struct net_device *vdev, void *data)
5378 struct ixgbe_adapter *adapter = data;
5379 struct ixgbe_fwd_adapter *accel;
5381 if (!netif_is_macvlan(vdev))
5384 accel = macvlan_accel_priv(vdev);
5388 ixgbe_fwd_ring_up(adapter, accel);
5393 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5395 netdev_walk_all_upper_dev_rcu(adapter->netdev,
5396 ixgbe_macvlan_up, adapter);
5399 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5401 struct ixgbe_hw *hw = &adapter->hw;
5403 ixgbe_configure_pb(adapter);
5404 #ifdef CONFIG_IXGBE_DCB
5405 ixgbe_configure_dcb(adapter);
5408 * We must restore virtualization before VLANs or else
5409 * the VLVF registers will not be populated
5411 ixgbe_configure_virtualization(adapter);
5413 ixgbe_set_rx_mode(adapter->netdev);
5414 ixgbe_restore_vlan(adapter);
5415 ixgbe_ipsec_restore(adapter);
5417 switch (hw->mac.type) {
5418 case ixgbe_mac_82599EB:
5419 case ixgbe_mac_X540:
5420 hw->mac.ops.disable_rx_buff(hw);
5426 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5427 ixgbe_init_fdir_signature_82599(&adapter->hw,
5428 adapter->fdir_pballoc);
5429 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5430 ixgbe_init_fdir_perfect_82599(&adapter->hw,
5431 adapter->fdir_pballoc);
5432 ixgbe_fdir_filter_restore(adapter);
5435 switch (hw->mac.type) {
5436 case ixgbe_mac_82599EB:
5437 case ixgbe_mac_X540:
5438 hw->mac.ops.enable_rx_buff(hw);
5444 #ifdef CONFIG_IXGBE_DCA
5446 if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5447 ixgbe_setup_dca(adapter);
5448 #endif /* CONFIG_IXGBE_DCA */
5451 /* configure FCoE L2 filters, redirection table, and Rx control */
5452 ixgbe_configure_fcoe(adapter);
5454 #endif /* IXGBE_FCOE */
5455 ixgbe_configure_tx(adapter);
5456 ixgbe_configure_rx(adapter);
5457 ixgbe_configure_dfwd(adapter);
5461 * ixgbe_sfp_link_config - set up SFP+ link
5462 * @adapter: pointer to private adapter struct
5464 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5467 * We are assuming the worst case scenario here, and that
5468 * is that an SFP was inserted/removed after the reset
5469 * but before SFP detection was enabled. As such the best
5470 * solution is to just start searching as soon as we start
5472 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5473 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5475 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5476 adapter->sfp_poll_time = 0;
5480 * ixgbe_non_sfp_link_config - set up non-SFP+ link
5481 * @hw: pointer to private hardware struct
5483 * Returns 0 on success, negative on failure
5485 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5488 bool autoneg, link_up = false;
5489 int ret = IXGBE_ERR_LINK_SETUP;
5491 if (hw->mac.ops.check_link)
5492 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5497 speed = hw->phy.autoneg_advertised;
5498 if ((!speed) && (hw->mac.ops.get_link_capabilities))
5499 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5504 if (hw->mac.ops.setup_link)
5505 ret = hw->mac.ops.setup_link(hw, speed, link_up);
5510 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5512 struct ixgbe_hw *hw = &adapter->hw;
5515 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5516 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5518 gpie |= IXGBE_GPIE_EIAME;
5520 * use EIAM to auto-mask when MSI-X interrupt is asserted
5521 * this saves a register write for every interrupt
5523 switch (hw->mac.type) {
5524 case ixgbe_mac_82598EB:
5525 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5527 case ixgbe_mac_82599EB:
5528 case ixgbe_mac_X540:
5529 case ixgbe_mac_X550:
5530 case ixgbe_mac_X550EM_x:
5531 case ixgbe_mac_x550em_a:
5533 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5534 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5538 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
5539 * specifically only auto mask tx and rx interrupts */
5540 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5543 /* XXX: to interrupt immediately for EICS writes, enable this */
5544 /* gpie |= IXGBE_GPIE_EIMEN; */
5546 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5547 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5549 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5550 case IXGBE_82599_VMDQ_8Q_MASK:
5551 gpie |= IXGBE_GPIE_VTMODE_16;
5553 case IXGBE_82599_VMDQ_4Q_MASK:
5554 gpie |= IXGBE_GPIE_VTMODE_32;
5557 gpie |= IXGBE_GPIE_VTMODE_64;
5562 /* Enable Thermal over heat sensor interrupt */
5563 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5564 switch (adapter->hw.mac.type) {
5565 case ixgbe_mac_82599EB:
5566 gpie |= IXGBE_SDP0_GPIEN_8259X;
5573 /* Enable fan failure interrupt */
5574 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5575 gpie |= IXGBE_SDP1_GPIEN(hw);
5577 switch (hw->mac.type) {
5578 case ixgbe_mac_82599EB:
5579 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5581 case ixgbe_mac_X550EM_x:
5582 case ixgbe_mac_x550em_a:
5583 gpie |= IXGBE_SDP0_GPIEN_X540;
5589 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5592 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5594 struct ixgbe_hw *hw = &adapter->hw;
5598 ixgbe_get_hw_control(adapter);
5599 ixgbe_setup_gpie(adapter);
5601 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5602 ixgbe_configure_msix(adapter);
5604 ixgbe_configure_msi_and_legacy(adapter);
5606 /* enable the optics for 82599 SFP+ fiber */
5607 if (hw->mac.ops.enable_tx_laser)
5608 hw->mac.ops.enable_tx_laser(hw);
5610 if (hw->phy.ops.set_phy_power)
5611 hw->phy.ops.set_phy_power(hw, true);
5613 smp_mb__before_atomic();
5614 clear_bit(__IXGBE_DOWN, &adapter->state);
5615 ixgbe_napi_enable_all(adapter);
5617 if (ixgbe_is_sfp(hw)) {
5618 ixgbe_sfp_link_config(adapter);
5620 err = ixgbe_non_sfp_link_config(hw);
5622 e_err(probe, "link_config FAILED %d\n", err);
5625 /* clear any pending interrupts, may auto mask */
5626 IXGBE_READ_REG(hw, IXGBE_EICR);
5627 ixgbe_irq_enable(adapter, true, true);
5630 * If this adapter has a fan, check to see if we had a failure
5631 * before we enabled the interrupt.
5633 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5634 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5635 if (esdp & IXGBE_ESDP_SDP1)
5636 e_crit(drv, "Fan has stopped, replace the adapter\n");
5639 /* bring the link up in the watchdog, this could race with our first
5640 * link up interrupt but shouldn't be a problem */
5641 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5642 adapter->link_check_timeout = jiffies;
5643 mod_timer(&adapter->service_timer, jiffies);
5645 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5646 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5647 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5648 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5651 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5653 WARN_ON(in_interrupt());
5654 /* put off any impending NetWatchDogTimeout */
5655 netif_trans_update(adapter->netdev);
5657 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5658 usleep_range(1000, 2000);
5659 if (adapter->hw.phy.type == ixgbe_phy_fw)
5660 ixgbe_watchdog_link_is_down(adapter);
5661 ixgbe_down(adapter);
5663 * If SR-IOV enabled then wait a bit before bringing the adapter
5664 * back up to give the VFs time to respond to the reset. The
5665 * two second wait is based upon the watchdog timer cycle in
5668 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5671 clear_bit(__IXGBE_RESETTING, &adapter->state);
5674 void ixgbe_up(struct ixgbe_adapter *adapter)
5676 /* hardware has been reset, we need to reload some things */
5677 ixgbe_configure(adapter);
5679 ixgbe_up_complete(adapter);
5682 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5686 pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5688 switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5689 case IXGBE_PCIDEVCTRL2_17_34s:
5690 case IXGBE_PCIDEVCTRL2_4_8s:
5691 /* For now we cap the upper limit on delay to 2 seconds
5692 * as we end up going up to 34 seconds of delay in worst
5693 * case timeout value.
5695 case IXGBE_PCIDEVCTRL2_1_2s:
5696 return 2000000ul; /* 2.0 s */
5697 case IXGBE_PCIDEVCTRL2_260_520ms:
5698 return 520000ul; /* 520 ms */
5699 case IXGBE_PCIDEVCTRL2_65_130ms:
5700 return 130000ul; /* 130 ms */
5701 case IXGBE_PCIDEVCTRL2_16_32ms:
5702 return 32000ul; /* 32 ms */
5703 case IXGBE_PCIDEVCTRL2_1_2ms:
5704 return 2000ul; /* 2 ms */
5705 case IXGBE_PCIDEVCTRL2_50_100us:
5706 return 100ul; /* 100 us */
5707 case IXGBE_PCIDEVCTRL2_16_32ms_def:
5708 return 32000ul; /* 32 ms */
5713 /* We shouldn't need to hit this path, but just in case default as
5714 * though completion timeout is not supported and support 32ms.
5719 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5721 unsigned long wait_delay, delay_interval;
5722 struct ixgbe_hw *hw = &adapter->hw;
5726 /* disable receives */
5727 hw->mac.ops.disable_rx(hw);
5729 if (ixgbe_removed(hw->hw_addr))
5732 /* disable all enabled Rx queues */
5733 for (i = 0; i < adapter->num_rx_queues; i++) {
5734 struct ixgbe_ring *ring = adapter->rx_ring[i];
5735 u8 reg_idx = ring->reg_idx;
5737 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5738 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5739 rxdctl |= IXGBE_RXDCTL_SWFLSH;
5741 /* write value back with RXDCTL.ENABLE bit cleared */
5742 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5745 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5746 if (hw->mac.type == ixgbe_mac_82598EB &&
5747 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5750 /* Determine our minimum delay interval. We will increase this value
5751 * with each subsequent test. This way if the device returns quickly
5752 * we should spend as little time as possible waiting, however as
5753 * the time increases we will wait for larger periods of time.
5755 * The trick here is that we increase the interval using the
5756 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5757 * of that wait is that it totals up to 100x whatever interval we
5758 * choose. Since our minimum wait is 100us we can just divide the
5759 * total timeout by 100 to get our minimum delay interval.
5761 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5763 wait_loop = IXGBE_MAX_RX_DESC_POLL;
5764 wait_delay = delay_interval;
5766 while (wait_loop--) {
5767 usleep_range(wait_delay, wait_delay + 10);
5768 wait_delay += delay_interval * 2;
5771 /* OR together the reading of all the active RXDCTL registers,
5772 * and then test the result. We need the disable to complete
5773 * before we start freeing the memory and invalidating the
5776 for (i = 0; i < adapter->num_rx_queues; i++) {
5777 struct ixgbe_ring *ring = adapter->rx_ring[i];
5778 u8 reg_idx = ring->reg_idx;
5780 rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5783 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5788 "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5791 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5793 unsigned long wait_delay, delay_interval;
5794 struct ixgbe_hw *hw = &adapter->hw;
5798 if (ixgbe_removed(hw->hw_addr))
5801 /* disable all enabled Tx queues */
5802 for (i = 0; i < adapter->num_tx_queues; i++) {
5803 struct ixgbe_ring *ring = adapter->tx_ring[i];
5804 u8 reg_idx = ring->reg_idx;
5806 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5809 /* disable all enabled XDP Tx queues */
5810 for (i = 0; i < adapter->num_xdp_queues; i++) {
5811 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5812 u8 reg_idx = ring->reg_idx;
5814 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5817 /* If the link is not up there shouldn't be much in the way of
5818 * pending transactions. Those that are left will be flushed out
5819 * when the reset logic goes through the flush sequence to clean out
5820 * the pending Tx transactions.
5822 if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5823 goto dma_engine_disable;
5825 /* Determine our minimum delay interval. We will increase this value
5826 * with each subsequent test. This way if the device returns quickly
5827 * we should spend as little time as possible waiting, however as
5828 * the time increases we will wait for larger periods of time.
5830 * The trick here is that we increase the interval using the
5831 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5832 * of that wait is that it totals up to 100x whatever interval we
5833 * choose. Since our minimum wait is 100us we can just divide the
5834 * total timeout by 100 to get our minimum delay interval.
5836 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5838 wait_loop = IXGBE_MAX_RX_DESC_POLL;
5839 wait_delay = delay_interval;
5841 while (wait_loop--) {
5842 usleep_range(wait_delay, wait_delay + 10);
5843 wait_delay += delay_interval * 2;
5846 /* OR together the reading of all the active TXDCTL registers,
5847 * and then test the result. We need the disable to complete
5848 * before we start freeing the memory and invalidating the
5851 for (i = 0; i < adapter->num_tx_queues; i++) {
5852 struct ixgbe_ring *ring = adapter->tx_ring[i];
5853 u8 reg_idx = ring->reg_idx;
5855 txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5857 for (i = 0; i < adapter->num_xdp_queues; i++) {
5858 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5859 u8 reg_idx = ring->reg_idx;
5861 txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5864 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5865 goto dma_engine_disable;
5869 "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5872 /* Disable the Tx DMA engine on 82599 and later MAC */
5873 switch (hw->mac.type) {
5874 case ixgbe_mac_82599EB:
5875 case ixgbe_mac_X540:
5876 case ixgbe_mac_X550:
5877 case ixgbe_mac_X550EM_x:
5878 case ixgbe_mac_x550em_a:
5879 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5880 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5881 ~IXGBE_DMATXCTL_TE));
5888 void ixgbe_reset(struct ixgbe_adapter *adapter)
5890 struct ixgbe_hw *hw = &adapter->hw;
5891 struct net_device *netdev = adapter->netdev;
5894 if (ixgbe_removed(hw->hw_addr))
5896 /* lock SFP init bit to prevent race conditions with the watchdog */
5897 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5898 usleep_range(1000, 2000);
5900 /* clear all SFP and link config related flags while holding SFP_INIT */
5901 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5902 IXGBE_FLAG2_SFP_NEEDS_RESET);
5903 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5905 err = hw->mac.ops.init_hw(hw);
5908 case IXGBE_ERR_SFP_NOT_PRESENT:
5909 case IXGBE_ERR_SFP_NOT_SUPPORTED:
5911 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5912 e_dev_err("master disable timed out\n");
5914 case IXGBE_ERR_EEPROM_VERSION:
5915 /* We are running on a pre-production device, log a warning */
5916 e_dev_warn("This device is a pre-production adapter/LOM. "
5917 "Please be aware there may be issues associated with "
5918 "your hardware. If you are experiencing problems "
5919 "please contact your Intel or hardware "
5920 "representative who provided you with this "
5924 e_dev_err("Hardware Error: %d\n", err);
5927 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5929 /* flush entries out of MAC table */
5930 ixgbe_flush_sw_mac_table(adapter);
5931 __dev_uc_unsync(netdev, NULL);
5933 /* do not flush user set addresses */
5934 ixgbe_mac_set_default_filter(adapter);
5936 /* update SAN MAC vmdq pool selection */
5937 if (hw->mac.san_mac_rar_index)
5938 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5940 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5941 ixgbe_ptp_reset(adapter);
5943 if (hw->phy.ops.set_phy_power) {
5944 if (!netif_running(adapter->netdev) && !adapter->wol)
5945 hw->phy.ops.set_phy_power(hw, false);
5947 hw->phy.ops.set_phy_power(hw, true);
5952 * ixgbe_clean_tx_ring - Free Tx Buffers
5953 * @tx_ring: ring to be cleaned
5955 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5957 u16 i = tx_ring->next_to_clean;
5958 struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
5960 if (tx_ring->xsk_umem) {
5961 ixgbe_xsk_clean_tx_ring(tx_ring);
5965 while (i != tx_ring->next_to_use) {
5966 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
5968 /* Free all the Tx ring sk_buffs */
5969 if (ring_is_xdp(tx_ring))
5970 xdp_return_frame(tx_buffer->xdpf);
5972 dev_kfree_skb_any(tx_buffer->skb);
5974 /* unmap skb header data */
5975 dma_unmap_single(tx_ring->dev,
5976 dma_unmap_addr(tx_buffer, dma),
5977 dma_unmap_len(tx_buffer, len),
5980 /* check for eop_desc to determine the end of the packet */
5981 eop_desc = tx_buffer->next_to_watch;
5982 tx_desc = IXGBE_TX_DESC(tx_ring, i);
5984 /* unmap remaining buffers */
5985 while (tx_desc != eop_desc) {
5989 if (unlikely(i == tx_ring->count)) {
5991 tx_buffer = tx_ring->tx_buffer_info;
5992 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5995 /* unmap any remaining paged data */
5996 if (dma_unmap_len(tx_buffer, len))
5997 dma_unmap_page(tx_ring->dev,
5998 dma_unmap_addr(tx_buffer, dma),
5999 dma_unmap_len(tx_buffer, len),
6003 /* move us one more past the eop_desc for start of next pkt */
6006 if (unlikely(i == tx_ring->count)) {
6008 tx_buffer = tx_ring->tx_buffer_info;
6012 /* reset BQL for queue */
6013 if (!ring_is_xdp(tx_ring))
6014 netdev_tx_reset_queue(txring_txq(tx_ring));
6017 /* reset next_to_use and next_to_clean */
6018 tx_ring->next_to_use = 0;
6019 tx_ring->next_to_clean = 0;
6023 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6024 * @adapter: board private structure
6026 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6030 for (i = 0; i < adapter->num_rx_queues; i++)
6031 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6035 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6036 * @adapter: board private structure
6038 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6042 for (i = 0; i < adapter->num_tx_queues; i++)
6043 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6044 for (i = 0; i < adapter->num_xdp_queues; i++)
6045 ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6048 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6050 struct hlist_node *node2;
6051 struct ixgbe_fdir_filter *filter;
6053 spin_lock(&adapter->fdir_perfect_lock);
6055 hlist_for_each_entry_safe(filter, node2,
6056 &adapter->fdir_filter_list, fdir_node) {
6057 hlist_del(&filter->fdir_node);
6060 adapter->fdir_filter_count = 0;
6062 spin_unlock(&adapter->fdir_perfect_lock);
6065 void ixgbe_down(struct ixgbe_adapter *adapter)
6067 struct net_device *netdev = adapter->netdev;
6068 struct ixgbe_hw *hw = &adapter->hw;
6071 /* signal that we are down to the interrupt handler */
6072 if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6073 return; /* do nothing if already down */
6075 /* Shut off incoming Tx traffic */
6076 netif_tx_stop_all_queues(netdev);
6078 /* call carrier off first to avoid false dev_watchdog timeouts */
6079 netif_carrier_off(netdev);
6080 netif_tx_disable(netdev);
6083 ixgbe_disable_rx(adapter);
6085 /* synchronize_rcu() needed for pending XDP buffers to drain */
6086 if (adapter->xdp_ring[0])
6089 ixgbe_irq_disable(adapter);
6091 ixgbe_napi_disable_all(adapter);
6093 clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6094 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6095 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6097 del_timer_sync(&adapter->service_timer);
6099 if (adapter->num_vfs) {
6100 /* Clear EITR Select mapping */
6101 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6103 /* Mark all the VFs as inactive */
6104 for (i = 0 ; i < adapter->num_vfs; i++)
6105 adapter->vfinfo[i].clear_to_send = false;
6107 /* ping all the active vfs to let them know we are going down */
6108 ixgbe_ping_all_vfs(adapter);
6110 /* Disable all VFTE/VFRE TX/RX */
6111 ixgbe_disable_tx_rx(adapter);
6114 /* disable transmits in the hardware now that interrupts are off */
6115 ixgbe_disable_tx(adapter);
6117 if (!pci_channel_offline(adapter->pdev))
6118 ixgbe_reset(adapter);
6120 /* power down the optics for 82599 SFP+ fiber */
6121 if (hw->mac.ops.disable_tx_laser)
6122 hw->mac.ops.disable_tx_laser(hw);
6124 ixgbe_clean_all_tx_rings(adapter);
6125 ixgbe_clean_all_rx_rings(adapter);
6129 * ixgbe_eee_capable - helper function to determine EEE support on X550
6130 * @adapter: board private structure
6132 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6134 struct ixgbe_hw *hw = &adapter->hw;
6136 switch (hw->device_id) {
6137 case IXGBE_DEV_ID_X550EM_A_1G_T:
6138 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6139 if (!hw->phy.eee_speeds_supported)
6141 adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6142 if (!hw->phy.eee_speeds_advertised)
6144 adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6147 adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6148 adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6154 * ixgbe_tx_timeout - Respond to a Tx Hang
6155 * @netdev: network interface device structure
6157 static void ixgbe_tx_timeout(struct net_device *netdev)
6159 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6161 /* Do the reset outside of interrupt context */
6162 ixgbe_tx_timeout_reset(adapter);
6165 #ifdef CONFIG_IXGBE_DCB
6166 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6168 struct ixgbe_hw *hw = &adapter->hw;
6169 struct tc_configuration *tc;
6172 switch (hw->mac.type) {
6173 case ixgbe_mac_82598EB:
6174 case ixgbe_mac_82599EB:
6175 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6176 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6178 case ixgbe_mac_X540:
6179 case ixgbe_mac_X550:
6180 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6181 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6183 case ixgbe_mac_X550EM_x:
6184 case ixgbe_mac_x550em_a:
6186 adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6187 adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6191 /* Configure DCB traffic classes */
6192 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6193 tc = &adapter->dcb_cfg.tc_config[j];
6194 tc->path[DCB_TX_CONFIG].bwg_id = 0;
6195 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6196 tc->path[DCB_RX_CONFIG].bwg_id = 0;
6197 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6198 tc->dcb_pfc = pfc_disabled;
6201 /* Initialize default user to priority mapping, UPx->TC0 */
6202 tc = &adapter->dcb_cfg.tc_config[0];
6203 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6204 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6206 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6207 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6208 adapter->dcb_cfg.pfc_mode_enable = false;
6209 adapter->dcb_set_bitmap = 0x00;
6210 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6211 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6212 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6213 sizeof(adapter->temp_dcb_cfg));
6218 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6219 * @adapter: board private structure to initialize
6220 * @ii: pointer to ixgbe_info for device
6222 * ixgbe_sw_init initializes the Adapter private data structure.
