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 <generated/utsrelease.h>
32 #include <scsi/fc/fc_fcoe.h>
33 #include <net/udp_tunnel.h>
34 #include <net/pkt_cls.h>
35 #include <net/tc_act/tc_gact.h>
36 #include <net/tc_act/tc_mirred.h>
37 #include <net/vxlan.h>
39 #include <net/xdp_sock_drv.h>
43 #include "ixgbe_common.h"
44 #include "ixgbe_dcb_82599.h"
45 #include "ixgbe_phy.h"
46 #include "ixgbe_sriov.h"
47 #include "ixgbe_model.h"
48 #include "ixgbe_txrx_common.h"
50 char ixgbe_driver_name[] = "ixgbe";
51 static const char ixgbe_driver_string[] =
52 "Intel(R) 10 Gigabit PCI Express Network Driver";
54 char ixgbe_default_device_descr[] =
55 "Intel(R) 10 Gigabit Network Connection";
57 static char ixgbe_default_device_descr[] =
58 "Intel(R) 10 Gigabit Network Connection";
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");
168 static struct workqueue_struct *ixgbe_wq;
170 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
171 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *);
173 static const struct net_device_ops ixgbe_netdev_ops;
175 static bool netif_is_ixgbe(struct net_device *dev)
177 return dev && (dev->netdev_ops == &ixgbe_netdev_ops);
180 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
183 struct pci_dev *parent_dev;
184 struct pci_bus *parent_bus;
186 parent_bus = adapter->pdev->bus->parent;
190 parent_dev = parent_bus->self;
194 if (!pci_is_pcie(parent_dev))
197 pcie_capability_read_word(parent_dev, reg, value);
198 if (*value == IXGBE_FAILED_READ_CFG_WORD &&
199 ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
204 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
206 struct ixgbe_hw *hw = &adapter->hw;
210 hw->bus.type = ixgbe_bus_type_pci_express;
212 /* Get the negotiated link width and speed from PCI config space of the
213 * parent, as this device is behind a switch
215 err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
217 /* assume caller will handle error case */
221 hw->bus.width = ixgbe_convert_bus_width(link_status);
222 hw->bus.speed = ixgbe_convert_bus_speed(link_status);
228 * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
229 * @hw: hw specific details
231 * This function is used by probe to determine whether a device's PCI-Express
232 * bandwidth details should be gathered from the parent bus instead of from the
233 * device. Used to ensure that various locations all have the correct device ID
236 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
238 switch (hw->device_id) {
239 case IXGBE_DEV_ID_82599_SFP_SF_QP:
240 case IXGBE_DEV_ID_82599_QSFP_SF_QP:
247 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
250 struct ixgbe_hw *hw = &adapter->hw;
251 struct pci_dev *pdev;
253 /* Some devices are not connected over PCIe and thus do not negotiate
254 * speed. These devices do not have valid bus info, and thus any report
255 * we generate may not be correct.
257 if (hw->bus.type == ixgbe_bus_type_internal)
260 /* determine whether to use the parent device */
261 if (ixgbe_pcie_from_parent(&adapter->hw))
262 pdev = adapter->pdev->bus->parent->self;
264 pdev = adapter->pdev;
266 pcie_print_link_status(pdev);
269 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
271 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
272 !test_bit(__IXGBE_REMOVING, &adapter->state) &&
273 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
274 queue_work(ixgbe_wq, &adapter->service_task);
277 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
279 struct ixgbe_adapter *adapter = hw->back;
284 e_dev_err("Adapter removed\n");
285 if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
286 ixgbe_service_event_schedule(adapter);
289 static u32 ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
291 u8 __iomem *reg_addr;
295 reg_addr = READ_ONCE(hw->hw_addr);
296 if (ixgbe_removed(reg_addr))
297 return IXGBE_FAILED_READ_REG;
299 /* Register read of 0xFFFFFFF can indicate the adapter has been removed,
300 * so perform several status register reads to determine if the adapter
303 for (i = 0; i < IXGBE_FAILED_READ_RETRIES; i++) {
304 value = readl(reg_addr + IXGBE_STATUS);
305 if (value != IXGBE_FAILED_READ_REG)
310 if (value == IXGBE_FAILED_READ_REG)
311 ixgbe_remove_adapter(hw);
313 value = readl(reg_addr + reg);
318 * ixgbe_read_reg - Read from device register
319 * @hw: hw specific details
320 * @reg: offset of register to read
322 * Returns : value read or IXGBE_FAILED_READ_REG if removed
324 * This function is used to read device registers. It checks for device
325 * removal by confirming any read that returns all ones by checking the
326 * status register value for all ones. This function avoids reading from
327 * the hardware if a removal was previously detected in which case it
328 * returns IXGBE_FAILED_READ_REG (all ones).
330 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
332 u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
335 if (ixgbe_removed(reg_addr))
336 return IXGBE_FAILED_READ_REG;
337 if (unlikely(hw->phy.nw_mng_if_sel &
338 IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE)) {
339 struct ixgbe_adapter *adapter;
342 for (i = 0; i < 200; ++i) {
343 value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY);
345 goto writes_completed;
346 if (value == IXGBE_FAILED_READ_REG) {
347 ixgbe_remove_adapter(hw);
348 return IXGBE_FAILED_READ_REG;
354 e_warn(hw, "register writes incomplete %08x\n", value);
358 value = readl(reg_addr + reg);
359 if (unlikely(value == IXGBE_FAILED_READ_REG))
360 value = ixgbe_check_remove(hw, reg);
364 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
368 pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
369 if (value == IXGBE_FAILED_READ_CFG_WORD) {
370 ixgbe_remove_adapter(hw);
376 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
378 struct ixgbe_adapter *adapter = hw->back;
381 if (ixgbe_removed(hw->hw_addr))
382 return IXGBE_FAILED_READ_CFG_WORD;
383 pci_read_config_word(adapter->pdev, reg, &value);
384 if (value == IXGBE_FAILED_READ_CFG_WORD &&
385 ixgbe_check_cfg_remove(hw, adapter->pdev))
386 return IXGBE_FAILED_READ_CFG_WORD;
390 #ifdef CONFIG_PCI_IOV
391 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
393 struct ixgbe_adapter *adapter = hw->back;
396 if (ixgbe_removed(hw->hw_addr))
397 return IXGBE_FAILED_READ_CFG_DWORD;
398 pci_read_config_dword(adapter->pdev, reg, &value);
399 if (value == IXGBE_FAILED_READ_CFG_DWORD &&
400 ixgbe_check_cfg_remove(hw, adapter->pdev))
401 return IXGBE_FAILED_READ_CFG_DWORD;
404 #endif /* CONFIG_PCI_IOV */
406 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
408 struct ixgbe_adapter *adapter = hw->back;
410 if (ixgbe_removed(hw->hw_addr))
412 pci_write_config_word(adapter->pdev, reg, value);
415 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
417 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
419 /* flush memory to make sure state is correct before next watchdog */
420 smp_mb__before_atomic();
421 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
424 struct ixgbe_reg_info {
429 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
431 /* General Registers */
432 {IXGBE_CTRL, "CTRL"},
433 {IXGBE_STATUS, "STATUS"},
434 {IXGBE_CTRL_EXT, "CTRL_EXT"},
436 /* Interrupt Registers */
437 {IXGBE_EICR, "EICR"},
440 {IXGBE_SRRCTL(0), "SRRCTL"},
441 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
442 {IXGBE_RDLEN(0), "RDLEN"},
443 {IXGBE_RDH(0), "RDH"},
444 {IXGBE_RDT(0), "RDT"},
445 {IXGBE_RXDCTL(0), "RXDCTL"},
446 {IXGBE_RDBAL(0), "RDBAL"},
447 {IXGBE_RDBAH(0), "RDBAH"},
450 {IXGBE_TDBAL(0), "TDBAL"},
451 {IXGBE_TDBAH(0), "TDBAH"},
452 {IXGBE_TDLEN(0), "TDLEN"},
453 {IXGBE_TDH(0), "TDH"},
454 {IXGBE_TDT(0), "TDT"},
455 {IXGBE_TXDCTL(0), "TXDCTL"},
457 /* List Terminator */
463 * ixgbe_regdump - register printout routine
465 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
471 switch (reginfo->ofs) {
472 case IXGBE_SRRCTL(0):
473 for (i = 0; i < 64; i++)
474 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
476 case IXGBE_DCA_RXCTRL(0):
477 for (i = 0; i < 64; i++)
478 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
481 for (i = 0; i < 64; i++)
482 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
485 for (i = 0; i < 64; i++)
486 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
489 for (i = 0; i < 64; i++)
490 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
492 case IXGBE_RXDCTL(0):
493 for (i = 0; i < 64; i++)
494 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
497 for (i = 0; i < 64; i++)
498 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
501 for (i = 0; i < 64; i++)
502 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
505 for (i = 0; i < 64; i++)
506 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
509 for (i = 0; i < 64; i++)
510 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
513 for (i = 0; i < 64; i++)
514 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
517 for (i = 0; i < 64; i++)
518 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
521 for (i = 0; i < 64; i++)
522 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
524 case IXGBE_TXDCTL(0):
525 for (i = 0; i < 64; i++)
526 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
529 pr_info("%-15s %08x\n",
530 reginfo->name, IXGBE_READ_REG(hw, reginfo->ofs));
540 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i, i + 7);
541 for (j = 0; j < 8; j++)
542 p += sprintf(p, " %08x", regs[i++]);
543 pr_err("%-15s%s\n", rname, buf);
548 static void ixgbe_print_buffer(struct ixgbe_ring *ring, int n)
550 struct ixgbe_tx_buffer *tx_buffer;
552 tx_buffer = &ring->tx_buffer_info[ring->next_to_clean];
553 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
554 n, ring->next_to_use, ring->next_to_clean,
555 (u64)dma_unmap_addr(tx_buffer, dma),
556 dma_unmap_len(tx_buffer, len),
557 tx_buffer->next_to_watch,
558 (u64)tx_buffer->time_stamp);
562 * ixgbe_dump - Print registers, tx-rings and rx-rings
564 static void ixgbe_dump(struct ixgbe_adapter *adapter)
566 struct net_device *netdev = adapter->netdev;
567 struct ixgbe_hw *hw = &adapter->hw;
568 struct ixgbe_reg_info *reginfo;
570 struct ixgbe_ring *ring;
571 struct ixgbe_tx_buffer *tx_buffer;
572 union ixgbe_adv_tx_desc *tx_desc;
573 struct my_u0 { u64 a; u64 b; } *u0;
574 struct ixgbe_ring *rx_ring;
575 union ixgbe_adv_rx_desc *rx_desc;
576 struct ixgbe_rx_buffer *rx_buffer_info;
579 if (!netif_msg_hw(adapter))
582 /* Print netdevice Info */
584 dev_info(&adapter->pdev->dev, "Net device Info\n");
585 pr_info("Device Name state "
587 pr_info("%-15s %016lX %016lX\n",
590 dev_trans_start(netdev));
593 /* Print Registers */
594 dev_info(&adapter->pdev->dev, "Register Dump\n");
595 pr_info(" Register Name Value\n");
596 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
597 reginfo->name; reginfo++) {
598 ixgbe_regdump(hw, reginfo);
601 /* Print TX Ring Summary */
602 if (!netdev || !netif_running(netdev))
605 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
606 pr_info(" %s %s %s %s\n",
607 "Queue [NTU] [NTC] [bi(ntc)->dma ]",
608 "leng", "ntw", "timestamp");
609 for (n = 0; n < adapter->num_tx_queues; n++) {
610 ring = adapter->tx_ring[n];
611 ixgbe_print_buffer(ring, n);
614 for (n = 0; n < adapter->num_xdp_queues; n++) {
615 ring = adapter->xdp_ring[n];
616 ixgbe_print_buffer(ring, n);
620 if (!netif_msg_tx_done(adapter))
621 goto rx_ring_summary;
623 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
625 /* Transmit Descriptor Formats
627 * 82598 Advanced Transmit Descriptor
628 * +--------------------------------------------------------------+
629 * 0 | Buffer Address [63:0] |
630 * +--------------------------------------------------------------+
631 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
632 * +--------------------------------------------------------------+
633 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
635 * 82598 Advanced Transmit Descriptor (Write-Back Format)
636 * +--------------------------------------------------------------+
638 * +--------------------------------------------------------------+
639 * 8 | RSV | STA | NXTSEQ |
640 * +--------------------------------------------------------------+
643 * 82599+ Advanced Transmit Descriptor
644 * +--------------------------------------------------------------+
645 * 0 | Buffer Address [63:0] |
646 * +--------------------------------------------------------------+
647 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN |
648 * +--------------------------------------------------------------+
649 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0
651 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
652 * +--------------------------------------------------------------+
654 * +--------------------------------------------------------------+
655 * 8 | RSV | STA | RSV |
656 * +--------------------------------------------------------------+
660 for (n = 0; n < adapter->num_tx_queues; n++) {
661 ring = adapter->tx_ring[n];
662 pr_info("------------------------------------\n");
663 pr_info("TX QUEUE INDEX = %d\n", ring->queue_index);
664 pr_info("------------------------------------\n");
665 pr_info("%s%s %s %s %s %s\n",
666 "T [desc] [address 63:0 ] ",
667 "[PlPOIdStDDt Ln] [bi->dma ] ",
668 "leng", "ntw", "timestamp", "bi->skb");
670 for (i = 0; ring->desc && (i < ring->count); i++) {
671 tx_desc = IXGBE_TX_DESC(ring, i);
672 tx_buffer = &ring->tx_buffer_info[i];
673 u0 = (struct my_u0 *)tx_desc;
674 if (dma_unmap_len(tx_buffer, len) > 0) {
675 const char *ring_desc;
677 if (i == ring->next_to_use &&
678 i == ring->next_to_clean)
679 ring_desc = " NTC/U";
680 else if (i == ring->next_to_use)
682 else if (i == ring->next_to_clean)
686 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p%s",
688 le64_to_cpu((__force __le64)u0->a),
689 le64_to_cpu((__force __le64)u0->b),
690 (u64)dma_unmap_addr(tx_buffer, dma),
691 dma_unmap_len(tx_buffer, len),
692 tx_buffer->next_to_watch,
693 (u64)tx_buffer->time_stamp,
697 if (netif_msg_pktdata(adapter) &&
699 print_hex_dump(KERN_INFO, "",
700 DUMP_PREFIX_ADDRESS, 16, 1,
701 tx_buffer->skb->data,
702 dma_unmap_len(tx_buffer, len),
708 /* Print RX Rings Summary */
710 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
711 pr_info("Queue [NTU] [NTC]\n");
712 for (n = 0; n < adapter->num_rx_queues; n++) {
713 rx_ring = adapter->rx_ring[n];
714 pr_info("%5d %5X %5X\n",
715 n, rx_ring->next_to_use, rx_ring->next_to_clean);
719 if (!netif_msg_rx_status(adapter))
722 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
724 /* Receive Descriptor Formats
726 * 82598 Advanced Receive Descriptor (Read) Format
728 * +-----------------------------------------------------+
729 * 0 | Packet Buffer Address [63:1] |A0/NSE|
730 * +----------------------------------------------+------+
731 * 8 | Header Buffer Address [63:1] | DD |
732 * +-----------------------------------------------------+
735 * 82598 Advanced Receive Descriptor (Write-Back) Format
737 * 63 48 47 32 31 30 21 20 16 15 4 3 0
738 * +------------------------------------------------------+
739 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS |
740 * | Packet | IP | | | | Type | Type |
741 * | Checksum | Ident | | | | | |
742 * +------------------------------------------------------+
743 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
744 * +------------------------------------------------------+
745 * 63 48 47 32 31 20 19 0
747 * 82599+ Advanced Receive Descriptor (Read) Format
749 * +-----------------------------------------------------+
750 * 0 | Packet Buffer Address [63:1] |A0/NSE|
751 * +----------------------------------------------+------+
752 * 8 | Header Buffer Address [63:1] | DD |
753 * +-----------------------------------------------------+
756 * 82599+ Advanced Receive Descriptor (Write-Back) Format
758 * 63 48 47 32 31 30 21 20 17 16 4 3 0
759 * +------------------------------------------------------+
760 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS |
761 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type |
762 * |/ Flow Dir Flt ID | | | | | |
763 * +------------------------------------------------------+
764 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
765 * +------------------------------------------------------+
766 * 63 48 47 32 31 20 19 0
769 for (n = 0; n < adapter->num_rx_queues; n++) {
770 rx_ring = adapter->rx_ring[n];
771 pr_info("------------------------------------\n");
772 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
773 pr_info("------------------------------------\n");
775 "R [desc] [ PktBuf A0] ",
776 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ",
777 "<-- Adv Rx Read format");
779 "RWB[desc] [PcsmIpSHl PtRs] ",
780 "[vl er S cks ln] ---------------- [bi->skb ] ",
781 "<-- Adv Rx Write-Back format");
783 for (i = 0; i < rx_ring->count; i++) {
784 const char *ring_desc;
786 if (i == rx_ring->next_to_use)
788 else if (i == rx_ring->next_to_clean)
793 rx_buffer_info = &rx_ring->rx_buffer_info[i];
794 rx_desc = IXGBE_RX_DESC(rx_ring, i);
795 u0 = (struct my_u0 *)rx_desc;
796 if (rx_desc->wb.upper.length) {
797 /* Descriptor Done */
798 pr_info("RWB[0x%03X] %016llX %016llX ---------------- %p%s\n",
800 le64_to_cpu((__force __le64)u0->a),
801 le64_to_cpu((__force __le64)u0->b),
805 pr_info("R [0x%03X] %016llX %016llX %016llX %p%s\n",
807 le64_to_cpu((__force __le64)u0->a),
808 le64_to_cpu((__force __le64)u0->b),
809 (u64)rx_buffer_info->dma,
813 if (netif_msg_pktdata(adapter) &&
814 rx_buffer_info->dma) {
815 print_hex_dump(KERN_INFO, "",
816 DUMP_PREFIX_ADDRESS, 16, 1,
817 page_address(rx_buffer_info->page) +
818 rx_buffer_info->page_offset,
819 ixgbe_rx_bufsz(rx_ring), true);
826 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
830 /* Let firmware take over control of h/w */
831 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
832 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
833 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
836 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
840 /* Let firmware know the driver has taken over */
841 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
842 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
843 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
847 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
848 * @adapter: pointer to adapter struct
849 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
850 * @queue: queue to map the corresponding interrupt to
851 * @msix_vector: the vector to map to the corresponding queue
854 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
855 u8 queue, u8 msix_vector)
858 struct ixgbe_hw *hw = &adapter->hw;
859 switch (hw->mac.type) {
860 case ixgbe_mac_82598EB:
861 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
864 index = (((direction * 64) + queue) >> 2) & 0x1F;
865 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
866 ivar &= ~(0xFF << (8 * (queue & 0x3)));
867 ivar |= (msix_vector << (8 * (queue & 0x3)));
868 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
870 case ixgbe_mac_82599EB:
873 case ixgbe_mac_X550EM_x:
874 case ixgbe_mac_x550em_a:
875 if (direction == -1) {
877 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
878 index = ((queue & 1) * 8);
879 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
880 ivar &= ~(0xFF << index);
881 ivar |= (msix_vector << index);
882 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
885 /* tx or rx causes */
886 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
887 index = ((16 * (queue & 1)) + (8 * direction));
888 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
889 ivar &= ~(0xFF << index);
890 ivar |= (msix_vector << index);
891 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
899 void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
904 switch (adapter->hw.mac.type) {
905 case ixgbe_mac_82598EB:
906 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
907 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
909 case ixgbe_mac_82599EB:
912 case ixgbe_mac_X550EM_x:
913 case ixgbe_mac_x550em_a:
914 mask = (qmask & 0xFFFFFFFF);
915 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
916 mask = (qmask >> 32);
917 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
924 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
926 struct ixgbe_hw *hw = &adapter->hw;
927 struct ixgbe_hw_stats *hwstats = &adapter->stats;
931 if ((hw->fc.current_mode != ixgbe_fc_full) &&
932 (hw->fc.current_mode != ixgbe_fc_rx_pause))
935 switch (hw->mac.type) {
936 case ixgbe_mac_82598EB:
937 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
940 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
942 hwstats->lxoffrxc += data;
944 /* refill credits (no tx hang) if we received xoff */
948 for (i = 0; i < adapter->num_tx_queues; i++)
949 clear_bit(__IXGBE_HANG_CHECK_ARMED,
950 &adapter->tx_ring[i]->state);
952 for (i = 0; i < adapter->num_xdp_queues; i++)
953 clear_bit(__IXGBE_HANG_CHECK_ARMED,
954 &adapter->xdp_ring[i]->state);
957 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
959 struct ixgbe_hw *hw = &adapter->hw;
960 struct ixgbe_hw_stats *hwstats = &adapter->stats;
964 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
966 if (adapter->ixgbe_ieee_pfc)
967 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
969 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
970 ixgbe_update_xoff_rx_lfc(adapter);
974 /* update stats for each tc, only valid with PFC enabled */
975 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
978 switch (hw->mac.type) {
979 case ixgbe_mac_82598EB:
980 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
983 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
985 hwstats->pxoffrxc[i] += pxoffrxc;
986 /* Get the TC for given UP */
987 tc = netdev_get_prio_tc_map(adapter->netdev, i);
988 xoff[tc] += pxoffrxc;
991 /* disarm tx queues that have received xoff frames */
992 for (i = 0; i < adapter->num_tx_queues; i++) {
993 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
995 tc = tx_ring->dcb_tc;
997 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1000 for (i = 0; i < adapter->num_xdp_queues; i++) {
1001 struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
1003 tc = xdp_ring->dcb_tc;
1005 clear_bit(__IXGBE_HANG_CHECK_ARMED, &xdp_ring->state);
1009 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1011 return ring->stats.packets;
1014 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1016 unsigned int head, tail;
1018 head = ring->next_to_clean;
1019 tail = ring->next_to_use;
1021 return ((head <= tail) ? tail : tail + ring->count) - head;
1024 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1026 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1027 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1028 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1030 clear_check_for_tx_hang(tx_ring);
1033 * Check for a hung queue, but be thorough. This verifies
1034 * that a transmit has been completed since the previous
1035 * check AND there is at least one packet pending. The
1036 * ARMED bit is set to indicate a potential hang. The
1037 * bit is cleared if a pause frame is received to remove
1038 * false hang detection due to PFC or 802.3x frames. By
1039 * requiring this to fail twice we avoid races with
1040 * pfc clearing the ARMED bit and conditions where we
1041 * run the check_tx_hang logic with a transmit completion
1042 * pending but without time to complete it yet.
1044 if (tx_done_old == tx_done && tx_pending)
1045 /* make sure it is true for two checks in a row */
1046 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1048 /* update completed stats and continue */
1049 tx_ring->tx_stats.tx_done_old = tx_done;
1050 /* reset the countdown */
1051 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1057 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1058 * @adapter: driver private struct
1060 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1063 /* Do the reset outside of interrupt context */
1064 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1065 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
1066 e_warn(drv, "initiating reset due to tx timeout\n");
1067 ixgbe_service_event_schedule(adapter);
1072 * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1073 * @netdev: network interface device structure
1074 * @queue_index: Tx queue to set
1075 * @maxrate: desired maximum transmit bitrate
1077 static int ixgbe_tx_maxrate(struct net_device *netdev,
1078 int queue_index, u32 maxrate)
1080 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1081 struct ixgbe_hw *hw = &adapter->hw;
1082 u32 bcnrc_val = ixgbe_link_mbps(adapter);
1087 /* Calculate the rate factor values to set */
1088 bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT;
1089 bcnrc_val /= maxrate;
1091 /* clear everything but the rate factor */
1092 bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK |
1093 IXGBE_RTTBCNRC_RF_DEC_MASK;
1095 /* enable the rate scheduler */
1096 bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA;
1098 IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index);
1099 IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val);
1105 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1106 * @q_vector: structure containing interrupt and ring information
1107 * @tx_ring: tx ring to clean
1108 * @napi_budget: Used to determine if we are in netpoll
1110 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1111 struct ixgbe_ring *tx_ring, int napi_budget)
1113 struct ixgbe_adapter *adapter = q_vector->adapter;
1114 struct ixgbe_tx_buffer *tx_buffer;
1115 union ixgbe_adv_tx_desc *tx_desc;
1116 unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
1117 unsigned int budget = q_vector->tx.work_limit;
1118 unsigned int i = tx_ring->next_to_clean;
1120 if (test_bit(__IXGBE_DOWN, &adapter->state))
1123 tx_buffer = &tx_ring->tx_buffer_info[i];
1124 tx_desc = IXGBE_TX_DESC(tx_ring, i);
1125 i -= tx_ring->count;
1128 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1130 /* if next_to_watch is not set then there is no work pending */
1134 /* prevent any other reads prior to eop_desc */
1137 /* if DD is not set pending work has not been completed */
1138 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1141 /* clear next_to_watch to prevent false hangs */
1142 tx_buffer->next_to_watch = NULL;
1144 /* update the statistics for this packet */
1145 total_bytes += tx_buffer->bytecount;
1146 total_packets += tx_buffer->gso_segs;
1147 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
1151 if (ring_is_xdp(tx_ring))
1152 xdp_return_frame(tx_buffer->xdpf);
1154 napi_consume_skb(tx_buffer->skb, napi_budget);
1156 /* unmap skb header data */
1157 dma_unmap_single(tx_ring->dev,
1158 dma_unmap_addr(tx_buffer, dma),
1159 dma_unmap_len(tx_buffer, len),
1162 /* clear tx_buffer data */
1163 dma_unmap_len_set(tx_buffer, len, 0);
1165 /* unmap remaining buffers */
1166 while (tx_desc != eop_desc) {
1171 i -= tx_ring->count;
1172 tx_buffer = tx_ring->tx_buffer_info;
1173 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1176 /* unmap any remaining paged data */
1177 if (dma_unmap_len(tx_buffer, len)) {
1178 dma_unmap_page(tx_ring->dev,
1179 dma_unmap_addr(tx_buffer, dma),
1180 dma_unmap_len(tx_buffer, len),
1182 dma_unmap_len_set(tx_buffer, len, 0);
1186 /* move us one more past the eop_desc for start of next pkt */
1191 i -= tx_ring->count;
1192 tx_buffer = tx_ring->tx_buffer_info;
1193 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1196 /* issue prefetch for next Tx descriptor */
1199 /* update budget accounting */
1201 } while (likely(budget));
1203 i += tx_ring->count;
1204 tx_ring->next_to_clean = i;
1205 u64_stats_update_begin(&tx_ring->syncp);
1206 tx_ring->stats.bytes += total_bytes;
1207 tx_ring->stats.packets += total_packets;
1208 u64_stats_update_end(&tx_ring->syncp);
1209 q_vector->tx.total_bytes += total_bytes;
1210 q_vector->tx.total_packets += total_packets;
1211 adapter->tx_ipsec += total_ipsec;
1213 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1214 /* schedule immediate reset if we believe we hung */
1215 struct ixgbe_hw *hw = &adapter->hw;
1216 e_err(drv, "Detected Tx Unit Hang %s\n"
1218 " TDH, TDT <%x>, <%x>\n"
1219 " next_to_use <%x>\n"
1220 " next_to_clean <%x>\n"
1221 "tx_buffer_info[next_to_clean]\n"
1222 " time_stamp <%lx>\n"
1224 ring_is_xdp(tx_ring) ? "(XDP)" : "",
1225 tx_ring->queue_index,
1226 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1227 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1228 tx_ring->next_to_use, i,
1229 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1231 if (!ring_is_xdp(tx_ring))
1232 netif_stop_subqueue(tx_ring->netdev,
1233 tx_ring->queue_index);
1236 "tx hang %d detected on queue %d, resetting adapter\n",
1237 adapter->tx_timeout_count + 1, tx_ring->queue_index);
1239 /* schedule immediate reset if we believe we hung */
1240 ixgbe_tx_timeout_reset(adapter);
1242 /* the adapter is about to reset, no point in enabling stuff */
1246 if (ring_is_xdp(tx_ring))
1249 netdev_tx_completed_queue(txring_txq(tx_ring),
1250 total_packets, total_bytes);
1252 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1253 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1254 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1255 /* Make sure that anybody stopping the queue after this
1256 * sees the new next_to_clean.