6223 * Fields are initialized based on PCI device information and
6224 * OS network device settings (MTU size).
6226 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6227 const struct ixgbe_info *ii)
6229 struct ixgbe_hw *hw = &adapter->hw;
6230 struct pci_dev *pdev = adapter->pdev;
6231 unsigned int rss, fdir;
6235 /* PCI config space info */
6237 hw->vendor_id = pdev->vendor;
6238 hw->device_id = pdev->device;
6239 hw->revision_id = pdev->revision;
6240 hw->subsystem_vendor_id = pdev->subsystem_vendor;
6241 hw->subsystem_device_id = pdev->subsystem_device;
6243 /* get_invariants needs the device IDs */
6244 ii->get_invariants(hw);
6246 /* Set common capability flags and settings */
6247 rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6248 adapter->ring_feature[RING_F_RSS].limit = rss;
6249 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6250 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6251 adapter->atr_sample_rate = 20;
6252 fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6253 adapter->ring_feature[RING_F_FDIR].limit = fdir;
6254 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6255 adapter->ring_feature[RING_F_VMDQ].limit = 1;
6256 #ifdef CONFIG_IXGBE_DCA
6257 adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6259 #ifdef CONFIG_IXGBE_DCB
6260 adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6261 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6264 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6265 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6266 #ifdef CONFIG_IXGBE_DCB
6267 /* Default traffic class to use for FCoE */
6268 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6269 #endif /* CONFIG_IXGBE_DCB */
6270 #endif /* IXGBE_FCOE */
6272 /* initialize static ixgbe jump table entries */
6273 adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6275 if (!adapter->jump_tables[0])
6277 adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6279 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6280 adapter->jump_tables[i] = NULL;
6282 adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6283 sizeof(struct ixgbe_mac_addr),
6285 if (!adapter->mac_table)
6288 if (ixgbe_init_rss_key(adapter))
6291 /* Set MAC specific capability flags and exceptions */
6292 switch (hw->mac.type) {
6293 case ixgbe_mac_82598EB:
6294 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6296 if (hw->device_id == IXGBE_DEV_ID_82598AT)
6297 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6299 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6300 adapter->ring_feature[RING_F_FDIR].limit = 0;
6301 adapter->atr_sample_rate = 0;
6302 adapter->fdir_pballoc = 0;
6304 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6305 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6306 #ifdef CONFIG_IXGBE_DCB
6307 adapter->fcoe.up = 0;
6308 #endif /* IXGBE_DCB */
6309 #endif /* IXGBE_FCOE */
6311 case ixgbe_mac_82599EB:
6312 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6313 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6315 case ixgbe_mac_X540:
6316 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6317 if (fwsm & IXGBE_FWSM_TS_ENABLED)
6318 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6320 case ixgbe_mac_x550em_a:
6321 adapter->flags |= IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE;
6322 switch (hw->device_id) {
6323 case IXGBE_DEV_ID_X550EM_A_1G_T:
6324 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6325 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6331 case ixgbe_mac_X550EM_x:
6332 #ifdef CONFIG_IXGBE_DCB
6333 adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6336 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6337 #ifdef CONFIG_IXGBE_DCB
6338 adapter->fcoe.up = 0;
6339 #endif /* IXGBE_DCB */
6340 #endif /* IXGBE_FCOE */
6342 case ixgbe_mac_X550:
6343 if (hw->mac.type == ixgbe_mac_X550)
6344 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6345 #ifdef CONFIG_IXGBE_DCA
6346 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6348 adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
6355 /* FCoE support exists, always init the FCoE lock */
6356 spin_lock_init(&adapter->fcoe.lock);
6359 /* n-tuple support exists, always init our spinlock */
6360 spin_lock_init(&adapter->fdir_perfect_lock);
6362 #ifdef CONFIG_IXGBE_DCB
6363 ixgbe_init_dcb(adapter);
6365 ixgbe_init_ipsec_offload(adapter);
6367 /* default flow control settings */
6368 hw->fc.requested_mode = ixgbe_fc_full;
6369 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
6370 ixgbe_pbthresh_setup(adapter);
6371 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6372 hw->fc.send_xon = true;
6373 hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6375 #ifdef CONFIG_PCI_IOV
6377 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6379 /* assign number of SR-IOV VFs */
6380 if (hw->mac.type != ixgbe_mac_82598EB) {
6381 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6383 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6386 #endif /* CONFIG_PCI_IOV */
6388 /* enable itr by default in dynamic mode */
6389 adapter->rx_itr_setting = 1;
6390 adapter->tx_itr_setting = 1;
6392 /* set default ring sizes */
6393 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6394 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6396 /* set default work limits */
6397 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6399 /* initialize eeprom parameters */
6400 if (ixgbe_init_eeprom_params_generic(hw)) {
6401 e_dev_err("EEPROM initialization failed\n");
6405 /* PF holds first pool slot */
6406 set_bit(0, adapter->fwd_bitmask);
6407 set_bit(__IXGBE_DOWN, &adapter->state);
6413 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6414 * @tx_ring: tx descriptor ring (for a specific queue) to setup
6416 * Return 0 on success, negative on failure
6418 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6420 struct device *dev = tx_ring->dev;
6421 int orig_node = dev_to_node(dev);
6422 int ring_node = NUMA_NO_NODE;
6425 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6427 if (tx_ring->q_vector)
6428 ring_node = tx_ring->q_vector->numa_node;
6430 tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6431 if (!tx_ring->tx_buffer_info)
6432 tx_ring->tx_buffer_info = vmalloc(size);
6433 if (!tx_ring->tx_buffer_info)
6436 /* round up to nearest 4K */
6437 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6438 tx_ring->size = ALIGN(tx_ring->size, 4096);
6440 set_dev_node(dev, ring_node);
6441 tx_ring->desc = dma_alloc_coherent(dev,
6445 set_dev_node(dev, orig_node);
6447 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6448 &tx_ring->dma, GFP_KERNEL);
6452 tx_ring->next_to_use = 0;
6453 tx_ring->next_to_clean = 0;
6457 vfree(tx_ring->tx_buffer_info);
6458 tx_ring->tx_buffer_info = NULL;
6459 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6464 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6465 * @adapter: board private structure
6467 * If this function returns with an error, then it's possible one or
6468 * more of the rings is populated (while the rest are not). It is the
6469 * callers duty to clean those orphaned rings.
6471 * Return 0 on success, negative on failure
6473 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6475 int i, j = 0, err = 0;
6477 for (i = 0; i < adapter->num_tx_queues; i++) {
6478 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6482 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6485 for (j = 0; j < adapter->num_xdp_queues; j++) {
6486 err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6490 e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6496 /* rewind the index freeing the rings as we go */
6498 ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6500 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6505 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6506 * @adapter: pointer to ixgbe_adapter
6507 * @rx_ring: rx descriptor ring (for a specific queue) to setup
6509 * Returns 0 on success, negative on failure
6511 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6512 struct ixgbe_ring *rx_ring)
6514 struct device *dev = rx_ring->dev;
6515 int orig_node = dev_to_node(dev);
6516 int ring_node = NUMA_NO_NODE;
6519 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6521 if (rx_ring->q_vector)
6522 ring_node = rx_ring->q_vector->numa_node;
6524 rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6525 if (!rx_ring->rx_buffer_info)
6526 rx_ring->rx_buffer_info = vmalloc(size);
6527 if (!rx_ring->rx_buffer_info)
6530 /* Round up to nearest 4K */
6531 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6532 rx_ring->size = ALIGN(rx_ring->size, 4096);
6534 set_dev_node(dev, ring_node);
6535 rx_ring->desc = dma_alloc_coherent(dev,
6539 set_dev_node(dev, orig_node);
6541 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6542 &rx_ring->dma, GFP_KERNEL);
6546 rx_ring->next_to_clean = 0;
6547 rx_ring->next_to_use = 0;
6549 /* XDP RX-queue info */
6550 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6551 rx_ring->queue_index) < 0)
6554 rx_ring->xdp_prog = adapter->xdp_prog;
6558 vfree(rx_ring->rx_buffer_info);
6559 rx_ring->rx_buffer_info = NULL;
6560 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6565 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6566 * @adapter: board private structure
6568 * If this function returns with an error, then it's possible one or
6569 * more of the rings is populated (while the rest are not). It is the
6570 * callers duty to clean those orphaned rings.
6572 * Return 0 on success, negative on failure
6574 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6578 for (i = 0; i < adapter->num_rx_queues; i++) {
6579 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6583 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6588 err = ixgbe_setup_fcoe_ddp_resources(adapter);
6593 /* rewind the index freeing the rings as we go */
6595 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6600 * ixgbe_free_tx_resources - Free Tx Resources per Queue
6601 * @tx_ring: Tx descriptor ring for a specific queue
6603 * Free all transmit software resources
6605 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6607 ixgbe_clean_tx_ring(tx_ring);
6609 vfree(tx_ring->tx_buffer_info);
6610 tx_ring->tx_buffer_info = NULL;
6612 /* if not set, then don't free */
6616 dma_free_coherent(tx_ring->dev, tx_ring->size,
6617 tx_ring->desc, tx_ring->dma);
6619 tx_ring->desc = NULL;
6623 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6624 * @adapter: board private structure
6626 * Free all transmit software resources
6628 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6632 for (i = 0; i < adapter->num_tx_queues; i++)
6633 if (adapter->tx_ring[i]->desc)
6634 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6635 for (i = 0; i < adapter->num_xdp_queues; i++)
6636 if (adapter->xdp_ring[i]->desc)
6637 ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6641 * ixgbe_free_rx_resources - Free Rx Resources
6642 * @rx_ring: ring to clean the resources from
6644 * Free all receive software resources
6646 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6648 ixgbe_clean_rx_ring(rx_ring);
6650 rx_ring->xdp_prog = NULL;
6651 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6652 vfree(rx_ring->rx_buffer_info);
6653 rx_ring->rx_buffer_info = NULL;
6655 /* if not set, then don't free */
6659 dma_free_coherent(rx_ring->dev, rx_ring->size,
6660 rx_ring->desc, rx_ring->dma);
6662 rx_ring->desc = NULL;
6666 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6667 * @adapter: board private structure
6669 * Free all receive software resources
6671 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6676 ixgbe_free_fcoe_ddp_resources(adapter);
6679 for (i = 0; i < adapter->num_rx_queues; i++)
6680 if (adapter->rx_ring[i]->desc)
6681 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6685 * ixgbe_change_mtu - Change the Maximum Transfer Unit
6686 * @netdev: network interface device structure
6687 * @new_mtu: new value for maximum frame size
6689 * Returns 0 on success, negative on failure
6691 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6693 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6695 if (adapter->xdp_prog) {
6696 int new_frame_size = new_mtu + ETH_HLEN + ETH_FCS_LEN +
6700 for (i = 0; i < adapter->num_rx_queues; i++) {
6701 struct ixgbe_ring *ring = adapter->rx_ring[i];
6703 if (new_frame_size > ixgbe_rx_bufsz(ring)) {
6704 e_warn(probe, "Requested MTU size is not supported with XDP\n");
6711 * For 82599EB we cannot allow legacy VFs to enable their receive
6712 * paths when MTU greater than 1500 is configured. So display a
6713 * warning that legacy VFs will be disabled.
6715 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6716 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6717 (new_mtu > ETH_DATA_LEN))
6718 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6720 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
6722 /* must set new MTU before calling down or up */
6723 netdev->mtu = new_mtu;
6725 if (netif_running(netdev))
6726 ixgbe_reinit_locked(adapter);
6732 * ixgbe_open - Called when a network interface is made active
6733 * @netdev: network interface device structure
6735 * Returns 0 on success, negative value on failure
6737 * The open entry point is called when a network interface is made
6738 * active by the system (IFF_UP). At this point all resources needed
6739 * for transmit and receive operations are allocated, the interrupt
6740 * handler is registered with the OS, the watchdog timer is started,
6741 * and the stack is notified that the interface is ready.
6743 int ixgbe_open(struct net_device *netdev)
6745 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6746 struct ixgbe_hw *hw = &adapter->hw;
6749 /* disallow open during test */
6750 if (test_bit(__IXGBE_TESTING, &adapter->state))
6753 netif_carrier_off(netdev);
6755 /* allocate transmit descriptors */
6756 err = ixgbe_setup_all_tx_resources(adapter);
6760 /* allocate receive descriptors */
6761 err = ixgbe_setup_all_rx_resources(adapter);
6765 ixgbe_configure(adapter);
6767 err = ixgbe_request_irq(adapter);
6771 /* Notify the stack of the actual queue counts. */
6772 queues = adapter->num_tx_queues;
6773 err = netif_set_real_num_tx_queues(netdev, queues);
6775 goto err_set_queues;
6777 queues = adapter->num_rx_queues;
6778 err = netif_set_real_num_rx_queues(netdev, queues);
6780 goto err_set_queues;
6782 ixgbe_ptp_init(adapter);
6784 ixgbe_up_complete(adapter);
6786 ixgbe_clear_udp_tunnel_port(adapter, IXGBE_VXLANCTRL_ALL_UDPPORT_MASK);
6787 udp_tunnel_get_rx_info(netdev);
6792 ixgbe_free_irq(adapter);
6794 ixgbe_free_all_rx_resources(adapter);
6795 if (hw->phy.ops.set_phy_power && !adapter->wol)
6796 hw->phy.ops.set_phy_power(&adapter->hw, false);
6798 ixgbe_free_all_tx_resources(adapter);
6800 ixgbe_reset(adapter);
6805 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6807 ixgbe_ptp_suspend(adapter);
6809 if (adapter->hw.phy.ops.enter_lplu) {
6810 adapter->hw.phy.reset_disable = true;
6811 ixgbe_down(adapter);
6812 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6813 adapter->hw.phy.reset_disable = false;
6815 ixgbe_down(adapter);
6818 ixgbe_free_irq(adapter);
6820 ixgbe_free_all_tx_resources(adapter);
6821 ixgbe_free_all_rx_resources(adapter);
6825 * ixgbe_close - Disables a network interface
6826 * @netdev: network interface device structure
6828 * Returns 0, this is not allowed to fail
6830 * The close entry point is called when an interface is de-activated
6831 * by the OS. The hardware is still under the drivers control, but
6832 * needs to be disabled. A global MAC reset is issued to stop the
6833 * hardware, and all transmit and receive resources are freed.
6835 int ixgbe_close(struct net_device *netdev)
6837 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6839 ixgbe_ptp_stop(adapter);
6841 if (netif_device_present(netdev))
6842 ixgbe_close_suspend(adapter);
6844 ixgbe_fdir_filter_exit(adapter);
6846 ixgbe_release_hw_control(adapter);
6852 static int ixgbe_resume(struct pci_dev *pdev)
6854 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6855 struct net_device *netdev = adapter->netdev;
6858 adapter->hw.hw_addr = adapter->io_addr;
6859 pci_set_power_state(pdev, PCI_D0);
6860 pci_restore_state(pdev);
6862 * pci_restore_state clears dev->state_saved so call
6863 * pci_save_state to restore it.
6865 pci_save_state(pdev);
6867 err = pci_enable_device_mem(pdev);
6869 e_dev_err("Cannot enable PCI device from suspend\n");
6872 smp_mb__before_atomic();
6873 clear_bit(__IXGBE_DISABLED, &adapter->state);
6874 pci_set_master(pdev);
6876 pci_wake_from_d3(pdev, false);
6878 ixgbe_reset(adapter);
6880 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6883 err = ixgbe_init_interrupt_scheme(adapter);
6884 if (!err && netif_running(netdev))
6885 err = ixgbe_open(netdev);
6889 netif_device_attach(netdev);
6894 #endif /* CONFIG_PM */
6896 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6898 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6899 struct net_device *netdev = adapter->netdev;
6900 struct ixgbe_hw *hw = &adapter->hw;
6902 u32 wufc = adapter->wol;
6908 netif_device_detach(netdev);
6910 if (netif_running(netdev))
6911 ixgbe_close_suspend(adapter);
6913 ixgbe_clear_interrupt_scheme(adapter);
6917 retval = pci_save_state(pdev);
6922 if (hw->mac.ops.stop_link_on_d3)
6923 hw->mac.ops.stop_link_on_d3(hw);
6928 ixgbe_set_rx_mode(netdev);
6930 /* enable the optics for 82599 SFP+ fiber as we can WoL */
6931 if (hw->mac.ops.enable_tx_laser)
6932 hw->mac.ops.enable_tx_laser(hw);
6934 /* enable the reception of multicast packets */
6935 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6936 fctrl |= IXGBE_FCTRL_MPE;
6937 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6939 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6940 ctrl |= IXGBE_CTRL_GIO_DIS;
6941 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6943 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6945 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6946 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6949 switch (hw->mac.type) {
6950 case ixgbe_mac_82598EB:
6951 pci_wake_from_d3(pdev, false);
6953 case ixgbe_mac_82599EB:
6954 case ixgbe_mac_X540:
6955 case ixgbe_mac_X550:
6956 case ixgbe_mac_X550EM_x:
6957 case ixgbe_mac_x550em_a:
6958 pci_wake_from_d3(pdev, !!wufc);
6964 *enable_wake = !!wufc;
6965 if (hw->phy.ops.set_phy_power && !*enable_wake)
6966 hw->phy.ops.set_phy_power(hw, false);
6968 ixgbe_release_hw_control(adapter);
6970 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
6971 pci_disable_device(pdev);
6977 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
6982 retval = __ixgbe_shutdown(pdev, &wake);
6987 pci_prepare_to_sleep(pdev);
6989 pci_wake_from_d3(pdev, false);
6990 pci_set_power_state(pdev, PCI_D3hot);
6995 #endif /* CONFIG_PM */
6997 static void ixgbe_shutdown(struct pci_dev *pdev)
7001 __ixgbe_shutdown(pdev, &wake);
7003 if (system_state == SYSTEM_POWER_OFF) {
7004 pci_wake_from_d3(pdev, wake);
7005 pci_set_power_state(pdev, PCI_D3hot);
7010 * ixgbe_update_stats - Update the board statistics counters.
7011 * @adapter: board private structure
7013 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7015 struct net_device *netdev = adapter->netdev;
7016 struct ixgbe_hw *hw = &adapter->hw;
7017 struct ixgbe_hw_stats *hwstats = &adapter->stats;
7019 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7020 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7021 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7022 u64 alloc_rx_page = 0;
7023 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7025 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7026 test_bit(__IXGBE_RESETTING, &adapter->state))
7029 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7032 for (i = 0; i < adapter->num_rx_queues; i++) {
7033 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7034 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7036 adapter->rsc_total_count = rsc_count;
7037 adapter->rsc_total_flush = rsc_flush;
7040 for (i = 0; i < adapter->num_rx_queues; i++) {
7041 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
7042 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7043 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7044 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7045 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7046 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7047 bytes += rx_ring->stats.bytes;
7048 packets += rx_ring->stats.packets;
7050 adapter->non_eop_descs = non_eop_descs;
7051 adapter->alloc_rx_page = alloc_rx_page;
7052 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7053 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7054 adapter->hw_csum_rx_error = hw_csum_rx_error;
7055 netdev->stats.rx_bytes = bytes;
7056 netdev->stats.rx_packets = packets;
7060 /* gather some stats to the adapter struct that are per queue */
7061 for (i = 0; i < adapter->num_tx_queues; i++) {
7062 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7063 restart_queue += tx_ring->tx_stats.restart_queue;
7064 tx_busy += tx_ring->tx_stats.tx_busy;
7065 bytes += tx_ring->stats.bytes;
7066 packets += tx_ring->stats.packets;
7068 for (i = 0; i < adapter->num_xdp_queues; i++) {
7069 struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
7071 restart_queue += xdp_ring->tx_stats.restart_queue;
7072 tx_busy += xdp_ring->tx_stats.tx_busy;
7073 bytes += xdp_ring->stats.bytes;
7074 packets += xdp_ring->stats.packets;
7076 adapter->restart_queue = restart_queue;
7077 adapter->tx_busy = tx_busy;
7078 netdev->stats.tx_bytes = bytes;
7079 netdev->stats.tx_packets = packets;
7081 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7083 /* 8 register reads */
7084 for (i = 0; i < 8; i++) {
7085 /* for packet buffers not used, the register should read 0 */
7086 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7088 hwstats->mpc[i] += mpc;
7089 total_mpc += hwstats->mpc[i];
7090 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7091 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7092 switch (hw->mac.type) {
7093 case ixgbe_mac_82598EB:
7094 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7095 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7096 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7097 hwstats->pxonrxc[i] +=
7098 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7100 case ixgbe_mac_82599EB:
7101 case ixgbe_mac_X540:
7102 case ixgbe_mac_X550:
7103 case ixgbe_mac_X550EM_x:
7104 case ixgbe_mac_x550em_a:
7105 hwstats->pxonrxc[i] +=
7106 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7113 /*16 register reads */
7114 for (i = 0; i < 16; i++) {
7115 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7116 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7117 if ((hw->mac.type == ixgbe_mac_82599EB) ||
7118 (hw->mac.type == ixgbe_mac_X540) ||
7119 (hw->mac.type == ixgbe_mac_X550) ||
7120 (hw->mac.type == ixgbe_mac_X550EM_x) ||
7121 (hw->mac.type == ixgbe_mac_x550em_a)) {
7122 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7123 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7124 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7125 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7129 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7130 /* work around hardware counting issue */
7131 hwstats->gprc -= missed_rx;
7133 ixgbe_update_xoff_received(adapter);
7135 /* 82598 hardware only has a 32 bit counter in the high register */
7136 switch (hw->mac.type) {
7137 case ixgbe_mac_82598EB:
7138 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7139 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7140 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7141 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7143 case ixgbe_mac_X540:
7144 case ixgbe_mac_X550:
7145 case ixgbe_mac_X550EM_x:
7146 case ixgbe_mac_x550em_a:
7147 /* OS2BMC stats are X540 and later */
7148 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7149 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7150 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7151 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7153 case ixgbe_mac_82599EB:
7154 for (i = 0; i < 16; i++)
7155 adapter->hw_rx_no_dma_resources +=
7156 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7157 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7158 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7159 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7160 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7161 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7162 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7163 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7164 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7165 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7167 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7168 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7169 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7170 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7171 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7172 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7173 /* Add up per cpu counters for total ddp aloc fail */
7174 if (adapter->fcoe.ddp_pool) {
7175 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7176 struct ixgbe_fcoe_ddp_pool *ddp_pool;
7178 u64 noddp = 0, noddp_ext_buff = 0;
7179 for_each_possible_cpu(cpu) {
7180 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7181 noddp += ddp_pool->noddp;
7182 noddp_ext_buff += ddp_pool->noddp_ext_buff;
7184 hwstats->fcoe_noddp = noddp;
7185 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7187 #endif /* IXGBE_FCOE */
7192 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7193 hwstats->bprc += bprc;
7194 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7195 if (hw->mac.type == ixgbe_mac_82598EB)
7196 hwstats->mprc -= bprc;
7197 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7198 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7199 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7200 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7201 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7202 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7203 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7204 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7205 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7206 hwstats->lxontxc += lxon;
7207 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7208 hwstats->lxofftxc += lxoff;
7209 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7210 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7212 * 82598 errata - tx of flow control packets is included in tx counters
7214 xon_off_tot = lxon + lxoff;
7215 hwstats->gptc -= xon_off_tot;
7216 hwstats->mptc -= xon_off_tot;
7217 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7218 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7219 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7220 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7221 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7222 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7223 hwstats->ptc64 -= xon_off_tot;
7224 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7225 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7226 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7227 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7228 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7229 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7231 /* Fill out the OS statistics structure */
7232 netdev->stats.multicast = hwstats->mprc;
7235 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7236 netdev->stats.rx_dropped = 0;
7237 netdev->stats.rx_length_errors = hwstats->rlec;
7238 netdev->stats.rx_crc_errors = hwstats->crcerrs;
7239 netdev->stats.rx_missed_errors = total_mpc;
7243 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7244 * @adapter: pointer to the device adapter structure
7246 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7248 struct ixgbe_hw *hw = &adapter->hw;
7251 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7254 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7256 /* if interface is down do nothing */
7257 if (test_bit(__IXGBE_DOWN, &adapter->state))
7260 /* do nothing if we are not using signature filters */
7261 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7264 adapter->fdir_overflow++;
7266 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7267 for (i = 0; i < adapter->num_tx_queues; i++)
7268 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7269 &(adapter->tx_ring[i]->state));
7270 for (i = 0; i < adapter->num_xdp_queues; i++)
7271 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7272 &adapter->xdp_ring[i]->state);
7273 /* re-enable flow director interrupts */
7274 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7276 e_err(probe, "failed to finish FDIR re-initialization, "
7277 "ignored adding FDIR ATR filters\n");
7282 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7283 * @adapter: pointer to the device adapter structure
7285 * This function serves two purposes. First it strobes the interrupt lines
7286 * in order to make certain interrupts are occurring. Secondly it sets the
7287 * bits needed to check for TX hangs. As a result we should immediately
7288 * determine if a hang has occurred.
7290 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7292 struct ixgbe_hw *hw = &adapter->hw;
7296 /* If we're down, removing or resetting, just bail */
7297 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7298 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7299 test_bit(__IXGBE_RESETTING, &adapter->state))
7302 /* Force detection of hung controller */
7303 if (netif_carrier_ok(adapter->netdev)) {
7304 for (i = 0; i < adapter->num_tx_queues; i++)
7305 set_check_for_tx_hang(adapter->tx_ring[i]);
7306 for (i = 0; i < adapter->num_xdp_queues; i++)
7307 set_check_for_tx_hang(adapter->xdp_ring[i]);
7310 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7312 * for legacy and MSI interrupts don't set any bits
7313 * that are enabled for EIAM, because this operation
7314 * would set *both* EIMS and EICS for any bit in EIAM
7316 IXGBE_WRITE_REG(hw, IXGBE_EICS,
7317 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7319 /* get one bit for every active tx/rx interrupt vector */
7320 for (i = 0; i < adapter->num_q_vectors; i++) {
7321 struct ixgbe_q_vector *qv = adapter->q_vector[i];
7322 if (qv->rx.ring || qv->tx.ring)
7327 /* Cause software interrupt to ensure rings are cleaned */
7328 ixgbe_irq_rearm_queues(adapter, eics);
7332 * ixgbe_watchdog_update_link - update the link status
7333 * @adapter: pointer to the device adapter structure
7335 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7337 struct ixgbe_hw *hw = &adapter->hw;
7338 u32 link_speed = adapter->link_speed;
7339 bool link_up = adapter->link_up;
7340 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7342 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7345 if (hw->mac.ops.check_link) {
7346 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7348 /* always assume link is up, if no check link function */
7349 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7353 if (adapter->ixgbe_ieee_pfc)
7354 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7356 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7357 hw->mac.ops.fc_enable(hw);
7358 ixgbe_set_rx_drop_en(adapter);
7362 time_after(jiffies, (adapter->link_check_timeout +
7363 IXGBE_TRY_LINK_TIMEOUT))) {
7364 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7365 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7366 IXGBE_WRITE_FLUSH(hw);
7369 adapter->link_up = link_up;
7370 adapter->link_speed = link_speed;
7373 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7375 #ifdef CONFIG_IXGBE_DCB
7376 struct net_device *netdev = adapter->netdev;
7377 struct dcb_app app = {
7378 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7383 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7384 up = dcb_ieee_getapp_mask(netdev, &app);
7386 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7391 * ixgbe_watchdog_link_is_up - update netif_carrier status and
7392 * print link up message
7393 * @adapter: pointer to the device adapter structure
7395 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7397 struct net_device *netdev = adapter->netdev;
7398 struct ixgbe_hw *hw = &adapter->hw;
7399 u32 link_speed = adapter->link_speed;
7400 const char *speed_str;
7401 bool flow_rx, flow_tx;
7403 /* only continue if link was previously down */
7404 if (netif_carrier_ok(netdev))
7407 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7409 switch (hw->mac.type) {
7410 case ixgbe_mac_82598EB: {
7411 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7412 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7413 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7414 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7417 case ixgbe_mac_X540:
7418 case ixgbe_mac_X550:
7419 case ixgbe_mac_X550EM_x:
7420 case ixgbe_mac_x550em_a:
7421 case ixgbe_mac_82599EB: {
7422 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7423 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7424 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7425 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7434 adapter->last_rx_ptp_check = jiffies;
7436 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7437 ixgbe_ptp_start_cyclecounter(adapter);
7439 switch (link_speed) {
7440 case IXGBE_LINK_SPEED_10GB_FULL:
7441 speed_str = "10 Gbps";
7443 case IXGBE_LINK_SPEED_5GB_FULL:
7444 speed_str = "5 Gbps";
7446 case IXGBE_LINK_SPEED_2_5GB_FULL:
7447 speed_str = "2.5 Gbps";
7449 case IXGBE_LINK_SPEED_1GB_FULL:
7450 speed_str = "1 Gbps";
7452 case IXGBE_LINK_SPEED_100_FULL:
7453 speed_str = "100 Mbps";
7455 case IXGBE_LINK_SPEED_10_FULL:
7456 speed_str = "10 Mbps";
7459 speed_str = "unknown speed";
7462 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7463 ((flow_rx && flow_tx) ? "RX/TX" :
7465 (flow_tx ? "TX" : "None"))));
7467 netif_carrier_on(netdev);
7468 ixgbe_check_vf_rate_limit(adapter);
7470 /* enable transmits */
7471 netif_tx_wake_all_queues(adapter->netdev);
7473 /* update the default user priority for VFs */
7474 ixgbe_update_default_up(adapter);
7476 /* ping all the active vfs to let them know link has changed */
7477 ixgbe_ping_all_vfs(adapter);
7481 * ixgbe_watchdog_link_is_down - update netif_carrier status and
7482 * print link down message
7483 * @adapter: pointer to the adapter structure
7485 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7487 struct net_device *netdev = adapter->netdev;
7488 struct ixgbe_hw *hw = &adapter->hw;
7490 adapter->link_up = false;
7491 adapter->link_speed = 0;
7493 /* only continue if link was up previously */
7494 if (!netif_carrier_ok(netdev))
7497 /* poll for SFP+ cable when link is down */
7498 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7499 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7501 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7502 ixgbe_ptp_start_cyclecounter(adapter);
7504 e_info(drv, "NIC Link is Down\n");
7505 netif_carrier_off(netdev);
7507 /* ping all the active vfs to let them know link has changed */
7508 ixgbe_ping_all_vfs(adapter);
7511 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7515 for (i = 0; i < adapter->num_tx_queues; i++) {
7516 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7518 if (tx_ring->next_to_use != tx_ring->next_to_clean)
7522 for (i = 0; i < adapter->num_xdp_queues; i++) {
7523 struct ixgbe_ring *ring = adapter->xdp_ring[i];
7525 if (ring->next_to_use != ring->next_to_clean)
7532 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7534 struct ixgbe_hw *hw = &adapter->hw;
7535 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7536 u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7540 if (!adapter->num_vfs)
7543 /* resetting the PF is only needed for MAC before X550 */
7544 if (hw->mac.type >= ixgbe_mac_X550)
7547 for (i = 0; i < adapter->num_vfs; i++) {
7548 for (j = 0; j < q_per_pool; j++) {
7551 h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7552 t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7563 * ixgbe_watchdog_flush_tx - flush queues on link down
7564 * @adapter: pointer to the device adapter structure
7566 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7568 if (!netif_carrier_ok(adapter->netdev)) {
7569 if (ixgbe_ring_tx_pending(adapter) ||
7570 ixgbe_vf_tx_pending(adapter)) {
7571 /* We've lost link, so the controller stops DMA,
7572 * but we've got queued Tx work that's never going
7573 * to get done, so reset controller to flush Tx.
7574 * (Do the reset outside of interrupt context).
7576 e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7577 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7582 #ifdef CONFIG_PCI_IOV
7583 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7585 struct ixgbe_hw *hw = &adapter->hw;
7586 struct pci_dev *pdev = adapter->pdev;
7590 if (!(netif_carrier_ok(adapter->netdev)))
7593 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7594 if (gpc) /* If incrementing then no need for the check below */
7596 /* Check to see if a bad DMA write target from an errant or
7597 * malicious VF has caused a PCIe error. If so then we can
7598 * issue a VFLR to the offending VF(s) and then resume without
7599 * requesting a full slot reset.
7605 /* check status reg for all VFs owned by this PF */
7606 for (vf = 0; vf < adapter->num_vfs; ++vf) {
7607 struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7612 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7613 if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7614 status_reg & PCI_STATUS_REC_MASTER_ABORT)
7619 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7623 /* Do not perform spoof check for 82598 or if not in IOV mode */
7624 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7625 adapter->num_vfs == 0)
7628 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7631 * ssvpc register is cleared on read, if zero then no
7632 * spoofed packets in the last interval.