1259 if (__netif_subqueue_stopped(tx_ring->netdev,
1260 tx_ring->queue_index)
1261 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1262 netif_wake_subqueue(tx_ring->netdev,
1263 tx_ring->queue_index);
1264 ++tx_ring->tx_stats.restart_queue;
1271 #ifdef CONFIG_IXGBE_DCA
1272 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1273 struct ixgbe_ring *tx_ring,
1276 struct ixgbe_hw *hw = &adapter->hw;
1280 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1281 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1283 switch (hw->mac.type) {
1284 case ixgbe_mac_82598EB:
1285 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1287 case ixgbe_mac_82599EB:
1288 case ixgbe_mac_X540:
1289 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1290 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1293 /* for unknown hardware do not write register */
1298 * We can enable relaxed ordering for reads, but not writes when
1299 * DCA is enabled. This is due to a known issue in some chipsets
1300 * which will cause the DCA tag to be cleared.
1302 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1303 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1304 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1306 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1309 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1310 struct ixgbe_ring *rx_ring,
1313 struct ixgbe_hw *hw = &adapter->hw;
1315 u8 reg_idx = rx_ring->reg_idx;
1317 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1318 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1320 switch (hw->mac.type) {
1321 case ixgbe_mac_82599EB:
1322 case ixgbe_mac_X540:
1323 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1330 * We can enable relaxed ordering for reads, but not writes when
1331 * DCA is enabled. This is due to a known issue in some chipsets
1332 * which will cause the DCA tag to be cleared.
1334 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1335 IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1336 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1338 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1341 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1343 struct ixgbe_adapter *adapter = q_vector->adapter;
1344 struct ixgbe_ring *ring;
1345 int cpu = get_cpu();
1347 if (q_vector->cpu == cpu)
1350 ixgbe_for_each_ring(ring, q_vector->tx)
1351 ixgbe_update_tx_dca(adapter, ring, cpu);
1353 ixgbe_for_each_ring(ring, q_vector->rx)
1354 ixgbe_update_rx_dca(adapter, ring, cpu);
1356 q_vector->cpu = cpu;
1361 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1365 /* always use CB2 mode, difference is masked in the CB driver */
1366 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1367 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1368 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1370 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1371 IXGBE_DCA_CTRL_DCA_DISABLE);
1373 for (i = 0; i < adapter->num_q_vectors; i++) {
1374 adapter->q_vector[i]->cpu = -1;
1375 ixgbe_update_dca(adapter->q_vector[i]);
1379 static int __ixgbe_notify_dca(struct device *dev, void *data)
1381 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1382 unsigned long event = *(unsigned long *)data;
1384 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1388 case DCA_PROVIDER_ADD:
1389 /* if we're already enabled, don't do it again */
1390 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1392 if (dca_add_requester(dev) == 0) {
1393 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1394 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1395 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1398 fallthrough; /* DCA is disabled. */
1399 case DCA_PROVIDER_REMOVE:
1400 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1401 dca_remove_requester(dev);
1402 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1403 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1404 IXGBE_DCA_CTRL_DCA_DISABLE);
1412 #endif /* CONFIG_IXGBE_DCA */
1414 #define IXGBE_RSS_L4_TYPES_MASK \
1415 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1416 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1417 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1418 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1420 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1421 union ixgbe_adv_rx_desc *rx_desc,
1422 struct sk_buff *skb)
1426 if (!(ring->netdev->features & NETIF_F_RXHASH))
1429 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1430 IXGBE_RXDADV_RSSTYPE_MASK;
1435 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1436 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1437 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1442 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1443 * @ring: structure containing ring specific data
1444 * @rx_desc: advanced rx descriptor
1446 * Returns : true if it is FCoE pkt
1448 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1449 union ixgbe_adv_rx_desc *rx_desc)
1451 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1453 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1454 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1455 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1456 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1459 #endif /* IXGBE_FCOE */
1461 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1462 * @ring: structure containing ring specific data
1463 * @rx_desc: current Rx descriptor being processed
1464 * @skb: skb currently being received and modified
1466 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1467 union ixgbe_adv_rx_desc *rx_desc,
1468 struct sk_buff *skb)
1470 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1471 bool encap_pkt = false;
1473 skb_checksum_none_assert(skb);
1475 /* Rx csum disabled */
1476 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1479 /* check for VXLAN and Geneve packets */
1480 if (pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) {
1482 skb->encapsulation = 1;
1485 /* if IP and error */
1486 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1487 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1488 ring->rx_stats.csum_err++;
1492 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1495 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1497 * 82599 errata, UDP frames with a 0 checksum can be marked as
1500 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1501 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1504 ring->rx_stats.csum_err++;
1508 /* It must be a TCP or UDP packet with a valid checksum */
1509 skb->ip_summed = CHECKSUM_UNNECESSARY;
1511 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1514 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1515 skb->ip_summed = CHECKSUM_NONE;
1518 /* If we checked the outer header let the stack know */
1519 skb->csum_level = 1;
1523 static inline unsigned int ixgbe_rx_offset(struct ixgbe_ring *rx_ring)
1525 return ring_uses_build_skb(rx_ring) ? IXGBE_SKB_PAD : 0;
1528 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1529 struct ixgbe_rx_buffer *bi)
1531 struct page *page = bi->page;
1534 /* since we are recycling buffers we should seldom need to alloc */
1538 /* alloc new page for storage */
1539 page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1540 if (unlikely(!page)) {
1541 rx_ring->rx_stats.alloc_rx_page_failed++;
1545 /* map page for use */
1546 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1547 ixgbe_rx_pg_size(rx_ring),
1552 * if mapping failed free memory back to system since
1553 * there isn't much point in holding memory we can't use
1555 if (dma_mapping_error(rx_ring->dev, dma)) {
1556 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1558 rx_ring->rx_stats.alloc_rx_page_failed++;
1564 bi->page_offset = ixgbe_rx_offset(rx_ring);
1565 page_ref_add(page, USHRT_MAX - 1);
1566 bi->pagecnt_bias = USHRT_MAX;
1567 rx_ring->rx_stats.alloc_rx_page++;
1573 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1574 * @rx_ring: ring to place buffers on
1575 * @cleaned_count: number of buffers to replace
1577 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1579 union ixgbe_adv_rx_desc *rx_desc;
1580 struct ixgbe_rx_buffer *bi;
1581 u16 i = rx_ring->next_to_use;
1588 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1589 bi = &rx_ring->rx_buffer_info[i];
1590 i -= rx_ring->count;
1592 bufsz = ixgbe_rx_bufsz(rx_ring);
1595 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1598 /* sync the buffer for use by the device */
1599 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1600 bi->page_offset, bufsz,
1604 * Refresh the desc even if buffer_addrs didn't change
1605 * because each write-back erases this info.
1607 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1613 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1614 bi = rx_ring->rx_buffer_info;
1615 i -= rx_ring->count;
1618 /* clear the length for the next_to_use descriptor */
1619 rx_desc->wb.upper.length = 0;
1622 } while (cleaned_count);
1624 i += rx_ring->count;
1626 if (rx_ring->next_to_use != i) {
1627 rx_ring->next_to_use = i;
1629 /* update next to alloc since we have filled the ring */
1630 rx_ring->next_to_alloc = i;
1632 /* Force memory writes to complete before letting h/w
1633 * know there are new descriptors to fetch. (Only
1634 * applicable for weak-ordered memory model archs,
1638 writel(i, rx_ring->tail);
1642 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1643 struct sk_buff *skb)
1645 u16 hdr_len = skb_headlen(skb);
1647 /* set gso_size to avoid messing up TCP MSS */
1648 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1649 IXGBE_CB(skb)->append_cnt);
1650 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1653 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1654 struct sk_buff *skb)
1656 /* if append_cnt is 0 then frame is not RSC */
1657 if (!IXGBE_CB(skb)->append_cnt)
1660 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1661 rx_ring->rx_stats.rsc_flush++;
1663 ixgbe_set_rsc_gso_size(rx_ring, skb);
1665 /* gso_size is computed using append_cnt so always clear it last */
1666 IXGBE_CB(skb)->append_cnt = 0;
1670 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1671 * @rx_ring: rx descriptor ring packet is being transacted on
1672 * @rx_desc: pointer to the EOP Rx descriptor
1673 * @skb: pointer to current skb being populated
1675 * This function checks the ring, descriptor, and packet information in
1676 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1677 * other fields within the skb.
1679 void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1680 union ixgbe_adv_rx_desc *rx_desc,
1681 struct sk_buff *skb)
1683 struct net_device *dev = rx_ring->netdev;
1684 u32 flags = rx_ring->q_vector->adapter->flags;
1686 ixgbe_update_rsc_stats(rx_ring, skb);
1688 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1690 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1692 if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1693 ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1695 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1696 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1697 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1698 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1701 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
1702 ixgbe_ipsec_rx(rx_ring, rx_desc, skb);
1704 /* record Rx queue, or update MACVLAN statistics */
1705 if (netif_is_ixgbe(dev))
1706 skb_record_rx_queue(skb, rx_ring->queue_index);
1708 macvlan_count_rx(netdev_priv(dev), skb->len + ETH_HLEN, true,
1711 skb->protocol = eth_type_trans(skb, dev);
1714 void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1715 struct sk_buff *skb)
1717 napi_gro_receive(&q_vector->napi, skb);
1721 * ixgbe_is_non_eop - process handling of non-EOP buffers
1722 * @rx_ring: Rx ring being processed
1723 * @rx_desc: Rx descriptor for current buffer
1724 * @skb: Current socket buffer containing buffer in progress
1726 * This function updates next to clean. If the buffer is an EOP buffer
1727 * this function exits returning false, otherwise it will place the
1728 * sk_buff in the next buffer to be chained and return true indicating
1729 * that this is in fact a non-EOP buffer.
1731 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1732 union ixgbe_adv_rx_desc *rx_desc,
1733 struct sk_buff *skb)
1735 u32 ntc = rx_ring->next_to_clean + 1;
1737 /* fetch, update, and store next to clean */
1738 ntc = (ntc < rx_ring->count) ? ntc : 0;
1739 rx_ring->next_to_clean = ntc;
1741 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1743 /* update RSC append count if present */
1744 if (ring_is_rsc_enabled(rx_ring)) {
1745 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1746 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1748 if (unlikely(rsc_enabled)) {
1749 u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1751 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1752 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1754 /* update ntc based on RSC value */
1755 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1756 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1757 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1761 /* if we are the last buffer then there is nothing else to do */
1762 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1765 /* place skb in next buffer to be received */
1766 rx_ring->rx_buffer_info[ntc].skb = skb;
1767 rx_ring->rx_stats.non_eop_descs++;
1773 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1774 * @rx_ring: rx descriptor ring packet is being transacted on
1775 * @skb: pointer to current skb being adjusted
1777 * This function is an ixgbe specific version of __pskb_pull_tail. The
1778 * main difference between this version and the original function is that
1779 * this function can make several assumptions about the state of things
1780 * that allow for significant optimizations versus the standard function.
1781 * As a result we can do things like drop a frag and maintain an accurate
1782 * truesize for the skb.
1784 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1785 struct sk_buff *skb)
1787 skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1789 unsigned int pull_len;
1792 * it is valid to use page_address instead of kmap since we are
1793 * working with pages allocated out of the lomem pool per
1794 * alloc_page(GFP_ATOMIC)
1796 va = skb_frag_address(frag);
1799 * we need the header to contain the greater of either ETH_HLEN or
1800 * 60 bytes if the skb->len is less than 60 for skb_pad.
1802 pull_len = eth_get_headlen(skb->dev, va, IXGBE_RX_HDR_SIZE);
1804 /* align pull length to size of long to optimize memcpy performance */
1805 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1807 /* update all of the pointers */
1808 skb_frag_size_sub(frag, pull_len);
1809 skb_frag_off_add(frag, pull_len);
1810 skb->data_len -= pull_len;
1811 skb->tail += pull_len;
1815 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1816 * @rx_ring: rx descriptor ring packet is being transacted on
1817 * @skb: pointer to current skb being updated
1819 * This function provides a basic DMA sync up for the first fragment of an
1820 * skb. The reason for doing this is that the first fragment cannot be
1821 * unmapped until we have reached the end of packet descriptor for a buffer
1824 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1825 struct sk_buff *skb)
1827 if (ring_uses_build_skb(rx_ring)) {
1828 unsigned long mask = (unsigned long)ixgbe_rx_pg_size(rx_ring) - 1;
1829 unsigned long offset = (unsigned long)(skb->data) & mask;
1831 dma_sync_single_range_for_cpu(rx_ring->dev,
1837 skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1839 dma_sync_single_range_for_cpu(rx_ring->dev,
1842 skb_frag_size(frag),
1846 /* If the page was released, just unmap it. */
1847 if (unlikely(IXGBE_CB(skb)->page_released)) {
1848 dma_unmap_page_attrs(rx_ring->dev, IXGBE_CB(skb)->dma,
1849 ixgbe_rx_pg_size(rx_ring),
1856 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1857 * @rx_ring: rx descriptor ring packet is being transacted on
1858 * @rx_desc: pointer to the EOP Rx descriptor
1859 * @skb: pointer to current skb being fixed
1861 * Check if the skb is valid in the XDP case it will be an error pointer.
1862 * Return true in this case to abort processing and advance to next
1865 * Check for corrupted packet headers caused by senders on the local L2
1866 * embedded NIC switch not setting up their Tx Descriptors right. These
1867 * should be very rare.
1869 * Also address the case where we are pulling data in on pages only
1870 * and as such no data is present in the skb header.
1872 * In addition if skb is not at least 60 bytes we need to pad it so that
1873 * it is large enough to qualify as a valid Ethernet frame.
1875 * Returns true if an error was encountered and skb was freed.
1877 bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1878 union ixgbe_adv_rx_desc *rx_desc,
1879 struct sk_buff *skb)
1881 struct net_device *netdev = rx_ring->netdev;
1883 /* XDP packets use error pointer so abort at this point */
1887 /* Verify netdev is present, and that packet does not have any
1888 * errors that would be unacceptable to the netdev.
1891 (unlikely(ixgbe_test_staterr(rx_desc,
1892 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1893 !(netdev->features & NETIF_F_RXALL)))) {
1894 dev_kfree_skb_any(skb);
1898 /* place header in linear portion of buffer */
1899 if (!skb_headlen(skb))
1900 ixgbe_pull_tail(rx_ring, skb);
1903 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1904 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1908 /* if eth_skb_pad returns an error the skb was freed */
1909 if (eth_skb_pad(skb))
1916 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1917 * @rx_ring: rx descriptor ring to store buffers on
1918 * @old_buff: donor buffer to have page reused
1920 * Synchronizes page for reuse by the adapter
1922 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1923 struct ixgbe_rx_buffer *old_buff)
1925 struct ixgbe_rx_buffer *new_buff;
1926 u16 nta = rx_ring->next_to_alloc;
1928 new_buff = &rx_ring->rx_buffer_info[nta];
1930 /* update, and store next to alloc */
1932 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1934 /* Transfer page from old buffer to new buffer.
1935 * Move each member individually to avoid possible store
1936 * forwarding stalls and unnecessary copy of skb.
1938 new_buff->dma = old_buff->dma;
1939 new_buff->page = old_buff->page;
1940 new_buff->page_offset = old_buff->page_offset;
1941 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
1944 static inline bool ixgbe_page_is_reserved(struct page *page)
1946 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1949 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer,
1950 int rx_buffer_pgcnt)
1952 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1953 struct page *page = rx_buffer->page;
1955 /* avoid re-using remote pages */
1956 if (unlikely(ixgbe_page_is_reserved(page)))
1959 #if (PAGE_SIZE < 8192)
1960 /* if we are only owner of page we can reuse it */
1961 if (unlikely((rx_buffer_pgcnt - pagecnt_bias) > 1))
1964 /* The last offset is a bit aggressive in that we assume the
1965 * worst case of FCoE being enabled and using a 3K buffer.
1966 * However this should have minimal impact as the 1K extra is
1967 * still less than one buffer in size.
1969 #define IXGBE_LAST_OFFSET \
1970 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1971 if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1975 /* If we have drained the page fragment pool we need to update
1976 * the pagecnt_bias and page count so that we fully restock the
1977 * number of references the driver holds.
1979 if (unlikely(pagecnt_bias == 1)) {
1980 page_ref_add(page, USHRT_MAX - 1);
1981 rx_buffer->pagecnt_bias = USHRT_MAX;
1988 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1989 * @rx_ring: rx descriptor ring to transact packets on
1990 * @rx_buffer: buffer containing page to add
1991 * @skb: sk_buff to place the data into
1992 * @size: size of data in rx_buffer
1994 * This function will add the data contained in rx_buffer->page to the skb.
1995 * This is done either through a direct copy if the data in the buffer is
1996 * less than the skb header size, otherwise it will just attach the page as
1997 * a frag to the skb.
1999 * The function will then update the page offset if necessary and return
2000 * true if the buffer can be reused by the adapter.
2002 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
2003 struct ixgbe_rx_buffer *rx_buffer,
2004 struct sk_buff *skb,
2007 #if (PAGE_SIZE < 8192)
2008 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2010 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2011 SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2012 SKB_DATA_ALIGN(size);
2014 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2015 rx_buffer->page_offset, size, truesize);
2016 #if (PAGE_SIZE < 8192)
2017 rx_buffer->page_offset ^= truesize;
2019 rx_buffer->page_offset += truesize;
2023 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2024 union ixgbe_adv_rx_desc *rx_desc,
2025 struct sk_buff **skb,
2026 const unsigned int size,
2027 int *rx_buffer_pgcnt)
2029 struct ixgbe_rx_buffer *rx_buffer;
2031 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2033 #if (PAGE_SIZE < 8192)
2034 page_count(rx_buffer->page);
2038 prefetchw(rx_buffer->page);
2039 *skb = rx_buffer->skb;
2041 /* Delay unmapping of the first packet. It carries the header
2042 * information, HW may still access the header after the writeback.
2043 * Only unmap it when EOP is reached
2045 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2050 ixgbe_dma_sync_frag(rx_ring, *skb);
2053 /* we are reusing so sync this buffer for CPU use */
2054 dma_sync_single_range_for_cpu(rx_ring->dev,
2056 rx_buffer->page_offset,
2060 rx_buffer->pagecnt_bias--;
2065 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2066 struct ixgbe_rx_buffer *rx_buffer,
2067 struct sk_buff *skb,
2068 int rx_buffer_pgcnt)
2070 if (ixgbe_can_reuse_rx_page(rx_buffer, rx_buffer_pgcnt)) {
2071 /* hand second half of page back to the ring */
2072 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2074 if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2075 /* the page has been released from the ring */
2076 IXGBE_CB(skb)->page_released = true;
2078 /* we are not reusing the buffer so unmap it */
2079 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2080 ixgbe_rx_pg_size(rx_ring),
2084 __page_frag_cache_drain(rx_buffer->page,
2085 rx_buffer->pagecnt_bias);
2088 /* clear contents of rx_buffer */
2089 rx_buffer->page = NULL;
2090 rx_buffer->skb = NULL;
2093 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2094 struct ixgbe_rx_buffer *rx_buffer,
2095 struct xdp_buff *xdp,
2096 union ixgbe_adv_rx_desc *rx_desc)
2098 unsigned int size = xdp->data_end - xdp->data;
2099 #if (PAGE_SIZE < 8192)
2100 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2102 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2103 xdp->data_hard_start);
2105 struct sk_buff *skb;
2107 /* prefetch first cache line of first page */
2108 net_prefetch(xdp->data);
2110 /* Note, we get here by enabling legacy-rx via:
2112 * ethtool --set-priv-flags <dev> legacy-rx on
2114 * In this mode, we currently get 0 extra XDP headroom as
2115 * opposed to having legacy-rx off, where we process XDP
2116 * packets going to stack via ixgbe_build_skb(). The latter
2117 * provides us currently with 192 bytes of headroom.
2119 * For ixgbe_construct_skb() mode it means that the
2120 * xdp->data_meta will always point to xdp->data, since
2121 * the helper cannot expand the head. Should this ever
2122 * change in future for legacy-rx mode on, then lets also
2123 * add xdp->data_meta handling here.
2126 /* allocate a skb to store the frags */
2127 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2131 if (size > IXGBE_RX_HDR_SIZE) {
2132 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2133 IXGBE_CB(skb)->dma = rx_buffer->dma;
2135 skb_add_rx_frag(skb, 0, rx_buffer->page,
2136 xdp->data - page_address(rx_buffer->page),
2138 #if (PAGE_SIZE < 8192)
2139 rx_buffer->page_offset ^= truesize;
2141 rx_buffer->page_offset += truesize;
2144 memcpy(__skb_put(skb, size),
2145 xdp->data, ALIGN(size, sizeof(long)));
2146 rx_buffer->pagecnt_bias++;
2152 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2153 struct ixgbe_rx_buffer *rx_buffer,
2154 struct xdp_buff *xdp,
2155 union ixgbe_adv_rx_desc *rx_desc)
2157 unsigned int metasize = xdp->data - xdp->data_meta;
2158 #if (PAGE_SIZE < 8192)
2159 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2161 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2162 SKB_DATA_ALIGN(xdp->data_end -
2163 xdp->data_hard_start);
2165 struct sk_buff *skb;
2167 /* Prefetch first cache line of first page. If xdp->data_meta
2168 * is unused, this points extactly as xdp->data, otherwise we
2169 * likely have a consumer accessing first few bytes of meta
2170 * data, and then actual data.
2172 net_prefetch(xdp->data_meta);
2174 /* build an skb to around the page buffer */
2175 skb = build_skb(xdp->data_hard_start, truesize);
2179 /* update pointers within the skb to store the data */
2180 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2181 __skb_put(skb, xdp->data_end - xdp->data);
2183 skb_metadata_set(skb, metasize);
2185 /* record DMA address if this is the start of a chain of buffers */
2186 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2187 IXGBE_CB(skb)->dma = rx_buffer->dma;
2189 /* update buffer offset */
2190 #if (PAGE_SIZE < 8192)
2191 rx_buffer->page_offset ^= truesize;
2193 rx_buffer->page_offset += truesize;
2199 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2200 struct ixgbe_ring *rx_ring,
2201 struct xdp_buff *xdp)
2203 int err, result = IXGBE_XDP_PASS;
2204 struct bpf_prog *xdp_prog;
2205 struct xdp_frame *xdpf;
2209 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2214 prefetchw(xdp->data_hard_start); /* xdp_frame write */
2216 act = bpf_prog_run_xdp(xdp_prog, xdp);
2221 xdpf = xdp_convert_buff_to_frame(xdp);
2222 if (unlikely(!xdpf))
2224 result = ixgbe_xmit_xdp_ring(adapter, xdpf);
2225 if (result == IXGBE_XDP_CONSUMED)
2229 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2232 result = IXGBE_XDP_REDIR;
2235 bpf_warn_invalid_xdp_action(act);
2239 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2240 fallthrough; /* handle aborts by dropping packet */
2242 result = IXGBE_XDP_CONSUMED;
2247 return ERR_PTR(-result);
2250 static unsigned int ixgbe_rx_frame_truesize(struct ixgbe_ring *rx_ring,
2253 unsigned int truesize;
2255 #if (PAGE_SIZE < 8192)
2256 truesize = ixgbe_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
2258 truesize = ring_uses_build_skb(rx_ring) ?
2259 SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) +
2260 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
2261 SKB_DATA_ALIGN(size);
2266 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2267 struct ixgbe_rx_buffer *rx_buffer,
2270 unsigned int truesize = ixgbe_rx_frame_truesize(rx_ring, size);
2271 #if (PAGE_SIZE < 8192)
2272 rx_buffer->page_offset ^= truesize;
2274 rx_buffer->page_offset += truesize;
2279 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2280 * @q_vector: structure containing interrupt and ring information
2281 * @rx_ring: rx descriptor ring to transact packets on
2282 * @budget: Total limit on number of packets to process
2284 * This function provides a "bounce buffer" approach to Rx interrupt
2285 * processing. The advantage to this is that on systems that have
2286 * expensive overhead for IOMMU access this provides a means of avoiding
2287 * it by maintaining the mapping of the page to the syste.
2289 * Returns amount of work completed
2291 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2292 struct ixgbe_ring *rx_ring,
2295 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2296 struct ixgbe_adapter *adapter = q_vector->adapter;
2299 unsigned int mss = 0;
2300 #endif /* IXGBE_FCOE */
2301 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2302 unsigned int xdp_xmit = 0;
2303 struct xdp_buff xdp;
2305 xdp.rxq = &rx_ring->xdp_rxq;
2307 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
2308 #if (PAGE_SIZE < 8192)
2309 xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, 0);
2312 while (likely(total_rx_packets < budget)) {
2313 union ixgbe_adv_rx_desc *rx_desc;
2314 struct ixgbe_rx_buffer *rx_buffer;
2315 struct sk_buff *skb;
2316 int rx_buffer_pgcnt;
2319 /* return some buffers to hardware, one at a time is too slow */
2320 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2321 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2325 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2326 size = le16_to_cpu(rx_desc->wb.upper.length);
2330 /* This memory barrier is needed to keep us from reading
2331 * any other fields out of the rx_desc until we know the
2332 * descriptor has been written back
2336 rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size, &rx_buffer_pgcnt);
2338 /* retrieve a buffer from the ring */
2340 xdp.data = page_address(rx_buffer->page) +
2341 rx_buffer->page_offset;
2342 xdp.data_meta = xdp.data;
2343 xdp.data_hard_start = xdp.data -
2344 ixgbe_rx_offset(rx_ring);
2345 xdp.data_end = xdp.data + size;
2346 #if (PAGE_SIZE > 4096)
2347 /* At larger PAGE_SIZE, frame_sz depend on len size */
2348 xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, size);
2350 skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2354 unsigned int xdp_res = -PTR_ERR(skb);
2356 if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2357 xdp_xmit |= xdp_res;
2358 ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2360 rx_buffer->pagecnt_bias++;
2363 total_rx_bytes += size;
2365 ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2366 } else if (ring_uses_build_skb(rx_ring)) {
2367 skb = ixgbe_build_skb(rx_ring, rx_buffer,
2370 skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2374 /* exit if we failed to retrieve a buffer */
2376 rx_ring->rx_stats.alloc_rx_buff_failed++;
2377 rx_buffer->pagecnt_bias++;
2381 ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb, rx_buffer_pgcnt);
2384 /* place incomplete frames back on ring for completion */
2385 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2388 /* verify the packet layout is correct */
2389 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2392 /* probably a little skewed due to removing CRC */
2393 total_rx_bytes += skb->len;
2395 /* populate checksum, timestamp, VLAN, and protocol */
2396 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2399 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2400 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2401 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2402 /* include DDPed FCoE data */
2403 if (ddp_bytes > 0) {
2405 mss = rx_ring->netdev->mtu -
2406 sizeof(struct fcoe_hdr) -
2407 sizeof(struct fc_frame_header) -
2408 sizeof(struct fcoe_crc_eof);
2412 total_rx_bytes += ddp_bytes;
2413 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2417 dev_kfree_skb_any(skb);
2422 #endif /* IXGBE_FCOE */
2423 ixgbe_rx_skb(q_vector, skb);
2425 /* update budget accounting */
2429 if (xdp_xmit & IXGBE_XDP_REDIR)
2432 if (xdp_xmit & IXGBE_XDP_TX) {
2433 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2435 /* Force memory writes to complete before letting h/w
2436 * know there are new descriptors to fetch.
2439 writel(ring->next_to_use, ring->tail);
2442 u64_stats_update_begin(&rx_ring->syncp);
2443 rx_ring->stats.packets += total_rx_packets;
2444 rx_ring->stats.bytes += total_rx_bytes;
2445 u64_stats_update_end(&rx_ring->syncp);
2446 q_vector->rx.total_packets += total_rx_packets;
2447 q_vector->rx.total_bytes += total_rx_bytes;
2449 return total_rx_packets;
2453 * ixgbe_configure_msix - Configure MSI-X hardware
2454 * @adapter: board private structure
2456 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2459 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2461 struct ixgbe_q_vector *q_vector;
2465 /* Populate MSIX to EITR Select */
2466 if (adapter->num_vfs > 32) {
2467 u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2468 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2472 * Populate the IVAR table and set the ITR values to the
2473 * corresponding register.
2475 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2476 struct ixgbe_ring *ring;
2477 q_vector = adapter->q_vector[v_idx];
2479 ixgbe_for_each_ring(ring, q_vector->rx)
2480 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2482 ixgbe_for_each_ring(ring, q_vector->tx)
2483 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2485 ixgbe_write_eitr(q_vector);
2488 switch (adapter->hw.mac.type) {
2489 case ixgbe_mac_82598EB:
2490 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2493 case ixgbe_mac_82599EB:
2494 case ixgbe_mac_X540:
2495 case ixgbe_mac_X550:
2496 case ixgbe_mac_X550EM_x:
2497 case ixgbe_mac_x550em_a:
2498 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2503 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2505 /* set up to autoclear timer, and the vectors */
2506 mask = IXGBE_EIMS_ENABLE_MASK;
2507 mask &= ~(IXGBE_EIMS_OTHER |
2508 IXGBE_EIMS_MAILBOX |
2511 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2515 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2516 * @q_vector: structure containing interrupt and ring information
2517 * @ring_container: structure containing ring performance data
2519 * Stores a new ITR value based on packets and byte
2520 * counts during the last interrupt. The advantage of per interrupt
2521 * computation is faster updates and more accurate ITR for the current
2522 * traffic pattern. Constants in this function were computed
2523 * based on theoretical maximum wire speed and thresholds were set based
2524 * on testing data as well as attempting to minimize response time
2525 * while increasing bulk throughput.
2527 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2528 struct ixgbe_ring_container *ring_container)
2530 unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2531 IXGBE_ITR_ADAPTIVE_LATENCY;
2532 unsigned int avg_wire_size, packets, bytes;
2533 unsigned long next_update = jiffies;
2535 /* If we don't have any rings just leave ourselves set for maximum
2536 * possible latency so we take ourselves out of the equation.
2538 if (!ring_container->ring)
2541 /* If we didn't update within up to 1 - 2 jiffies we can assume
2542 * that either packets are coming in so slow there hasn't been
2543 * any work, or that there is so much work that NAPI is dealing
2544 * with interrupt moderation and we don't need to do anything.
2546 if (time_after(next_update, ring_container->next_update))
2549 packets = ring_container->total_packets;
2551 /* We have no packets to actually measure against. This means
2552 * either one of the other queues on this vector is active or
2553 * we are a Tx queue doing TSO with too high of an interrupt rate.
2555 * When this occurs just tick up our delay by the minimum value
2556 * and hope that this extra delay will prevent us from being called
2557 * without any work on our queue.
2560 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2561 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2562 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2563 itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2567 bytes = ring_container->total_bytes;
2569 /* If packets are less than 4 or bytes are less than 9000 assume
2570 * insufficient data to use bulk rate limiting approach. We are
2571 * likely latency driven.
2573 if (packets < 4 && bytes < 9000) {
2574 itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2575 goto adjust_by_size;
2578 /* Between 4 and 48 we can assume that our current interrupt delay
2579 * is only slightly too low. As such we should increase it by a small
2583 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2584 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2585 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2589 /* Between 48 and 96 is our "goldilocks" zone where we are working
2590 * out "just right". Just report that our current ITR is good for us.
2593 itr = q_vector->itr >> 2;
2597 /* If packet count is 96 or greater we are likely looking at a slight
2598 * overrun of the delay we want. Try halving our delay to see if that
2599 * will cut the number of packets in half per interrupt.
2601 if (packets < 256) {
2602 itr = q_vector->itr >> 3;
2603 if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2604 itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2608 /* The paths below assume we are dealing with a bulk ITR since number
2609 * of packets is 256 or greater. We are just going to have to compute
2610 * a value and try to bring the count under control, though for smaller
2611 * packet sizes there isn't much we can do as NAPI polling will likely
2612 * be kicking in sooner rather than later.
2614 itr = IXGBE_ITR_ADAPTIVE_BULK;
2617 /* If packet counts are 256 or greater we can assume we have a gross
2618 * overestimation of what the rate should be. Instead of trying to fine
2619 * tune it just use the formula below to try and dial in an exact value
2620 * give the current packet size of the frame.
2622 avg_wire_size = bytes / packets;
2624 /* The following is a crude approximation of:
2625 * wmem_default / (size + overhead) = desired_pkts_per_int
2626 * rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2627 * (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2629 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2630 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2633 * (170 * (size + 24)) / (size + 640) = ITR
2635 * We first do some math on the packet size and then finally bitshift
2636 * by 8 after rounding up. We also have to account for PCIe link speed
2637 * difference as ITR scales based on this.
2639 if (avg_wire_size <= 60) {
2640 /* Start at 50k ints/sec */
2641 avg_wire_size = 5120;
2642 } else if (avg_wire_size <= 316) {
2643 /* 50K ints/sec to 16K ints/sec */
2644 avg_wire_size *= 40;
2645 avg_wire_size += 2720;
2646 } else if (avg_wire_size <= 1084) {
2647 /* 16K ints/sec to 9.2K ints/sec */
2648 avg_wire_size *= 15;
2649 avg_wire_size += 11452;
2650 } else if (avg_wire_size < 1968) {
2651 /* 9.2K ints/sec to 8K ints/sec */
2653 avg_wire_size += 22420;
2655 /* plateau at a limit of 8K ints/sec */
2656 avg_wire_size = 32256;
2659 /* If we are in low latency mode half our delay which doubles the rate
2660 * to somewhere between 100K to 16K ints/sec
2662 if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2663 avg_wire_size >>= 1;
2665 /* Resultant value is 256 times larger than it needs to be. This
2666 * gives us room to adjust the value as needed to either increase
2667 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2669 * Use addition as we have already recorded the new latency flag
2670 * for the ITR value.
2672 switch (q_vector->adapter->link_speed) {
2673 case IXGBE_LINK_SPEED_10GB_FULL:
2674 case IXGBE_LINK_SPEED_100_FULL:
2676 itr += DIV_ROUND_UP(avg_wire_size,
2677 IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2678 IXGBE_ITR_ADAPTIVE_MIN_INC;
2680 case IXGBE_LINK_SPEED_2_5GB_FULL:
2681 case IXGBE_LINK_SPEED_1GB_FULL:
2682 case IXGBE_LINK_SPEED_10_FULL:
2683 if (avg_wire_size > 8064)
2684 avg_wire_size = 8064;
2685 itr += DIV_ROUND_UP(avg_wire_size,
2686 IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2687 IXGBE_ITR_ADAPTIVE_MIN_INC;
2692 /* write back value */
2693 ring_container->itr = itr;
2695 /* next update should occur within next jiffy */
2696 ring_container->next_update = next_update + 1;
2698 ring_container->total_bytes = 0;
2699 ring_container->total_packets = 0;
2703 * ixgbe_write_eitr - write EITR register in hardware specific way
2704 * @q_vector: structure containing interrupt and ring information
2706 * This function is made to be called by ethtool and by the driver
2707 * when it needs to update EITR registers at runtime. Hardware
2708 * specific quirks/differences are taken care of here.
2710 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2712 struct ixgbe_adapter *adapter = q_vector->adapter;
2713 struct ixgbe_hw *hw = &adapter->hw;
2714 int v_idx = q_vector->v_idx;
2715 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2717 switch (adapter->hw.mac.type) {
2718 case ixgbe_mac_82598EB:
2719 /* must write high and low 16 bits to reset counter */
2720 itr_reg |= (itr_reg << 16);
2722 case ixgbe_mac_82599EB:
2723 case ixgbe_mac_X540:
2724 case ixgbe_mac_X550:
2725 case ixgbe_mac_X550EM_x:
2726 case ixgbe_mac_x550em_a:
2728 * set the WDIS bit to not clear the timer bits and cause an
2729 * immediate assertion of the interrupt
2731 itr_reg |= IXGBE_EITR_CNT_WDIS;
2736 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2739 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2743 ixgbe_update_itr(q_vector, &q_vector->tx);
2744 ixgbe_update_itr(q_vector, &q_vector->rx);
2746 /* use the smallest value of new ITR delay calculations */
2747 new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2749 /* Clear latency flag if set, shift into correct position */
2750 new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2753 if (new_itr != q_vector->itr) {
2754 /* save the algorithm value here */
2755 q_vector->itr = new_itr;
2757 ixgbe_write_eitr(q_vector);
2762 * ixgbe_check_overtemp_subtask - check for over temperature
2763 * @adapter: pointer to adapter
2765 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2767 struct ixgbe_hw *hw = &adapter->hw;
2768 u32 eicr = adapter->interrupt_event;
2770 if (test_bit(__IXGBE_DOWN, &adapter->state))
2773 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2776 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2778 switch (hw->device_id) {
2779 case IXGBE_DEV_ID_82599_T3_LOM:
2781 * Since the warning interrupt is for both ports
2782 * we don't have to check if:
2783 * - This interrupt wasn't for our port.
2784 * - We may have missed the interrupt so always have to
2785 * check if we got a LSC
2787 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2788 !(eicr & IXGBE_EICR_LSC))
2791 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2793 bool link_up = false;
2795 hw->mac.ops.check_link(hw, &speed, &link_up, false);
2801 /* Check if this is not due to overtemp */
2802 if (!hw->phy.ops.check_overtemp(hw))
2806 case IXGBE_DEV_ID_X550EM_A_1G_T:
2807 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2808 if (!hw->phy.ops.check_overtemp(hw))
2812 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2814 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2818 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2820 adapter->interrupt_event = 0;
2823 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2825 struct ixgbe_hw *hw = &adapter->hw;
2827 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2828 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2829 e_crit(probe, "Fan has stopped, replace the adapter\n");
2830 /* write to clear the interrupt */
2831 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2835 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2837 struct ixgbe_hw *hw = &adapter->hw;
2839 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2842 switch (adapter->hw.mac.type) {
2843 case ixgbe_mac_82599EB:
2845 * Need to check link state so complete overtemp check
2848 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2849 (eicr & IXGBE_EICR_LSC)) &&
2850 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2851 adapter->interrupt_event = eicr;
2852 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2853 ixgbe_service_event_schedule(adapter);
2857 case ixgbe_mac_x550em_a:
2858 if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2859 adapter->interrupt_event = eicr;
2860 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2861 ixgbe_service_event_schedule(adapter);
2862 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2863 IXGBE_EICR_GPI_SDP0_X550EM_a);
2864 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2865 IXGBE_EICR_GPI_SDP0_X550EM_a);
2868 case ixgbe_mac_X550:
2869 case ixgbe_mac_X540:
2870 if (!(eicr & IXGBE_EICR_TS))
2877 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2880 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2882 switch (hw->mac.type) {
2883 case ixgbe_mac_82598EB:
2884 if (hw->phy.type == ixgbe_phy_nl)
2887 case ixgbe_mac_82599EB:
2888 case ixgbe_mac_X550EM_x:
2889 case ixgbe_mac_x550em_a:
2890 switch (hw->mac.ops.get_media_type(hw)) {
2891 case ixgbe_media_type_fiber:
2892 case ixgbe_media_type_fiber_qsfp:
2902 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2904 struct ixgbe_hw *hw = &adapter->hw;
2905 u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2907 if (!ixgbe_is_sfp(hw))
2910 /* Later MAC's use different SDP */
2911 if (hw->mac.type >= ixgbe_mac_X540)
2912 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2914 if (eicr & eicr_mask) {
2915 /* Clear the interrupt */
2916 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2917 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2918 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2919 adapter->sfp_poll_time = 0;
2920 ixgbe_service_event_schedule(adapter);
2924 if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2925 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2926 /* Clear the interrupt */
2927 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2928 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2929 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2930 ixgbe_service_event_schedule(adapter);
2935 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2937 struct ixgbe_hw *hw = &adapter->hw;
2940 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2941 adapter->link_check_timeout = jiffies;
2942 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2943 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2944 IXGBE_WRITE_FLUSH(hw);
2945 ixgbe_service_event_schedule(adapter);
2949 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2952 struct ixgbe_hw *hw = &adapter->hw;
2955 switch (hw->mac.type) {
2956 case ixgbe_mac_82598EB:
2957 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2958 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2960 case ixgbe_mac_82599EB:
2961 case ixgbe_mac_X540:
2962 case ixgbe_mac_X550:
2963 case ixgbe_mac_X550EM_x:
2964 case ixgbe_mac_x550em_a:
2965 mask = (qmask & 0xFFFFFFFF);
2967 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2968 mask = (qmask >> 32);
2970 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2975 /* skip the flush */
2979 * ixgbe_irq_enable - Enable default interrupt generation settings
2980 * @adapter: board private structure
2981 * @queues: enable irqs for queues
2982 * @flush: flush register write
2984 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2987 struct ixgbe_hw *hw = &adapter->hw;
2988 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2990 /* don't reenable LSC while waiting for link */
2991 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2992 mask &= ~IXGBE_EIMS_LSC;
2994 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2995 switch (adapter->hw.mac.type) {
2996 case ixgbe_mac_82599EB:
2997 mask |= IXGBE_EIMS_GPI_SDP0(hw);
2999 case ixgbe_mac_X540:
3000 case ixgbe_mac_X550:
3001 case ixgbe_mac_X550EM_x:
3002 case ixgbe_mac_x550em_a:
3003 mask |= IXGBE_EIMS_TS;
3008 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3009 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3010 switch (adapter->hw.mac.type) {
3011 case ixgbe_mac_82599EB:
3012 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3013 mask |= IXGBE_EIMS_GPI_SDP2(hw);
3015 case ixgbe_mac_X540:
3016 case ixgbe_mac_X550:
3017 case ixgbe_mac_X550EM_x:
3018 case ixgbe_mac_x550em_a:
3019 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3020 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3021 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3022 mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3023 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3024 mask |= IXGBE_EICR_GPI_SDP0_X540;
3025 mask |= IXGBE_EIMS_ECC;
3026 mask |= IXGBE_EIMS_MAILBOX;
3032 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3033 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3034 mask |= IXGBE_EIMS_FLOW_DIR;
3036 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3038 ixgbe_irq_enable_queues(adapter, ~0);
3040 IXGBE_WRITE_FLUSH(&adapter->hw);
3043 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3045 struct ixgbe_adapter *adapter = data;
3046 struct ixgbe_hw *hw = &adapter->hw;
3050 * Workaround for Silicon errata. Use clear-by-write instead
3051 * of clear-by-read. Reading with EICS will return the
3052 * interrupt causes without clearing, which later be done
3053 * with the write to EICR.
3055 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3057 /* The lower 16bits of the EICR register are for the queue interrupts
3058 * which should be masked here in order to not accidentally clear them if
3059 * the bits are high when ixgbe_msix_other is called. There is a race
3060 * condition otherwise which results in possible performance loss
3061 * especially if the ixgbe_msix_other interrupt is triggering
3062 * consistently (as it would when PPS is turned on for the X540 device)
3066 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3068 if (eicr & IXGBE_EICR_LSC)
3069 ixgbe_check_lsc(adapter);
3071 if (eicr & IXGBE_EICR_MAILBOX)
3072 ixgbe_msg_task(adapter);
3074 switch (hw->mac.type) {
3075 case ixgbe_mac_82599EB:
3076 case ixgbe_mac_X540:
3077 case ixgbe_mac_X550:
3078 case ixgbe_mac_X550EM_x:
3079 case ixgbe_mac_x550em_a:
3080 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3081 (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3082 adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3083 ixgbe_service_event_schedule(adapter);
3084 IXGBE_WRITE_REG(hw, IXGBE_EICR,
3085 IXGBE_EICR_GPI_SDP0_X540);
3087 if (eicr & IXGBE_EICR_ECC) {
3088 e_info(link, "Received ECC Err, initiating reset\n");
3089 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3090 ixgbe_service_event_schedule(adapter);
3091 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3093 /* Handle Flow Director Full threshold interrupt */
3094 if (eicr & IXGBE_EICR_FLOW_DIR) {
3095 int reinit_count = 0;
3097 for (i = 0; i < adapter->num_tx_queues; i++) {
3098 struct ixgbe_ring *ring = adapter->tx_ring[i];
3099 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3104 /* no more flow director interrupts until after init */
3105 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3106 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3107 ixgbe_service_event_schedule(adapter);
3110 ixgbe_check_sfp_event(adapter, eicr);
3111 ixgbe_check_overtemp_event(adapter, eicr);
3117 ixgbe_check_fan_failure(adapter, eicr);
3119 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3120 ixgbe_ptp_check_pps_event(adapter);
3122 /* re-enable the original interrupt state, no lsc, no queues */
3123 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3124 ixgbe_irq_enable(adapter, false, false);
3129 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3131 struct ixgbe_q_vector *q_vector = data;
3133 /* EIAM disabled interrupts (on this vector) for us */
3135 if (q_vector->rx.ring || q_vector->tx.ring)
3136 napi_schedule_irqoff(&q_vector->napi);
3142 * ixgbe_poll - NAPI Rx polling callback
3143 * @napi: structure for representing this polling device
3144 * @budget: how many packets driver is allowed to clean
3146 * This function is used for legacy and MSI, NAPI mode
3148 int ixgbe_poll(struct napi_struct *napi, int budget)
3150 struct ixgbe_q_vector *q_vector =
3151 container_of(napi, struct ixgbe_q_vector, napi);
3152 struct ixgbe_adapter *adapter = q_vector->adapter;
3153 struct ixgbe_ring *ring;
3154 int per_ring_budget, work_done = 0;
3155 bool clean_complete = true;
3157 #ifdef CONFIG_IXGBE_DCA
3158 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3159 ixgbe_update_dca(q_vector);
3162 ixgbe_for_each_ring(ring, q_vector->tx) {
3163 bool wd = ring->xsk_pool ?
3164 ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3165 ixgbe_clean_tx_irq(q_vector, ring, budget);
3168 clean_complete = false;
3171 /* Exit if we are called by netpoll */
3175 /* attempt to distribute budget to each queue fairly, but don't allow
3176 * the budget to go below 1 because we'll exit polling */
3177 if (q_vector->rx.count > 1)
3178 per_ring_budget = max(budget/q_vector->rx.count, 1);
3180 per_ring_budget = budget;
3182 ixgbe_for_each_ring(ring, q_vector->rx) {
3183 int cleaned = ring->xsk_pool ?
3184 ixgbe_clean_rx_irq_zc(q_vector, ring,
3186 ixgbe_clean_rx_irq(q_vector, ring,
3189 work_done += cleaned;
3190 if (cleaned >= per_ring_budget)
3191 clean_complete = false;
3194 /* If all work not completed, return budget and keep polling */
3195 if (!clean_complete)
3198 /* all work done, exit the polling mode */
3199 if (likely(napi_complete_done(napi, work_done))) {
3200 if (adapter->rx_itr_setting & 1)
3201 ixgbe_set_itr(q_vector);
3202 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3203 ixgbe_irq_enable_queues(adapter,
3204 BIT_ULL(q_vector->v_idx));
3207 return min(work_done, budget - 1);
3211 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3212 * @adapter: board private structure
3214 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3215 * interrupts from the kernel.
3217 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3219 struct net_device *netdev = adapter->netdev;
3220 unsigned int ri = 0, ti = 0;
3223 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3224 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3225 struct msix_entry *entry = &adapter->msix_entries[vector];
3227 if (q_vector->tx.ring && q_vector->rx.ring) {
3228 snprintf(q_vector->name, sizeof(q_vector->name),
3229 "%s-TxRx-%u", netdev->name, ri++);
3231 } else if (q_vector->rx.ring) {
3232 snprintf(q_vector->name, sizeof(q_vector->name),
3233 "%s-rx-%u", netdev->name, ri++);
3234 } else if (q_vector->tx.ring) {
3235 snprintf(q_vector->name, sizeof(q_vector->name),
3236 "%s-tx-%u", netdev->name, ti++);
3238 /* skip this unused q_vector */
3241 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3242 q_vector->name, q_vector);
3244 e_err(probe, "request_irq failed for MSIX interrupt "
3245 "Error: %d\n", err);
3246 goto free_queue_irqs;
3248 /* If Flow Director is enabled, set interrupt affinity */
3249 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3250 /* assign the mask for this irq */
3251 irq_set_affinity_hint(entry->vector,
3252 &q_vector->affinity_mask);
3256 err = request_irq(adapter->msix_entries[vector].vector,
3257 ixgbe_msix_other, 0, netdev->name, adapter);
3259 e_err(probe, "request_irq for msix_other failed: %d\n", err);
3260 goto free_queue_irqs;
3268 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3270 free_irq(adapter->msix_entries[vector].vector,
3271 adapter->q_vector[vector]);
3273 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3274 pci_disable_msix(adapter->pdev);
3275 kfree(adapter->msix_entries);
3276 adapter->msix_entries = NULL;
3281 * ixgbe_intr - legacy mode Interrupt Handler
3282 * @irq: interrupt number
3283 * @data: pointer to a network interface device structure
3285 static irqreturn_t ixgbe_intr(int irq, void *data)
3287 struct ixgbe_adapter *adapter = data;
3288 struct ixgbe_hw *hw = &adapter->hw;
3289 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3293 * Workaround for silicon errata #26 on 82598. Mask the interrupt
3294 * before the read of EICR.
3296 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3298 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3299 * therefore no explicit interrupt disable is necessary */
3300 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3303 * shared interrupt alert!
3304 * make sure interrupts are enabled because the read will
3305 * have disabled interrupts due to EIAM
3306 * finish the workaround of silicon errata on 82598. Unmask
3307 * the interrupt that we masked before the EICR read.
3309 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3310 ixgbe_irq_enable(adapter, true, true);
3311 return IRQ_NONE; /* Not our interrupt */
3314 if (eicr & IXGBE_EICR_LSC)
3315 ixgbe_check_lsc(adapter);
3317 switch (hw->mac.type) {
3318 case ixgbe_mac_82599EB:
3319 ixgbe_check_sfp_event(adapter, eicr);
3321 case ixgbe_mac_X540:
3322 case ixgbe_mac_X550:
3323 case ixgbe_mac_X550EM_x:
3324 case ixgbe_mac_x550em_a:
3325 if (eicr & IXGBE_EICR_ECC) {
3326 e_info(link, "Received ECC Err, initiating reset\n");
3327 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3328 ixgbe_service_event_schedule(adapter);
3329 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3331 ixgbe_check_overtemp_event(adapter, eicr);
3337 ixgbe_check_fan_failure(adapter, eicr);
3338 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3339 ixgbe_ptp_check_pps_event(adapter);
3341 /* would disable interrupts here but EIAM disabled it */
3342 napi_schedule_irqoff(&q_vector->napi);
3345 * re-enable link(maybe) and non-queue interrupts, no flush.
3346 * ixgbe_poll will re-enable the queue interrupts
3348 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3349 ixgbe_irq_enable(adapter, false, false);
3355 * ixgbe_request_irq - initialize interrupts
3356 * @adapter: board private structure
3358 * Attempts to configure interrupts using the best available
3359 * capabilities of the hardware and kernel.
3361 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3363 struct net_device *netdev = adapter->netdev;
3366 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3367 err = ixgbe_request_msix_irqs(adapter);
3368 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3369 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3370 netdev->name, adapter);
3372 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3373 netdev->name, adapter);
3376 e_err(probe, "request_irq failed, Error %d\n", err);
3381 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3385 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3386 free_irq(adapter->pdev->irq, adapter);
3390 if (!adapter->msix_entries)
3393 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3394 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3395 struct msix_entry *entry = &adapter->msix_entries[vector];
3397 /* free only the irqs that were actually requested */
3398 if (!q_vector->rx.ring && !q_vector->tx.ring)
3401 /* clear the affinity_mask in the IRQ descriptor */
3402 irq_set_affinity_hint(entry->vector, NULL);
3404 free_irq(entry->vector, q_vector);
3407 free_irq(adapter->msix_entries[vector].vector, adapter);
3411 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3412 * @adapter: board private structure
3414 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3416 switch (adapter->hw.mac.type) {
3417 case ixgbe_mac_82598EB:
3418 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3420 case ixgbe_mac_82599EB:
3421 case ixgbe_mac_X540:
3422 case ixgbe_mac_X550:
3423 case ixgbe_mac_X550EM_x:
3424 case ixgbe_mac_x550em_a:
3425 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3426 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3427 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3432 IXGBE_WRITE_FLUSH(&adapter->hw);
3433 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3436 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3437 synchronize_irq(adapter->msix_entries[vector].vector);
3439 synchronize_irq(adapter->msix_entries[vector++].vector);
3441 synchronize_irq(adapter->pdev->irq);
3446 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3447 * @adapter: board private structure
3450 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3452 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3454 ixgbe_write_eitr(q_vector);
3456 ixgbe_set_ivar(adapter, 0, 0, 0);
3457 ixgbe_set_ivar(adapter, 1, 0, 0);
3459 e_info(hw, "Legacy interrupt IVAR setup done\n");
3463 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3464 * @adapter: board private structure
3465 * @ring: structure containing ring specific data
3467 * Configure the Tx descriptor ring after a reset.
3469 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3470 struct ixgbe_ring *ring)
3472 struct ixgbe_hw *hw = &adapter->hw;
3473 u64 tdba = ring->dma;
3475 u32 txdctl = IXGBE_TXDCTL_ENABLE;
3476 u8 reg_idx = ring->reg_idx;
3478 ring->xsk_pool = NULL;
3479 if (ring_is_xdp(ring))
3480 ring->xsk_pool = ixgbe_xsk_pool(adapter, ring);
3482 /* disable queue to avoid issues while updating state */
3483 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3484 IXGBE_WRITE_FLUSH(hw);
3486 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3487 (tdba & DMA_BIT_MASK(32)));
3488 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3489 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3490 ring->count * sizeof(union ixgbe_adv_tx_desc));
3491 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3492 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3493 ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3496 * set WTHRESH to encourage burst writeback, it should not be set
3497 * higher than 1 when:
3498 * - ITR is 0 as it could cause false TX hangs
3499 * - ITR is set to > 100k int/sec and BQL is enabled
3501 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3502 * to or less than the number of on chip descriptors, which is
3505 if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3506 txdctl |= 1u << 16; /* WTHRESH = 1 */
3508 txdctl |= 8u << 16; /* WTHRESH = 8 */
3511 * Setting PTHRESH to 32 both improves performance
3512 * and avoids a TX hang with DFP enabled
3514 txdctl |= (1u << 8) | /* HTHRESH = 1 */
3515 32; /* PTHRESH = 32 */
3517 /* reinitialize flowdirector state */
3518 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3519 ring->atr_sample_rate = adapter->atr_sample_rate;
3520 ring->atr_count = 0;
3521 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3523 ring->atr_sample_rate = 0;
3526 /* initialize XPS */
3527 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3528 struct ixgbe_q_vector *q_vector = ring->q_vector;
3531 netif_set_xps_queue(ring->netdev,
3532 &q_vector->affinity_mask,
3536 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3538 /* reinitialize tx_buffer_info */
3539 memset(ring->tx_buffer_info, 0,
3540 sizeof(struct ixgbe_tx_buffer) * ring->count);
3543 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3545 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3546 if (hw->mac.type == ixgbe_mac_82598EB &&
3547 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3550 /* poll to verify queue is enabled */
3552 usleep_range(1000, 2000);
3553 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3554 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3556 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3559 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3561 struct ixgbe_hw *hw = &adapter->hw;
3563 u8 tcs = adapter->hw_tcs;
3565 if (hw->mac.type == ixgbe_mac_82598EB)
3568 /* disable the arbiter while setting MTQC */
3569 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3570 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3571 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3573 /* set transmit pool layout */
3574 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3575 mtqc = IXGBE_MTQC_VT_ENA;
3577 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3579 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3580 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3581 IXGBE_82599_VMDQ_4Q_MASK)
3582 mtqc |= IXGBE_MTQC_32VF;
3584 mtqc |= IXGBE_MTQC_64VF;
3587 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3588 } else if (tcs > 1) {
3589 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3591 u8 max_txq = adapter->num_tx_queues +
3592 adapter->num_xdp_queues;
3594 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3596 mtqc = IXGBE_MTQC_64Q_1PB;
3600 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3602 /* Enable Security TX Buffer IFG for multiple pb */
3604 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3605 sectx |= IXGBE_SECTX_DCB;
3606 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3609 /* re-enable the arbiter */
3610 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3611 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3615 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3616 * @adapter: board private structure
3618 * Configure the Tx unit of the MAC after a reset.
3620 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3622 struct ixgbe_hw *hw = &adapter->hw;
3626 ixgbe_setup_mtqc(adapter);
3628 if (hw->mac.type != ixgbe_mac_82598EB) {
3629 /* DMATXCTL.EN must be before Tx queues are enabled */
3630 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3631 dmatxctl |= IXGBE_DMATXCTL_TE;
3632 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3635 /* Setup the HW Tx Head and Tail descriptor pointers */
3636 for (i = 0; i < adapter->num_tx_queues; i++)
3637 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3638 for (i = 0; i < adapter->num_xdp_queues; i++)
3639 ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3642 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3643 struct ixgbe_ring *ring)
3645 struct ixgbe_hw *hw = &adapter->hw;
3646 u8 reg_idx = ring->reg_idx;
3647 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3649 srrctl |= IXGBE_SRRCTL_DROP_EN;
3651 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3654 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3655 struct ixgbe_ring *ring)
3657 struct ixgbe_hw *hw = &adapter->hw;
3658 u8 reg_idx = ring->reg_idx;
3659 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3661 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3663 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3666 #ifdef CONFIG_IXGBE_DCB
3667 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3669 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3673 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3675 if (adapter->ixgbe_ieee_pfc)
3676 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3679 * We should set the drop enable bit if:
3682 * Number of Rx queues > 1 and flow control is disabled
3684 * This allows us to avoid head of line blocking for security
3685 * and performance reasons.