7637 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7640 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7645 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7648 #endif /* CONFIG_PCI_IOV */
7652 * ixgbe_watchdog_subtask - check and bring link up
7653 * @adapter: pointer to the device adapter structure
7655 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7657 /* if interface is down, removing or resetting, do nothing */
7658 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7659 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7660 test_bit(__IXGBE_RESETTING, &adapter->state))
7663 ixgbe_watchdog_update_link(adapter);
7665 if (adapter->link_up)
7666 ixgbe_watchdog_link_is_up(adapter);
7668 ixgbe_watchdog_link_is_down(adapter);
7670 ixgbe_check_for_bad_vf(adapter);
7671 ixgbe_spoof_check(adapter);
7672 ixgbe_update_stats(adapter);
7674 ixgbe_watchdog_flush_tx(adapter);
7678 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7679 * @adapter: the ixgbe adapter structure
7681 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7683 struct ixgbe_hw *hw = &adapter->hw;
7686 /* not searching for SFP so there is nothing to do here */
7687 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7688 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7691 if (adapter->sfp_poll_time &&
7692 time_after(adapter->sfp_poll_time, jiffies))
7693 return; /* If not yet time to poll for SFP */
7695 /* someone else is in init, wait until next service event */
7696 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7699 adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7701 err = hw->phy.ops.identify_sfp(hw);
7702 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7705 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7706 /* If no cable is present, then we need to reset
7707 * the next time we find a good cable. */
7708 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7715 /* exit if reset not needed */
7716 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7719 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7722 * A module may be identified correctly, but the EEPROM may not have
7723 * support for that module. setup_sfp() will fail in that case, so
7724 * we should not allow that module to load.
7726 if (hw->mac.type == ixgbe_mac_82598EB)
7727 err = hw->phy.ops.reset(hw);
7729 err = hw->mac.ops.setup_sfp(hw);
7731 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7734 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7735 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7738 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7740 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7741 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7742 e_dev_err("failed to initialize because an unsupported "
7743 "SFP+ module type was detected.\n");
7744 e_dev_err("Reload the driver after installing a "
7745 "supported module.\n");
7746 unregister_netdev(adapter->netdev);
7751 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7752 * @adapter: the ixgbe adapter structure
7754 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7756 struct ixgbe_hw *hw = &adapter->hw;
7759 bool autoneg = false;
7761 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7764 /* someone else is in init, wait until next service event */
7765 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7768 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7770 hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7772 /* advertise highest capable link speed */
7773 if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7774 speed = IXGBE_LINK_SPEED_10GB_FULL;
7776 speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7777 IXGBE_LINK_SPEED_1GB_FULL);
7779 if (hw->mac.ops.setup_link)
7780 hw->mac.ops.setup_link(hw, speed, true);
7782 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7783 adapter->link_check_timeout = jiffies;
7784 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7788 * ixgbe_service_timer - Timer Call-back
7789 * @t: pointer to timer_list structure
7791 static void ixgbe_service_timer(struct timer_list *t)
7793 struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7794 unsigned long next_event_offset;
7796 /* poll faster when waiting for link */
7797 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7798 next_event_offset = HZ / 10;
7800 next_event_offset = HZ * 2;
7802 /* Reset the timer */
7803 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7805 ixgbe_service_event_schedule(adapter);
7808 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7810 struct ixgbe_hw *hw = &adapter->hw;
7813 if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7816 adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7818 if (!hw->phy.ops.handle_lasi)
7821 status = hw->phy.ops.handle_lasi(&adapter->hw);
7822 if (status != IXGBE_ERR_OVERTEMP)
7825 e_crit(drv, "%s\n", ixgbe_overheat_msg);
7828 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7830 if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7834 /* If we're already down, removing or resetting, just bail */
7835 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7836 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7837 test_bit(__IXGBE_RESETTING, &adapter->state)) {
7842 ixgbe_dump(adapter);
7843 netdev_err(adapter->netdev, "Reset adapter\n");
7844 adapter->tx_timeout_count++;
7846 ixgbe_reinit_locked(adapter);
7851 * ixgbe_check_fw_error - Check firmware for errors
7852 * @adapter: the adapter private structure
7854 * Check firmware errors in register FWSM
7856 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
7858 struct ixgbe_hw *hw = &adapter->hw;
7861 /* read fwsm.ext_err_ind register and log errors */
7862 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
7864 if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
7865 !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
7866 e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7869 if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
7870 e_dev_err("Firmware recovery mode detected. Limiting functionality. Refer to the Intel(R) Ethernet Adapters and Devices User Guide for details on firmware recovery mode.\n");
7878 * ixgbe_service_task - manages and runs subtasks
7879 * @work: pointer to work_struct containing our data
7881 static void ixgbe_service_task(struct work_struct *work)
7883 struct ixgbe_adapter *adapter = container_of(work,
7884 struct ixgbe_adapter,
7886 if (ixgbe_removed(adapter->hw.hw_addr)) {
7887 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7889 ixgbe_down(adapter);
7892 ixgbe_service_event_complete(adapter);
7895 if (ixgbe_check_fw_error(adapter)) {
7896 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7898 unregister_netdev(adapter->netdev);
7901 ixgbe_service_event_complete(adapter);
7904 if (adapter->flags2 & IXGBE_FLAG2_UDP_TUN_REREG_NEEDED) {
7906 adapter->flags2 &= ~IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
7907 udp_tunnel_get_rx_info(adapter->netdev);
7910 ixgbe_reset_subtask(adapter);
7911 ixgbe_phy_interrupt_subtask(adapter);
7912 ixgbe_sfp_detection_subtask(adapter);
7913 ixgbe_sfp_link_config_subtask(adapter);
7914 ixgbe_check_overtemp_subtask(adapter);
7915 ixgbe_watchdog_subtask(adapter);
7916 ixgbe_fdir_reinit_subtask(adapter);
7917 ixgbe_check_hang_subtask(adapter);
7919 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7920 ixgbe_ptp_overflow_check(adapter);
7921 if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
7922 ixgbe_ptp_rx_hang(adapter);
7923 ixgbe_ptp_tx_hang(adapter);
7926 ixgbe_service_event_complete(adapter);
7929 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7930 struct ixgbe_tx_buffer *first,
7932 struct ixgbe_ipsec_tx_data *itd)
7934 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7935 struct sk_buff *skb = first->skb;
7945 u32 paylen, l4_offset;
7946 u32 fceof_saidx = 0;
7949 if (skb->ip_summed != CHECKSUM_PARTIAL)
7952 if (!skb_is_gso(skb))
7955 err = skb_cow_head(skb, 0);
7959 if (eth_p_mpls(first->protocol))
7960 ip.hdr = skb_inner_network_header(skb);
7962 ip.hdr = skb_network_header(skb);
7963 l4.hdr = skb_checksum_start(skb);
7965 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7966 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
7968 /* initialize outer IP header fields */
7969 if (ip.v4->version == 4) {
7970 unsigned char *csum_start = skb_checksum_start(skb);
7971 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
7972 int len = csum_start - trans_start;
7974 /* IP header will have to cancel out any data that
7975 * is not a part of the outer IP header, so set to
7976 * a reverse csum if needed, else init check to 0.
7978 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
7979 csum_fold(csum_partial(trans_start,
7981 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7984 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7985 IXGBE_TX_FLAGS_CSUM |
7986 IXGBE_TX_FLAGS_IPV4;
7988 ip.v6->payload_len = 0;
7989 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7990 IXGBE_TX_FLAGS_CSUM;
7993 /* determine offset of inner transport header */
7994 l4_offset = l4.hdr - skb->data;
7996 /* compute length of segmentation header */
7997 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
7999 /* remove payload length from inner checksum */
8000 paylen = skb->len - l4_offset;
8001 csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
8003 /* update gso size and bytecount with header size */
8004 first->gso_segs = skb_shinfo(skb)->gso_segs;
8005 first->bytecount += (first->gso_segs - 1) * *hdr_len;
8007 /* mss_l4len_id: use 0 as index for TSO */
8008 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8009 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8011 fceof_saidx |= itd->sa_idx;
8012 type_tucmd |= itd->flags | itd->trailer_len;
8014 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8015 vlan_macip_lens = l4.hdr - ip.hdr;
8016 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8017 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8019 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8025 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb)
8027 unsigned int offset = 0;
8029 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
8031 return offset == skb_checksum_start_offset(skb);
8034 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8035 struct ixgbe_tx_buffer *first,
8036 struct ixgbe_ipsec_tx_data *itd)
8038 struct sk_buff *skb = first->skb;
8039 u32 vlan_macip_lens = 0;
8040 u32 fceof_saidx = 0;
8043 if (skb->ip_summed != CHECKSUM_PARTIAL) {
8045 if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8046 IXGBE_TX_FLAGS_CC)))
8051 switch (skb->csum_offset) {
8052 case offsetof(struct tcphdr, check):
8053 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8055 case offsetof(struct udphdr, check):
8057 case offsetof(struct sctphdr, checksum):
8058 /* validate that this is actually an SCTP request */
8059 if (((first->protocol == htons(ETH_P_IP)) &&
8060 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
8061 ((first->protocol == htons(ETH_P_IPV6)) &&
8062 ixgbe_ipv6_csum_is_sctp(skb))) {
8063 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8068 skb_checksum_help(skb);
8072 /* update TX checksum flag */
8073 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8074 vlan_macip_lens = skb_checksum_start_offset(skb) -
8075 skb_network_offset(skb);
8077 /* vlan_macip_lens: MACLEN, VLAN tag */
8078 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8079 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8081 fceof_saidx |= itd->sa_idx;
8082 type_tucmd |= itd->flags | itd->trailer_len;
8084 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8087 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8088 ((_flag <= _result) ? \
8089 ((u32)(_input & _flag) * (_result / _flag)) : \
8090 ((u32)(_input & _flag) / (_flag / _result)))
8092 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8094 /* set type for advanced descriptor with frame checksum insertion */
8095 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8096 IXGBE_ADVTXD_DCMD_DEXT |
8097 IXGBE_ADVTXD_DCMD_IFCS;
8099 /* set HW vlan bit if vlan is present */
8100 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8101 IXGBE_ADVTXD_DCMD_VLE);
8103 /* set segmentation enable bits for TSO/FSO */
8104 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8105 IXGBE_ADVTXD_DCMD_TSE);
8107 /* set timestamp bit if present */
8108 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8109 IXGBE_ADVTXD_MAC_TSTAMP);
8111 /* insert frame checksum */
8112 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8117 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8118 u32 tx_flags, unsigned int paylen)
8120 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8122 /* enable L4 checksum for TSO and TX checksum offload */
8123 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8124 IXGBE_TX_FLAGS_CSUM,
8125 IXGBE_ADVTXD_POPTS_TXSM);
8127 /* enable IPv4 checksum for TSO */
8128 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8129 IXGBE_TX_FLAGS_IPV4,
8130 IXGBE_ADVTXD_POPTS_IXSM);
8133 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8134 IXGBE_TX_FLAGS_IPSEC,
8135 IXGBE_ADVTXD_POPTS_IPSEC);
8138 * Check Context must be set if Tx switch is enabled, which it
8139 * always is for case where virtual functions are running
8141 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8145 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8148 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8150 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8152 /* Herbert's original patch had:
8153 * smp_mb__after_netif_stop_queue();
8154 * but since that doesn't exist yet, just open code it.
8158 /* We need to check again in a case another CPU has just
8159 * made room available.
8161 if (likely(ixgbe_desc_unused(tx_ring) < size))
8164 /* A reprieve! - use start_queue because it doesn't call schedule */
8165 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8166 ++tx_ring->tx_stats.restart_queue;
8170 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8172 if (likely(ixgbe_desc_unused(tx_ring) >= size))
8175 return __ixgbe_maybe_stop_tx(tx_ring, size);
8178 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8179 struct ixgbe_tx_buffer *first,
8182 struct sk_buff *skb = first->skb;
8183 struct ixgbe_tx_buffer *tx_buffer;
8184 union ixgbe_adv_tx_desc *tx_desc;
8185 struct skb_frag_struct *frag;
8187 unsigned int data_len, size;
8188 u32 tx_flags = first->tx_flags;
8189 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8190 u16 i = tx_ring->next_to_use;
8192 tx_desc = IXGBE_TX_DESC(tx_ring, i);
8194 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8196 size = skb_headlen(skb);
8197 data_len = skb->data_len;
8200 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8201 if (data_len < sizeof(struct fcoe_crc_eof)) {
8202 size -= sizeof(struct fcoe_crc_eof) - data_len;
8205 data_len -= sizeof(struct fcoe_crc_eof);
8210 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8214 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8215 if (dma_mapping_error(tx_ring->dev, dma))
8218 /* record length, and DMA address */
8219 dma_unmap_len_set(tx_buffer, len, size);
8220 dma_unmap_addr_set(tx_buffer, dma, dma);
8222 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8224 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8225 tx_desc->read.cmd_type_len =
8226 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8230 if (i == tx_ring->count) {
8231 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8234 tx_desc->read.olinfo_status = 0;
8236 dma += IXGBE_MAX_DATA_PER_TXD;
8237 size -= IXGBE_MAX_DATA_PER_TXD;
8239 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8242 if (likely(!data_len))
8245 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8249 if (i == tx_ring->count) {
8250 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8253 tx_desc->read.olinfo_status = 0;
8256 size = min_t(unsigned int, data_len, skb_frag_size(frag));
8258 size = skb_frag_size(frag);
8262 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8265 tx_buffer = &tx_ring->tx_buffer_info[i];
8268 /* write last descriptor with RS and EOP bits */
8269 cmd_type |= size | IXGBE_TXD_CMD;
8270 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8272 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8274 /* set the timestamp */
8275 first->time_stamp = jiffies;
8277 skb_tx_timestamp(skb);
8280 * Force memory writes to complete before letting h/w know there
8281 * are new descriptors to fetch. (Only applicable for weak-ordered
8282 * memory model archs, such as IA-64).
8284 * We also need this memory barrier to make certain all of the
8285 * status bits have been updated before next_to_watch is written.
8289 /* set next_to_watch value indicating a packet is present */
8290 first->next_to_watch = tx_desc;
8293 if (i == tx_ring->count)
8296 tx_ring->next_to_use = i;
8298 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8300 if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
8301 writel(i, tx_ring->tail);
8303 /* we need this if more than one processor can write to our tail
8304 * at a time, it synchronizes IO on IA64/Altix systems
8311 dev_err(tx_ring->dev, "TX DMA map failed\n");
8313 /* clear dma mappings for failed tx_buffer_info map */
8315 tx_buffer = &tx_ring->tx_buffer_info[i];
8316 if (dma_unmap_len(tx_buffer, len))
8317 dma_unmap_page(tx_ring->dev,
8318 dma_unmap_addr(tx_buffer, dma),
8319 dma_unmap_len(tx_buffer, len),
8321 dma_unmap_len_set(tx_buffer, len, 0);
8322 if (tx_buffer == first)
8325 i += tx_ring->count;
8329 dev_kfree_skb_any(first->skb);
8332 tx_ring->next_to_use = i;
8337 static void ixgbe_atr(struct ixgbe_ring *ring,
8338 struct ixgbe_tx_buffer *first)
8340 struct ixgbe_q_vector *q_vector = ring->q_vector;
8341 union ixgbe_atr_hash_dword input = { .dword = 0 };
8342 union ixgbe_atr_hash_dword common = { .dword = 0 };
8344 unsigned char *network;
8346 struct ipv6hdr *ipv6;
8350 struct sk_buff *skb;
8354 /* if ring doesn't have a interrupt vector, cannot perform ATR */
8358 /* do nothing if sampling is disabled */
8359 if (!ring->atr_sample_rate)
8364 /* currently only IPv4/IPv6 with TCP is supported */
8365 if ((first->protocol != htons(ETH_P_IP)) &&
8366 (first->protocol != htons(ETH_P_IPV6)))
8369 /* snag network header to get L4 type and address */
8371 hdr.network = skb_network_header(skb);
8372 if (unlikely(hdr.network <= skb->data))
8374 if (skb->encapsulation &&
8375 first->protocol == htons(ETH_P_IP) &&
8376 hdr.ipv4->protocol == IPPROTO_UDP) {
8377 struct ixgbe_adapter *adapter = q_vector->adapter;
8379 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8383 /* verify the port is recognized as VXLAN */
8384 if (adapter->vxlan_port &&
8385 udp_hdr(skb)->dest == adapter->vxlan_port)
8386 hdr.network = skb_inner_network_header(skb);
8388 if (adapter->geneve_port &&
8389 udp_hdr(skb)->dest == adapter->geneve_port)
8390 hdr.network = skb_inner_network_header(skb);
8393 /* Make sure we have at least [minimum IPv4 header + TCP]
8394 * or [IPv6 header] bytes
8396 if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8399 /* Currently only IPv4/IPv6 with TCP is supported */
8400 switch (hdr.ipv4->version) {
8402 /* access ihl as u8 to avoid unaligned access on ia64 */
8403 hlen = (hdr.network[0] & 0x0F) << 2;
8404 l4_proto = hdr.ipv4->protocol;
8407 hlen = hdr.network - skb->data;
8408 l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8409 hlen -= hdr.network - skb->data;
8415 if (l4_proto != IPPROTO_TCP)
8418 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8419 hlen + sizeof(struct tcphdr)))
8422 th = (struct tcphdr *)(hdr.network + hlen);
8424 /* skip this packet since the socket is closing */
8428 /* sample on all syn packets or once every atr sample count */
8429 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8432 /* reset sample count */
8433 ring->atr_count = 0;
8435 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8438 * src and dst are inverted, think how the receiver sees them
8440 * The input is broken into two sections, a non-compressed section
8441 * containing vm_pool, vlan_id, and flow_type. The rest of the data
8442 * is XORed together and stored in the compressed dword.
8444 input.formatted.vlan_id = vlan_id;
8447 * since src port and flex bytes occupy the same word XOR them together
8448 * and write the value to source port portion of compressed dword
8450 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8451 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8453 common.port.src ^= th->dest ^ first->protocol;
8454 common.port.dst ^= th->source;
8456 switch (hdr.ipv4->version) {
8458 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8459 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8462 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8463 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8464 hdr.ipv6->saddr.s6_addr32[1] ^
8465 hdr.ipv6->saddr.s6_addr32[2] ^
8466 hdr.ipv6->saddr.s6_addr32[3] ^
8467 hdr.ipv6->daddr.s6_addr32[0] ^
8468 hdr.ipv6->daddr.s6_addr32[1] ^
8469 hdr.ipv6->daddr.s6_addr32[2] ^
8470 hdr.ipv6->daddr.s6_addr32[3];
8476 if (hdr.network != skb_network_header(skb))
8477 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8479 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
8480 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8481 input, common, ring->queue_index);
8485 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8486 struct net_device *sb_dev,
8487 select_queue_fallback_t fallback)
8489 struct ixgbe_adapter *adapter;
8490 struct ixgbe_ring_feature *f;
8494 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8495 struct net_device *vdev = sb_dev;
8497 txq = vdev->tc_to_txq[tc].offset;
8498 txq += reciprocal_scale(skb_get_hash(skb),
8499 vdev->tc_to_txq[tc].count);
8505 * only execute the code below if protocol is FCoE
8506 * or FIP and we have FCoE enabled on the adapter
8508 switch (vlan_get_protocol(skb)) {
8509 case htons(ETH_P_FCOE):
8510 case htons(ETH_P_FIP):
8511 adapter = netdev_priv(dev);
8513 if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8517 return fallback(dev, skb, sb_dev);
8520 f = &adapter->ring_feature[RING_F_FCOE];
8522 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8525 while (txq >= f->indices)
8528 return txq + f->offset;
8532 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8533 struct xdp_frame *xdpf)
8535 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8536 struct ixgbe_tx_buffer *tx_buffer;
8537 union ixgbe_adv_tx_desc *tx_desc;
8544 if (unlikely(!ixgbe_desc_unused(ring)))
8545 return IXGBE_XDP_CONSUMED;
8547 dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
8548 if (dma_mapping_error(ring->dev, dma))
8549 return IXGBE_XDP_CONSUMED;
8551 /* record the location of the first descriptor for this packet */
8552 tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8553 tx_buffer->bytecount = len;
8554 tx_buffer->gso_segs = 1;
8555 tx_buffer->protocol = 0;
8557 i = ring->next_to_use;
8558 tx_desc = IXGBE_TX_DESC(ring, i);
8560 dma_unmap_len_set(tx_buffer, len, len);
8561 dma_unmap_addr_set(tx_buffer, dma, dma);
8562 tx_buffer->xdpf = xdpf;
8564 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8566 /* put descriptor type bits */
8567 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8568 IXGBE_ADVTXD_DCMD_DEXT |
8569 IXGBE_ADVTXD_DCMD_IFCS;
8570 cmd_type |= len | IXGBE_TXD_CMD;
8571 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8572 tx_desc->read.olinfo_status =
8573 cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8575 /* Avoid any potential race with xdp_xmit and cleanup */
8578 /* set next_to_watch value indicating a packet is present */
8580 if (i == ring->count)
8583 tx_buffer->next_to_watch = tx_desc;
8584 ring->next_to_use = i;
8586 return IXGBE_XDP_TX;
8589 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8590 struct ixgbe_adapter *adapter,
8591 struct ixgbe_ring *tx_ring)
8593 struct ixgbe_tx_buffer *first;
8597 u16 count = TXD_USE_COUNT(skb_headlen(skb));
8598 struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8599 __be16 protocol = skb->protocol;
8603 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8604 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8605 * + 2 desc gap to keep tail from touching head,
8606 * + 1 desc for context descriptor,
8607 * otherwise try next time
8609 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8610 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
8612 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8613 tx_ring->tx_stats.tx_busy++;
8614 return NETDEV_TX_BUSY;
8617 /* record the location of the first descriptor for this packet */
8618 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8620 first->bytecount = skb->len;
8621 first->gso_segs = 1;
8623 /* if we have a HW VLAN tag being added default to the HW one */
8624 if (skb_vlan_tag_present(skb)) {
8625 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8626 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8627 /* else if it is a SW VLAN check the next protocol and store the tag */
8628 } else if (protocol == htons(ETH_P_8021Q)) {
8629 struct vlan_hdr *vhdr, _vhdr;
8630 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8634 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8635 IXGBE_TX_FLAGS_VLAN_SHIFT;
8636 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8638 protocol = vlan_get_protocol(skb);
8640 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8641 adapter->ptp_clock) {
8642 if (!test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8644 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8645 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8647 /* schedule check for Tx timestamp */
8648 adapter->ptp_tx_skb = skb_get(skb);
8649 adapter->ptp_tx_start = jiffies;
8650 schedule_work(&adapter->ptp_tx_work);
8652 adapter->tx_hwtstamp_skipped++;
8656 #ifdef CONFIG_PCI_IOV
8658 * Use the l2switch_enable flag - would be false if the DMA
8659 * Tx switch had been disabled.
8661 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8662 tx_flags |= IXGBE_TX_FLAGS_CC;
8665 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8666 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8667 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8668 (skb->priority != TC_PRIO_CONTROL))) {
8669 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8670 tx_flags |= (skb->priority & 0x7) <<
8671 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8672 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8673 struct vlan_ethhdr *vhdr;
8675 if (skb_cow_head(skb, 0))
8677 vhdr = (struct vlan_ethhdr *)skb->data;
8678 vhdr->h_vlan_TCI = htons(tx_flags >>
8679 IXGBE_TX_FLAGS_VLAN_SHIFT);
8681 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8685 /* record initial flags and protocol */
8686 first->tx_flags = tx_flags;
8687 first->protocol = protocol;
8690 /* setup tx offload for FCoE */
8691 if ((protocol == htons(ETH_P_FCOE)) &&
8692 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8693 tso = ixgbe_fso(tx_ring, first, &hdr_len);
8700 #endif /* IXGBE_FCOE */
8702 #ifdef CONFIG_IXGBE_IPSEC
8703 if (secpath_exists(skb) &&
8704 !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8707 tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8711 ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8713 /* add the ATR filter if ATR is on */
8714 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8715 ixgbe_atr(tx_ring, first);
8719 #endif /* IXGBE_FCOE */
8720 if (ixgbe_tx_map(tx_ring, first, hdr_len))
8721 goto cleanup_tx_timestamp;
8723 return NETDEV_TX_OK;
8726 dev_kfree_skb_any(first->skb);
8728 cleanup_tx_timestamp:
8729 if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8730 dev_kfree_skb_any(adapter->ptp_tx_skb);
8731 adapter->ptp_tx_skb = NULL;
8732 cancel_work_sync(&adapter->ptp_tx_work);
8733 clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8736 return NETDEV_TX_OK;
8739 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8740 struct net_device *netdev,
8741 struct ixgbe_ring *ring)
8743 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8744 struct ixgbe_ring *tx_ring;
8747 * The minimum packet size for olinfo paylen is 17 so pad the skb
8748 * in order to meet this minimum size requirement.
8750 if (skb_put_padto(skb, 17))
8751 return NETDEV_TX_OK;
8753 tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping];
8754 if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8755 return NETDEV_TX_BUSY;
8757 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8760 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8761 struct net_device *netdev)
8763 return __ixgbe_xmit_frame(skb, netdev, NULL);
8767 * ixgbe_set_mac - Change the Ethernet Address of the NIC
8768 * @netdev: network interface device structure
8769 * @p: pointer to an address structure
8771 * Returns 0 on success, negative on failure
8773 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8775 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8776 struct ixgbe_hw *hw = &adapter->hw;
8777 struct sockaddr *addr = p;
8779 if (!is_valid_ether_addr(addr->sa_data))
8780 return -EADDRNOTAVAIL;
8782 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8783 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8785 ixgbe_mac_set_default_filter(adapter);
8791 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8793 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8794 struct ixgbe_hw *hw = &adapter->hw;
8798 if (adapter->mii_bus) {
8801 if (devad != MDIO_DEVAD_NONE)
8802 regnum |= (devad << 16) | MII_ADDR_C45;
8804 return mdiobus_read(adapter->mii_bus, prtad, regnum);
8807 if (prtad != hw->phy.mdio.prtad)
8809 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8815 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8816 u16 addr, u16 value)
8818 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8819 struct ixgbe_hw *hw = &adapter->hw;
8821 if (adapter->mii_bus) {
8824 if (devad != MDIO_DEVAD_NONE)
8825 regnum |= (devad << 16) | MII_ADDR_C45;
8827 return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8830 if (prtad != hw->phy.mdio.prtad)
8832 return hw->phy.ops.write_reg(hw, addr, devad, value);
8835 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8837 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8841 return ixgbe_ptp_set_ts_config(adapter, req);
8843 return ixgbe_ptp_get_ts_config(adapter, req);
8845 if (!adapter->hw.phy.ops.read_reg)
8849 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8854 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8856 * @dev: network interface device structure
8858 * Returns non-zero on failure
8860 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8863 struct ixgbe_adapter *adapter = netdev_priv(dev);
8864 struct ixgbe_hw *hw = &adapter->hw;
8866 if (is_valid_ether_addr(hw->mac.san_addr)) {
8868 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8871 /* update SAN MAC vmdq pool selection */
8872 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8878 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8880 * @dev: network interface device structure
8882 * Returns non-zero on failure
8884 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8887 struct ixgbe_adapter *adapter = netdev_priv(dev);
8888 struct ixgbe_mac_info *mac = &adapter->hw.mac;
8890 if (is_valid_ether_addr(mac->san_addr)) {
8892 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8898 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8899 struct ixgbe_ring *ring)
8906 start = u64_stats_fetch_begin_irq(&ring->syncp);
8907 packets = ring->stats.packets;
8908 bytes = ring->stats.bytes;
8909 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8910 stats->tx_packets += packets;
8911 stats->tx_bytes += bytes;
8915 static void ixgbe_get_stats64(struct net_device *netdev,
8916 struct rtnl_link_stats64 *stats)
8918 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8922 for (i = 0; i < adapter->num_rx_queues; i++) {
8923 struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8929 start = u64_stats_fetch_begin_irq(&ring->syncp);
8930 packets = ring->stats.packets;
8931 bytes = ring->stats.bytes;
8932 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8933 stats->rx_packets += packets;
8934 stats->rx_bytes += bytes;
8938 for (i = 0; i < adapter->num_tx_queues; i++) {
8939 struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8941 ixgbe_get_ring_stats64(stats, ring);
8943 for (i = 0; i < adapter->num_xdp_queues; i++) {
8944 struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8946 ixgbe_get_ring_stats64(stats, ring);
8950 /* following stats updated by ixgbe_watchdog_task() */
8951 stats->multicast = netdev->stats.multicast;
8952 stats->rx_errors = netdev->stats.rx_errors;
8953 stats->rx_length_errors = netdev->stats.rx_length_errors;
8954 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
8955 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
8958 #ifdef CONFIG_IXGBE_DCB
8960 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8961 * @adapter: pointer to ixgbe_adapter
8962 * @tc: number of traffic classes currently enabled
8964 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8965 * 802.1Q priority maps to a packet buffer that exists.
8967 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8969 struct ixgbe_hw *hw = &adapter->hw;
8973 /* 82598 have a static priority to TC mapping that can not
8974 * be changed so no validation is needed.
8976 if (hw->mac.type == ixgbe_mac_82598EB)
8979 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
8982 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
8983 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
8985 /* If up2tc is out of bounds default to zero */
8987 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
8991 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
8997 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
8998 * @adapter: Pointer to adapter struct
9000 * Populate the netdev user priority to tc map
9002 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
9004 struct net_device *dev = adapter->netdev;
9005 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9006 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9009 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9012 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9013 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9015 tc = ets->prio_tc[prio];
9017 netdev_set_prio_tc_map(dev, prio, tc);
9021 #endif /* CONFIG_IXGBE_DCB */
9022 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev, void *data)
9024 struct ixgbe_adapter *adapter = data;
9025 struct ixgbe_fwd_adapter *accel;
9028 /* we only care about macvlans... */
9029 if (!netif_is_macvlan(vdev))
9032 /* that have hardware offload enabled... */
9033 accel = macvlan_accel_priv(vdev);
9037 /* If we can relocate to a different bit do so */
9038 pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9039 if (pool < adapter->num_rx_pools) {
9040 set_bit(pool, adapter->fwd_bitmask);
9045 /* if we cannot find a free pool then disable the offload */
9046 netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9047 macvlan_release_l2fw_offload(vdev);
9049 /* unbind the queues and drop the subordinate channel config */
9050 netdev_unbind_sb_channel(adapter->netdev, vdev);
9051 netdev_set_sb_channel(vdev, 0);
9058 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9060 struct ixgbe_adapter *adapter = netdev_priv(dev);
9062 /* flush any stale bits out of the fwd bitmask */
9063 bitmap_clear(adapter->fwd_bitmask, 1, 63);
9065 /* walk through upper devices reassigning pools */
9066 netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9071 * ixgbe_setup_tc - configure net_device for multiple traffic classes
9073 * @dev: net device to configure
9074 * @tc: number of traffic classes to enable
9076 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9078 struct ixgbe_adapter *adapter = netdev_priv(dev);
9079 struct ixgbe_hw *hw = &adapter->hw;
9081 /* Hardware supports up to 8 traffic classes */
9082 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9085 if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9088 /* Hardware has to reinitialize queues and interrupts to
9089 * match packet buffer alignment. Unfortunately, the
9090 * hardware is not flexible enough to do this dynamically.
9092 if (netif_running(dev))
9095 ixgbe_reset(adapter);
9097 ixgbe_clear_interrupt_scheme(adapter);
9099 #ifdef CONFIG_IXGBE_DCB
9101 if (adapter->xdp_prog) {
9102 e_warn(probe, "DCB is not supported with XDP\n");
9104 ixgbe_init_interrupt_scheme(adapter);
9105 if (netif_running(dev))
9110 netdev_set_num_tc(dev, tc);
9111 ixgbe_set_prio_tc_map(adapter);
9113 adapter->hw_tcs = tc;
9114 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9116 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9117 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9118 adapter->hw.fc.requested_mode = ixgbe_fc_none;
9121 netdev_reset_tc(dev);
9123 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9124 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9126 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9127 adapter->hw_tcs = tc;
9129 adapter->temp_dcb_cfg.pfc_mode_enable = false;
9130 adapter->dcb_cfg.pfc_mode_enable = false;
9133 ixgbe_validate_rtr(adapter, tc);
9135 #endif /* CONFIG_IXGBE_DCB */
9136 ixgbe_init_interrupt_scheme(adapter);
9138 ixgbe_defrag_macvlan_pools(dev);
9140 if (netif_running(dev))
9141 return ixgbe_open(dev);
9146 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9147 struct tc_cls_u32_offload *cls)
9149 u32 hdl = cls->knode.handle;
9150 u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9151 u32 loc = cls->knode.handle & 0xfffff;
9153 struct ixgbe_jump_table *jump = NULL;
9155 if (loc > IXGBE_MAX_HW_ENTRIES)
9158 if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9161 /* Clear this filter in the link data it is associated with */
9162 if (uhtid != 0x800) {
9163 jump = adapter->jump_tables[uhtid];
9166 if (!test_bit(loc - 1, jump->child_loc_map))
9168 clear_bit(loc - 1, jump->child_loc_map);
9171 /* Check if the filter being deleted is a link */
9172 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9173 jump = adapter->jump_tables[i];
9174 if (jump && jump->link_hdl == hdl) {
9175 /* Delete filters in the hardware in the child hash
9176 * table associated with this link
9178 for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9179 if (!test_bit(j, jump->child_loc_map))
9181 spin_lock(&adapter->fdir_perfect_lock);
9182 err = ixgbe_update_ethtool_fdir_entry(adapter,
9185 spin_unlock(&adapter->fdir_perfect_lock);
9186 clear_bit(j, jump->child_loc_map);
9188 /* Remove resources for this link */
9192 adapter->jump_tables[i] = NULL;
9197 spin_lock(&adapter->fdir_perfect_lock);
9198 err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9199 spin_unlock(&adapter->fdir_perfect_lock);
9203 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9204 struct tc_cls_u32_offload *cls)
9206 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9208 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9211 /* This ixgbe devices do not support hash tables at the moment
9212 * so abort when given hash tables.