3687 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3688 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3689 for (i = 0; i < adapter->num_rx_queues; i++)
3690 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3692 for (i = 0; i < adapter->num_rx_queues; i++)
3693 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3697 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3699 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3700 struct ixgbe_ring *rx_ring)
3702 struct ixgbe_hw *hw = &adapter->hw;
3704 u8 reg_idx = rx_ring->reg_idx;
3706 if (hw->mac.type == ixgbe_mac_82598EB) {
3707 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3710 * if VMDq is not active we must program one srrctl register
3711 * per RSS queue since we have enabled RDRXCTL.MVMEN
3716 /* configure header buffer length, needed for RSC */
3717 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3719 /* configure the packet buffer length */
3720 if (rx_ring->xsk_pool) {
3721 u32 xsk_buf_len = xsk_pool_get_rx_frame_size(rx_ring->xsk_pool);
3723 /* If the MAC support setting RXDCTL.RLPML, the
3724 * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3725 * RXDCTL.RLPML is set to the actual UMEM buffer
3726 * size. If not, then we are stuck with a 1k buffer
3727 * size resolution. In this case frames larger than
3728 * the UMEM buffer size viewed in a 1k resolution will
3731 if (hw->mac.type != ixgbe_mac_82599EB)
3732 srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3734 srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3735 } else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3736 srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3738 srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3741 /* configure descriptor type */
3742 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3744 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3748 * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3749 * @adapter: device handle
3751 * - 82598/82599/X540: 128
3752 * - X550(non-SRIOV mode): 512
3753 * - X550(SRIOV mode): 64
3755 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3757 if (adapter->hw.mac.type < ixgbe_mac_X550)
3759 else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3766 * ixgbe_store_key - Write the RSS key to HW
3767 * @adapter: device handle
3769 * Write the RSS key stored in adapter.rss_key to HW.
3771 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3773 struct ixgbe_hw *hw = &adapter->hw;
3776 for (i = 0; i < 10; i++)
3777 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3781 * ixgbe_init_rss_key - Initialize adapter RSS key
3782 * @adapter: device handle
3784 * Allocates and initializes the RSS key if it is not allocated.
3786 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3790 if (!adapter->rss_key) {
3791 rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3792 if (unlikely(!rss_key))
3795 netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3796 adapter->rss_key = rss_key;
3803 * ixgbe_store_reta - Write the RETA table to HW
3804 * @adapter: device handle
3806 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3808 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3810 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3811 struct ixgbe_hw *hw = &adapter->hw;
3814 u8 *indir_tbl = adapter->rss_indir_tbl;
3816 /* Fill out the redirection table as follows:
3817 * - 82598: 8 bit wide entries containing pair of 4 bit RSS
3819 * - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3820 * - X550: 8 bit wide entries containing 6 bit RSS index
3822 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3823 indices_multi = 0x11;
3825 indices_multi = 0x1;
3827 /* Write redirection table to HW */
3828 for (i = 0; i < reta_entries; i++) {
3829 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3832 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3834 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3842 * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3843 * @adapter: device handle
3845 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3847 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3849 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3850 struct ixgbe_hw *hw = &adapter->hw;
3853 /* Write redirection table to HW */
3854 for (i = 0; i < reta_entries; i++) {
3855 u16 pool = adapter->num_rx_pools;
3857 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3863 IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3869 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3872 u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3873 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3875 /* Program table for at least 4 queues w/ SR-IOV so that VFs can
3876 * make full use of any rings they may have. We will use the
3877 * PSRTYPE register to control how many rings we use within the PF.
3879 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3882 /* Fill out hash function seeds */
3883 ixgbe_store_key(adapter);
3885 /* Fill out redirection table */
3886 memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3888 for (i = 0, j = 0; i < reta_entries; i++, j++) {
3892 adapter->rss_indir_tbl[i] = j;
3895 ixgbe_store_reta(adapter);
3898 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3900 struct ixgbe_hw *hw = &adapter->hw;
3901 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3904 /* Fill out hash function seeds */
3905 for (i = 0; i < 10; i++) {
3906 u16 pool = adapter->num_rx_pools;
3910 IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3911 *(adapter->rss_key + i));
3914 /* Fill out the redirection table */
3915 for (i = 0, j = 0; i < 64; i++, j++) {
3919 adapter->rss_indir_tbl[i] = j;
3922 ixgbe_store_vfreta(adapter);
3925 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3927 struct ixgbe_hw *hw = &adapter->hw;
3928 u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3931 /* Disable indicating checksum in descriptor, enables RSS hash */
3932 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3933 rxcsum |= IXGBE_RXCSUM_PCSD;
3934 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3936 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3937 if (adapter->ring_feature[RING_F_RSS].mask)
3938 mrqc = IXGBE_MRQC_RSSEN;
3940 u8 tcs = adapter->hw_tcs;
3942 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3944 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3946 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3947 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3948 IXGBE_82599_VMDQ_4Q_MASK)
3949 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3951 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3953 /* Enable L3/L4 for Tx Switched packets only for X550,
3954 * older devices do not support this feature
3956 if (hw->mac.type >= ixgbe_mac_X550)
3957 mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3960 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3962 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3964 mrqc = IXGBE_MRQC_RSSEN;
3968 /* Perform hash on these packet types */
3969 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3970 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3971 IXGBE_MRQC_RSS_FIELD_IPV6 |
3972 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3974 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3975 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3976 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3977 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3979 if ((hw->mac.type >= ixgbe_mac_X550) &&
3980 (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3981 u16 pool = adapter->num_rx_pools;
3983 /* Enable VF RSS mode */
3984 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3985 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3987 /* Setup RSS through the VF registers */
3988 ixgbe_setup_vfreta(adapter);
3989 vfmrqc = IXGBE_MRQC_RSSEN;
3990 vfmrqc |= rss_field;
3994 IXGBE_PFVFMRQC(VMDQ_P(pool)),
3997 ixgbe_setup_reta(adapter);
3999 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4004 * ixgbe_configure_rscctl - enable RSC for the indicated ring
4005 * @adapter: address of board private structure
4006 * @ring: structure containing ring specific data
4008 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4009 struct ixgbe_ring *ring)
4011 struct ixgbe_hw *hw = &adapter->hw;
4013 u8 reg_idx = ring->reg_idx;
4015 if (!ring_is_rsc_enabled(ring))
4018 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4019 rscctrl |= IXGBE_RSCCTL_RSCEN;
4021 * we must limit the number of descriptors so that the
4022 * total size of max desc * buf_len is not greater
4025 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4026 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4029 #define IXGBE_MAX_RX_DESC_POLL 10
4030 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4031 struct ixgbe_ring *ring)
4033 struct ixgbe_hw *hw = &adapter->hw;
4034 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4036 u8 reg_idx = ring->reg_idx;
4038 if (ixgbe_removed(hw->hw_addr))
4040 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4041 if (hw->mac.type == ixgbe_mac_82598EB &&
4042 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4046 usleep_range(1000, 2000);
4047 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4048 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4051 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4052 "the polling period\n", reg_idx);
4056 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4057 struct ixgbe_ring *ring)
4059 struct ixgbe_hw *hw = &adapter->hw;
4060 union ixgbe_adv_rx_desc *rx_desc;
4061 u64 rdba = ring->dma;
4063 u8 reg_idx = ring->reg_idx;
4065 xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4066 ring->xsk_pool = ixgbe_xsk_pool(adapter, ring);
4067 if (ring->xsk_pool) {
4068 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4069 MEM_TYPE_XSK_BUFF_POOL,
4071 xsk_pool_set_rxq_info(ring->xsk_pool, &ring->xdp_rxq);
4073 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4074 MEM_TYPE_PAGE_SHARED, NULL));
4077 /* disable queue to avoid use of these values while updating state */
4078 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4079 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4081 /* write value back with RXDCTL.ENABLE bit cleared */
4082 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4083 IXGBE_WRITE_FLUSH(hw);
4085 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4086 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4087 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4088 ring->count * sizeof(union ixgbe_adv_rx_desc));
4089 /* Force flushing of IXGBE_RDLEN to prevent MDD */
4090 IXGBE_WRITE_FLUSH(hw);
4092 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4093 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4094 ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4096 ixgbe_configure_srrctl(adapter, ring);
4097 ixgbe_configure_rscctl(adapter, ring);
4099 if (hw->mac.type == ixgbe_mac_82598EB) {
4101 * enable cache line friendly hardware writes:
4102 * PTHRESH=32 descriptors (half the internal cache),
4103 * this also removes ugly rx_no_buffer_count increment
4104 * HTHRESH=4 descriptors (to minimize latency on fetch)
4105 * WTHRESH=8 burst writeback up to two cache lines
4107 rxdctl &= ~0x3FFFFF;
4109 #if (PAGE_SIZE < 8192)
4110 /* RXDCTL.RLPML does not work on 82599 */
4111 } else if (hw->mac.type != ixgbe_mac_82599EB) {
4112 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4113 IXGBE_RXDCTL_RLPML_EN);
4115 /* Limit the maximum frame size so we don't overrun the skb.
4116 * This can happen in SRIOV mode when the MTU of the VF is
4117 * higher than the MTU of the PF.
4119 if (ring_uses_build_skb(ring) &&
4120 !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4121 rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4122 IXGBE_RXDCTL_RLPML_EN;
4126 if (ring->xsk_pool && hw->mac.type != ixgbe_mac_82599EB) {
4127 u32 xsk_buf_len = xsk_pool_get_rx_frame_size(ring->xsk_pool);
4129 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4130 IXGBE_RXDCTL_RLPML_EN);
4131 rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4133 ring->rx_buf_len = xsk_buf_len;
4136 /* initialize rx_buffer_info */
4137 memset(ring->rx_buffer_info, 0,
4138 sizeof(struct ixgbe_rx_buffer) * ring->count);
4140 /* initialize Rx descriptor 0 */
4141 rx_desc = IXGBE_RX_DESC(ring, 0);
4142 rx_desc->wb.upper.length = 0;
4144 /* enable receive descriptor ring */
4145 rxdctl |= IXGBE_RXDCTL_ENABLE;
4146 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4148 ixgbe_rx_desc_queue_enable(adapter, ring);
4150 ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4152 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4155 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4157 struct ixgbe_hw *hw = &adapter->hw;
4158 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4159 u16 pool = adapter->num_rx_pools;
4161 /* PSRTYPE must be initialized in non 82598 adapters */
4162 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4163 IXGBE_PSRTYPE_UDPHDR |
4164 IXGBE_PSRTYPE_IPV4HDR |
4165 IXGBE_PSRTYPE_L2HDR |
4166 IXGBE_PSRTYPE_IPV6HDR;
4168 if (hw->mac.type == ixgbe_mac_82598EB)
4172 psrtype |= 2u << 29;
4174 psrtype |= 1u << 29;
4177 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4180 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4182 struct ixgbe_hw *hw = &adapter->hw;
4183 u16 pool = adapter->num_rx_pools;
4184 u32 reg_offset, vf_shift, vmolr;
4185 u32 gcr_ext, vmdctl;
4188 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4191 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4192 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4193 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4194 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4195 vmdctl |= IXGBE_VT_CTL_REPLEN;
4196 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4198 /* accept untagged packets until a vlan tag is
4199 * specifically set for the VMDQ queue/pool
4201 vmolr = IXGBE_VMOLR_AUPE;
4203 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4205 vf_shift = VMDQ_P(0) % 32;
4206 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4208 /* Enable only the PF's pool for Tx/Rx */
4209 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4210 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4211 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4212 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4213 if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4214 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4216 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4217 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4219 /* clear VLAN promisc flag so VFTA will be updated if necessary */
4220 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4223 * Set up VF register offsets for selected VT Mode,
4224 * i.e. 32 or 64 VFs for SR-IOV
4226 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4227 case IXGBE_82599_VMDQ_8Q_MASK:
4228 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4230 case IXGBE_82599_VMDQ_4Q_MASK:
4231 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4234 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4238 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4240 for (i = 0; i < adapter->num_vfs; i++) {
4241 /* configure spoof checking */
4242 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4243 adapter->vfinfo[i].spoofchk_enabled);
4245 /* Enable/Disable RSS query feature */
4246 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4247 adapter->vfinfo[i].rss_query_enabled);
4251 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4253 struct ixgbe_hw *hw = &adapter->hw;
4254 struct net_device *netdev = adapter->netdev;
4255 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4256 struct ixgbe_ring *rx_ring;
4261 /* adjust max frame to be able to do baby jumbo for FCoE */
4262 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4263 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4264 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4266 #endif /* IXGBE_FCOE */
4268 /* adjust max frame to be at least the size of a standard frame */
4269 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4270 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4272 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4273 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4274 mhadd &= ~IXGBE_MHADD_MFS_MASK;
4275 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4277 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4280 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4281 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4282 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4283 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4286 * Setup the HW Rx Head and Tail Descriptor Pointers and
4287 * the Base and Length of the Rx Descriptor Ring
4289 for (i = 0; i < adapter->num_rx_queues; i++) {
4290 rx_ring = adapter->rx_ring[i];
4292 clear_ring_rsc_enabled(rx_ring);
4293 clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4294 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4296 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4297 set_ring_rsc_enabled(rx_ring);
4299 if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4300 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4302 if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4305 set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4307 #if (PAGE_SIZE < 8192)
4308 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4309 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4311 if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4312 (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4313 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4318 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4320 struct ixgbe_hw *hw = &adapter->hw;
4321 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4323 switch (hw->mac.type) {
4324 case ixgbe_mac_82598EB:
4326 * For VMDq support of different descriptor types or
4327 * buffer sizes through the use of multiple SRRCTL
4328 * registers, RDRXCTL.MVMEN must be set to 1
4330 * also, the manual doesn't mention it clearly but DCA hints
4331 * will only use queue 0's tags unless this bit is set. Side
4332 * effects of setting this bit are only that SRRCTL must be
4333 * fully programmed [0..15]
4335 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4337 case ixgbe_mac_X550:
4338 case ixgbe_mac_X550EM_x:
4339 case ixgbe_mac_x550em_a:
4340 if (adapter->num_vfs)
4341 rdrxctl |= IXGBE_RDRXCTL_PSP;
4343 case ixgbe_mac_82599EB:
4344 case ixgbe_mac_X540:
4345 /* Disable RSC for ACK packets */
4346 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4347 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4348 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4349 /* hardware requires some bits to be set by default */
4350 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4351 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4354 /* We should do nothing since we don't know this hardware */
4358 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4362 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4363 * @adapter: board private structure
4365 * Configure the Rx unit of the MAC after a reset.
4367 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4369 struct ixgbe_hw *hw = &adapter->hw;
4373 /* disable receives while setting up the descriptors */
4374 hw->mac.ops.disable_rx(hw);
4376 ixgbe_setup_psrtype(adapter);
4377 ixgbe_setup_rdrxctl(adapter);
4380 rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4381 rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4382 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4383 rfctl |= IXGBE_RFCTL_RSC_DIS;
4385 /* disable NFS filtering */
4386 rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4387 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4389 /* Program registers for the distribution of queues */
4390 ixgbe_setup_mrqc(adapter);
4392 /* set_rx_buffer_len must be called before ring initialization */
4393 ixgbe_set_rx_buffer_len(adapter);
4396 * Setup the HW Rx Head and Tail Descriptor Pointers and
4397 * the Base and Length of the Rx Descriptor Ring
4399 for (i = 0; i < adapter->num_rx_queues; i++)
4400 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4402 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4403 /* disable drop enable for 82598 parts */
4404 if (hw->mac.type == ixgbe_mac_82598EB)
4405 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4407 /* enable all receives */
4408 rxctrl |= IXGBE_RXCTRL_RXEN;
4409 hw->mac.ops.enable_rx_dma(hw, rxctrl);
4412 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4413 __be16 proto, u16 vid)
4415 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4416 struct ixgbe_hw *hw = &adapter->hw;
4418 /* add VID to filter table */
4419 if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4420 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4422 set_bit(vid, adapter->active_vlans);
4427 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4432 /* short cut the special case */
4436 /* Search for the vlan id in the VLVF entries */
4437 for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4438 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4439 if ((vlvf & VLAN_VID_MASK) == vlan)
4446 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4448 struct ixgbe_hw *hw = &adapter->hw;
4452 idx = ixgbe_find_vlvf_entry(hw, vid);
4456 /* See if any other pools are set for this VLAN filter
4457 * entry other than the PF.
4459 word = idx * 2 + (VMDQ_P(0) / 32);
4460 bits = ~BIT(VMDQ_P(0) % 32);
4461 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4463 /* Disable the filter so this falls into the default pool. */
4464 if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4465 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4466 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4467 IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4471 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4472 __be16 proto, u16 vid)
4474 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4475 struct ixgbe_hw *hw = &adapter->hw;
4477 /* remove VID from filter table */
4478 if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4479 hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4481 clear_bit(vid, adapter->active_vlans);
4487 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4488 * @adapter: driver data
4490 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4492 struct ixgbe_hw *hw = &adapter->hw;
4496 switch (hw->mac.type) {
4497 case ixgbe_mac_82598EB:
4498 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4499 vlnctrl &= ~IXGBE_VLNCTRL_VME;
4500 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4502 case ixgbe_mac_82599EB:
4503 case ixgbe_mac_X540:
4504 case ixgbe_mac_X550:
4505 case ixgbe_mac_X550EM_x:
4506 case ixgbe_mac_x550em_a:
4507 for (i = 0; i < adapter->num_rx_queues; i++) {
4508 struct ixgbe_ring *ring = adapter->rx_ring[i];
4510 if (!netif_is_ixgbe(ring->netdev))
4514 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4515 vlnctrl &= ~IXGBE_RXDCTL_VME;
4516 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4525 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4526 * @adapter: driver data
4528 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4530 struct ixgbe_hw *hw = &adapter->hw;
4534 switch (hw->mac.type) {
4535 case ixgbe_mac_82598EB:
4536 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4537 vlnctrl |= IXGBE_VLNCTRL_VME;
4538 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4540 case ixgbe_mac_82599EB:
4541 case ixgbe_mac_X540:
4542 case ixgbe_mac_X550:
4543 case ixgbe_mac_X550EM_x:
4544 case ixgbe_mac_x550em_a:
4545 for (i = 0; i < adapter->num_rx_queues; i++) {
4546 struct ixgbe_ring *ring = adapter->rx_ring[i];
4548 if (!netif_is_ixgbe(ring->netdev))
4552 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4553 vlnctrl |= IXGBE_RXDCTL_VME;
4554 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4562 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4564 struct ixgbe_hw *hw = &adapter->hw;
4567 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4569 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4570 /* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4571 vlnctrl |= IXGBE_VLNCTRL_VFE;
4572 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4574 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4575 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4579 /* Nothing to do for 82598 */
4580 if (hw->mac.type == ixgbe_mac_82598EB)
4583 /* We are already in VLAN promisc, nothing to do */
4584 if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4587 /* Set flag so we don't redo unnecessary work */
4588 adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4590 /* Add PF to all active pools */
4591 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4592 u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4593 u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4595 vlvfb |= BIT(VMDQ_P(0) % 32);
4596 IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4599 /* Set all bits in the VLAN filter table array */
4600 for (i = hw->mac.vft_size; i--;)
4601 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4604 #define VFTA_BLOCK_SIZE 8
4605 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4607 struct ixgbe_hw *hw = &adapter->hw;
4608 u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4609 u32 vid_start = vfta_offset * 32;
4610 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4611 u32 i, vid, word, bits;
4613 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4614 u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4616 /* pull VLAN ID from VLVF */
4617 vid = vlvf & VLAN_VID_MASK;
4619 /* only concern outselves with a certain range */
4620 if (vid < vid_start || vid >= vid_end)
4624 /* record VLAN ID in VFTA */
4625 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4627 /* if PF is part of this then continue */
4628 if (test_bit(vid, adapter->active_vlans))
4632 /* remove PF from the pool */
4633 word = i * 2 + VMDQ_P(0) / 32;
4634 bits = ~BIT(VMDQ_P(0) % 32);
4635 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4636 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4639 /* extract values from active_vlans and write back to VFTA */
4640 for (i = VFTA_BLOCK_SIZE; i--;) {
4641 vid = (vfta_offset + i) * 32;
4642 word = vid / BITS_PER_LONG;
4643 bits = vid % BITS_PER_LONG;
4645 vfta[i] |= adapter->active_vlans[word] >> bits;
4647 IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4651 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4653 struct ixgbe_hw *hw = &adapter->hw;
4656 /* Set VLAN filtering to enabled */
4657 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4658 vlnctrl |= IXGBE_VLNCTRL_VFE;
4659 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4661 if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4662 hw->mac.type == ixgbe_mac_82598EB)
4665 /* We are not in VLAN promisc, nothing to do */
4666 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4669 /* Set flag so we don't redo unnecessary work */
4670 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4672 for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4673 ixgbe_scrub_vfta(adapter, i);
4676 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4680 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4682 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4683 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4687 * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4688 * @netdev: network interface device structure
4690 * Writes multicast address list to the MTA hash table.
4691 * Returns: -ENOMEM on failure
4692 * 0 on no addresses written
4693 * X on writing X addresses to MTA
4695 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4697 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4698 struct ixgbe_hw *hw = &adapter->hw;
4700 if (!netif_running(netdev))
4703 if (hw->mac.ops.update_mc_addr_list)
4704 hw->mac.ops.update_mc_addr_list(hw, netdev);
4708 #ifdef CONFIG_PCI_IOV
4709 ixgbe_restore_vf_multicasts(adapter);
4712 return netdev_mc_count(netdev);
4715 #ifdef CONFIG_PCI_IOV
4716 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4718 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4719 struct ixgbe_hw *hw = &adapter->hw;
4722 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4723 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4725 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4726 hw->mac.ops.set_rar(hw, i,
4731 hw->mac.ops.clear_rar(hw, i);
4736 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4738 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4739 struct ixgbe_hw *hw = &adapter->hw;
4742 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4743 if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4746 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4748 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4749 hw->mac.ops.set_rar(hw, i,
4754 hw->mac.ops.clear_rar(hw, i);
4758 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4760 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4761 struct ixgbe_hw *hw = &adapter->hw;
4764 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4765 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4766 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4769 ixgbe_sync_mac_table(adapter);
4772 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4774 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4775 struct ixgbe_hw *hw = &adapter->hw;
4778 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4779 /* do not count default RAR as available */
4780 if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4783 /* only count unused and addresses that belong to us */
4784 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4785 if (mac_table->pool != pool)
4795 /* this function destroys the first RAR entry */
4796 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4798 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4799 struct ixgbe_hw *hw = &adapter->hw;
4801 memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4802 mac_table->pool = VMDQ_P(0);
4804 mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4806 hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4810 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4811 const u8 *addr, u16 pool)
4813 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4814 struct ixgbe_hw *hw = &adapter->hw;
4817 if (is_zero_ether_addr(addr))
4820 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4821 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4824 ether_addr_copy(mac_table->addr, addr);
4825 mac_table->pool = pool;
4827 mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4828 IXGBE_MAC_STATE_IN_USE;
4830 ixgbe_sync_mac_table(adapter);
4838 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4839 const u8 *addr, u16 pool)
4841 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4842 struct ixgbe_hw *hw = &adapter->hw;
4845 if (is_zero_ether_addr(addr))
4848 /* search table for addr, if found clear IN_USE flag and sync */
4849 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4850 /* we can only delete an entry if it is in use */
4851 if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4853 /* we only care about entries that belong to the given pool */
4854 if (mac_table->pool != pool)
4856 /* we only care about a specific MAC address */
4857 if (!ether_addr_equal(addr, mac_table->addr))
4860 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4861 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4863 ixgbe_sync_mac_table(adapter);
4871 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4873 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4876 ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4878 return min_t(int, ret, 0);
4881 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4883 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4885 ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4891 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4892 * @netdev: network interface device structure
4894 * The set_rx_method entry point is called whenever the unicast/multicast
4895 * address list or the network interface flags are updated. This routine is
4896 * responsible for configuring the hardware for proper unicast, multicast and
4899 void ixgbe_set_rx_mode(struct net_device *netdev)
4901 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4902 struct ixgbe_hw *hw = &adapter->hw;
4903 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4904 netdev_features_t features = netdev->features;
4907 /* Check for Promiscuous and All Multicast modes */
4908 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4910 /* set all bits that we expect to always be set */
4911 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4912 fctrl |= IXGBE_FCTRL_BAM;
4913 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4914 fctrl |= IXGBE_FCTRL_PMCF;
4916 /* clear the bits we are changing the status of */
4917 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4918 if (netdev->flags & IFF_PROMISC) {
4919 hw->addr_ctrl.user_set_promisc = true;
4920 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4921 vmolr |= IXGBE_VMOLR_MPE;
4922 features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4924 if (netdev->flags & IFF_ALLMULTI) {
4925 fctrl |= IXGBE_FCTRL_MPE;
4926 vmolr |= IXGBE_VMOLR_MPE;
4928 hw->addr_ctrl.user_set_promisc = false;
4932 * Write addresses to available RAR registers, if there is not
4933 * sufficient space to store all the addresses then enable
4934 * unicast promiscuous mode
4936 if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4937 fctrl |= IXGBE_FCTRL_UPE;
4938 vmolr |= IXGBE_VMOLR_ROPE;
4941 /* Write addresses to the MTA, if the attempt fails
4942 * then we should just turn on promiscuous mode so
4943 * that we can at least receive multicast traffic
4945 count = ixgbe_write_mc_addr_list(netdev);
4947 fctrl |= IXGBE_FCTRL_MPE;
4948 vmolr |= IXGBE_VMOLR_MPE;
4950 vmolr |= IXGBE_VMOLR_ROMPE;
4953 if (hw->mac.type != ixgbe_mac_82598EB) {
4954 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4955 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4957 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4960 /* This is useful for sniffing bad packets. */
4961 if (features & NETIF_F_RXALL) {
4962 /* UPE and MPE will be handled by normal PROMISC logic
4963 * in e1000e_set_rx_mode */
4964 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4965 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4966 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4968 fctrl &= ~(IXGBE_FCTRL_DPF);
4969 /* NOTE: VLAN filtering is disabled by setting PROMISC */
4972 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4974 if (features & NETIF_F_HW_VLAN_CTAG_RX)
4975 ixgbe_vlan_strip_enable(adapter);
4977 ixgbe_vlan_strip_disable(adapter);
4979 if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4980 ixgbe_vlan_promisc_disable(adapter);
4982 ixgbe_vlan_promisc_enable(adapter);
4985 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4989 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4990 napi_enable(&adapter->q_vector[q_idx]->napi);
4993 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4997 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4998 napi_disable(&adapter->q_vector[q_idx]->napi);
5001 static int ixgbe_udp_tunnel_sync(struct net_device *dev, unsigned int table)
5003 struct ixgbe_adapter *adapter = netdev_priv(dev);
5004 struct ixgbe_hw *hw = &adapter->hw;
5005 struct udp_tunnel_info ti;
5007 udp_tunnel_nic_get_port(dev, table, 0, &ti);
5008 if (ti.type == UDP_TUNNEL_TYPE_VXLAN)
5009 adapter->vxlan_port = ti.port;
5011 adapter->geneve_port = ti.port;
5013 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL,
5014 ntohs(adapter->vxlan_port) |
5015 ntohs(adapter->geneve_port) <<
5016 IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT);
5020 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550 = {
5021 .sync_table = ixgbe_udp_tunnel_sync,
5022 .flags = UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5024 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
5028 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550em_a = {
5029 .sync_table = ixgbe_udp_tunnel_sync,
5030 .flags = UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5032 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
5033 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_GENEVE, },
5037 #ifdef CONFIG_IXGBE_DCB
5039 * ixgbe_configure_dcb - Configure DCB hardware
5040 * @adapter: ixgbe adapter struct
5042 * This is called by the driver on open to configure the DCB hardware.
5043 * This is also called by the gennetlink interface when reconfiguring
5046 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5048 struct ixgbe_hw *hw = &adapter->hw;
5049 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5051 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5052 if (hw->mac.type == ixgbe_mac_82598EB)
5053 netif_set_gso_max_size(adapter->netdev, 65536);
5057 if (hw->mac.type == ixgbe_mac_82598EB)
5058 netif_set_gso_max_size(adapter->netdev, 32768);
5061 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5062 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5065 /* reconfigure the hardware */
5066 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5067 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5069 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5071 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5072 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5073 ixgbe_dcb_hw_ets(&adapter->hw,
5074 adapter->ixgbe_ieee_ets,
5076 ixgbe_dcb_hw_pfc_config(&adapter->hw,
5077 adapter->ixgbe_ieee_pfc->pfc_en,
5078 adapter->ixgbe_ieee_ets->prio_tc);
5081 /* Enable RSS Hash per TC */
5082 if (hw->mac.type != ixgbe_mac_82598EB) {
5084 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5091 /* write msb to all 8 TCs in one write */
5092 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5097 /* Additional bittime to account for IXGBE framing */
5098 #define IXGBE_ETH_FRAMING 20
5101 * ixgbe_hpbthresh - calculate high water mark for flow control
5103 * @adapter: board private structure to calculate for
5104 * @pb: packet buffer to calculate
5106 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5108 struct ixgbe_hw *hw = &adapter->hw;
5109 struct net_device *dev = adapter->netdev;
5110 int link, tc, kb, marker;
5113 /* Calculate max LAN frame size */
5114 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5117 /* FCoE traffic class uses FCOE jumbo frames */
5118 if ((dev->features & NETIF_F_FCOE_MTU) &&
5119 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5120 (pb == ixgbe_fcoe_get_tc(adapter)))
5121 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5124 /* Calculate delay value for device */
5125 switch (hw->mac.type) {
5126 case ixgbe_mac_X540:
5127 case ixgbe_mac_X550:
5128 case ixgbe_mac_X550EM_x:
5129 case ixgbe_mac_x550em_a:
5130 dv_id = IXGBE_DV_X540(link, tc);
5133 dv_id = IXGBE_DV(link, tc);
5137 /* Loopback switch introduces additional latency */
5138 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5139 dv_id += IXGBE_B2BT(tc);
5141 /* Delay value is calculated in bit times convert to KB */
5142 kb = IXGBE_BT2KB(dv_id);
5143 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5145 marker = rx_pba - kb;
5147 /* It is possible that the packet buffer is not large enough
5148 * to provide required headroom. In this case throw an error
5149 * to user and a do the best we can.
5152 e_warn(drv, "Packet Buffer(%i) can not provide enough"
5153 "headroom to support flow control."
5154 "Decrease MTU or number of traffic classes\n", pb);
5162 * ixgbe_lpbthresh - calculate low water mark for for flow control
5164 * @adapter: board private structure to calculate for
5165 * @pb: packet buffer to calculate
5167 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5169 struct ixgbe_hw *hw = &adapter->hw;
5170 struct net_device *dev = adapter->netdev;
5174 /* Calculate max LAN frame size */
5175 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5178 /* FCoE traffic class uses FCOE jumbo frames */
5179 if ((dev->features & NETIF_F_FCOE_MTU) &&
5180 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5181 (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5182 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5185 /* Calculate delay value for device */
5186 switch (hw->mac.type) {
5187 case ixgbe_mac_X540:
5188 case ixgbe_mac_X550:
5189 case ixgbe_mac_X550EM_x:
5190 case ixgbe_mac_x550em_a:
5191 dv_id = IXGBE_LOW_DV_X540(tc);
5194 dv_id = IXGBE_LOW_DV(tc);
5198 /* Delay value is calculated in bit times convert to KB */
5199 return IXGBE_BT2KB(dv_id);
5203 * ixgbe_pbthresh_setup - calculate and setup high low water marks
5205 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5207 struct ixgbe_hw *hw = &adapter->hw;
5208 int num_tc = adapter->hw_tcs;
5214 for (i = 0; i < num_tc; i++) {
5215 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5216 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5218 /* Low water marks must not be larger than high water marks */
5219 if (hw->fc.low_water[i] > hw->fc.high_water[i])
5220 hw->fc.low_water[i] = 0;
5223 for (; i < MAX_TRAFFIC_CLASS; i++)
5224 hw->fc.high_water[i] = 0;
5227 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5229 struct ixgbe_hw *hw = &adapter->hw;
5231 u8 tc = adapter->hw_tcs;
5233 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5234 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5235 hdrm = 32 << adapter->fdir_pballoc;
5239 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5240 ixgbe_pbthresh_setup(adapter);
5243 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5245 struct ixgbe_hw *hw = &adapter->hw;
5246 struct hlist_node *node2;
5247 struct ixgbe_fdir_filter *filter;
5250 spin_lock(&adapter->fdir_perfect_lock);
5252 if (!hlist_empty(&adapter->fdir_filter_list))
5253 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5255 hlist_for_each_entry_safe(filter, node2,
5256 &adapter->fdir_filter_list, fdir_node) {
5257 if (filter->action == IXGBE_FDIR_DROP_QUEUE) {
5258 queue = IXGBE_FDIR_DROP_QUEUE;
5260 u32 ring = ethtool_get_flow_spec_ring(filter->action);
5261 u8 vf = ethtool_get_flow_spec_ring_vf(filter->action);
5263 if (!vf && (ring >= adapter->num_rx_queues)) {
5264 e_err(drv, "FDIR restore failed without VF, ring: %u\n",
5268 ((vf > adapter->num_vfs) ||
5269 ring >= adapter->num_rx_queues_per_pool)) {
5270 e_err(drv, "FDIR restore failed with VF, vf: %hhu, ring: %u\n",
5275 /* Map the ring onto the absolute queue index */
5277 queue = adapter->rx_ring[ring]->reg_idx;
5280 adapter->num_rx_queues_per_pool) + ring;
5283 ixgbe_fdir_write_perfect_filter_82599(hw,
5284 &filter->filter, filter->sw_idx, queue);
5287 spin_unlock(&adapter->fdir_perfect_lock);
5291 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5292 * @rx_ring: ring to free buffers from
5294 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5296 u16 i = rx_ring->next_to_clean;
5297 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5299 if (rx_ring->xsk_pool) {
5300 ixgbe_xsk_clean_rx_ring(rx_ring);
5304 /* Free all the Rx ring sk_buffs */
5305 while (i != rx_ring->next_to_alloc) {
5306 if (rx_buffer->skb) {
5307 struct sk_buff *skb = rx_buffer->skb;
5308 if (IXGBE_CB(skb)->page_released)
5309 dma_unmap_page_attrs(rx_ring->dev,
5311 ixgbe_rx_pg_size(rx_ring),
5317 /* Invalidate cache lines that may have been written to by
5318 * device so that we avoid corrupting memory.
5320 dma_sync_single_range_for_cpu(rx_ring->dev,
5322 rx_buffer->page_offset,
5323 ixgbe_rx_bufsz(rx_ring),
5326 /* free resources associated with mapping */
5327 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5328 ixgbe_rx_pg_size(rx_ring),
5331 __page_frag_cache_drain(rx_buffer->page,
5332 rx_buffer->pagecnt_bias);
5336 if (i == rx_ring->count) {
5338 rx_buffer = rx_ring->rx_buffer_info;
5343 rx_ring->next_to_alloc = 0;
5344 rx_ring->next_to_clean = 0;
5345 rx_ring->next_to_use = 0;
5348 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5349 struct ixgbe_fwd_adapter *accel)
5351 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5352 int num_tc = netdev_get_num_tc(adapter->netdev);
5353 struct net_device *vdev = accel->netdev;
5356 baseq = accel->pool * adapter->num_rx_queues_per_pool;
5357 netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5358 accel->pool, adapter->num_rx_pools,
5359 baseq, baseq + adapter->num_rx_queues_per_pool);
5361 accel->rx_base_queue = baseq;
5362 accel->tx_base_queue = baseq;
5364 /* record configuration for macvlan interface in vdev */
5365 for (i = 0; i < num_tc; i++)
5366 netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5367 i, rss_i, baseq + (rss_i * i));
5369 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5370 adapter->rx_ring[baseq + i]->netdev = vdev;
5372 /* Guarantee all rings are updated before we update the
5373 * MAC address filter.
5377 /* ixgbe_add_mac_filter will return an index if it succeeds, so we
5378 * need to only treat it as an error value if it is negative.
5380 err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5381 VMDQ_P(accel->pool));
5385 /* if we cannot add the MAC rule then disable the offload */
5386 macvlan_release_l2fw_offload(vdev);
5388 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5389 adapter->rx_ring[baseq + i]->netdev = NULL;
5391 netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5393 /* unbind the queues and drop the subordinate channel config */
5394 netdev_unbind_sb_channel(adapter->netdev, vdev);
5395 netdev_set_sb_channel(vdev, 0);
5397 clear_bit(accel->pool, adapter->fwd_bitmask);
5403 static int ixgbe_macvlan_up(struct net_device *vdev,
5404 struct netdev_nested_priv *priv)
5406 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)priv->data;
5407 struct ixgbe_fwd_adapter *accel;
5409 if (!netif_is_macvlan(vdev))
5412 accel = macvlan_accel_priv(vdev);
5416 ixgbe_fwd_ring_up(adapter, accel);
5421 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5423 struct netdev_nested_priv priv = {
5424 .data = (void *)adapter,
5427 netdev_walk_all_upper_dev_rcu(adapter->netdev,
5428 ixgbe_macvlan_up, &priv);
5431 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5433 struct ixgbe_hw *hw = &adapter->hw;
5435 ixgbe_configure_pb(adapter);
5436 #ifdef CONFIG_IXGBE_DCB
5437 ixgbe_configure_dcb(adapter);
5440 * We must restore virtualization before VLANs or else
5441 * the VLVF registers will not be populated
5443 ixgbe_configure_virtualization(adapter);
5445 ixgbe_set_rx_mode(adapter->netdev);
5446 ixgbe_restore_vlan(adapter);
5447 ixgbe_ipsec_restore(adapter);
5449 switch (hw->mac.type) {
5450 case ixgbe_mac_82599EB:
5451 case ixgbe_mac_X540:
5452 hw->mac.ops.disable_rx_buff(hw);
5458 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5459 ixgbe_init_fdir_signature_82599(&adapter->hw,
5460 adapter->fdir_pballoc);
5461 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5462 ixgbe_init_fdir_perfect_82599(&adapter->hw,
5463 adapter->fdir_pballoc);
5464 ixgbe_fdir_filter_restore(adapter);
5467 switch (hw->mac.type) {
5468 case ixgbe_mac_82599EB:
5469 case ixgbe_mac_X540:
5470 hw->mac.ops.enable_rx_buff(hw);
5476 #ifdef CONFIG_IXGBE_DCA
5478 if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5479 ixgbe_setup_dca(adapter);
5480 #endif /* CONFIG_IXGBE_DCA */
5483 /* configure FCoE L2 filters, redirection table, and Rx control */
5484 ixgbe_configure_fcoe(adapter);
5486 #endif /* IXGBE_FCOE */
5487 ixgbe_configure_tx(adapter);
5488 ixgbe_configure_rx(adapter);
5489 ixgbe_configure_dfwd(adapter);
5493 * ixgbe_sfp_link_config - set up SFP+ link
5494 * @adapter: pointer to private adapter struct
5496 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5499 * We are assuming the worst case scenario here, and that
5500 * is that an SFP was inserted/removed after the reset
5501 * but before SFP detection was enabled. As such the best
5502 * solution is to just start searching as soon as we start
5504 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5505 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5507 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5508 adapter->sfp_poll_time = 0;
5512 * ixgbe_non_sfp_link_config - set up non-SFP+ link
5513 * @hw: pointer to private hardware struct
5515 * Returns 0 on success, negative on failure
5517 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5520 bool autoneg, link_up = false;
5523 if (hw->mac.ops.check_link)
5524 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5529 speed = hw->phy.autoneg_advertised;
5530 if (!speed && hw->mac.ops.get_link_capabilities) {
5531 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5533 /* remove NBASE-T speeds from default autonegotiation
5534 * to accommodate broken network switches in the field
5535 * which cannot cope with advertised NBASE-T speeds
5537 speed &= ~(IXGBE_LINK_SPEED_5GB_FULL |
5538 IXGBE_LINK_SPEED_2_5GB_FULL);
5544 if (hw->mac.ops.setup_link)
5545 ret = hw->mac.ops.setup_link(hw, speed, link_up);
5550 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5552 struct ixgbe_hw *hw = &adapter->hw;
5555 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5556 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5558 gpie |= IXGBE_GPIE_EIAME;
5560 * use EIAM to auto-mask when MSI-X interrupt is asserted
5561 * this saves a register write for every interrupt
5563 switch (hw->mac.type) {
5564 case ixgbe_mac_82598EB:
5565 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5567 case ixgbe_mac_82599EB:
5568 case ixgbe_mac_X540:
5569 case ixgbe_mac_X550:
5570 case ixgbe_mac_X550EM_x:
5571 case ixgbe_mac_x550em_a:
5573 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5574 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5578 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
5579 * specifically only auto mask tx and rx interrupts */
5580 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5583 /* XXX: to interrupt immediately for EICS writes, enable this */
5584 /* gpie |= IXGBE_GPIE_EIMEN; */
5586 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5587 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5589 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5590 case IXGBE_82599_VMDQ_8Q_MASK:
5591 gpie |= IXGBE_GPIE_VTMODE_16;
5593 case IXGBE_82599_VMDQ_4Q_MASK:
5594 gpie |= IXGBE_GPIE_VTMODE_32;
5597 gpie |= IXGBE_GPIE_VTMODE_64;
5602 /* Enable Thermal over heat sensor interrupt */
5603 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5604 switch (adapter->hw.mac.type) {
5605 case ixgbe_mac_82599EB:
5606 gpie |= IXGBE_SDP0_GPIEN_8259X;
5613 /* Enable fan failure interrupt */
5614 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5615 gpie |= IXGBE_SDP1_GPIEN(hw);
5617 switch (hw->mac.type) {
5618 case ixgbe_mac_82599EB:
5619 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5621 case ixgbe_mac_X550EM_x:
5622 case ixgbe_mac_x550em_a:
5623 gpie |= IXGBE_SDP0_GPIEN_X540;
5629 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5632 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5634 struct ixgbe_hw *hw = &adapter->hw;
5638 ixgbe_get_hw_control(adapter);
5639 ixgbe_setup_gpie(adapter);
5641 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5642 ixgbe_configure_msix(adapter);
5644 ixgbe_configure_msi_and_legacy(adapter);
5646 /* enable the optics for 82599 SFP+ fiber */
5647 if (hw->mac.ops.enable_tx_laser)
5648 hw->mac.ops.enable_tx_laser(hw);
5650 if (hw->phy.ops.set_phy_power)
5651 hw->phy.ops.set_phy_power(hw, true);
5653 smp_mb__before_atomic();
5654 clear_bit(__IXGBE_DOWN, &adapter->state);
5655 ixgbe_napi_enable_all(adapter);
5657 if (ixgbe_is_sfp(hw)) {
5658 ixgbe_sfp_link_config(adapter);
5660 err = ixgbe_non_sfp_link_config(hw);
5662 e_err(probe, "link_config FAILED %d\n", err);
5665 /* clear any pending interrupts, may auto mask */
5666 IXGBE_READ_REG(hw, IXGBE_EICR);
5667 ixgbe_irq_enable(adapter, true, true);
5670 * If this adapter has a fan, check to see if we had a failure
5671 * before we enabled the interrupt.
5673 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5674 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5675 if (esdp & IXGBE_ESDP_SDP1)
5676 e_crit(drv, "Fan has stopped, replace the adapter\n");
5679 /* bring the link up in the watchdog, this could race with our first
5680 * link up interrupt but shouldn't be a problem */
5681 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5682 adapter->link_check_timeout = jiffies;
5683 mod_timer(&adapter->service_timer, jiffies);
5685 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5686 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5687 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5688 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5691 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5693 /* put off any impending NetWatchDogTimeout */
5694 netif_trans_update(adapter->netdev);
5696 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5697 usleep_range(1000, 2000);
5698 if (adapter->hw.phy.type == ixgbe_phy_fw)
5699 ixgbe_watchdog_link_is_down(adapter);
5700 ixgbe_down(adapter);
5702 * If SR-IOV enabled then wait a bit before bringing the adapter
5703 * back up to give the VFs time to respond to the reset. The
5704 * two second wait is based upon the watchdog timer cycle in
5707 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5710 clear_bit(__IXGBE_RESETTING, &adapter->state);
5713 void ixgbe_up(struct ixgbe_adapter *adapter)
5715 /* hardware has been reset, we need to reload some things */
5716 ixgbe_configure(adapter);
5718 ixgbe_up_complete(adapter);
5721 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5725 pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5727 switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5728 case IXGBE_PCIDEVCTRL2_17_34s:
5729 case IXGBE_PCIDEVCTRL2_4_8s:
5730 /* For now we cap the upper limit on delay to 2 seconds
5731 * as we end up going up to 34 seconds of delay in worst
5732 * case timeout value.
5734 case IXGBE_PCIDEVCTRL2_1_2s:
5735 return 2000000ul; /* 2.0 s */
5736 case IXGBE_PCIDEVCTRL2_260_520ms:
5737 return 520000ul; /* 520 ms */
5738 case IXGBE_PCIDEVCTRL2_65_130ms:
5739 return 130000ul; /* 130 ms */
5740 case IXGBE_PCIDEVCTRL2_16_32ms:
5741 return 32000ul; /* 32 ms */
5742 case IXGBE_PCIDEVCTRL2_1_2ms:
5743 return 2000ul; /* 2 ms */
5744 case IXGBE_PCIDEVCTRL2_50_100us:
5745 return 100ul; /* 100 us */
5746 case IXGBE_PCIDEVCTRL2_16_32ms_def:
5747 return 32000ul; /* 32 ms */
5752 /* We shouldn't need to hit this path, but just in case default as
5753 * though completion timeout is not supported and support 32ms.
5758 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5760 unsigned long wait_delay, delay_interval;
5761 struct ixgbe_hw *hw = &adapter->hw;
5765 /* disable receives */
5766 hw->mac.ops.disable_rx(hw);
5768 if (ixgbe_removed(hw->hw_addr))
5771 /* disable all enabled Rx queues */
5772 for (i = 0; i < adapter->num_rx_queues; i++) {
5773 struct ixgbe_ring *ring = adapter->rx_ring[i];
5774 u8 reg_idx = ring->reg_idx;
5776 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5777 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5778 rxdctl |= IXGBE_RXDCTL_SWFLSH;
5780 /* write value back with RXDCTL.ENABLE bit cleared */
5781 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5784 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5785 if (hw->mac.type == ixgbe_mac_82598EB &&
5786 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5789 /* Determine our minimum delay interval. We will increase this value
5790 * with each subsequent test. This way if the device returns quickly
5791 * we should spend as little time as possible waiting, however as
5792 * the time increases we will wait for larger periods of time.
5794 * The trick here is that we increase the interval using the
5795 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5796 * of that wait is that it totals up to 100x whatever interval we
5797 * choose. Since our minimum wait is 100us we can just divide the
5798 * total timeout by 100 to get our minimum delay interval.
5800 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5802 wait_loop = IXGBE_MAX_RX_DESC_POLL;
5803 wait_delay = delay_interval;
5805 while (wait_loop--) {
5806 usleep_range(wait_delay, wait_delay + 10);
5807 wait_delay += delay_interval * 2;
5810 /* OR together the reading of all the active RXDCTL registers,
5811 * and then test the result. We need the disable to complete
5812 * before we start freeing the memory and invalidating the
5815 for (i = 0; i < adapter->num_rx_queues; i++) {
5816 struct ixgbe_ring *ring = adapter->rx_ring[i];
5817 u8 reg_idx = ring->reg_idx;
5819 rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5822 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5827 "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5830 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5832 unsigned long wait_delay, delay_interval;
5833 struct ixgbe_hw *hw = &adapter->hw;
5837 if (ixgbe_removed(hw->hw_addr))
5840 /* disable all enabled Tx queues */
5841 for (i = 0; i < adapter->num_tx_queues; i++) {
5842 struct ixgbe_ring *ring = adapter->tx_ring[i];
5843 u8 reg_idx = ring->reg_idx;
5845 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5848 /* disable all enabled XDP Tx queues */
5849 for (i = 0; i < adapter->num_xdp_queues; i++) {
5850 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5851 u8 reg_idx = ring->reg_idx;
5853 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5856 /* If the link is not up there shouldn't be much in the way of
5857 * pending transactions. Those that are left will be flushed out
5858 * when the reset logic goes through the flush sequence to clean out
5859 * the pending Tx transactions.
5861 if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5862 goto dma_engine_disable;
5864 /* Determine our minimum delay interval. We will increase this value
5865 * with each subsequent test. This way if the device returns quickly
5866 * we should spend as little time as possible waiting, however as
5867 * the time increases we will wait for larger periods of time.
5869 * The trick here is that we increase the interval using the
5870 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5871 * of that wait is that it totals up to 100x whatever interval we
5872 * choose. Since our minimum wait is 100us we can just divide the
5873 * total timeout by 100 to get our minimum delay interval.
5875 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5877 wait_loop = IXGBE_MAX_RX_DESC_POLL;
5878 wait_delay = delay_interval;
5880 while (wait_loop--) {
5881 usleep_range(wait_delay, wait_delay + 10);
5882 wait_delay += delay_interval * 2;
5885 /* OR together the reading of all the active TXDCTL registers,
5886 * and then test the result. We need the disable to complete
5887 * before we start freeing the memory and invalidating the
5890 for (i = 0; i < adapter->num_tx_queues; i++) {
5891 struct ixgbe_ring *ring = adapter->tx_ring[i];
5892 u8 reg_idx = ring->reg_idx;
5894 txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5896 for (i = 0; i < adapter->num_xdp_queues; i++) {
5897 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5898 u8 reg_idx = ring->reg_idx;
5900 txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5903 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5904 goto dma_engine_disable;
5908 "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5911 /* Disable the Tx DMA engine on 82599 and later MAC */
5912 switch (hw->mac.type) {
5913 case ixgbe_mac_82599EB:
5914 case ixgbe_mac_X540:
5915 case ixgbe_mac_X550:
5916 case ixgbe_mac_X550EM_x:
5917 case ixgbe_mac_x550em_a:
5918 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5919 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5920 ~IXGBE_DMATXCTL_TE));
5927 void ixgbe_reset(struct ixgbe_adapter *adapter)
5929 struct ixgbe_hw *hw = &adapter->hw;
5930 struct net_device *netdev = adapter->netdev;
5933 if (ixgbe_removed(hw->hw_addr))
5935 /* lock SFP init bit to prevent race conditions with the watchdog */
5936 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5937 usleep_range(1000, 2000);
5939 /* clear all SFP and link config related flags while holding SFP_INIT */
5940 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5941 IXGBE_FLAG2_SFP_NEEDS_RESET);
5942 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5944 err = hw->mac.ops.init_hw(hw);
5951 e_dev_err("primary disable timed out\n");
5954 /* We are running on a pre-production device, log a warning */
5955 e_dev_warn("This device is a pre-production adapter/LOM. "
5956 "Please be aware there may be issues associated with "
5957 "your hardware. If you are experiencing problems "
5958 "please contact your Intel or hardware "
5959 "representative who provided you with this "
5963 e_dev_err("Hardware Error: %d\n", err);
5966 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5968 /* flush entries out of MAC table */
5969 ixgbe_flush_sw_mac_table(adapter);
5970 __dev_uc_unsync(netdev, NULL);
5972 /* do not flush user set addresses */
5973 ixgbe_mac_set_default_filter(adapter);
5975 /* update SAN MAC vmdq pool selection */
5976 if (hw->mac.san_mac_rar_index)
5977 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5979 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5980 ixgbe_ptp_reset(adapter);
5982 if (hw->phy.ops.set_phy_power) {
5983 if (!netif_running(adapter->netdev) && !adapter->wol)
5984 hw->phy.ops.set_phy_power(hw, false);
5986 hw->phy.ops.set_phy_power(hw, true);
5991 * ixgbe_clean_tx_ring - Free Tx Buffers
5992 * @tx_ring: ring to be cleaned
5994 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5996 u16 i = tx_ring->next_to_clean;
5997 struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
5999 if (tx_ring->xsk_pool) {
6000 ixgbe_xsk_clean_tx_ring(tx_ring);
6004 while (i != tx_ring->next_to_use) {
6005 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
6007 /* Free all the Tx ring sk_buffs */
6008 if (ring_is_xdp(tx_ring))
6009 xdp_return_frame(tx_buffer->xdpf);
6011 dev_kfree_skb_any(tx_buffer->skb);
6013 /* unmap skb header data */
6014 dma_unmap_single(tx_ring->dev,
6015 dma_unmap_addr(tx_buffer, dma),
6016 dma_unmap_len(tx_buffer, len),
6019 /* check for eop_desc to determine the end of the packet */
6020 eop_desc = tx_buffer->next_to_watch;
6021 tx_desc = IXGBE_TX_DESC(tx_ring, i);
6023 /* unmap remaining buffers */
6024 while (tx_desc != eop_desc) {
6028 if (unlikely(i == tx_ring->count)) {
6030 tx_buffer = tx_ring->tx_buffer_info;
6031 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6034 /* unmap any remaining paged data */
6035 if (dma_unmap_len(tx_buffer, len))
6036 dma_unmap_page(tx_ring->dev,
6037 dma_unmap_addr(tx_buffer, dma),
6038 dma_unmap_len(tx_buffer, len),
6042 /* move us one more past the eop_desc for start of next pkt */
6045 if (unlikely(i == tx_ring->count)) {
6047 tx_buffer = tx_ring->tx_buffer_info;
6051 /* reset BQL for queue */
6052 if (!ring_is_xdp(tx_ring))
6053 netdev_tx_reset_queue(txring_txq(tx_ring));
6056 /* reset next_to_use and next_to_clean */
6057 tx_ring->next_to_use = 0;
6058 tx_ring->next_to_clean = 0;
6062 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6063 * @adapter: board private structure
6065 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6069 for (i = 0; i < adapter->num_rx_queues; i++)
6070 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6074 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6075 * @adapter: board private structure
6077 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6081 for (i = 0; i < adapter->num_tx_queues; i++)
6082 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6083 for (i = 0; i < adapter->num_xdp_queues; i++)
6084 ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6087 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6089 struct hlist_node *node2;
6090 struct ixgbe_fdir_filter *filter;
6092 spin_lock(&adapter->fdir_perfect_lock);
6094 hlist_for_each_entry_safe(filter, node2,
6095 &adapter->fdir_filter_list, fdir_node) {
6096 hlist_del(&filter->fdir_node);
6099 adapter->fdir_filter_count = 0;
6101 spin_unlock(&adapter->fdir_perfect_lock);
6104 void ixgbe_down(struct ixgbe_adapter *adapter)
6106 struct net_device *netdev = adapter->netdev;
6107 struct ixgbe_hw *hw = &adapter->hw;
6110 /* signal that we are down to the interrupt handler */
6111 if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6112 return; /* do nothing if already down */
6114 /* Shut off incoming Tx traffic */
6115 netif_tx_stop_all_queues(netdev);
6117 /* call carrier off first to avoid false dev_watchdog timeouts */
6118 netif_carrier_off(netdev);
6119 netif_tx_disable(netdev);
6122 ixgbe_disable_rx(adapter);
6124 /* synchronize_rcu() needed for pending XDP buffers to drain */
6125 if (adapter->xdp_ring[0])
6128 ixgbe_irq_disable(adapter);
6130 ixgbe_napi_disable_all(adapter);
6132 clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6133 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6134 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6136 del_timer_sync(&adapter->service_timer);
6138 if (adapter->num_vfs) {
6139 /* Clear EITR Select mapping */
6140 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6142 /* Mark all the VFs as inactive */
6143 for (i = 0 ; i < adapter->num_vfs; i++)
6144 adapter->vfinfo[i].clear_to_send = false;
6146 /* ping all the active vfs to let them know we are going down */
6147 ixgbe_ping_all_vfs(adapter);
6149 /* Disable all VFTE/VFRE TX/RX */
6150 ixgbe_disable_tx_rx(adapter);
6153 /* disable transmits in the hardware now that interrupts are off */
6154 ixgbe_disable_tx(adapter);
6156 if (!pci_channel_offline(adapter->pdev))
6157 ixgbe_reset(adapter);
6159 /* power down the optics for 82599 SFP+ fiber */
6160 if (hw->mac.ops.disable_tx_laser)
6161 hw->mac.ops.disable_tx_laser(hw);
6163 ixgbe_clean_all_tx_rings(adapter);
6164 ixgbe_clean_all_rx_rings(adapter);
6168 * ixgbe_eee_capable - helper function to determine EEE support on X550
6169 * @adapter: board private structure
6171 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6173 struct ixgbe_hw *hw = &adapter->hw;
6175 switch (hw->device_id) {
6176 case IXGBE_DEV_ID_X550EM_A_1G_T:
6177 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6178 if (!hw->phy.eee_speeds_supported)
6180 adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6181 if (!hw->phy.eee_speeds_advertised)
6183 adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6186 adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6187 adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6193 * ixgbe_tx_timeout - Respond to a Tx Hang
6194 * @netdev: network interface device structure
6195 * @txqueue: queue number that timed out
6197 static void ixgbe_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
6199 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6201 /* Do the reset outside of interrupt context */
6202 ixgbe_tx_timeout_reset(adapter);
6205 #ifdef CONFIG_IXGBE_DCB
6206 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6208 struct ixgbe_hw *hw = &adapter->hw;
6209 struct tc_configuration *tc;
6212 switch (hw->mac.type) {
6213 case ixgbe_mac_82598EB:
6214 case ixgbe_mac_82599EB:
6215 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6216 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6218 case ixgbe_mac_X540:
6219 case ixgbe_mac_X550:
6220 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6221 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6223 case ixgbe_mac_X550EM_x:
6224 case ixgbe_mac_x550em_a:
6226 adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6227 adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6231 /* Configure DCB traffic classes */
6232 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6233 tc = &adapter->dcb_cfg.tc_config[j];
6234 tc->path[DCB_TX_CONFIG].bwg_id = 0;
6235 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6236 tc->path[DCB_RX_CONFIG].bwg_id = 0;
6237 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6238 tc->dcb_pfc = pfc_disabled;
6241 /* Initialize default user to priority mapping, UPx->TC0 */
6242 tc = &adapter->dcb_cfg.tc_config[0];
6243 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6244 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6246 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6247 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6248 adapter->dcb_cfg.pfc_mode_enable = false;
6249 adapter->dcb_set_bitmap = 0x00;
6250 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6251 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6252 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6253 sizeof(adapter->temp_dcb_cfg));
6258 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6259 * @adapter: board private structure to initialize
6260 * @ii: pointer to ixgbe_info for device
6262 * ixgbe_sw_init initializes the Adapter private data structure.