9214 if (cls->hnode.divisor > 0)
9217 set_bit(uhtid - 1, &adapter->tables);
9221 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9222 struct tc_cls_u32_offload *cls)
9224 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9226 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9229 clear_bit(uhtid - 1, &adapter->tables);
9233 #ifdef CONFIG_NET_CLS_ACT
9234 struct upper_walk_data {
9235 struct ixgbe_adapter *adapter;
9241 static int get_macvlan_queue(struct net_device *upper, void *_data)
9243 if (netif_is_macvlan(upper)) {
9244 struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9245 struct upper_walk_data *data = _data;
9246 struct ixgbe_adapter *adapter = data->adapter;
9247 int ifindex = data->ifindex;
9249 if (vadapter && upper->ifindex == ifindex) {
9250 data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9251 data->action = data->queue;
9259 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9260 u8 *queue, u64 *action)
9262 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9263 unsigned int num_vfs = adapter->num_vfs, vf;
9264 struct upper_walk_data data;
9265 struct net_device *upper;
9267 /* redirect to a SRIOV VF */
9268 for (vf = 0; vf < num_vfs; ++vf) {
9269 upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9270 if (upper->ifindex == ifindex) {
9271 *queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9273 *action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9278 /* redirect to a offloaded macvlan netdev */
9279 data.adapter = adapter;
9280 data.ifindex = ifindex;
9283 if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9284 get_macvlan_queue, &data)) {
9285 *action = data.action;
9286 *queue = data.queue;
9294 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9295 struct tcf_exts *exts, u64 *action, u8 *queue)
9297 const struct tc_action *a;
9300 if (!tcf_exts_has_actions(exts))
9303 tcf_exts_for_each_action(i, a, exts) {
9305 if (is_tcf_gact_shot(a)) {
9306 *action = IXGBE_FDIR_DROP_QUEUE;
9307 *queue = IXGBE_FDIR_DROP_QUEUE;
9311 /* Redirect to a VF or a offloaded macvlan */
9312 if (is_tcf_mirred_egress_redirect(a)) {
9313 struct net_device *dev = tcf_mirred_dev(a);
9317 return handle_redirect_action(adapter, dev->ifindex,
9327 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9328 struct tcf_exts *exts, u64 *action, u8 *queue)
9332 #endif /* CONFIG_NET_CLS_ACT */
9334 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9335 union ixgbe_atr_input *mask,
9336 struct tc_cls_u32_offload *cls,
9337 struct ixgbe_mat_field *field_ptr,
9338 struct ixgbe_nexthdr *nexthdr)
9342 bool found_entry = false, found_jump_field = false;
9344 for (i = 0; i < cls->knode.sel->nkeys; i++) {
9345 off = cls->knode.sel->keys[i].off;
9346 val = cls->knode.sel->keys[i].val;
9347 m = cls->knode.sel->keys[i].mask;
9349 for (j = 0; field_ptr[j].val; j++) {
9350 if (field_ptr[j].off == off) {
9351 field_ptr[j].val(input, mask, (__force u32)val,
9353 input->filter.formatted.flow_type |=
9360 if (nexthdr->off == cls->knode.sel->keys[i].off &&
9362 (__force u32)cls->knode.sel->keys[i].val &&
9364 (__force u32)cls->knode.sel->keys[i].mask)
9365 found_jump_field = true;
9371 if (nexthdr && !found_jump_field)
9377 mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9378 IXGBE_ATR_L4TYPE_MASK;
9380 if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9381 mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9386 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9387 struct tc_cls_u32_offload *cls)
9389 __be16 protocol = cls->common.protocol;
9390 u32 loc = cls->knode.handle & 0xfffff;
9391 struct ixgbe_hw *hw = &adapter->hw;
9392 struct ixgbe_mat_field *field_ptr;
9393 struct ixgbe_fdir_filter *input = NULL;
9394 union ixgbe_atr_input *mask = NULL;
9395 struct ixgbe_jump_table *jump = NULL;
9396 int i, err = -EINVAL;
9398 u32 uhtid, link_uhtid;
9400 uhtid = TC_U32_USERHTID(cls->knode.handle);
9401 link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9403 /* At the moment cls_u32 jumps to network layer and skips past
9404 * L2 headers. The canonical method to match L2 frames is to use
9405 * negative values. However this is error prone at best but really
9406 * just broken because there is no way to "know" what sort of hdr
9407 * is in front of the network layer. Fix cls_u32 to support L2
9408 * headers when needed.
9410 if (protocol != htons(ETH_P_IP))
9413 if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9414 e_err(drv, "Location out of range\n");
9418 /* cls u32 is a graph starting at root node 0x800. The driver tracks
9419 * links and also the fields used to advance the parser across each
9420 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9421 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9422 * To add support for new nodes update ixgbe_model.h parse structures
9423 * this function _should_ be generic try not to hardcode values here.
9425 if (uhtid == 0x800) {
9426 field_ptr = (adapter->jump_tables[0])->mat;
9428 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9430 if (!adapter->jump_tables[uhtid])
9432 field_ptr = (adapter->jump_tables[uhtid])->mat;
9438 /* At this point we know the field_ptr is valid and need to either
9439 * build cls_u32 link or attach filter. Because adding a link to
9440 * a handle that does not exist is invalid and the same for adding
9441 * rules to handles that don't exist.
9445 struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9447 if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9450 if (!test_bit(link_uhtid - 1, &adapter->tables))
9453 /* Multiple filters as links to the same hash table are not
9454 * supported. To add a new filter with the same next header
9455 * but different match/jump conditions, create a new hash table
9458 if (adapter->jump_tables[link_uhtid] &&
9459 (adapter->jump_tables[link_uhtid])->link_hdl) {
9460 e_err(drv, "Link filter exists for link: %x\n",
9465 for (i = 0; nexthdr[i].jump; i++) {
9466 if (nexthdr[i].o != cls->knode.sel->offoff ||
9467 nexthdr[i].s != cls->knode.sel->offshift ||
9469 (__force u32)cls->knode.sel->offmask)
9472 jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9475 input = kzalloc(sizeof(*input), GFP_KERNEL);
9480 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9485 jump->input = input;
9487 jump->link_hdl = cls->knode.handle;
9489 err = ixgbe_clsu32_build_input(input, mask, cls,
9490 field_ptr, &nexthdr[i]);
9492 jump->mat = nexthdr[i].jump;
9493 adapter->jump_tables[link_uhtid] = jump;
9500 input = kzalloc(sizeof(*input), GFP_KERNEL);
9503 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9509 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9510 if ((adapter->jump_tables[uhtid])->input)
9511 memcpy(input, (adapter->jump_tables[uhtid])->input,
9513 if ((adapter->jump_tables[uhtid])->mask)
9514 memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9517 /* Lookup in all child hash tables if this location is already
9518 * filled with a filter
9520 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9521 struct ixgbe_jump_table *link = adapter->jump_tables[i];
9523 if (link && (test_bit(loc - 1, link->child_loc_map))) {
9524 e_err(drv, "Filter exists in location: %x\n",
9531 err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9535 err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9540 input->sw_idx = loc;
9542 spin_lock(&adapter->fdir_perfect_lock);
9544 if (hlist_empty(&adapter->fdir_filter_list)) {
9545 memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9546 err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9548 goto err_out_w_lock;
9549 } else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9551 goto err_out_w_lock;
9554 ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9555 err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9556 input->sw_idx, queue);
9558 ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9559 spin_unlock(&adapter->fdir_perfect_lock);
9561 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9562 set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9567 spin_unlock(&adapter->fdir_perfect_lock);
9577 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9578 struct tc_cls_u32_offload *cls_u32)
9580 switch (cls_u32->command) {
9581 case TC_CLSU32_NEW_KNODE:
9582 case TC_CLSU32_REPLACE_KNODE:
9583 return ixgbe_configure_clsu32(adapter, cls_u32);
9584 case TC_CLSU32_DELETE_KNODE:
9585 return ixgbe_delete_clsu32(adapter, cls_u32);
9586 case TC_CLSU32_NEW_HNODE:
9587 case TC_CLSU32_REPLACE_HNODE:
9588 return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9589 case TC_CLSU32_DELETE_HNODE:
9590 return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9596 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9599 struct ixgbe_adapter *adapter = cb_priv;
9601 if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9605 case TC_SETUP_CLSU32:
9606 return ixgbe_setup_tc_cls_u32(adapter, type_data);
9612 static int ixgbe_setup_tc_block(struct net_device *dev,
9613 struct tc_block_offload *f)
9615 struct ixgbe_adapter *adapter = netdev_priv(dev);
9617 if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
9620 switch (f->command) {
9622 return tcf_block_cb_register(f->block, ixgbe_setup_tc_block_cb,
9623 adapter, adapter, f->extack);
9624 case TC_BLOCK_UNBIND:
9625 tcf_block_cb_unregister(f->block, ixgbe_setup_tc_block_cb,
9633 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9634 struct tc_mqprio_qopt *mqprio)
9636 mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9637 return ixgbe_setup_tc(dev, mqprio->num_tc);
9640 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9644 case TC_SETUP_BLOCK:
9645 return ixgbe_setup_tc_block(dev, type_data);
9646 case TC_SETUP_QDISC_MQPRIO:
9647 return ixgbe_setup_tc_mqprio(dev, type_data);
9653 #ifdef CONFIG_PCI_IOV
9654 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9656 struct net_device *netdev = adapter->netdev;
9659 ixgbe_setup_tc(netdev, adapter->hw_tcs);
9664 void ixgbe_do_reset(struct net_device *netdev)
9666 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9668 if (netif_running(netdev))
9669 ixgbe_reinit_locked(adapter);
9671 ixgbe_reset(adapter);
9674 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9675 netdev_features_t features)
9677 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9679 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9680 if (!(features & NETIF_F_RXCSUM))
9681 features &= ~NETIF_F_LRO;
9683 /* Turn off LRO if not RSC capable */
9684 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9685 features &= ~NETIF_F_LRO;
9687 if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9688 e_dev_err("LRO is not supported with XDP\n");
9689 features &= ~NETIF_F_LRO;
9695 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9697 int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9700 /* go back to full RSS if we're not running SR-IOV */
9701 if (!adapter->ring_feature[RING_F_VMDQ].offset)
9702 adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9703 IXGBE_FLAG_SRIOV_ENABLED);
9705 adapter->ring_feature[RING_F_RSS].limit = rss;
9706 adapter->ring_feature[RING_F_VMDQ].limit = 1;
9708 ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9711 static int ixgbe_set_features(struct net_device *netdev,
9712 netdev_features_t features)
9714 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9715 netdev_features_t changed = netdev->features ^ features;
9716 bool need_reset = false;
9718 /* Make sure RSC matches LRO, reset if change */
9719 if (!(features & NETIF_F_LRO)) {
9720 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9722 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9723 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9724 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9725 if (adapter->rx_itr_setting == 1 ||
9726 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9727 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9729 } else if ((changed ^ features) & NETIF_F_LRO) {
9730 e_info(probe, "rx-usecs set too low, "
9736 * Check if Flow Director n-tuple support or hw_tc support was
9737 * enabled or disabled. If the state changed, we need to reset.
9739 if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9740 /* turn off ATR, enable perfect filters and reset */
9741 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9744 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9745 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9747 /* turn off perfect filters, enable ATR and reset */
9748 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9751 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9753 /* We cannot enable ATR if SR-IOV is enabled */
9754 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9755 /* We cannot enable ATR if we have 2 or more tcs */
9756 (adapter->hw_tcs > 1) ||
9757 /* We cannot enable ATR if RSS is disabled */
9758 (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9759 /* A sample rate of 0 indicates ATR disabled */
9760 (!adapter->atr_sample_rate))
9761 ; /* do nothing not supported */
9762 else /* otherwise supported and set the flag */
9763 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9766 if (changed & NETIF_F_RXALL)
9769 netdev->features = features;
9771 if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) {
9772 if (features & NETIF_F_RXCSUM) {
9773 adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9775 u32 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9777 ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9781 if ((adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)) {
9782 if (features & NETIF_F_RXCSUM) {
9783 adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9785 u32 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9787 ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9791 if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9792 ixgbe_reset_l2fw_offload(adapter);
9793 else if (need_reset)
9794 ixgbe_do_reset(netdev);
9795 else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9796 NETIF_F_HW_VLAN_CTAG_FILTER))
9797 ixgbe_set_rx_mode(netdev);
9803 * ixgbe_add_udp_tunnel_port - Get notifications about adding UDP tunnel ports
9804 * @dev: The port's netdev
9805 * @ti: Tunnel endpoint information
9807 static void ixgbe_add_udp_tunnel_port(struct net_device *dev,
9808 struct udp_tunnel_info *ti)
9810 struct ixgbe_adapter *adapter = netdev_priv(dev);
9811 struct ixgbe_hw *hw = &adapter->hw;
9812 __be16 port = ti->port;
9816 if (ti->sa_family != AF_INET)
9820 case UDP_TUNNEL_TYPE_VXLAN:
9821 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9824 if (adapter->vxlan_port == port)
9827 if (adapter->vxlan_port) {
9829 "VXLAN port %d set, not adding port %d\n",
9830 ntohs(adapter->vxlan_port),
9835 adapter->vxlan_port = port;
9837 case UDP_TUNNEL_TYPE_GENEVE:
9838 if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9841 if (adapter->geneve_port == port)
9844 if (adapter->geneve_port) {
9846 "GENEVE port %d set, not adding port %d\n",
9847 ntohs(adapter->geneve_port),
9852 port_shift = IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT;
9853 adapter->geneve_port = port;
9859 reg = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) | ntohs(port) << port_shift;
9860 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, reg);
9864 * ixgbe_del_udp_tunnel_port - Get notifications about removing UDP tunnel ports
9865 * @dev: The port's netdev
9866 * @ti: Tunnel endpoint information
9868 static void ixgbe_del_udp_tunnel_port(struct net_device *dev,
9869 struct udp_tunnel_info *ti)
9871 struct ixgbe_adapter *adapter = netdev_priv(dev);
9874 if (ti->type != UDP_TUNNEL_TYPE_VXLAN &&
9875 ti->type != UDP_TUNNEL_TYPE_GENEVE)
9878 if (ti->sa_family != AF_INET)
9882 case UDP_TUNNEL_TYPE_VXLAN:
9883 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9886 if (adapter->vxlan_port != ti->port) {
9887 netdev_info(dev, "VXLAN port %d not found\n",
9892 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9894 case UDP_TUNNEL_TYPE_GENEVE:
9895 if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9898 if (adapter->geneve_port != ti->port) {
9899 netdev_info(dev, "GENEVE port %d not found\n",
9904 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9910 ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9911 adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9914 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9915 struct net_device *dev,
9916 const unsigned char *addr, u16 vid,
9918 struct netlink_ext_ack *extack)
9920 /* guarantee we can provide a unique filter for the unicast address */
9921 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9922 struct ixgbe_adapter *adapter = netdev_priv(dev);
9923 u16 pool = VMDQ_P(0);
9925 if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9929 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9933 * ixgbe_configure_bridge_mode - set various bridge modes
9934 * @adapter: the private structure
9935 * @mode: requested bridge mode
9937 * Configure some settings require for various bridge modes.
9939 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9942 struct ixgbe_hw *hw = &adapter->hw;
9943 unsigned int p, num_pools;
9947 case BRIDGE_MODE_VEPA:
9948 /* disable Tx loopback, rely on switch hairpin mode */
9949 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9951 /* must enable Rx switching replication to allow multicast
9952 * packet reception on all VFs, and to enable source address
9955 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9956 vmdctl |= IXGBE_VT_CTL_REPLEN;
9957 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9959 /* enable Rx source address pruning. Note, this requires
9960 * replication to be enabled or else it does nothing.
9962 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9963 for (p = 0; p < num_pools; p++) {
9964 if (hw->mac.ops.set_source_address_pruning)
9965 hw->mac.ops.set_source_address_pruning(hw,
9970 case BRIDGE_MODE_VEB:
9971 /* enable Tx loopback for internal VF/PF communication */
9972 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9973 IXGBE_PFDTXGSWC_VT_LBEN);
9975 /* disable Rx switching replication unless we have SR-IOV
9978 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9979 if (!adapter->num_vfs)
9980 vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9981 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9983 /* disable Rx source address pruning, since we don't expect to
9984 * be receiving external loopback of our transmitted frames.
9986 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9987 for (p = 0; p < num_pools; p++) {
9988 if (hw->mac.ops.set_source_address_pruning)
9989 hw->mac.ops.set_source_address_pruning(hw,
9998 adapter->bridge_mode = mode;
10000 e_info(drv, "enabling bridge mode: %s\n",
10001 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
10006 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
10007 struct nlmsghdr *nlh, u16 flags,
10008 struct netlink_ext_ack *extack)
10010 struct ixgbe_adapter *adapter = netdev_priv(dev);
10011 struct nlattr *attr, *br_spec;
10014 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10015 return -EOPNOTSUPP;
10017 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
10021 nla_for_each_nested(attr, br_spec, rem) {
10025 if (nla_type(attr) != IFLA_BRIDGE_MODE)
10028 if (nla_len(attr) < sizeof(mode))
10031 mode = nla_get_u16(attr);
10032 status = ixgbe_configure_bridge_mode(adapter, mode);
10042 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
10043 struct net_device *dev,
10044 u32 filter_mask, int nlflags)
10046 struct ixgbe_adapter *adapter = netdev_priv(dev);
10048 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10051 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
10052 adapter->bridge_mode, 0, 0, nlflags,
10053 filter_mask, NULL);
10056 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
10058 struct ixgbe_adapter *adapter = netdev_priv(pdev);
10059 struct ixgbe_fwd_adapter *accel;
10060 int tcs = adapter->hw_tcs ? : 1;
10063 if (adapter->xdp_prog) {
10064 e_warn(probe, "L2FW offload is not supported with XDP\n");
10065 return ERR_PTR(-EINVAL);
10068 /* The hardware supported by ixgbe only filters on the destination MAC
10069 * address. In order to avoid issues we only support offloading modes
10070 * where the hardware can actually provide the functionality.
10072 if (!macvlan_supports_dest_filter(vdev))
10073 return ERR_PTR(-EMEDIUMTYPE);
10075 /* We need to lock down the macvlan to be a single queue device so that
10076 * we can reuse the tc_to_txq field in the macvlan netdev to represent
10077 * the queue mapping to our netdev.
10079 if (netif_is_multiqueue(vdev))
10080 return ERR_PTR(-ERANGE);
10082 pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
10083 if (pool == adapter->num_rx_pools) {
10084 u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
10085 u16 reserved_pools;
10087 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
10088 adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
10089 adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
10090 return ERR_PTR(-EBUSY);
10092 /* Hardware has a limited number of available pools. Each VF,
10093 * and the PF require a pool. Check to ensure we don't
10094 * attempt to use more then the available number of pools.
10096 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
10097 return ERR_PTR(-EBUSY);
10099 /* Enable VMDq flag so device will be set in VM mode */
10100 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
10101 IXGBE_FLAG_SRIOV_ENABLED;
10103 /* Try to reserve as many queues per pool as possible,
10104 * we start with the configurations that support 4 queues
10105 * per pools, followed by 2, and then by just 1 per pool.
10107 if (used_pools < 32 && adapter->num_rx_pools < 16)
10108 reserved_pools = min_t(u16,
10110 16 - adapter->num_rx_pools);
10111 else if (adapter->num_rx_pools < 32)
10112 reserved_pools = min_t(u16,
10114 32 - adapter->num_rx_pools);
10116 reserved_pools = 64 - used_pools;
10119 if (!reserved_pools)
10120 return ERR_PTR(-EBUSY);
10122 adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
10124 /* Force reinit of ring allocation with VMDQ enabled */
10125 err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
10127 return ERR_PTR(err);
10129 if (pool >= adapter->num_rx_pools)
10130 return ERR_PTR(-ENOMEM);
10133 accel = kzalloc(sizeof(*accel), GFP_KERNEL);
10135 return ERR_PTR(-ENOMEM);
10137 set_bit(pool, adapter->fwd_bitmask);
10138 netdev_set_sb_channel(vdev, pool);
10139 accel->pool = pool;
10140 accel->netdev = vdev;
10142 if (!netif_running(pdev))
10145 err = ixgbe_fwd_ring_up(adapter, accel);
10147 return ERR_PTR(err);
10152 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10154 struct ixgbe_fwd_adapter *accel = priv;
10155 struct ixgbe_adapter *adapter = netdev_priv(pdev);
10156 unsigned int rxbase = accel->rx_base_queue;
10159 /* delete unicast filter associated with offloaded interface */
10160 ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10161 VMDQ_P(accel->pool));
10163 /* Allow remaining Rx packets to get flushed out of the
10164 * Rx FIFO before we drop the netdev for the ring.
10166 usleep_range(10000, 20000);
10168 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10169 struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10170 struct ixgbe_q_vector *qv = ring->q_vector;
10172 /* Make sure we aren't processing any packets and clear
10173 * netdev to shut down the ring.
10175 if (netif_running(adapter->netdev))
10176 napi_synchronize(&qv->napi);
10177 ring->netdev = NULL;
10180 /* unbind the queues and drop the subordinate channel config */
10181 netdev_unbind_sb_channel(pdev, accel->netdev);
10182 netdev_set_sb_channel(accel->netdev, 0);
10184 clear_bit(accel->pool, adapter->fwd_bitmask);
10188 #define IXGBE_MAX_MAC_HDR_LEN 127
10189 #define IXGBE_MAX_NETWORK_HDR_LEN 511
10191 static netdev_features_t
10192 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10193 netdev_features_t features)
10195 unsigned int network_hdr_len, mac_hdr_len;
10197 /* Make certain the headers can be described by a context descriptor */
10198 mac_hdr_len = skb_network_header(skb) - skb->data;
10199 if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10200 return features & ~(NETIF_F_HW_CSUM |
10202 NETIF_F_HW_VLAN_CTAG_TX |
10206 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10207 if (unlikely(network_hdr_len > IXGBE_MAX_NETWORK_HDR_LEN))
10208 return features & ~(NETIF_F_HW_CSUM |
10213 /* We can only support IPV4 TSO in tunnels if we can mangle the
10214 * inner IP ID field, so strip TSO if MANGLEID is not supported.
10215 * IPsec offoad sets skb->encapsulation but still can handle
10216 * the TSO, so it's the exception.
10218 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10219 #ifdef CONFIG_IXGBE_IPSEC
10220 if (!secpath_exists(skb))
10222 features &= ~NETIF_F_TSO;
10228 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10230 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10231 struct ixgbe_adapter *adapter = netdev_priv(dev);
10232 struct bpf_prog *old_prog;
10235 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10238 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10241 /* verify ixgbe ring attributes are sufficient for XDP */
10242 for (i = 0; i < adapter->num_rx_queues; i++) {
10243 struct ixgbe_ring *ring = adapter->rx_ring[i];
10245 if (ring_is_rsc_enabled(ring))
10248 if (frame_size > ixgbe_rx_bufsz(ring))
10252 if (nr_cpu_ids > MAX_XDP_QUEUES)
10255 old_prog = xchg(&adapter->xdp_prog, prog);
10256 need_reset = (!!prog != !!old_prog);
10258 /* If transitioning XDP modes reconfigure rings */
10260 int err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10263 rcu_assign_pointer(adapter->xdp_prog, old_prog);
10267 for (i = 0; i < adapter->num_rx_queues; i++)
10268 (void)xchg(&adapter->rx_ring[i]->xdp_prog,
10269 adapter->xdp_prog);
10273 bpf_prog_put(old_prog);
10275 /* Kick start the NAPI context if there is an AF_XDP socket open
10276 * on that queue id. This so that receiving will start.
10278 if (need_reset && prog)
10279 for (i = 0; i < adapter->num_rx_queues; i++)
10280 if (adapter->xdp_ring[i]->xsk_umem)
10281 (void)ixgbe_xsk_async_xmit(adapter->netdev, i);
10286 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10288 struct ixgbe_adapter *adapter = netdev_priv(dev);
10290 switch (xdp->command) {
10291 case XDP_SETUP_PROG:
10292 return ixgbe_xdp_setup(dev, xdp->prog);
10293 case XDP_QUERY_PROG:
10294 xdp->prog_id = adapter->xdp_prog ?
10295 adapter->xdp_prog->aux->id : 0;
10297 case XDP_SETUP_XSK_UMEM:
10298 return ixgbe_xsk_umem_setup(adapter, xdp->xsk.umem,
10299 xdp->xsk.queue_id);
10306 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10308 /* Force memory writes to complete before letting h/w know there
10309 * are new descriptors to fetch.
10312 writel(ring->next_to_use, ring->tail);
10315 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10316 struct xdp_frame **frames, u32 flags)
10318 struct ixgbe_adapter *adapter = netdev_priv(dev);
10319 struct ixgbe_ring *ring;
10323 if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10326 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10329 /* During program transitions its possible adapter->xdp_prog is assigned
10330 * but ring has not been configured yet. In this case simply abort xmit.
10332 ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10333 if (unlikely(!ring))
10336 if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10339 for (i = 0; i < n; i++) {
10340 struct xdp_frame *xdpf = frames[i];
10343 err = ixgbe_xmit_xdp_ring(adapter, xdpf);
10344 if (err != IXGBE_XDP_TX) {
10345 xdp_return_frame_rx_napi(xdpf);
10350 if (unlikely(flags & XDP_XMIT_FLUSH))
10351 ixgbe_xdp_ring_update_tail(ring);
10356 static const struct net_device_ops ixgbe_netdev_ops = {
10357 .ndo_open = ixgbe_open,
10358 .ndo_stop = ixgbe_close,
10359 .ndo_start_xmit = ixgbe_xmit_frame,
10360 .ndo_set_rx_mode = ixgbe_set_rx_mode,
10361 .ndo_validate_addr = eth_validate_addr,
10362 .ndo_set_mac_address = ixgbe_set_mac,
10363 .ndo_change_mtu = ixgbe_change_mtu,
10364 .ndo_tx_timeout = ixgbe_tx_timeout,
10365 .ndo_set_tx_maxrate = ixgbe_tx_maxrate,
10366 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
10367 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
10368 .ndo_do_ioctl = ixgbe_ioctl,
10369 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
10370 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
10371 .ndo_set_vf_rate = ixgbe_ndo_set_vf_bw,
10372 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
10373 .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10374 .ndo_set_vf_trust = ixgbe_ndo_set_vf_trust,
10375 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
10376 .ndo_get_stats64 = ixgbe_get_stats64,
10377 .ndo_setup_tc = __ixgbe_setup_tc,
10379 .ndo_select_queue = ixgbe_select_queue,
10380 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10381 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10382 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10383 .ndo_fcoe_enable = ixgbe_fcoe_enable,
10384 .ndo_fcoe_disable = ixgbe_fcoe_disable,
10385 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10386 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10387 #endif /* IXGBE_FCOE */
10388 .ndo_set_features = ixgbe_set_features,
10389 .ndo_fix_features = ixgbe_fix_features,
10390 .ndo_fdb_add = ixgbe_ndo_fdb_add,
10391 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
10392 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
10393 .ndo_dfwd_add_station = ixgbe_fwd_add,
10394 .ndo_dfwd_del_station = ixgbe_fwd_del,
10395 .ndo_udp_tunnel_add = ixgbe_add_udp_tunnel_port,
10396 .ndo_udp_tunnel_del = ixgbe_del_udp_tunnel_port,
10397 .ndo_features_check = ixgbe_features_check,
10398 .ndo_bpf = ixgbe_xdp,
10399 .ndo_xdp_xmit = ixgbe_xdp_xmit,
10400 .ndo_xsk_async_xmit = ixgbe_xsk_async_xmit,
10403 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10404 struct ixgbe_ring *tx_ring)
10406 unsigned long wait_delay, delay_interval;
10407 struct ixgbe_hw *hw = &adapter->hw;
10408 u8 reg_idx = tx_ring->reg_idx;
10412 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10414 /* delay mechanism from ixgbe_disable_tx */
10415 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10417 wait_loop = IXGBE_MAX_RX_DESC_POLL;
10418 wait_delay = delay_interval;
10420 while (wait_loop--) {
10421 usleep_range(wait_delay, wait_delay + 10);
10422 wait_delay += delay_interval * 2;
10423 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10425 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10429 e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10432 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10433 struct ixgbe_ring *tx_ring)
10435 set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10436 ixgbe_disable_txr_hw(adapter, tx_ring);
10439 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10440 struct ixgbe_ring *rx_ring)
10442 unsigned long wait_delay, delay_interval;
10443 struct ixgbe_hw *hw = &adapter->hw;
10444 u8 reg_idx = rx_ring->reg_idx;
10448 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10449 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10450 rxdctl |= IXGBE_RXDCTL_SWFLSH;
10452 /* write value back with RXDCTL.ENABLE bit cleared */
10453 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10455 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10456 if (hw->mac.type == ixgbe_mac_82598EB &&
10457 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10460 /* delay mechanism from ixgbe_disable_rx */
10461 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10463 wait_loop = IXGBE_MAX_RX_DESC_POLL;
10464 wait_delay = delay_interval;
10466 while (wait_loop--) {
10467 usleep_range(wait_delay, wait_delay + 10);
10468 wait_delay += delay_interval * 2;
10469 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10471 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10475 e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10478 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10480 memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10481 memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10484 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10486 memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10487 memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10491 * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10492 * @adapter: adapter structure
10493 * @ring: ring index
10495 * This function disables a certain Rx/Tx/XDP Tx ring. The function
10496 * assumes that the netdev is running.
10498 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10500 struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10502 rx_ring = adapter->rx_ring[ring];
10503 tx_ring = adapter->tx_ring[ring];
10504 xdp_ring = adapter->xdp_ring[ring];
10506 ixgbe_disable_txr(adapter, tx_ring);
10508 ixgbe_disable_txr(adapter, xdp_ring);
10509 ixgbe_disable_rxr_hw(adapter, rx_ring);
10514 /* Rx/Tx/XDP Tx share the same napi context. */
10515 napi_disable(&rx_ring->q_vector->napi);
10517 ixgbe_clean_tx_ring(tx_ring);
10519 ixgbe_clean_tx_ring(xdp_ring);
10520 ixgbe_clean_rx_ring(rx_ring);
10522 ixgbe_reset_txr_stats(tx_ring);
10524 ixgbe_reset_txr_stats(xdp_ring);
10525 ixgbe_reset_rxr_stats(rx_ring);
10529 * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10530 * @adapter: adapter structure
10531 * @ring: ring index
10533 * This function enables a certain Rx/Tx/XDP Tx ring. The function
10534 * assumes that the netdev is running.
10536 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10538 struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10540 rx_ring = adapter->rx_ring[ring];
10541 tx_ring = adapter->tx_ring[ring];
10542 xdp_ring = adapter->xdp_ring[ring];
10544 /* Rx/Tx/XDP Tx share the same napi context. */
10545 napi_enable(&rx_ring->q_vector->napi);
10547 ixgbe_configure_tx_ring(adapter, tx_ring);
10549 ixgbe_configure_tx_ring(adapter, xdp_ring);
10550 ixgbe_configure_rx_ring(adapter, rx_ring);
10552 clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10554 clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10558 * ixgbe_enumerate_functions - Get the number of ports this device has
10559 * @adapter: adapter structure
10561 * This function enumerates the phsyical functions co-located on a single slot,
10562 * in order to determine how many ports a device has. This is most useful in
10563 * determining the required GT/s of PCIe bandwidth necessary for optimal
10566 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10568 struct pci_dev *entry, *pdev = adapter->pdev;
10571 /* Some cards can not use the generic count PCIe functions method,
10572 * because they are behind a parent switch, so we hardcode these with
10573 * the correct number of functions.
10575 if (ixgbe_pcie_from_parent(&adapter->hw))
10578 list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10579 /* don't count virtual functions */
10580 if (entry->is_virtfn)
10583 /* When the devices on the bus don't all match our device ID,
10584 * we can't reliably determine the correct number of
10585 * functions. This can occur if a function has been direct
10586 * attached to a virtual machine using VT-d, for example. In
10587 * this case, simply return -1 to indicate this.
10589 if ((entry->vendor != pdev->vendor) ||
10590 (entry->device != pdev->device))
10600 * ixgbe_wol_supported - Check whether device supports WoL
10601 * @adapter: the adapter private structure
10602 * @device_id: the device ID
10603 * @subdevice_id: the subsystem device ID
10605 * This function is used by probe and ethtool to determine
10606 * which devices have WoL support
10609 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10612 struct ixgbe_hw *hw = &adapter->hw;
10613 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10615 /* WOL not supported on 82598 */
10616 if (hw->mac.type == ixgbe_mac_82598EB)
10619 /* check eeprom to see if WOL is enabled for X540 and newer */
10620 if (hw->mac.type >= ixgbe_mac_X540) {
10621 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10622 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10623 (hw->bus.func == 0)))
10627 /* WOL is determined based on device IDs for 82599 MACs */
10628 switch (device_id) {
10629 case IXGBE_DEV_ID_82599_SFP:
10630 /* Only these subdevices could supports WOL */
10631 switch (subdevice_id) {
10632 case IXGBE_SUBDEV_ID_82599_560FLR:
10633 case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10634 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10635 case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10636 /* only support first port */
10637 if (hw->bus.func != 0)
10640 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10641 case IXGBE_SUBDEV_ID_82599_SFP:
10642 case IXGBE_SUBDEV_ID_82599_RNDC:
10643 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10644 case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10645 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10646 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10650 case IXGBE_DEV_ID_82599EN_SFP:
10651 /* Only these subdevices support WOL */
10652 switch (subdevice_id) {
10653 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10657 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10658 /* All except this subdevice support WOL */
10659 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10662 case IXGBE_DEV_ID_82599_KX4:
10672 * ixgbe_set_fw_version - Set FW version
10673 * @adapter: the adapter private structure
10675 * This function is used by probe and ethtool to determine the FW version to
10676 * format to display. The FW version is taken from the EEPROM/NVM.
10678 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10680 struct ixgbe_hw *hw = &adapter->hw;
10681 struct ixgbe_nvm_version nvm_ver;
10683 ixgbe_get_oem_prod_version(hw, &nvm_ver);
10684 if (nvm_ver.oem_valid) {
10685 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10686 "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10687 nvm_ver.oem_release);
10691 ixgbe_get_etk_id(hw, &nvm_ver);
10692 ixgbe_get_orom_version(hw, &nvm_ver);
10694 if (nvm_ver.or_valid) {
10695 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10696 "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10697 nvm_ver.or_build, nvm_ver.or_patch);
10701 /* Set ETrack ID format */
10702 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10703 "0x%08x", nvm_ver.etk_id);
10707 * ixgbe_probe - Device Initialization Routine
10708 * @pdev: PCI device information struct
10709 * @ent: entry in ixgbe_pci_tbl
10711 * Returns 0 on success, negative on failure
10713 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10714 * The OS initialization, configuring of the adapter private structure,
10715 * and a hardware reset occur.
10717 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10719 struct net_device *netdev;
10720 struct ixgbe_adapter *adapter = NULL;
10721 struct ixgbe_hw *hw;
10722 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10723 int i, err, pci_using_dac, expected_gts;
10724 unsigned int indices = MAX_TX_QUEUES;
10725 u8 part_str[IXGBE_PBANUM_LENGTH];
10726 bool disable_dev = false;
10732 /* Catch broken hardware that put the wrong VF device ID in
10733 * the PCIe SR-IOV capability.