6263 * Fields are initialized based on PCI device information and
6264 * OS network device settings (MTU size).
6266 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6267 const struct ixgbe_info *ii)
6269 struct ixgbe_hw *hw = &adapter->hw;
6270 struct pci_dev *pdev = adapter->pdev;
6271 unsigned int rss, fdir;
6275 /* PCI config space info */
6277 hw->vendor_id = pdev->vendor;
6278 hw->device_id = pdev->device;
6279 hw->revision_id = pdev->revision;
6280 hw->subsystem_vendor_id = pdev->subsystem_vendor;
6281 hw->subsystem_device_id = pdev->subsystem_device;
6283 /* get_invariants needs the device IDs */
6284 ii->get_invariants(hw);
6286 /* Set common capability flags and settings */
6287 rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6288 adapter->ring_feature[RING_F_RSS].limit = rss;
6289 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6290 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6291 adapter->atr_sample_rate = 20;
6292 fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6293 adapter->ring_feature[RING_F_FDIR].limit = fdir;
6294 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6295 adapter->ring_feature[RING_F_VMDQ].limit = 1;
6296 #ifdef CONFIG_IXGBE_DCA
6297 adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6299 #ifdef CONFIG_IXGBE_DCB
6300 adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6301 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6304 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6305 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6306 #ifdef CONFIG_IXGBE_DCB
6307 /* Default traffic class to use for FCoE */
6308 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6309 #endif /* CONFIG_IXGBE_DCB */
6310 #endif /* IXGBE_FCOE */
6312 /* initialize static ixgbe jump table entries */
6313 adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6315 if (!adapter->jump_tables[0])
6317 adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6319 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6320 adapter->jump_tables[i] = NULL;
6322 adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6323 sizeof(struct ixgbe_mac_addr),
6325 if (!adapter->mac_table)
6328 if (ixgbe_init_rss_key(adapter))
6331 adapter->af_xdp_zc_qps = bitmap_zalloc(MAX_XDP_QUEUES, GFP_KERNEL);
6332 if (!adapter->af_xdp_zc_qps)
6335 /* Set MAC specific capability flags and exceptions */
6336 switch (hw->mac.type) {
6337 case ixgbe_mac_82598EB:
6338 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6340 if (hw->device_id == IXGBE_DEV_ID_82598AT)
6341 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6343 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6344 adapter->ring_feature[RING_F_FDIR].limit = 0;
6345 adapter->atr_sample_rate = 0;
6346 adapter->fdir_pballoc = 0;
6348 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6349 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6350 #ifdef CONFIG_IXGBE_DCB
6351 adapter->fcoe.up = 0;
6352 #endif /* IXGBE_DCB */
6353 #endif /* IXGBE_FCOE */
6355 case ixgbe_mac_82599EB:
6356 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6357 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6359 case ixgbe_mac_X540:
6360 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6361 if (fwsm & IXGBE_FWSM_TS_ENABLED)
6362 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6364 case ixgbe_mac_x550em_a:
6365 switch (hw->device_id) {
6366 case IXGBE_DEV_ID_X550EM_A_1G_T:
6367 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6368 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6374 case ixgbe_mac_X550EM_x:
6375 #ifdef CONFIG_IXGBE_DCB
6376 adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6379 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6380 #ifdef CONFIG_IXGBE_DCB
6381 adapter->fcoe.up = 0;
6382 #endif /* IXGBE_DCB */
6383 #endif /* IXGBE_FCOE */
6385 case ixgbe_mac_X550:
6386 if (hw->mac.type == ixgbe_mac_X550)
6387 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6388 #ifdef CONFIG_IXGBE_DCA
6389 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6397 /* FCoE support exists, always init the FCoE lock */
6398 spin_lock_init(&adapter->fcoe.lock);
6401 /* n-tuple support exists, always init our spinlock */
6402 spin_lock_init(&adapter->fdir_perfect_lock);
6404 /* init spinlock to avoid concurrency of VF resources */
6405 spin_lock_init(&adapter->vfs_lock);
6407 #ifdef CONFIG_IXGBE_DCB
6408 ixgbe_init_dcb(adapter);
6410 ixgbe_init_ipsec_offload(adapter);
6412 /* default flow control settings */
6413 hw->fc.requested_mode = ixgbe_fc_full;
6414 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
6415 ixgbe_pbthresh_setup(adapter);
6416 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6417 hw->fc.send_xon = true;
6418 hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6420 #ifdef CONFIG_PCI_IOV
6422 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6424 /* assign number of SR-IOV VFs */
6425 if (hw->mac.type != ixgbe_mac_82598EB) {
6426 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6428 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6431 #endif /* CONFIG_PCI_IOV */
6433 /* enable itr by default in dynamic mode */
6434 adapter->rx_itr_setting = 1;
6435 adapter->tx_itr_setting = 1;
6437 /* set default ring sizes */
6438 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6439 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6441 /* set default work limits */
6442 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6444 /* initialize eeprom parameters */
6445 if (ixgbe_init_eeprom_params_generic(hw)) {
6446 e_dev_err("EEPROM initialization failed\n");
6450 /* PF holds first pool slot */
6451 set_bit(0, adapter->fwd_bitmask);
6452 set_bit(__IXGBE_DOWN, &adapter->state);
6458 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6459 * @tx_ring: tx descriptor ring (for a specific queue) to setup
6461 * Return 0 on success, negative on failure
6463 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6465 struct device *dev = tx_ring->dev;
6466 int orig_node = dev_to_node(dev);
6467 int ring_node = NUMA_NO_NODE;
6470 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6472 if (tx_ring->q_vector)
6473 ring_node = tx_ring->q_vector->numa_node;
6475 tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6476 if (!tx_ring->tx_buffer_info)
6477 tx_ring->tx_buffer_info = vmalloc(size);
6478 if (!tx_ring->tx_buffer_info)
6481 /* round up to nearest 4K */
6482 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6483 tx_ring->size = ALIGN(tx_ring->size, 4096);
6485 set_dev_node(dev, ring_node);
6486 tx_ring->desc = dma_alloc_coherent(dev,
6490 set_dev_node(dev, orig_node);
6492 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6493 &tx_ring->dma, GFP_KERNEL);
6497 tx_ring->next_to_use = 0;
6498 tx_ring->next_to_clean = 0;
6502 vfree(tx_ring->tx_buffer_info);
6503 tx_ring->tx_buffer_info = NULL;
6504 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6509 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6510 * @adapter: board private structure
6512 * If this function returns with an error, then it's possible one or
6513 * more of the rings is populated (while the rest are not). It is the
6514 * callers duty to clean those orphaned rings.
6516 * Return 0 on success, negative on failure
6518 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6520 int i, j = 0, err = 0;
6522 for (i = 0; i < adapter->num_tx_queues; i++) {
6523 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6527 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6530 for (j = 0; j < adapter->num_xdp_queues; j++) {
6531 err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6535 e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6541 /* rewind the index freeing the rings as we go */
6543 ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6545 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6550 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6551 * @adapter: pointer to ixgbe_adapter
6552 * @rx_ring: rx descriptor ring (for a specific queue) to setup
6554 * Returns 0 on success, negative on failure
6556 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6557 struct ixgbe_ring *rx_ring)
6559 struct device *dev = rx_ring->dev;
6560 int orig_node = dev_to_node(dev);
6561 int ring_node = NUMA_NO_NODE;
6564 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6566 if (rx_ring->q_vector)
6567 ring_node = rx_ring->q_vector->numa_node;
6569 rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6570 if (!rx_ring->rx_buffer_info)
6571 rx_ring->rx_buffer_info = vmalloc(size);
6572 if (!rx_ring->rx_buffer_info)
6575 /* Round up to nearest 4K */
6576 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6577 rx_ring->size = ALIGN(rx_ring->size, 4096);
6579 set_dev_node(dev, ring_node);
6580 rx_ring->desc = dma_alloc_coherent(dev,
6584 set_dev_node(dev, orig_node);
6586 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6587 &rx_ring->dma, GFP_KERNEL);
6591 rx_ring->next_to_clean = 0;
6592 rx_ring->next_to_use = 0;
6594 /* XDP RX-queue info */
6595 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6596 rx_ring->queue_index) < 0)
6599 rx_ring->xdp_prog = adapter->xdp_prog;
6603 vfree(rx_ring->rx_buffer_info);
6604 rx_ring->rx_buffer_info = NULL;
6605 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6610 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6611 * @adapter: board private structure
6613 * If this function returns with an error, then it's possible one or
6614 * more of the rings is populated (while the rest are not). It is the
6615 * callers duty to clean those orphaned rings.
6617 * Return 0 on success, negative on failure
6619 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6623 for (i = 0; i < adapter->num_rx_queues; i++) {
6624 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6628 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6633 err = ixgbe_setup_fcoe_ddp_resources(adapter);
6638 /* rewind the index freeing the rings as we go */
6640 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6645 * ixgbe_free_tx_resources - Free Tx Resources per Queue
6646 * @tx_ring: Tx descriptor ring for a specific queue
6648 * Free all transmit software resources
6650 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6652 ixgbe_clean_tx_ring(tx_ring);
6654 vfree(tx_ring->tx_buffer_info);
6655 tx_ring->tx_buffer_info = NULL;
6657 /* if not set, then don't free */
6661 dma_free_coherent(tx_ring->dev, tx_ring->size,
6662 tx_ring->desc, tx_ring->dma);
6664 tx_ring->desc = NULL;
6668 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6669 * @adapter: board private structure
6671 * Free all transmit software resources
6673 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6677 for (i = 0; i < adapter->num_tx_queues; i++)
6678 if (adapter->tx_ring[i]->desc)
6679 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6680 for (i = 0; i < adapter->num_xdp_queues; i++)
6681 if (adapter->xdp_ring[i]->desc)
6682 ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6686 * ixgbe_free_rx_resources - Free Rx Resources
6687 * @rx_ring: ring to clean the resources from
6689 * Free all receive software resources
6691 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6693 ixgbe_clean_rx_ring(rx_ring);
6695 rx_ring->xdp_prog = NULL;
6696 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6697 vfree(rx_ring->rx_buffer_info);
6698 rx_ring->rx_buffer_info = NULL;
6700 /* if not set, then don't free */
6704 dma_free_coherent(rx_ring->dev, rx_ring->size,
6705 rx_ring->desc, rx_ring->dma);
6707 rx_ring->desc = NULL;
6711 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6712 * @adapter: board private structure
6714 * Free all receive software resources
6716 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6721 ixgbe_free_fcoe_ddp_resources(adapter);
6724 for (i = 0; i < adapter->num_rx_queues; i++)
6725 if (adapter->rx_ring[i]->desc)
6726 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6730 * ixgbe_max_xdp_frame_size - returns the maximum allowed frame size for XDP
6731 * @adapter: device handle, pointer to adapter
6733 static int ixgbe_max_xdp_frame_size(struct ixgbe_adapter *adapter)
6735 if (PAGE_SIZE >= 8192 || adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
6736 return IXGBE_RXBUFFER_2K;
6738 return IXGBE_RXBUFFER_3K;
6742 * ixgbe_change_mtu - Change the Maximum Transfer Unit
6743 * @netdev: network interface device structure
6744 * @new_mtu: new value for maximum frame size
6746 * Returns 0 on success, negative on failure
6748 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6750 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6752 if (ixgbe_enabled_xdp_adapter(adapter)) {
6753 int new_frame_size = new_mtu + IXGBE_PKT_HDR_PAD;
6755 if (new_frame_size > ixgbe_max_xdp_frame_size(adapter)) {
6756 e_warn(probe, "Requested MTU size is not supported with XDP\n");
6762 * For 82599EB we cannot allow legacy VFs to enable their receive
6763 * paths when MTU greater than 1500 is configured. So display a
6764 * warning that legacy VFs will be disabled.
6766 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6767 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6768 (new_mtu > ETH_DATA_LEN))
6769 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6771 netdev_dbg(netdev, "changing MTU from %d to %d\n",
6772 netdev->mtu, new_mtu);
6774 /* must set new MTU before calling down or up */
6775 netdev->mtu = new_mtu;
6777 if (netif_running(netdev))
6778 ixgbe_reinit_locked(adapter);
6784 * ixgbe_open - Called when a network interface is made active
6785 * @netdev: network interface device structure
6787 * Returns 0 on success, negative value on failure
6789 * The open entry point is called when a network interface is made
6790 * active by the system (IFF_UP). At this point all resources needed
6791 * for transmit and receive operations are allocated, the interrupt
6792 * handler is registered with the OS, the watchdog timer is started,
6793 * and the stack is notified that the interface is ready.
6795 int ixgbe_open(struct net_device *netdev)
6797 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6798 struct ixgbe_hw *hw = &adapter->hw;
6801 /* disallow open during test */
6802 if (test_bit(__IXGBE_TESTING, &adapter->state))
6805 netif_carrier_off(netdev);
6807 /* allocate transmit descriptors */
6808 err = ixgbe_setup_all_tx_resources(adapter);
6812 /* allocate receive descriptors */
6813 err = ixgbe_setup_all_rx_resources(adapter);
6817 ixgbe_configure(adapter);
6819 err = ixgbe_request_irq(adapter);
6823 /* Notify the stack of the actual queue counts. */
6824 queues = adapter->num_tx_queues;
6825 err = netif_set_real_num_tx_queues(netdev, queues);
6827 goto err_set_queues;
6829 queues = adapter->num_rx_queues;
6830 err = netif_set_real_num_rx_queues(netdev, queues);
6832 goto err_set_queues;
6834 ixgbe_ptp_init(adapter);
6836 ixgbe_up_complete(adapter);
6838 udp_tunnel_nic_reset_ntf(netdev);
6843 ixgbe_free_irq(adapter);
6845 ixgbe_free_all_rx_resources(adapter);
6846 if (hw->phy.ops.set_phy_power && !adapter->wol)
6847 hw->phy.ops.set_phy_power(&adapter->hw, false);
6849 ixgbe_free_all_tx_resources(adapter);
6851 ixgbe_reset(adapter);
6856 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6858 ixgbe_ptp_suspend(adapter);
6860 if (adapter->hw.phy.ops.enter_lplu) {
6861 adapter->hw.phy.reset_disable = true;
6862 ixgbe_down(adapter);
6863 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6864 adapter->hw.phy.reset_disable = false;
6866 ixgbe_down(adapter);
6869 ixgbe_free_irq(adapter);
6871 ixgbe_free_all_tx_resources(adapter);
6872 ixgbe_free_all_rx_resources(adapter);
6876 * ixgbe_close - Disables a network interface
6877 * @netdev: network interface device structure
6879 * Returns 0, this is not allowed to fail
6881 * The close entry point is called when an interface is de-activated
6882 * by the OS. The hardware is still under the drivers control, but
6883 * needs to be disabled. A global MAC reset is issued to stop the
6884 * hardware, and all transmit and receive resources are freed.
6886 int ixgbe_close(struct net_device *netdev)
6888 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6890 ixgbe_ptp_stop(adapter);
6892 if (netif_device_present(netdev))
6893 ixgbe_close_suspend(adapter);
6895 ixgbe_fdir_filter_exit(adapter);
6897 ixgbe_release_hw_control(adapter);
6902 static int __maybe_unused ixgbe_resume(struct device *dev_d)
6904 struct pci_dev *pdev = to_pci_dev(dev_d);
6905 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6906 struct net_device *netdev = adapter->netdev;
6909 adapter->hw.hw_addr = adapter->io_addr;
6911 err = pci_enable_device_mem(pdev);
6913 e_dev_err("Cannot enable PCI device from suspend\n");
6916 smp_mb__before_atomic();
6917 clear_bit(__IXGBE_DISABLED, &adapter->state);
6918 pci_set_master(pdev);
6920 device_wakeup_disable(dev_d);
6922 ixgbe_reset(adapter);
6924 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6927 err = ixgbe_init_interrupt_scheme(adapter);
6928 if (!err && netif_running(netdev))
6929 err = ixgbe_open(netdev);
6933 netif_device_attach(netdev);
6939 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6941 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6942 struct net_device *netdev = adapter->netdev;
6943 struct ixgbe_hw *hw = &adapter->hw;
6945 u32 wufc = adapter->wol;
6948 netif_device_detach(netdev);
6950 if (netif_running(netdev))
6951 ixgbe_close_suspend(adapter);
6953 ixgbe_clear_interrupt_scheme(adapter);
6956 if (hw->mac.ops.stop_link_on_d3)
6957 hw->mac.ops.stop_link_on_d3(hw);
6962 ixgbe_set_rx_mode(netdev);
6964 /* enable the optics for 82599 SFP+ fiber as we can WoL */
6965 if (hw->mac.ops.enable_tx_laser)
6966 hw->mac.ops.enable_tx_laser(hw);
6968 /* enable the reception of multicast packets */
6969 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6970 fctrl |= IXGBE_FCTRL_MPE;
6971 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6973 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6974 ctrl |= IXGBE_CTRL_GIO_DIS;
6975 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6977 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6979 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6980 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6983 switch (hw->mac.type) {
6984 case ixgbe_mac_82598EB:
6985 pci_wake_from_d3(pdev, false);
6987 case ixgbe_mac_82599EB:
6988 case ixgbe_mac_X540:
6989 case ixgbe_mac_X550:
6990 case ixgbe_mac_X550EM_x:
6991 case ixgbe_mac_x550em_a:
6992 pci_wake_from_d3(pdev, !!wufc);
6998 *enable_wake = !!wufc;
6999 if (hw->phy.ops.set_phy_power && !*enable_wake)
7000 hw->phy.ops.set_phy_power(hw, false);
7002 ixgbe_release_hw_control(adapter);
7004 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
7005 pci_disable_device(pdev);
7010 static int __maybe_unused ixgbe_suspend(struct device *dev_d)
7012 struct pci_dev *pdev = to_pci_dev(dev_d);
7016 retval = __ixgbe_shutdown(pdev, &wake);
7018 device_set_wakeup_enable(dev_d, wake);
7023 static void ixgbe_shutdown(struct pci_dev *pdev)
7027 __ixgbe_shutdown(pdev, &wake);
7029 if (system_state == SYSTEM_POWER_OFF) {
7030 pci_wake_from_d3(pdev, wake);
7031 pci_set_power_state(pdev, PCI_D3hot);
7036 * ixgbe_update_stats - Update the board statistics counters.
7037 * @adapter: board private structure
7039 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7041 struct net_device *netdev = adapter->netdev;
7042 struct ixgbe_hw *hw = &adapter->hw;
7043 struct ixgbe_hw_stats *hwstats = &adapter->stats;
7045 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7046 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7047 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7048 u64 alloc_rx_page = 0;
7049 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7051 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7052 test_bit(__IXGBE_RESETTING, &adapter->state))
7055 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7058 for (i = 0; i < adapter->num_rx_queues; i++) {
7059 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7060 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7062 adapter->rsc_total_count = rsc_count;
7063 adapter->rsc_total_flush = rsc_flush;
7066 for (i = 0; i < adapter->num_rx_queues; i++) {
7067 struct ixgbe_ring *rx_ring = READ_ONCE(adapter->rx_ring[i]);
7071 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7072 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7073 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7074 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7075 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7076 bytes += rx_ring->stats.bytes;
7077 packets += rx_ring->stats.packets;
7079 adapter->non_eop_descs = non_eop_descs;
7080 adapter->alloc_rx_page = alloc_rx_page;
7081 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7082 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7083 adapter->hw_csum_rx_error = hw_csum_rx_error;
7084 netdev->stats.rx_bytes = bytes;
7085 netdev->stats.rx_packets = packets;
7089 /* gather some stats to the adapter struct that are per queue */
7090 for (i = 0; i < adapter->num_tx_queues; i++) {
7091 struct ixgbe_ring *tx_ring = READ_ONCE(adapter->tx_ring[i]);
7095 restart_queue += tx_ring->tx_stats.restart_queue;
7096 tx_busy += tx_ring->tx_stats.tx_busy;
7097 bytes += tx_ring->stats.bytes;
7098 packets += tx_ring->stats.packets;
7100 for (i = 0; i < adapter->num_xdp_queues; i++) {
7101 struct ixgbe_ring *xdp_ring = READ_ONCE(adapter->xdp_ring[i]);
7105 restart_queue += xdp_ring->tx_stats.restart_queue;
7106 tx_busy += xdp_ring->tx_stats.tx_busy;
7107 bytes += xdp_ring->stats.bytes;
7108 packets += xdp_ring->stats.packets;
7110 adapter->restart_queue = restart_queue;
7111 adapter->tx_busy = tx_busy;
7112 netdev->stats.tx_bytes = bytes;
7113 netdev->stats.tx_packets = packets;
7115 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7117 /* 8 register reads */
7118 for (i = 0; i < 8; i++) {
7119 /* for packet buffers not used, the register should read 0 */
7120 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7122 hwstats->mpc[i] += mpc;
7123 total_mpc += hwstats->mpc[i];
7124 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7125 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7126 switch (hw->mac.type) {
7127 case ixgbe_mac_82598EB:
7128 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7129 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7130 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7131 hwstats->pxonrxc[i] +=
7132 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7134 case ixgbe_mac_82599EB:
7135 case ixgbe_mac_X540:
7136 case ixgbe_mac_X550:
7137 case ixgbe_mac_X550EM_x:
7138 case ixgbe_mac_x550em_a:
7139 hwstats->pxonrxc[i] +=
7140 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7147 /*16 register reads */
7148 for (i = 0; i < 16; i++) {
7149 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7150 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7151 if ((hw->mac.type == ixgbe_mac_82599EB) ||
7152 (hw->mac.type == ixgbe_mac_X540) ||
7153 (hw->mac.type == ixgbe_mac_X550) ||
7154 (hw->mac.type == ixgbe_mac_X550EM_x) ||
7155 (hw->mac.type == ixgbe_mac_x550em_a)) {
7156 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7157 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7158 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7159 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7163 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7164 /* work around hardware counting issue */
7165 hwstats->gprc -= missed_rx;
7167 ixgbe_update_xoff_received(adapter);
7169 /* 82598 hardware only has a 32 bit counter in the high register */
7170 switch (hw->mac.type) {
7171 case ixgbe_mac_82598EB:
7172 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7173 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7174 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7175 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7177 case ixgbe_mac_X540:
7178 case ixgbe_mac_X550:
7179 case ixgbe_mac_X550EM_x:
7180 case ixgbe_mac_x550em_a:
7181 /* OS2BMC stats are X540 and later */
7182 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7183 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7184 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7185 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7187 case ixgbe_mac_82599EB:
7188 for (i = 0; i < 16; i++)
7189 adapter->hw_rx_no_dma_resources +=
7190 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7191 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7192 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7193 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7194 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7195 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7196 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7197 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7198 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7199 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7201 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7202 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7203 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7204 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7205 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7206 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7207 /* Add up per cpu counters for total ddp aloc fail */
7208 if (adapter->fcoe.ddp_pool) {
7209 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7210 struct ixgbe_fcoe_ddp_pool *ddp_pool;
7212 u64 noddp = 0, noddp_ext_buff = 0;
7213 for_each_possible_cpu(cpu) {
7214 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7215 noddp += ddp_pool->noddp;
7216 noddp_ext_buff += ddp_pool->noddp_ext_buff;
7218 hwstats->fcoe_noddp = noddp;
7219 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7221 #endif /* IXGBE_FCOE */
7226 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7227 hwstats->bprc += bprc;
7228 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7229 if (hw->mac.type == ixgbe_mac_82598EB)
7230 hwstats->mprc -= bprc;
7231 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7232 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7233 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7234 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7235 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7236 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7237 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7238 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7239 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7240 hwstats->lxontxc += lxon;
7241 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7242 hwstats->lxofftxc += lxoff;
7243 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7244 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7246 * 82598 errata - tx of flow control packets is included in tx counters
7248 xon_off_tot = lxon + lxoff;
7249 hwstats->gptc -= xon_off_tot;
7250 hwstats->mptc -= xon_off_tot;
7251 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7252 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7253 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7254 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7255 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7256 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7257 hwstats->ptc64 -= xon_off_tot;
7258 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7259 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7260 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7261 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7262 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7263 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7265 /* Fill out the OS statistics structure */
7266 netdev->stats.multicast = hwstats->mprc;
7269 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7270 netdev->stats.rx_dropped = 0;
7271 netdev->stats.rx_length_errors = hwstats->rlec;
7272 netdev->stats.rx_crc_errors = hwstats->crcerrs;
7273 netdev->stats.rx_missed_errors = total_mpc;
7277 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7278 * @adapter: pointer to the device adapter structure
7280 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7282 struct ixgbe_hw *hw = &adapter->hw;
7285 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7288 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7290 /* if interface is down do nothing */
7291 if (test_bit(__IXGBE_DOWN, &adapter->state))
7294 /* do nothing if we are not using signature filters */
7295 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7298 adapter->fdir_overflow++;
7300 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7301 for (i = 0; i < adapter->num_tx_queues; i++)
7302 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7303 &(adapter->tx_ring[i]->state));
7304 for (i = 0; i < adapter->num_xdp_queues; i++)
7305 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7306 &adapter->xdp_ring[i]->state);
7307 /* re-enable flow director interrupts */
7308 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7310 e_err(probe, "failed to finish FDIR re-initialization, "
7311 "ignored adding FDIR ATR filters\n");
7316 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7317 * @adapter: pointer to the device adapter structure
7319 * This function serves two purposes. First it strobes the interrupt lines
7320 * in order to make certain interrupts are occurring. Secondly it sets the
7321 * bits needed to check for TX hangs. As a result we should immediately
7322 * determine if a hang has occurred.