10735 if (pdev->is_virtfn) {
10736 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10737 pci_name(pdev), pdev->vendor, pdev->device);
10741 err = pci_enable_device_mem(pdev);
10745 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10748 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10750 dev_err(&pdev->dev,
10751 "No usable DMA configuration, aborting\n");
10757 err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10759 dev_err(&pdev->dev,
10760 "pci_request_selected_regions failed 0x%x\n", err);
10764 pci_enable_pcie_error_reporting(pdev);
10766 pci_set_master(pdev);
10767 pci_save_state(pdev);
10769 if (ii->mac == ixgbe_mac_82598EB) {
10770 #ifdef CONFIG_IXGBE_DCB
10771 /* 8 TC w/ 4 queues per TC */
10772 indices = 4 * MAX_TRAFFIC_CLASS;
10774 indices = IXGBE_MAX_RSS_INDICES;
10778 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10781 goto err_alloc_etherdev;
10784 SET_NETDEV_DEV(netdev, &pdev->dev);
10786 adapter = netdev_priv(netdev);
10788 adapter->netdev = netdev;
10789 adapter->pdev = pdev;
10791 hw->back = adapter;
10792 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10794 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10795 pci_resource_len(pdev, 0));
10796 adapter->io_addr = hw->hw_addr;
10797 if (!hw->hw_addr) {
10802 netdev->netdev_ops = &ixgbe_netdev_ops;
10803 ixgbe_set_ethtool_ops(netdev);
10804 netdev->watchdog_timeo = 5 * HZ;
10805 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10808 hw->mac.ops = *ii->mac_ops;
10809 hw->mac.type = ii->mac;
10810 hw->mvals = ii->mvals;
10812 hw->link.ops = *ii->link_ops;
10815 hw->eeprom.ops = *ii->eeprom_ops;
10816 eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10817 if (ixgbe_removed(hw->hw_addr)) {
10821 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10822 if (!(eec & BIT(8)))
10823 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10826 hw->phy.ops = *ii->phy_ops;
10827 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10828 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
10829 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10830 hw->phy.mdio.mmds = 0;
10831 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10832 hw->phy.mdio.dev = netdev;
10833 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10834 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10836 /* setup the private structure */
10837 err = ixgbe_sw_init(adapter, ii);
10841 /* Make sure the SWFW semaphore is in a valid state */
10842 if (hw->mac.ops.init_swfw_sync)
10843 hw->mac.ops.init_swfw_sync(hw);
10845 /* Make it possible the adapter to be woken up via WOL */
10846 switch (adapter->hw.mac.type) {
10847 case ixgbe_mac_82599EB:
10848 case ixgbe_mac_X540:
10849 case ixgbe_mac_X550:
10850 case ixgbe_mac_X550EM_x:
10851 case ixgbe_mac_x550em_a:
10852 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10859 * If there is a fan on this device and it has failed log the
10862 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10863 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10864 if (esdp & IXGBE_ESDP_SDP1)
10865 e_crit(probe, "Fan has stopped, replace the adapter\n");
10868 if (allow_unsupported_sfp)
10869 hw->allow_unsupported_sfp = allow_unsupported_sfp;
10871 /* reset_hw fills in the perm_addr as well */
10872 hw->phy.reset_if_overtemp = true;
10873 err = hw->mac.ops.reset_hw(hw);
10874 hw->phy.reset_if_overtemp = false;
10875 ixgbe_set_eee_capable(adapter);
10876 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10878 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10879 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10880 e_dev_err("Reload the driver after installing a supported module.\n");
10883 e_dev_err("HW Init failed: %d\n", err);
10887 #ifdef CONFIG_PCI_IOV
10888 /* SR-IOV not supported on the 82598 */
10889 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10892 ixgbe_init_mbx_params_pf(hw);
10893 hw->mbx.ops = ii->mbx_ops;
10894 pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10895 ixgbe_enable_sriov(adapter, max_vfs);
10899 netdev->features = NETIF_F_SG |
10906 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10907 NETIF_F_GSO_GRE_CSUM | \
10908 NETIF_F_GSO_IPXIP4 | \
10909 NETIF_F_GSO_IPXIP6 | \
10910 NETIF_F_GSO_UDP_TUNNEL | \
10911 NETIF_F_GSO_UDP_TUNNEL_CSUM)
10913 netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10914 netdev->features |= NETIF_F_GSO_PARTIAL |
10915 IXGBE_GSO_PARTIAL_FEATURES;
10917 if (hw->mac.type >= ixgbe_mac_82599EB)
10918 netdev->features |= NETIF_F_SCTP_CRC;
10920 #ifdef CONFIG_IXGBE_IPSEC
10921 #define IXGBE_ESP_FEATURES (NETIF_F_HW_ESP | \
10922 NETIF_F_HW_ESP_TX_CSUM | \
10925 if (adapter->ipsec)
10926 netdev->features |= IXGBE_ESP_FEATURES;
10928 /* copy netdev features into list of user selectable features */
10929 netdev->hw_features |= netdev->features |
10930 NETIF_F_HW_VLAN_CTAG_FILTER |
10931 NETIF_F_HW_VLAN_CTAG_RX |
10932 NETIF_F_HW_VLAN_CTAG_TX |
10934 NETIF_F_HW_L2FW_DOFFLOAD;
10936 if (hw->mac.type >= ixgbe_mac_82599EB)
10937 netdev->hw_features |= NETIF_F_NTUPLE |
10941 netdev->features |= NETIF_F_HIGHDMA;
10943 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10944 netdev->hw_enc_features |= netdev->vlan_features;
10945 netdev->mpls_features |= NETIF_F_SG |
10949 netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10951 /* set this bit last since it cannot be part of vlan_features */
10952 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10953 NETIF_F_HW_VLAN_CTAG_RX |
10954 NETIF_F_HW_VLAN_CTAG_TX;
10956 netdev->priv_flags |= IFF_UNICAST_FLT;
10957 netdev->priv_flags |= IFF_SUPP_NOFCS;
10959 /* MTU range: 68 - 9710 */
10960 netdev->min_mtu = ETH_MIN_MTU;
10961 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
10963 #ifdef CONFIG_IXGBE_DCB
10964 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
10965 netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
10969 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
10970 unsigned int fcoe_l;
10972 if (hw->mac.ops.get_device_caps) {
10973 hw->mac.ops.get_device_caps(hw, &device_caps);
10974 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
10975 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
10979 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
10980 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
10982 netdev->features |= NETIF_F_FSO |
10985 netdev->vlan_features |= NETIF_F_FSO |
10989 #endif /* IXGBE_FCOE */
10990 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
10991 netdev->hw_features |= NETIF_F_LRO;
10992 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
10993 netdev->features |= NETIF_F_LRO;
10995 if (ixgbe_check_fw_error(adapter)) {
11000 /* make sure the EEPROM is good */
11001 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
11002 e_dev_err("The EEPROM Checksum Is Not Valid\n");
11007 eth_platform_get_mac_address(&adapter->pdev->dev,
11008 adapter->hw.mac.perm_addr);
11010 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
11012 if (!is_valid_ether_addr(netdev->dev_addr)) {
11013 e_dev_err("invalid MAC address\n");
11018 /* Set hw->mac.addr to permanent MAC address */
11019 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
11020 ixgbe_mac_set_default_filter(adapter);
11022 timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
11024 if (ixgbe_removed(hw->hw_addr)) {
11028 INIT_WORK(&adapter->service_task, ixgbe_service_task);
11029 set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
11030 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
11032 err = ixgbe_init_interrupt_scheme(adapter);
11036 for (i = 0; i < adapter->num_rx_queues; i++)
11037 u64_stats_init(&adapter->rx_ring[i]->syncp);
11038 for (i = 0; i < adapter->num_tx_queues; i++)
11039 u64_stats_init(&adapter->tx_ring[i]->syncp);
11040 for (i = 0; i < adapter->num_xdp_queues; i++)
11041 u64_stats_init(&adapter->xdp_ring[i]->syncp);
11043 /* WOL not supported for all devices */
11045 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
11046 hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
11047 pdev->subsystem_device);
11048 if (hw->wol_enabled)
11049 adapter->wol = IXGBE_WUFC_MAG;
11051 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
11053 /* save off EEPROM version number */
11054 ixgbe_set_fw_version(adapter);
11056 /* pick up the PCI bus settings for reporting later */
11057 if (ixgbe_pcie_from_parent(hw))
11058 ixgbe_get_parent_bus_info(adapter);
11060 hw->mac.ops.get_bus_info(hw);
11062 /* calculate the expected PCIe bandwidth required for optimal
11063 * performance. Note that some older parts will never have enough
11064 * bandwidth due to being older generation PCIe parts. We clamp these
11065 * parts to ensure no warning is displayed if it can't be fixed.
11067 switch (hw->mac.type) {
11068 case ixgbe_mac_82598EB:
11069 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
11072 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
11076 /* don't check link if we failed to enumerate functions */
11077 if (expected_gts > 0)
11078 ixgbe_check_minimum_link(adapter, expected_gts);
11080 err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
11082 strlcpy(part_str, "Unknown", sizeof(part_str));
11083 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
11084 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
11085 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
11088 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
11089 hw->mac.type, hw->phy.type, part_str);
11091 e_dev_info("%pM\n", netdev->dev_addr);
11093 /* reset the hardware with the new settings */
11094 err = hw->mac.ops.start_hw(hw);
11095 if (err == IXGBE_ERR_EEPROM_VERSION) {
11096 /* We are running on a pre-production device, log a warning */
11097 e_dev_warn("This device is a pre-production adapter/LOM. "
11098 "Please be aware there may be issues associated "
11099 "with your hardware. If you are experiencing "
11100 "problems please contact your Intel or hardware "
11101 "representative who provided you with this "
11104 strcpy(netdev->name, "eth%d");
11105 pci_set_drvdata(pdev, adapter);
11106 err = register_netdev(netdev);
11111 /* power down the optics for 82599 SFP+ fiber */
11112 if (hw->mac.ops.disable_tx_laser)
11113 hw->mac.ops.disable_tx_laser(hw);
11115 /* carrier off reporting is important to ethtool even BEFORE open */
11116 netif_carrier_off(netdev);
11118 #ifdef CONFIG_IXGBE_DCA
11119 if (dca_add_requester(&pdev->dev) == 0) {
11120 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11121 ixgbe_setup_dca(adapter);
11124 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11125 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11126 for (i = 0; i < adapter->num_vfs; i++)
11127 ixgbe_vf_configuration(pdev, (i | 0x10000000));
11130 /* firmware requires driver version to be 0xFFFFFFFF
11131 * since os does not support feature
11133 if (hw->mac.ops.set_fw_drv_ver)
11134 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11135 sizeof(ixgbe_driver_version) - 1,
11136 ixgbe_driver_version);
11138 /* add san mac addr to netdev */
11139 ixgbe_add_sanmac_netdev(netdev);
11141 e_dev_info("%s\n", ixgbe_default_device_descr);
11143 #ifdef CONFIG_IXGBE_HWMON
11144 if (ixgbe_sysfs_init(adapter))
11145 e_err(probe, "failed to allocate sysfs resources\n");
11146 #endif /* CONFIG_IXGBE_HWMON */
11148 ixgbe_dbg_adapter_init(adapter);
11150 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11151 if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11152 hw->mac.ops.setup_link(hw,
11153 IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11156 ixgbe_mii_bus_init(hw);
11161 ixgbe_release_hw_control(adapter);
11162 ixgbe_clear_interrupt_scheme(adapter);
11164 ixgbe_disable_sriov(adapter);
11165 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11166 iounmap(adapter->io_addr);
11167 kfree(adapter->jump_tables[0]);
11168 kfree(adapter->mac_table);
11169 kfree(adapter->rss_key);
11171 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11172 free_netdev(netdev);
11173 err_alloc_etherdev:
11174 pci_release_mem_regions(pdev);
11177 if (!adapter || disable_dev)
11178 pci_disable_device(pdev);
11183 * ixgbe_remove - Device Removal Routine
11184 * @pdev: PCI device information struct
11186 * ixgbe_remove is called by the PCI subsystem to alert the driver
11187 * that it should release a PCI device. The could be caused by a
11188 * Hot-Plug event, or because the driver is going to be removed from
11191 static void ixgbe_remove(struct pci_dev *pdev)
11193 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11194 struct net_device *netdev;
11198 /* if !adapter then we already cleaned up in probe */
11202 netdev = adapter->netdev;
11203 ixgbe_dbg_adapter_exit(adapter);
11205 set_bit(__IXGBE_REMOVING, &adapter->state);
11206 cancel_work_sync(&adapter->service_task);
11208 if (adapter->mii_bus)
11209 mdiobus_unregister(adapter->mii_bus);
11211 #ifdef CONFIG_IXGBE_DCA
11212 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11213 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11214 dca_remove_requester(&pdev->dev);
11215 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11216 IXGBE_DCA_CTRL_DCA_DISABLE);
11220 #ifdef CONFIG_IXGBE_HWMON
11221 ixgbe_sysfs_exit(adapter);
11222 #endif /* CONFIG_IXGBE_HWMON */
11224 /* remove the added san mac */
11225 ixgbe_del_sanmac_netdev(netdev);
11227 #ifdef CONFIG_PCI_IOV
11228 ixgbe_disable_sriov(adapter);
11230 if (netdev->reg_state == NETREG_REGISTERED)
11231 unregister_netdev(netdev);
11233 ixgbe_stop_ipsec_offload(adapter);
11234 ixgbe_clear_interrupt_scheme(adapter);
11236 ixgbe_release_hw_control(adapter);
11239 kfree(adapter->ixgbe_ieee_pfc);
11240 kfree(adapter->ixgbe_ieee_ets);
11243 iounmap(adapter->io_addr);
11244 pci_release_mem_regions(pdev);
11246 e_dev_info("complete\n");
11248 for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11249 if (adapter->jump_tables[i]) {
11250 kfree(adapter->jump_tables[i]->input);
11251 kfree(adapter->jump_tables[i]->mask);
11253 kfree(adapter->jump_tables[i]);
11256 kfree(adapter->mac_table);
11257 kfree(adapter->rss_key);
11258 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11259 free_netdev(netdev);
11261 pci_disable_pcie_error_reporting(pdev);
11264 pci_disable_device(pdev);
11268 * ixgbe_io_error_detected - called when PCI error is detected
11269 * @pdev: Pointer to PCI device
11270 * @state: The current pci connection state
11272 * This function is called after a PCI bus error affecting
11273 * this device has been detected.
11275 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11276 pci_channel_state_t state)
11278 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11279 struct net_device *netdev = adapter->netdev;
11281 #ifdef CONFIG_PCI_IOV
11282 struct ixgbe_hw *hw = &adapter->hw;
11283 struct pci_dev *bdev, *vfdev;
11284 u32 dw0, dw1, dw2, dw3;
11286 u16 req_id, pf_func;
11288 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11289 adapter->num_vfs == 0)
11290 goto skip_bad_vf_detection;
11292 bdev = pdev->bus->self;
11293 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11294 bdev = bdev->bus->self;
11297 goto skip_bad_vf_detection;
11299 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11301 goto skip_bad_vf_detection;
11303 dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11304 dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11305 dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11306 dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11307 if (ixgbe_removed(hw->hw_addr))
11308 goto skip_bad_vf_detection;
11310 req_id = dw1 >> 16;
11311 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11312 if (!(req_id & 0x0080))
11313 goto skip_bad_vf_detection;
11315 pf_func = req_id & 0x01;
11316 if ((pf_func & 1) == (pdev->devfn & 1)) {
11317 unsigned int device_id;
11319 vf = (req_id & 0x7F) >> 1;
11320 e_dev_err("VF %d has caused a PCIe error\n", vf);
11321 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11322 "%8.8x\tdw3: %8.8x\n",
11323 dw0, dw1, dw2, dw3);
11324 switch (adapter->hw.mac.type) {
11325 case ixgbe_mac_82599EB:
11326 device_id = IXGBE_82599_VF_DEVICE_ID;
11328 case ixgbe_mac_X540:
11329 device_id = IXGBE_X540_VF_DEVICE_ID;
11331 case ixgbe_mac_X550:
11332 device_id = IXGBE_DEV_ID_X550_VF;
11334 case ixgbe_mac_X550EM_x:
11335 device_id = IXGBE_DEV_ID_X550EM_X_VF;
11337 case ixgbe_mac_x550em_a:
11338 device_id = IXGBE_DEV_ID_X550EM_A_VF;
11345 /* Find the pci device of the offending VF */
11346 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11348 if (vfdev->devfn == (req_id & 0xFF))
11350 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11354 * There's a slim chance the VF could have been hot plugged,
11355 * so if it is no longer present we don't need to issue the
11356 * VFLR. Just clean up the AER in that case.
11360 /* Free device reference count */
11361 pci_dev_put(vfdev);
11366 * Even though the error may have occurred on the other port
11367 * we still need to increment the vf error reference count for
11368 * both ports because the I/O resume function will be called
11369 * for both of them.
11371 adapter->vferr_refcount++;
11373 return PCI_ERS_RESULT_RECOVERED;
11375 skip_bad_vf_detection:
11376 #endif /* CONFIG_PCI_IOV */
11377 if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11378 return PCI_ERS_RESULT_DISCONNECT;
11380 if (!netif_device_present(netdev))
11381 return PCI_ERS_RESULT_DISCONNECT;
11384 netif_device_detach(netdev);
11386 if (netif_running(netdev))
11387 ixgbe_close_suspend(adapter);
11389 if (state == pci_channel_io_perm_failure) {
11391 return PCI_ERS_RESULT_DISCONNECT;
11394 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11395 pci_disable_device(pdev);
11398 /* Request a slot reset. */
11399 return PCI_ERS_RESULT_NEED_RESET;
11403 * ixgbe_io_slot_reset - called after the pci bus has been reset.
11404 * @pdev: Pointer to PCI device
11406 * Restart the card from scratch, as if from a cold-boot.
11408 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11410 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11411 pci_ers_result_t result;
11413 if (pci_enable_device_mem(pdev)) {
11414 e_err(probe, "Cannot re-enable PCI device after reset.\n");
11415 result = PCI_ERS_RESULT_DISCONNECT;
11417 smp_mb__before_atomic();
11418 clear_bit(__IXGBE_DISABLED, &adapter->state);
11419 adapter->hw.hw_addr = adapter->io_addr;
11420 pci_set_master(pdev);
11421 pci_restore_state(pdev);
11422 pci_save_state(pdev);
11424 pci_wake_from_d3(pdev, false);
11426 ixgbe_reset(adapter);
11427 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11428 result = PCI_ERS_RESULT_RECOVERED;
11435 * ixgbe_io_resume - called when traffic can start flowing again.
11436 * @pdev: Pointer to PCI device
11438 * This callback is called when the error recovery driver tells us that
11439 * its OK to resume normal operation.
11441 static void ixgbe_io_resume(struct pci_dev *pdev)
11443 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11444 struct net_device *netdev = adapter->netdev;
11446 #ifdef CONFIG_PCI_IOV
11447 if (adapter->vferr_refcount) {
11448 e_info(drv, "Resuming after VF err\n");
11449 adapter->vferr_refcount--;
11455 if (netif_running(netdev))
11456 ixgbe_open(netdev);
11458 netif_device_attach(netdev);
11462 static const struct pci_error_handlers ixgbe_err_handler = {
11463 .error_detected = ixgbe_io_error_detected,
11464 .slot_reset = ixgbe_io_slot_reset,
11465 .resume = ixgbe_io_resume,
11468 static struct pci_driver ixgbe_driver = {
11469 .name = ixgbe_driver_name,
11470 .id_table = ixgbe_pci_tbl,
11471 .probe = ixgbe_probe,
11472 .remove = ixgbe_remove,
11474 .suspend = ixgbe_suspend,
11475 .resume = ixgbe_resume,
11477 .shutdown = ixgbe_shutdown,
11478 .sriov_configure = ixgbe_pci_sriov_configure,
11479 .err_handler = &ixgbe_err_handler
11483 * ixgbe_init_module - Driver Registration Routine
11485 * ixgbe_init_module is the first routine called when the driver is
11486 * loaded. All it does is register with the PCI subsystem.
11488 static int __init ixgbe_init_module(void)
11491 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
11492 pr_info("%s\n", ixgbe_copyright);
11494 ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11496 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11502 ret = pci_register_driver(&ixgbe_driver);
11504 destroy_workqueue(ixgbe_wq);
11509 #ifdef CONFIG_IXGBE_DCA
11510 dca_register_notify(&dca_notifier);
11516 module_init(ixgbe_init_module);
11519 * ixgbe_exit_module - Driver Exit Cleanup Routine
11521 * ixgbe_exit_module is called just before the driver is removed
11524 static void __exit ixgbe_exit_module(void)
11526 #ifdef CONFIG_IXGBE_DCA
11527 dca_unregister_notify(&dca_notifier);
11529 pci_unregister_driver(&ixgbe_driver);
11533 destroy_workqueue(ixgbe_wq);
11538 #ifdef CONFIG_IXGBE_DCA
11539 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11544 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11545 __ixgbe_notify_dca);
11547 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11550 #endif /* CONFIG_IXGBE_DCA */
11552 module_exit(ixgbe_exit_module);