7324 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7326 struct ixgbe_hw *hw = &adapter->hw;
7330 /* If we're down, removing or resetting, just bail */
7331 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7332 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7333 test_bit(__IXGBE_RESETTING, &adapter->state))
7336 /* Force detection of hung controller */
7337 if (netif_carrier_ok(adapter->netdev)) {
7338 for (i = 0; i < adapter->num_tx_queues; i++)
7339 set_check_for_tx_hang(adapter->tx_ring[i]);
7340 for (i = 0; i < adapter->num_xdp_queues; i++)
7341 set_check_for_tx_hang(adapter->xdp_ring[i]);
7344 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7346 * for legacy and MSI interrupts don't set any bits
7347 * that are enabled for EIAM, because this operation
7348 * would set *both* EIMS and EICS for any bit in EIAM
7350 IXGBE_WRITE_REG(hw, IXGBE_EICS,
7351 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7353 /* get one bit for every active tx/rx interrupt vector */
7354 for (i = 0; i < adapter->num_q_vectors; i++) {
7355 struct ixgbe_q_vector *qv = adapter->q_vector[i];
7356 if (qv->rx.ring || qv->tx.ring)
7361 /* Cause software interrupt to ensure rings are cleaned */
7362 ixgbe_irq_rearm_queues(adapter, eics);
7366 * ixgbe_watchdog_update_link - update the link status
7367 * @adapter: pointer to the device adapter structure
7369 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7371 struct ixgbe_hw *hw = &adapter->hw;
7372 u32 link_speed = adapter->link_speed;
7373 bool link_up = adapter->link_up;
7374 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7376 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7379 if (hw->mac.ops.check_link) {
7380 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7382 /* always assume link is up, if no check link function */
7383 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7387 if (adapter->ixgbe_ieee_pfc)
7388 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7390 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7391 hw->mac.ops.fc_enable(hw);
7392 ixgbe_set_rx_drop_en(adapter);
7396 time_after(jiffies, (adapter->link_check_timeout +
7397 IXGBE_TRY_LINK_TIMEOUT))) {
7398 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7399 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7400 IXGBE_WRITE_FLUSH(hw);
7403 adapter->link_up = link_up;
7404 adapter->link_speed = link_speed;
7407 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7409 #ifdef CONFIG_IXGBE_DCB
7410 struct net_device *netdev = adapter->netdev;
7411 struct dcb_app app = {
7412 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7417 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7418 up = dcb_ieee_getapp_mask(netdev, &app);
7420 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7425 * ixgbe_watchdog_link_is_up - update netif_carrier status and
7426 * print link up message
7427 * @adapter: pointer to the device adapter structure
7429 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7431 struct net_device *netdev = adapter->netdev;
7432 struct ixgbe_hw *hw = &adapter->hw;
7433 u32 link_speed = adapter->link_speed;
7434 const char *speed_str;
7435 bool flow_rx, flow_tx;
7437 /* only continue if link was previously down */
7438 if (netif_carrier_ok(netdev))
7441 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7443 switch (hw->mac.type) {
7444 case ixgbe_mac_82598EB: {
7445 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7446 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7447 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7448 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7451 case ixgbe_mac_X540:
7452 case ixgbe_mac_X550:
7453 case ixgbe_mac_X550EM_x:
7454 case ixgbe_mac_x550em_a:
7455 case ixgbe_mac_82599EB: {
7456 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7457 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7458 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7459 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7468 adapter->last_rx_ptp_check = jiffies;
7470 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7471 ixgbe_ptp_start_cyclecounter(adapter);
7473 switch (link_speed) {
7474 case IXGBE_LINK_SPEED_10GB_FULL:
7475 speed_str = "10 Gbps";
7477 case IXGBE_LINK_SPEED_5GB_FULL:
7478 speed_str = "5 Gbps";
7480 case IXGBE_LINK_SPEED_2_5GB_FULL:
7481 speed_str = "2.5 Gbps";
7483 case IXGBE_LINK_SPEED_1GB_FULL:
7484 speed_str = "1 Gbps";
7486 case IXGBE_LINK_SPEED_100_FULL:
7487 speed_str = "100 Mbps";
7489 case IXGBE_LINK_SPEED_10_FULL:
7490 speed_str = "10 Mbps";
7493 speed_str = "unknown speed";
7496 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7497 ((flow_rx && flow_tx) ? "RX/TX" :
7499 (flow_tx ? "TX" : "None"))));
7501 netif_carrier_on(netdev);
7502 ixgbe_check_vf_rate_limit(adapter);
7504 /* enable transmits */
7505 netif_tx_wake_all_queues(adapter->netdev);
7507 /* update the default user priority for VFs */
7508 ixgbe_update_default_up(adapter);
7510 /* ping all the active vfs to let them know link has changed */
7511 ixgbe_ping_all_vfs(adapter);
7515 * ixgbe_watchdog_link_is_down - update netif_carrier status and
7516 * print link down message
7517 * @adapter: pointer to the adapter structure
7519 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7521 struct net_device *netdev = adapter->netdev;
7522 struct ixgbe_hw *hw = &adapter->hw;
7524 adapter->link_up = false;
7525 adapter->link_speed = 0;
7527 /* only continue if link was up previously */
7528 if (!netif_carrier_ok(netdev))
7531 /* poll for SFP+ cable when link is down */
7532 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7533 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7535 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7536 ixgbe_ptp_start_cyclecounter(adapter);
7538 e_info(drv, "NIC Link is Down\n");
7539 netif_carrier_off(netdev);
7541 /* ping all the active vfs to let them know link has changed */
7542 ixgbe_ping_all_vfs(adapter);
7545 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7549 for (i = 0; i < adapter->num_tx_queues; i++) {
7550 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7552 if (tx_ring->next_to_use != tx_ring->next_to_clean)
7556 for (i = 0; i < adapter->num_xdp_queues; i++) {
7557 struct ixgbe_ring *ring = adapter->xdp_ring[i];
7559 if (ring->next_to_use != ring->next_to_clean)
7566 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7568 struct ixgbe_hw *hw = &adapter->hw;
7569 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7570 u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7574 if (!adapter->num_vfs)
7577 /* resetting the PF is only needed for MAC before X550 */
7578 if (hw->mac.type >= ixgbe_mac_X550)
7581 for (i = 0; i < adapter->num_vfs; i++) {
7582 for (j = 0; j < q_per_pool; j++) {
7585 h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7586 t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7597 * ixgbe_watchdog_flush_tx - flush queues on link down
7598 * @adapter: pointer to the device adapter structure
7600 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7602 if (!netif_carrier_ok(adapter->netdev)) {
7603 if (ixgbe_ring_tx_pending(adapter) ||
7604 ixgbe_vf_tx_pending(adapter)) {
7605 /* We've lost link, so the controller stops DMA,
7606 * but we've got queued Tx work that's never going
7607 * to get done, so reset controller to flush Tx.
7608 * (Do the reset outside of interrupt context).
7610 e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7611 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7616 #ifdef CONFIG_PCI_IOV
7617 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7619 struct ixgbe_hw *hw = &adapter->hw;
7620 struct pci_dev *pdev = adapter->pdev;
7624 if (!(netif_carrier_ok(adapter->netdev)))
7627 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7628 if (gpc) /* If incrementing then no need for the check below */
7630 /* Check to see if a bad DMA write target from an errant or
7631 * malicious VF has caused a PCIe error. If so then we can
7632 * issue a VFLR to the offending VF(s) and then resume without
7633 * requesting a full slot reset.
7639 /* check status reg for all VFs owned by this PF */
7640 for (vf = 0; vf < adapter->num_vfs; ++vf) {
7641 struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7646 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7647 if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7648 status_reg & PCI_STATUS_REC_MASTER_ABORT)
7653 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7657 /* Do not perform spoof check for 82598 or if not in IOV mode */
7658 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7659 adapter->num_vfs == 0)
7662 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7665 * ssvpc register is cleared on read, if zero then no
7666 * spoofed packets in the last interval.
7671 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7674 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7679 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7682 #endif /* CONFIG_PCI_IOV */
7686 * ixgbe_watchdog_subtask - check and bring link up
7687 * @adapter: pointer to the device adapter structure
7689 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7691 /* if interface is down, removing or resetting, do nothing */
7692 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7693 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7694 test_bit(__IXGBE_RESETTING, &adapter->state))
7697 ixgbe_watchdog_update_link(adapter);
7699 if (adapter->link_up)
7700 ixgbe_watchdog_link_is_up(adapter);
7702 ixgbe_watchdog_link_is_down(adapter);
7704 ixgbe_check_for_bad_vf(adapter);
7705 ixgbe_spoof_check(adapter);
7706 ixgbe_update_stats(adapter);
7708 ixgbe_watchdog_flush_tx(adapter);
7712 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7713 * @adapter: the ixgbe adapter structure
7715 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7717 struct ixgbe_hw *hw = &adapter->hw;
7720 /* not searching for SFP so there is nothing to do here */
7721 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7722 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7725 if (adapter->sfp_poll_time &&
7726 time_after(adapter->sfp_poll_time, jiffies))
7727 return; /* If not yet time to poll for SFP */
7729 /* someone else is in init, wait until next service event */
7730 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7733 adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7735 err = hw->phy.ops.identify_sfp(hw);
7736 if (err == -EOPNOTSUPP)
7739 if (err == -ENOENT) {
7740 /* If no cable is present, then we need to reset
7741 * the next time we find a good cable. */
7742 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7749 /* exit if reset not needed */
7750 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7753 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7756 * A module may be identified correctly, but the EEPROM may not have
7757 * support for that module. setup_sfp() will fail in that case, so
7758 * we should not allow that module to load.
7760 if (hw->mac.type == ixgbe_mac_82598EB)
7761 err = hw->phy.ops.reset(hw);
7763 err = hw->mac.ops.setup_sfp(hw);
7765 if (err == -EOPNOTSUPP)
7768 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7769 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7772 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7774 if (err == -EOPNOTSUPP &&
7775 adapter->netdev->reg_state == NETREG_REGISTERED) {
7776 e_dev_err("failed to initialize because an unsupported "
7777 "SFP+ module type was detected.\n");
7778 e_dev_err("Reload the driver after installing a "
7779 "supported module.\n");
7780 unregister_netdev(adapter->netdev);
7785 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7786 * @adapter: the ixgbe adapter structure
7788 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7790 struct ixgbe_hw *hw = &adapter->hw;
7793 bool autoneg = false;
7795 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7798 /* someone else is in init, wait until next service event */
7799 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7802 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7804 hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7806 /* advertise highest capable link speed */
7807 if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7808 speed = IXGBE_LINK_SPEED_10GB_FULL;
7810 speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7811 IXGBE_LINK_SPEED_1GB_FULL);
7813 if (hw->mac.ops.setup_link)
7814 hw->mac.ops.setup_link(hw, speed, true);
7816 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7817 adapter->link_check_timeout = jiffies;
7818 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7822 * ixgbe_service_timer - Timer Call-back
7823 * @t: pointer to timer_list structure
7825 static void ixgbe_service_timer(struct timer_list *t)
7827 struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7828 unsigned long next_event_offset;
7830 /* poll faster when waiting for link */
7831 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7832 next_event_offset = HZ / 10;
7834 next_event_offset = HZ * 2;
7836 /* Reset the timer */
7837 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7839 ixgbe_service_event_schedule(adapter);
7842 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7844 struct ixgbe_hw *hw = &adapter->hw;
7847 if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7850 adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7852 if (!hw->phy.ops.handle_lasi)
7855 hw->phy.ops.handle_lasi(&adapter->hw, &overtemp);
7857 e_crit(drv, "%s\n", ixgbe_overheat_msg);
7860 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7862 if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7866 /* If we're already down, removing or resetting, just bail */
7867 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7868 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7869 test_bit(__IXGBE_RESETTING, &adapter->state)) {
7874 ixgbe_dump(adapter);
7875 netdev_err(adapter->netdev, "Reset adapter\n");
7876 adapter->tx_timeout_count++;
7878 ixgbe_reinit_locked(adapter);
7883 * ixgbe_check_fw_error - Check firmware for errors
7884 * @adapter: the adapter private structure
7886 * Check firmware errors in register FWSM
7888 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
7890 struct ixgbe_hw *hw = &adapter->hw;
7893 /* read fwsm.ext_err_ind register and log errors */
7894 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
7896 if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
7897 !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
7898 e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7901 if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
7902 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");
7910 * ixgbe_service_task - manages and runs subtasks
7911 * @work: pointer to work_struct containing our data
7913 static void ixgbe_service_task(struct work_struct *work)
7915 struct ixgbe_adapter *adapter = container_of(work,
7916 struct ixgbe_adapter,
7918 if (ixgbe_removed(adapter->hw.hw_addr)) {
7919 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7921 ixgbe_down(adapter);
7924 ixgbe_service_event_complete(adapter);
7927 if (ixgbe_check_fw_error(adapter)) {
7928 if (!test_bit(__IXGBE_DOWN, &adapter->state))
7929 unregister_netdev(adapter->netdev);
7930 ixgbe_service_event_complete(adapter);
7933 ixgbe_reset_subtask(adapter);
7934 ixgbe_phy_interrupt_subtask(adapter);
7935 ixgbe_sfp_detection_subtask(adapter);
7936 ixgbe_sfp_link_config_subtask(adapter);
7937 ixgbe_check_overtemp_subtask(adapter);
7938 ixgbe_watchdog_subtask(adapter);
7939 ixgbe_fdir_reinit_subtask(adapter);
7940 ixgbe_check_hang_subtask(adapter);
7942 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7943 ixgbe_ptp_overflow_check(adapter);
7944 if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
7945 ixgbe_ptp_rx_hang(adapter);
7946 ixgbe_ptp_tx_hang(adapter);
7949 ixgbe_service_event_complete(adapter);
7952 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7953 struct ixgbe_tx_buffer *first,
7955 struct ixgbe_ipsec_tx_data *itd)
7957 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7958 struct sk_buff *skb = first->skb;
7969 u32 paylen, l4_offset;
7970 u32 fceof_saidx = 0;
7973 if (skb->ip_summed != CHECKSUM_PARTIAL)
7976 if (!skb_is_gso(skb))
7979 err = skb_cow_head(skb, 0);
7983 if (eth_p_mpls(first->protocol))
7984 ip.hdr = skb_inner_network_header(skb);
7986 ip.hdr = skb_network_header(skb);
7987 l4.hdr = skb_checksum_start(skb);
7989 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7990 type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
7991 IXGBE_ADVTXD_TUCMD_L4T_UDP : IXGBE_ADVTXD_TUCMD_L4T_TCP;
7993 /* initialize outer IP header fields */
7994 if (ip.v4->version == 4) {
7995 unsigned char *csum_start = skb_checksum_start(skb);
7996 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
7997 int len = csum_start - trans_start;
7999 /* IP header will have to cancel out any data that
8000 * is not a part of the outer IP header, so set to
8001 * a reverse csum if needed, else init check to 0.
8003 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
8004 csum_fold(csum_partial(trans_start,
8006 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
8009 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8010 IXGBE_TX_FLAGS_CSUM |
8011 IXGBE_TX_FLAGS_IPV4;
8013 ip.v6->payload_len = 0;
8014 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8015 IXGBE_TX_FLAGS_CSUM;
8018 /* determine offset of inner transport header */
8019 l4_offset = l4.hdr - skb->data;
8021 /* remove payload length from inner checksum */
8022 paylen = skb->len - l4_offset;
8024 if (type_tucmd & IXGBE_ADVTXD_TUCMD_L4T_TCP) {
8025 /* compute length of segmentation header */
8026 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
8027 csum_replace_by_diff(&l4.tcp->check,
8028 (__force __wsum)htonl(paylen));
8030 /* compute length of segmentation header */
8031 *hdr_len = sizeof(*l4.udp) + l4_offset;
8032 csum_replace_by_diff(&l4.udp->check,
8033 (__force __wsum)htonl(paylen));
8036 /* update gso size and bytecount with header size */
8037 first->gso_segs = skb_shinfo(skb)->gso_segs;
8038 first->bytecount += (first->gso_segs - 1) * *hdr_len;
8040 /* mss_l4len_id: use 0 as index for TSO */
8041 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8042 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8044 fceof_saidx |= itd->sa_idx;
8045 type_tucmd |= itd->flags | itd->trailer_len;
8047 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8048 vlan_macip_lens = l4.hdr - ip.hdr;
8049 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8050 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8052 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8058 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb)
8060 unsigned int offset = 0;
8062 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
8064 return offset == skb_checksum_start_offset(skb);
8067 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8068 struct ixgbe_tx_buffer *first,
8069 struct ixgbe_ipsec_tx_data *itd)
8071 struct sk_buff *skb = first->skb;
8072 u32 vlan_macip_lens = 0;
8073 u32 fceof_saidx = 0;
8076 if (skb->ip_summed != CHECKSUM_PARTIAL) {
8078 if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8079 IXGBE_TX_FLAGS_CC)))
8084 switch (skb->csum_offset) {
8085 case offsetof(struct tcphdr, check):
8086 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8088 case offsetof(struct udphdr, check):
8090 case offsetof(struct sctphdr, checksum):
8091 /* validate that this is actually an SCTP request */
8092 if (((first->protocol == htons(ETH_P_IP)) &&
8093 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
8094 ((first->protocol == htons(ETH_P_IPV6)) &&
8095 ixgbe_ipv6_csum_is_sctp(skb))) {
8096 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8101 skb_checksum_help(skb);
8105 /* update TX checksum flag */
8106 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8107 vlan_macip_lens = skb_checksum_start_offset(skb) -
8108 skb_network_offset(skb);
8110 /* vlan_macip_lens: MACLEN, VLAN tag */
8111 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8112 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8114 fceof_saidx |= itd->sa_idx;
8115 type_tucmd |= itd->flags | itd->trailer_len;
8117 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8120 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8121 ((_flag <= _result) ? \
8122 ((u32)(_input & _flag) * (_result / _flag)) : \
8123 ((u32)(_input & _flag) / (_flag / _result)))
8125 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8127 /* set type for advanced descriptor with frame checksum insertion */
8128 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8129 IXGBE_ADVTXD_DCMD_DEXT |
8130 IXGBE_ADVTXD_DCMD_IFCS;
8132 /* set HW vlan bit if vlan is present */
8133 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8134 IXGBE_ADVTXD_DCMD_VLE);
8136 /* set segmentation enable bits for TSO/FSO */
8137 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8138 IXGBE_ADVTXD_DCMD_TSE);
8140 /* set timestamp bit if present */
8141 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8142 IXGBE_ADVTXD_MAC_TSTAMP);
8144 /* insert frame checksum */
8145 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8150 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8151 u32 tx_flags, unsigned int paylen)
8153 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8155 /* enable L4 checksum for TSO and TX checksum offload */
8156 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8157 IXGBE_TX_FLAGS_CSUM,
8158 IXGBE_ADVTXD_POPTS_TXSM);
8160 /* enable IPv4 checksum for TSO */
8161 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8162 IXGBE_TX_FLAGS_IPV4,
8163 IXGBE_ADVTXD_POPTS_IXSM);
8166 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8167 IXGBE_TX_FLAGS_IPSEC,
8168 IXGBE_ADVTXD_POPTS_IPSEC);
8171 * Check Context must be set if Tx switch is enabled, which it
8172 * always is for case where virtual functions are running
8174 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8178 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8181 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8183 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8185 /* Herbert's original patch had:
8186 * smp_mb__after_netif_stop_queue();
8187 * but since that doesn't exist yet, just open code it.
8191 /* We need to check again in a case another CPU has just
8192 * made room available.
8194 if (likely(ixgbe_desc_unused(tx_ring) < size))
8197 /* A reprieve! - use start_queue because it doesn't call schedule */
8198 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8199 ++tx_ring->tx_stats.restart_queue;
8203 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8205 if (likely(ixgbe_desc_unused(tx_ring) >= size))
8208 return __ixgbe_maybe_stop_tx(tx_ring, size);
8211 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8212 struct ixgbe_tx_buffer *first,
8215 struct sk_buff *skb = first->skb;
8216 struct ixgbe_tx_buffer *tx_buffer;
8217 union ixgbe_adv_tx_desc *tx_desc;
8220 unsigned int data_len, size;
8221 u32 tx_flags = first->tx_flags;
8222 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8223 u16 i = tx_ring->next_to_use;
8225 tx_desc = IXGBE_TX_DESC(tx_ring, i);
8227 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8229 size = skb_headlen(skb);
8230 data_len = skb->data_len;
8233 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8234 if (data_len < sizeof(struct fcoe_crc_eof)) {
8235 size -= sizeof(struct fcoe_crc_eof) - data_len;
8238 data_len -= sizeof(struct fcoe_crc_eof);
8243 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8247 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8248 if (dma_mapping_error(tx_ring->dev, dma))
8251 /* record length, and DMA address */
8252 dma_unmap_len_set(tx_buffer, len, size);
8253 dma_unmap_addr_set(tx_buffer, dma, dma);
8255 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8257 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8258 tx_desc->read.cmd_type_len =
8259 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8263 if (i == tx_ring->count) {
8264 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8267 tx_desc->read.olinfo_status = 0;
8269 dma += IXGBE_MAX_DATA_PER_TXD;
8270 size -= IXGBE_MAX_DATA_PER_TXD;
8272 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8275 if (likely(!data_len))
8278 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8282 if (i == tx_ring->count) {
8283 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8286 tx_desc->read.olinfo_status = 0;
8289 size = min_t(unsigned int, data_len, skb_frag_size(frag));
8291 size = skb_frag_size(frag);
8295 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8298 tx_buffer = &tx_ring->tx_buffer_info[i];
8301 /* write last descriptor with RS and EOP bits */
8302 cmd_type |= size | IXGBE_TXD_CMD;
8303 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8305 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8307 /* set the timestamp */
8308 first->time_stamp = jiffies;
8310 skb_tx_timestamp(skb);
8313 * Force memory writes to complete before letting h/w know there
8314 * are new descriptors to fetch. (Only applicable for weak-ordered
8315 * memory model archs, such as IA-64).
8317 * We also need this memory barrier to make certain all of the
8318 * status bits have been updated before next_to_watch is written.
8322 /* set next_to_watch value indicating a packet is present */
8323 first->next_to_watch = tx_desc;
8326 if (i == tx_ring->count)
8329 tx_ring->next_to_use = i;
8331 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8333 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
8334 writel(i, tx_ring->tail);
8339 dev_err(tx_ring->dev, "TX DMA map failed\n");
8341 /* clear dma mappings for failed tx_buffer_info map */
8343 tx_buffer = &tx_ring->tx_buffer_info[i];
8344 if (dma_unmap_len(tx_buffer, len))
8345 dma_unmap_page(tx_ring->dev,
8346 dma_unmap_addr(tx_buffer, dma),
8347 dma_unmap_len(tx_buffer, len),
8349 dma_unmap_len_set(tx_buffer, len, 0);
8350 if (tx_buffer == first)
8353 i += tx_ring->count;
8357 dev_kfree_skb_any(first->skb);
8360 tx_ring->next_to_use = i;
8365 static void ixgbe_atr(struct ixgbe_ring *ring,
8366 struct ixgbe_tx_buffer *first)
8368 struct ixgbe_q_vector *q_vector = ring->q_vector;
8369 union ixgbe_atr_hash_dword input = { .dword = 0 };
8370 union ixgbe_atr_hash_dword common = { .dword = 0 };
8372 unsigned char *network;
8374 struct ipv6hdr *ipv6;
8378 struct sk_buff *skb;
8382 /* if ring doesn't have a interrupt vector, cannot perform ATR */
8386 /* do nothing if sampling is disabled */
8387 if (!ring->atr_sample_rate)
8392 /* currently only IPv4/IPv6 with TCP is supported */
8393 if ((first->protocol != htons(ETH_P_IP)) &&
8394 (first->protocol != htons(ETH_P_IPV6)))
8397 /* snag network header to get L4 type and address */
8399 hdr.network = skb_network_header(skb);
8400 if (unlikely(hdr.network <= skb->data))
8402 if (skb->encapsulation &&
8403 first->protocol == htons(ETH_P_IP) &&
8404 hdr.ipv4->protocol == IPPROTO_UDP) {
8405 struct ixgbe_adapter *adapter = q_vector->adapter;
8407 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8411 /* verify the port is recognized as VXLAN */
8412 if (adapter->vxlan_port &&
8413 udp_hdr(skb)->dest == adapter->vxlan_port)
8414 hdr.network = skb_inner_network_header(skb);
8416 if (adapter->geneve_port &&
8417 udp_hdr(skb)->dest == adapter->geneve_port)
8418 hdr.network = skb_inner_network_header(skb);
8421 /* Make sure we have at least [minimum IPv4 header + TCP]
8422 * or [IPv6 header] bytes
8424 if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8427 /* Currently only IPv4/IPv6 with TCP is supported */
8428 switch (hdr.ipv4->version) {
8430 /* access ihl as u8 to avoid unaligned access on ia64 */
8431 hlen = (hdr.network[0] & 0x0F) << 2;
8432 l4_proto = hdr.ipv4->protocol;
8435 hlen = hdr.network - skb->data;
8436 l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8437 hlen -= hdr.network - skb->data;
8443 if (l4_proto != IPPROTO_TCP)
8446 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8447 hlen + sizeof(struct tcphdr)))
8450 th = (struct tcphdr *)(hdr.network + hlen);
8452 /* skip this packet since the socket is closing */
8456 /* sample on all syn packets or once every atr sample count */
8457 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8460 /* reset sample count */
8461 ring->atr_count = 0;
8463 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8466 * src and dst are inverted, think how the receiver sees them
8468 * The input is broken into two sections, a non-compressed section
8469 * containing vm_pool, vlan_id, and flow_type. The rest of the data
8470 * is XORed together and stored in the compressed dword.
8472 input.formatted.vlan_id = vlan_id;
8475 * since src port and flex bytes occupy the same word XOR them together
8476 * and write the value to source port portion of compressed dword
8478 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8479 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8481 common.port.src ^= th->dest ^ first->protocol;
8482 common.port.dst ^= th->source;
8484 switch (hdr.ipv4->version) {
8486 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8487 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8490 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8491 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8492 hdr.ipv6->saddr.s6_addr32[1] ^
8493 hdr.ipv6->saddr.s6_addr32[2] ^
8494 hdr.ipv6->saddr.s6_addr32[3] ^
8495 hdr.ipv6->daddr.s6_addr32[0] ^
8496 hdr.ipv6->daddr.s6_addr32[1] ^
8497 hdr.ipv6->daddr.s6_addr32[2] ^
8498 hdr.ipv6->daddr.s6_addr32[3];
8504 if (hdr.network != skb_network_header(skb))
8505 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8507 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
8508 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8509 input, common, ring->queue_index);
8513 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8514 struct net_device *sb_dev)
8516 struct ixgbe_adapter *adapter;
8517 struct ixgbe_ring_feature *f;
8521 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8522 struct net_device *vdev = sb_dev;
8524 txq = vdev->tc_to_txq[tc].offset;
8525 txq += reciprocal_scale(skb_get_hash(skb),
8526 vdev->tc_to_txq[tc].count);
8532 * only execute the code below if protocol is FCoE
8533 * or FIP and we have FCoE enabled on the adapter
8535 switch (vlan_get_protocol(skb)) {
8536 case htons(ETH_P_FCOE):
8537 case htons(ETH_P_FIP):
8538 adapter = netdev_priv(dev);
8540 if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8544 return netdev_pick_tx(dev, skb, sb_dev);
8547 f = &adapter->ring_feature[RING_F_FCOE];
8549 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8552 while (txq >= f->indices)
8555 return txq + f->offset;
8559 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8560 struct xdp_frame *xdpf)
8562 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8563 struct ixgbe_tx_buffer *tx_buffer;
8564 union ixgbe_adv_tx_desc *tx_desc;
8571 if (unlikely(!ixgbe_desc_unused(ring)))
8572 return IXGBE_XDP_CONSUMED;
8574 dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
8575 if (dma_mapping_error(ring->dev, dma))
8576 return IXGBE_XDP_CONSUMED;
8578 /* record the location of the first descriptor for this packet */
8579 tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8580 tx_buffer->bytecount = len;
8581 tx_buffer->gso_segs = 1;
8582 tx_buffer->protocol = 0;
8584 i = ring->next_to_use;
8585 tx_desc = IXGBE_TX_DESC(ring, i);
8587 dma_unmap_len_set(tx_buffer, len, len);
8588 dma_unmap_addr_set(tx_buffer, dma, dma);
8589 tx_buffer->xdpf = xdpf;
8591 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8593 /* put descriptor type bits */
8594 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8595 IXGBE_ADVTXD_DCMD_DEXT |
8596 IXGBE_ADVTXD_DCMD_IFCS;
8597 cmd_type |= len | IXGBE_TXD_CMD;
8598 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8599 tx_desc->read.olinfo_status =
8600 cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8602 /* Avoid any potential race with xdp_xmit and cleanup */
8605 /* set next_to_watch value indicating a packet is present */
8607 if (i == ring->count)
8610 tx_buffer->next_to_watch = tx_desc;
8611 ring->next_to_use = i;
8613 return IXGBE_XDP_TX;
8616 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8617 struct ixgbe_adapter *adapter,
8618 struct ixgbe_ring *tx_ring)
8620 struct ixgbe_tx_buffer *first;
8624 u16 count = TXD_USE_COUNT(skb_headlen(skb));
8625 struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8626 __be16 protocol = skb->protocol;
8630 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8631 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8632 * + 2 desc gap to keep tail from touching head,
8633 * + 1 desc for context descriptor,
8634 * otherwise try next time
8636 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8637 count += TXD_USE_COUNT(skb_frag_size(
8638 &skb_shinfo(skb)->frags[f]));
8640 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8641 tx_ring->tx_stats.tx_busy++;
8642 return NETDEV_TX_BUSY;
8645 /* record the location of the first descriptor for this packet */
8646 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8648 first->bytecount = skb->len;
8649 first->gso_segs = 1;
8651 /* if we have a HW VLAN tag being added default to the HW one */
8652 if (skb_vlan_tag_present(skb)) {
8653 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8654 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8655 /* else if it is a SW VLAN check the next protocol and store the tag */
8656 } else if (protocol == htons(ETH_P_8021Q)) {
8657 struct vlan_hdr *vhdr, _vhdr;
8658 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8662 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8663 IXGBE_TX_FLAGS_VLAN_SHIFT;
8664 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8666 protocol = vlan_get_protocol(skb);
8668 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8669 adapter->ptp_clock) {
8670 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
8671 !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8673 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8674 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8676 /* schedule check for Tx timestamp */
8677 adapter->ptp_tx_skb = skb_get(skb);
8678 adapter->ptp_tx_start = jiffies;
8679 schedule_work(&adapter->ptp_tx_work);
8681 adapter->tx_hwtstamp_skipped++;
8685 #ifdef CONFIG_PCI_IOV
8687 * Use the l2switch_enable flag - would be false if the DMA
8688 * Tx switch had been disabled.
8690 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8691 tx_flags |= IXGBE_TX_FLAGS_CC;
8694 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8695 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8696 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8697 (skb->priority != TC_PRIO_CONTROL))) {
8698 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8699 tx_flags |= (skb->priority & 0x7) <<
8700 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8701 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8702 struct vlan_ethhdr *vhdr;
8704 if (skb_cow_head(skb, 0))
8706 vhdr = skb_vlan_eth_hdr(skb);
8707 vhdr->h_vlan_TCI = htons(tx_flags >>
8708 IXGBE_TX_FLAGS_VLAN_SHIFT);
8710 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8714 /* record initial flags and protocol */
8715 first->tx_flags = tx_flags;
8716 first->protocol = protocol;
8719 /* setup tx offload for FCoE */
8720 if ((protocol == htons(ETH_P_FCOE)) &&
8721 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8722 tso = ixgbe_fso(tx_ring, first, &hdr_len);
8729 #endif /* IXGBE_FCOE */
8731 #ifdef CONFIG_IXGBE_IPSEC
8732 if (xfrm_offload(skb) &&
8733 !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8736 tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8740 ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8742 /* add the ATR filter if ATR is on */
8743 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8744 ixgbe_atr(tx_ring, first);
8748 #endif /* IXGBE_FCOE */
8749 if (ixgbe_tx_map(tx_ring, first, hdr_len))
8750 goto cleanup_tx_timestamp;
8752 return NETDEV_TX_OK;
8755 dev_kfree_skb_any(first->skb);
8757 cleanup_tx_timestamp:
8758 if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8759 dev_kfree_skb_any(adapter->ptp_tx_skb);
8760 adapter->ptp_tx_skb = NULL;
8761 cancel_work_sync(&adapter->ptp_tx_work);
8762 clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8765 return NETDEV_TX_OK;
8768 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8769 struct net_device *netdev,
8770 struct ixgbe_ring *ring)
8772 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8773 struct ixgbe_ring *tx_ring;
8776 * The minimum packet size for olinfo paylen is 17 so pad the skb
8777 * in order to meet this minimum size requirement.
8779 if (skb_put_padto(skb, 17))
8780 return NETDEV_TX_OK;
8782 tx_ring = ring ? ring : adapter->tx_ring[skb_get_queue_mapping(skb)];
8783 if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8784 return NETDEV_TX_BUSY;
8786 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8789 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8790 struct net_device *netdev)
8792 return __ixgbe_xmit_frame(skb, netdev, NULL);
8796 * ixgbe_set_mac - Change the Ethernet Address of the NIC
8797 * @netdev: network interface device structure
8798 * @p: pointer to an address structure
8800 * Returns 0 on success, negative on failure
8802 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8804 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8805 struct ixgbe_hw *hw = &adapter->hw;
8806 struct sockaddr *addr = p;
8808 if (!is_valid_ether_addr(addr->sa_data))
8809 return -EADDRNOTAVAIL;
8811 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8812 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8814 ixgbe_mac_set_default_filter(adapter);
8820 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8822 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8823 struct ixgbe_hw *hw = &adapter->hw;
8827 if (adapter->mii_bus) {
8830 if (devad != MDIO_DEVAD_NONE)
8831 regnum |= (devad << 16) | MII_ADDR_C45;
8833 return mdiobus_read(adapter->mii_bus, prtad, regnum);
8836 if (prtad != hw->phy.mdio.prtad)
8838 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8844 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8845 u16 addr, u16 value)
8847 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8848 struct ixgbe_hw *hw = &adapter->hw;
8850 if (adapter->mii_bus) {
8853 if (devad != MDIO_DEVAD_NONE)
8854 regnum |= (devad << 16) | MII_ADDR_C45;
8856 return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8859 if (prtad != hw->phy.mdio.prtad)
8861 return hw->phy.ops.write_reg(hw, addr, devad, value);
8864 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8866 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8870 return ixgbe_ptp_set_ts_config(adapter, req);
8872 return ixgbe_ptp_get_ts_config(adapter, req);
8874 if (!adapter->hw.phy.ops.read_reg)
8878 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8883 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8885 * @dev: network interface device structure
8887 * Returns non-zero on failure
8889 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8892 struct ixgbe_adapter *adapter = netdev_priv(dev);
8893 struct ixgbe_hw *hw = &adapter->hw;
8895 if (is_valid_ether_addr(hw->mac.san_addr)) {
8897 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8900 /* update SAN MAC vmdq pool selection */
8901 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8907 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8909 * @dev: network interface device structure
8911 * Returns non-zero on failure
8913 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8916 struct ixgbe_adapter *adapter = netdev_priv(dev);
8917 struct ixgbe_mac_info *mac = &adapter->hw.mac;
8919 if (is_valid_ether_addr(mac->san_addr)) {
8921 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8927 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8928 struct ixgbe_ring *ring)
8935 start = u64_stats_fetch_begin_irq(&ring->syncp);
8936 packets = ring->stats.packets;
8937 bytes = ring->stats.bytes;
8938 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8939 stats->tx_packets += packets;
8940 stats->tx_bytes += bytes;
8944 static void ixgbe_get_stats64(struct net_device *netdev,
8945 struct rtnl_link_stats64 *stats)
8947 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8951 for (i = 0; i < adapter->num_rx_queues; i++) {
8952 struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8958 start = u64_stats_fetch_begin_irq(&ring->syncp);
8959 packets = ring->stats.packets;
8960 bytes = ring->stats.bytes;
8961 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8962 stats->rx_packets += packets;
8963 stats->rx_bytes += bytes;
8967 for (i = 0; i < adapter->num_tx_queues; i++) {
8968 struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8970 ixgbe_get_ring_stats64(stats, ring);
8972 for (i = 0; i < adapter->num_xdp_queues; i++) {
8973 struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8975 ixgbe_get_ring_stats64(stats, ring);
8979 /* following stats updated by ixgbe_watchdog_task() */
8980 stats->multicast = netdev->stats.multicast;
8981 stats->rx_errors = netdev->stats.rx_errors;
8982 stats->rx_length_errors = netdev->stats.rx_length_errors;
8983 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
8984 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
8987 #ifdef CONFIG_IXGBE_DCB
8989 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8990 * @adapter: pointer to ixgbe_adapter
8991 * @tc: number of traffic classes currently enabled
8993 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8994 * 802.1Q priority maps to a packet buffer that exists.
8996 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8998 struct ixgbe_hw *hw = &adapter->hw;
9002 /* 82598 have a static priority to TC mapping that can not
9003 * be changed so no validation is needed.
9005 if (hw->mac.type == ixgbe_mac_82598EB)
9008 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
9011 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
9012 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
9014 /* If up2tc is out of bounds default to zero */
9016 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
9020 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
9026 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
9027 * @adapter: Pointer to adapter struct
9029 * Populate the netdev user priority to tc map
9031 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
9033 struct net_device *dev = adapter->netdev;
9034 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9035 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9038 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9041 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9042 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9044 tc = ets->prio_tc[prio];
9046 netdev_set_prio_tc_map(dev, prio, tc);
9050 #endif /* CONFIG_IXGBE_DCB */
9051 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev,
9052 struct netdev_nested_priv *priv)
9054 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)priv->data;
9055 struct ixgbe_fwd_adapter *accel;
9058 /* we only care about macvlans... */
9059 if (!netif_is_macvlan(vdev))
9062 /* that have hardware offload enabled... */
9063 accel = macvlan_accel_priv(vdev);
9067 /* If we can relocate to a different bit do so */
9068 pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9069 if (pool < adapter->num_rx_pools) {
9070 set_bit(pool, adapter->fwd_bitmask);
9075 /* if we cannot find a free pool then disable the offload */
9076 netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9077 macvlan_release_l2fw_offload(vdev);
9079 /* unbind the queues and drop the subordinate channel config */
9080 netdev_unbind_sb_channel(adapter->netdev, vdev);
9081 netdev_set_sb_channel(vdev, 0);
9088 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9090 struct ixgbe_adapter *adapter = netdev_priv(dev);
9091 struct netdev_nested_priv priv = {
9092 .data = (void *)adapter,
9095 /* flush any stale bits out of the fwd bitmask */
9096 bitmap_clear(adapter->fwd_bitmask, 1, 63);
9098 /* walk through upper devices reassigning pools */
9099 netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9104 * ixgbe_setup_tc - configure net_device for multiple traffic classes
9106 * @dev: net device to configure
9107 * @tc: number of traffic classes to enable
9109 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9111 struct ixgbe_adapter *adapter = netdev_priv(dev);
9112 struct ixgbe_hw *hw = &adapter->hw;
9114 /* Hardware supports up to 8 traffic classes */
9115 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9118 if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9121 /* Hardware has to reinitialize queues and interrupts to
9122 * match packet buffer alignment. Unfortunately, the
9123 * hardware is not flexible enough to do this dynamically.
9125 if (netif_running(dev))
9128 ixgbe_reset(adapter);
9130 ixgbe_clear_interrupt_scheme(adapter);
9132 #ifdef CONFIG_IXGBE_DCB
9134 if (adapter->xdp_prog) {
9135 e_warn(probe, "DCB is not supported with XDP\n");
9137 ixgbe_init_interrupt_scheme(adapter);
9138 if (netif_running(dev))
9143 netdev_set_num_tc(dev, tc);
9144 ixgbe_set_prio_tc_map(adapter);
9146 adapter->hw_tcs = tc;
9147 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9149 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9150 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9151 adapter->hw.fc.requested_mode = ixgbe_fc_none;
9154 netdev_reset_tc(dev);
9156 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9157 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9159 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9160 adapter->hw_tcs = tc;
9162 adapter->temp_dcb_cfg.pfc_mode_enable = false;
9163 adapter->dcb_cfg.pfc_mode_enable = false;
9166 ixgbe_validate_rtr(adapter, tc);
9168 #endif /* CONFIG_IXGBE_DCB */
9169 ixgbe_init_interrupt_scheme(adapter);
9171 ixgbe_defrag_macvlan_pools(dev);
9173 if (netif_running(dev))
9174 return ixgbe_open(dev);
9179 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9180 struct tc_cls_u32_offload *cls)
9182 u32 hdl = cls->knode.handle;
9183 u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9184 u32 loc = cls->knode.handle & 0xfffff;
9186 struct ixgbe_jump_table *jump = NULL;
9188 if (loc > IXGBE_MAX_HW_ENTRIES)
9191 if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9194 /* Clear this filter in the link data it is associated with */
9195 if (uhtid != 0x800) {
9196 jump = adapter->jump_tables[uhtid];
9199 if (!test_bit(loc - 1, jump->child_loc_map))
9201 clear_bit(loc - 1, jump->child_loc_map);
9204 /* Check if the filter being deleted is a link */
9205 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9206 jump = adapter->jump_tables[i];
9207 if (jump && jump->link_hdl == hdl) {
9208 /* Delete filters in the hardware in the child hash
9209 * table associated with this link
9211 for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9212 if (!test_bit(j, jump->child_loc_map))
9214 spin_lock(&adapter->fdir_perfect_lock);
9215 err = ixgbe_update_ethtool_fdir_entry(adapter,
9218 spin_unlock(&adapter->fdir_perfect_lock);
9219 clear_bit(j, jump->child_loc_map);
9221 /* Remove resources for this link */
9225 adapter->jump_tables[i] = NULL;
9230 spin_lock(&adapter->fdir_perfect_lock);
9231 err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9232 spin_unlock(&adapter->fdir_perfect_lock);
9236 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9237 struct tc_cls_u32_offload *cls)
9239 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9241 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9244 /* This ixgbe devices do not support hash tables at the moment
9245 * so abort when given hash tables.
9247 if (cls->hnode.divisor > 0)
9250 set_bit(uhtid - 1, &adapter->tables);
9254 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9255 struct tc_cls_u32_offload *cls)
9257 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9259 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9262 clear_bit(uhtid - 1, &adapter->tables);
9266 #ifdef CONFIG_NET_CLS_ACT
9267 struct upper_walk_data {
9268 struct ixgbe_adapter *adapter;
9274 static int get_macvlan_queue(struct net_device *upper,
9275 struct netdev_nested_priv *priv)
9277 if (netif_is_macvlan(upper)) {
9278 struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9279 struct ixgbe_adapter *adapter;
9280 struct upper_walk_data *data;
9283 data = (struct upper_walk_data *)priv->data;
9284 ifindex = data->ifindex;
9285 adapter = data->adapter;
9286 if (vadapter && upper->ifindex == ifindex) {
9287 data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9288 data->action = data->queue;
9296 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9297 u8 *queue, u64 *action)
9299 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9300 unsigned int num_vfs = adapter->num_vfs, vf;
9301 struct netdev_nested_priv priv;
9302 struct upper_walk_data data;
9303 struct net_device *upper;
9305 /* redirect to a SRIOV VF */
9306 for (vf = 0; vf < num_vfs; ++vf) {
9307 upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9308 if (upper->ifindex == ifindex) {
9309 *queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9311 *action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9316 /* redirect to a offloaded macvlan netdev */
9317 data.adapter = adapter;
9318 data.ifindex = ifindex;
9321 priv.data = (void *)&data;
9322 if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9323 get_macvlan_queue, &priv)) {
9324 *action = data.action;
9325 *queue = data.queue;
9333 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9334 struct tcf_exts *exts, u64 *action, u8 *queue)
9336 const struct tc_action *a;
9339 if (!tcf_exts_has_actions(exts))
9342 tcf_exts_for_each_action(i, a, exts) {
9344 if (is_tcf_gact_shot(a)) {
9345 *action = IXGBE_FDIR_DROP_QUEUE;
9346 *queue = IXGBE_FDIR_DROP_QUEUE;
9350 /* Redirect to a VF or a offloaded macvlan */
9351 if (is_tcf_mirred_egress_redirect(a)) {
9352 struct net_device *dev = tcf_mirred_dev(a);
9356 return handle_redirect_action(adapter, dev->ifindex,
9366 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9367 struct tcf_exts *exts, u64 *action, u8 *queue)
9371 #endif /* CONFIG_NET_CLS_ACT */
9373 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9374 union ixgbe_atr_input *mask,
9375 struct tc_cls_u32_offload *cls,
9376 struct ixgbe_mat_field *field_ptr,
9377 struct ixgbe_nexthdr *nexthdr)
9381 bool found_entry = false, found_jump_field = false;
9383 for (i = 0; i < cls->knode.sel->nkeys; i++) {
9384 off = cls->knode.sel->keys[i].off;
9385 val = cls->knode.sel->keys[i].val;
9386 m = cls->knode.sel->keys[i].mask;
9388 for (j = 0; field_ptr[j].val; j++) {
9389 if (field_ptr[j].off == off) {
9390 field_ptr[j].val(input, mask, (__force u32)val,
9392 input->filter.formatted.flow_type |=
9399 if (nexthdr->off == cls->knode.sel->keys[i].off &&
9401 (__force u32)cls->knode.sel->keys[i].val &&
9403 (__force u32)cls->knode.sel->keys[i].mask)
9404 found_jump_field = true;
9410 if (nexthdr && !found_jump_field)
9416 mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9417 IXGBE_ATR_L4TYPE_MASK;
9419 if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9420 mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9425 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9426 struct tc_cls_u32_offload *cls)
9428 __be16 protocol = cls->common.protocol;
9429 u32 loc = cls->knode.handle & 0xfffff;
9430 struct ixgbe_hw *hw = &adapter->hw;
9431 struct ixgbe_mat_field *field_ptr;
9432 struct ixgbe_fdir_filter *input = NULL;
9433 union ixgbe_atr_input *mask = NULL;
9434 struct ixgbe_jump_table *jump = NULL;
9435 int i, err = -EINVAL;
9437 u32 uhtid, link_uhtid;
9439 uhtid = TC_U32_USERHTID(cls->knode.handle);
9440 link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9442 /* At the moment cls_u32 jumps to network layer and skips past
9443 * L2 headers. The canonical method to match L2 frames is to use
9444 * negative values. However this is error prone at best but really
9445 * just broken because there is no way to "know" what sort of hdr
9446 * is in front of the network layer. Fix cls_u32 to support L2
9447 * headers when needed.
9449 if (protocol != htons(ETH_P_IP))
9452 if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9453 e_err(drv, "Location out of range\n");
9457 /* cls u32 is a graph starting at root node 0x800. The driver tracks
9458 * links and also the fields used to advance the parser across each
9459 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9460 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9461 * To add support for new nodes update ixgbe_model.h parse structures
9462 * this function _should_ be generic try not to hardcode values here.
9464 if (uhtid == 0x800) {
9465 field_ptr = (adapter->jump_tables[0])->mat;
9467 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9469 if (!adapter->jump_tables[uhtid])
9471 field_ptr = (adapter->jump_tables[uhtid])->mat;
9477 /* At this point we know the field_ptr is valid and need to either
9478 * build cls_u32 link or attach filter. Because adding a link to
9479 * a handle that does not exist is invalid and the same for adding
9480 * rules to handles that don't exist.
9484 struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9486 if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9489 if (!test_bit(link_uhtid - 1, &adapter->tables))
9492 /* Multiple filters as links to the same hash table are not
9493 * supported. To add a new filter with the same next header
9494 * but different match/jump conditions, create a new hash table
9497 if (adapter->jump_tables[link_uhtid] &&
9498 (adapter->jump_tables[link_uhtid])->link_hdl) {
9499 e_err(drv, "Link filter exists for link: %x\n",
9504 for (i = 0; nexthdr[i].jump; i++) {
9505 if (nexthdr[i].o != cls->knode.sel->offoff ||
9506 nexthdr[i].s != cls->knode.sel->offshift ||
9508 (__force u32)cls->knode.sel->offmask)
9511 jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9514 input = kzalloc(sizeof(*input), GFP_KERNEL);
9519 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9524 jump->input = input;
9526 jump->link_hdl = cls->knode.handle;
9528 err = ixgbe_clsu32_build_input(input, mask, cls,
9529 field_ptr, &nexthdr[i]);
9531 jump->mat = nexthdr[i].jump;
9532 adapter->jump_tables[link_uhtid] = jump;
9543 input = kzalloc(sizeof(*input), GFP_KERNEL);
9546 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9552 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9553 if ((adapter->jump_tables[uhtid])->input)
9554 memcpy(input, (adapter->jump_tables[uhtid])->input,
9556 if ((adapter->jump_tables[uhtid])->mask)
9557 memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9560 /* Lookup in all child hash tables if this location is already
9561 * filled with a filter
9563 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9564 struct ixgbe_jump_table *link = adapter->jump_tables[i];
9566 if (link && (test_bit(loc - 1, link->child_loc_map))) {
9567 e_err(drv, "Filter exists in location: %x\n",
9574 err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9578 err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9583 input->sw_idx = loc;
9585 spin_lock(&adapter->fdir_perfect_lock);
9587 if (hlist_empty(&adapter->fdir_filter_list)) {
9588 memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9589 err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9591 goto err_out_w_lock;
9592 } else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9594 goto err_out_w_lock;
9597 ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9598 err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9599 input->sw_idx, queue);
9601 goto err_out_w_lock;
9603 ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9604 spin_unlock(&adapter->fdir_perfect_lock);
9606 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9607 set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9612 spin_unlock(&adapter->fdir_perfect_lock);
9622 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9623 struct tc_cls_u32_offload *cls_u32)
9625 switch (cls_u32->command) {
9626 case TC_CLSU32_NEW_KNODE:
9627 case TC_CLSU32_REPLACE_KNODE:
9628 return ixgbe_configure_clsu32(adapter, cls_u32);
9629 case TC_CLSU32_DELETE_KNODE:
9630 return ixgbe_delete_clsu32(adapter, cls_u32);
9631 case TC_CLSU32_NEW_HNODE:
9632 case TC_CLSU32_REPLACE_HNODE:
9633 return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9634 case TC_CLSU32_DELETE_HNODE:
9635 return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9641 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9644 struct ixgbe_adapter *adapter = cb_priv;
9646 if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9650 case TC_SETUP_CLSU32:
9651 return ixgbe_setup_tc_cls_u32(adapter, type_data);
9657 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9658 struct tc_mqprio_qopt *mqprio)
9660 mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9661 return ixgbe_setup_tc(dev, mqprio->num_tc);
9664 static LIST_HEAD(ixgbe_block_cb_list);
9666 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9669 struct ixgbe_adapter *adapter = netdev_priv(dev);
9672 case TC_SETUP_BLOCK:
9673 return flow_block_cb_setup_simple(type_data,
9674 &ixgbe_block_cb_list,
9675 ixgbe_setup_tc_block_cb,
9676 adapter, adapter, true);
9677 case TC_SETUP_QDISC_MQPRIO:
9678 return ixgbe_setup_tc_mqprio(dev, type_data);
9684 #ifdef CONFIG_PCI_IOV
9685 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9687 struct net_device *netdev = adapter->netdev;
9690 ixgbe_setup_tc(netdev, adapter->hw_tcs);
9695 void ixgbe_do_reset(struct net_device *netdev)
9697 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9699 if (netif_running(netdev))
9700 ixgbe_reinit_locked(adapter);
9702 ixgbe_reset(adapter);
9705 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9706 netdev_features_t features)
9708 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9710 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9711 if (!(features & NETIF_F_RXCSUM))
9712 features &= ~NETIF_F_LRO;
9714 /* Turn off LRO if not RSC capable */
9715 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9716 features &= ~NETIF_F_LRO;
9718 if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9719 e_dev_err("LRO is not supported with XDP\n");
9720 features &= ~NETIF_F_LRO;
9726 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9728 int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9731 /* go back to full RSS if we're not running SR-IOV */
9732 if (!adapter->ring_feature[RING_F_VMDQ].offset)
9733 adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9734 IXGBE_FLAG_SRIOV_ENABLED);
9736 adapter->ring_feature[RING_F_RSS].limit = rss;
9737 adapter->ring_feature[RING_F_VMDQ].limit = 1;
9739 ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9742 static int ixgbe_set_features(struct net_device *netdev,
9743 netdev_features_t features)
9745 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9746 netdev_features_t changed = netdev->features ^ features;
9747 bool need_reset = false;
9749 /* Make sure RSC matches LRO, reset if change */
9750 if (!(features & NETIF_F_LRO)) {
9751 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9753 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9754 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9755 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9756 if (adapter->rx_itr_setting == 1 ||
9757 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9758 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9760 } else if ((changed ^ features) & NETIF_F_LRO) {
9761 e_info(probe, "rx-usecs set too low, "
9767 * Check if Flow Director n-tuple support or hw_tc support was
9768 * enabled or disabled. If the state changed, we need to reset.
9770 if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9771 /* turn off ATR, enable perfect filters and reset */
9772 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9775 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9776 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9778 /* turn off perfect filters, enable ATR and reset */
9779 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9782 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9784 /* We cannot enable ATR if SR-IOV is enabled */
9785 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9786 /* We cannot enable ATR if we have 2 or more tcs */
9787 (adapter->hw_tcs > 1) ||
9788 /* We cannot enable ATR if RSS is disabled */
9789 (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9790 /* A sample rate of 0 indicates ATR disabled */
9791 (!adapter->atr_sample_rate))
9792 ; /* do nothing not supported */
9793 else /* otherwise supported and set the flag */
9794 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9797 if (changed & NETIF_F_RXALL)
9800 netdev->features = features;
9802 if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9803 ixgbe_reset_l2fw_offload(adapter);
9804 else if (need_reset)
9805 ixgbe_do_reset(netdev);
9806 else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9807 NETIF_F_HW_VLAN_CTAG_FILTER))
9808 ixgbe_set_rx_mode(netdev);
9813 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9814 struct net_device *dev,
9815 const unsigned char *addr, u16 vid,
9817 struct netlink_ext_ack *extack)
9819 /* guarantee we can provide a unique filter for the unicast address */
9820 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9821 struct ixgbe_adapter *adapter = netdev_priv(dev);
9822 u16 pool = VMDQ_P(0);
9824 if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9828 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9832 * ixgbe_configure_bridge_mode - set various bridge modes
9833 * @adapter: the private structure
9834 * @mode: requested bridge mode
9836 * Configure some settings require for various bridge modes.
9838 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9841 struct ixgbe_hw *hw = &adapter->hw;
9842 unsigned int p, num_pools;
9846 case BRIDGE_MODE_VEPA:
9847 /* disable Tx loopback, rely on switch hairpin mode */
9848 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9850 /* must enable Rx switching replication to allow multicast
9851 * packet reception on all VFs, and to enable source address
9854 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9855 vmdctl |= IXGBE_VT_CTL_REPLEN;
9856 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9858 /* enable Rx source address pruning. Note, this requires
9859 * replication to be enabled or else it does nothing.
9861 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9862 for (p = 0; p < num_pools; p++) {
9863 if (hw->mac.ops.set_source_address_pruning)
9864 hw->mac.ops.set_source_address_pruning(hw,
9869 case BRIDGE_MODE_VEB:
9870 /* enable Tx loopback for internal VF/PF communication */
9871 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9872 IXGBE_PFDTXGSWC_VT_LBEN);
9874 /* disable Rx switching replication unless we have SR-IOV
9877 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9878 if (!adapter->num_vfs)
9879 vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9880 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9882 /* disable Rx source address pruning, since we don't expect to
9883 * be receiving external loopback of our transmitted frames.
9885 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9886 for (p = 0; p < num_pools; p++) {
9887 if (hw->mac.ops.set_source_address_pruning)
9888 hw->mac.ops.set_source_address_pruning(hw,
9897 adapter->bridge_mode = mode;
9899 e_info(drv, "enabling bridge mode: %s\n",
9900 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
9905 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
9906 struct nlmsghdr *nlh, u16 flags,
9907 struct netlink_ext_ack *extack)
9909 struct ixgbe_adapter *adapter = netdev_priv(dev);
9910 struct nlattr *attr, *br_spec;
9913 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9916 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
9920 nla_for_each_nested(attr, br_spec, rem) {
9924 if (nla_type(attr) != IFLA_BRIDGE_MODE)
9927 if (nla_len(attr) < sizeof(mode))
9930 mode = nla_get_u16(attr);
9931 status = ixgbe_configure_bridge_mode(adapter, mode);
9941 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
9942 struct net_device *dev,
9943 u32 filter_mask, int nlflags)
9945 struct ixgbe_adapter *adapter = netdev_priv(dev);
9947 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9950 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
9951 adapter->bridge_mode, 0, 0, nlflags,
9955 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
9957 struct ixgbe_adapter *adapter = netdev_priv(pdev);
9958 struct ixgbe_fwd_adapter *accel;
9959 int tcs = adapter->hw_tcs ? : 1;
9962 if (adapter->xdp_prog) {
9963 e_warn(probe, "L2FW offload is not supported with XDP\n");
9964 return ERR_PTR(-EINVAL);
9967 /* The hardware supported by ixgbe only filters on the destination MAC
9968 * address. In order to avoid issues we only support offloading modes
9969 * where the hardware can actually provide the functionality.
9971 if (!macvlan_supports_dest_filter(vdev))
9972 return ERR_PTR(-EMEDIUMTYPE);
9974 /* We need to lock down the macvlan to be a single queue device so that
9975 * we can reuse the tc_to_txq field in the macvlan netdev to represent
9976 * the queue mapping to our netdev.
9978 if (netif_is_multiqueue(vdev))
9979 return ERR_PTR(-ERANGE);
9981 pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9982 if (pool == adapter->num_rx_pools) {
9983 u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
9986 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
9987 adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
9988 adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
9989 return ERR_PTR(-EBUSY);
9991 /* Hardware has a limited number of available pools. Each VF,
9992 * and the PF require a pool. Check to ensure we don't
9993 * attempt to use more then the available number of pools.
9995 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
9996 return ERR_PTR(-EBUSY);
9998 /* Enable VMDq flag so device will be set in VM mode */
9999 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
10000 IXGBE_FLAG_SRIOV_ENABLED;
10002 /* Try to reserve as many queues per pool as possible,
10003 * we start with the configurations that support 4 queues
10004 * per pools, followed by 2, and then by just 1 per pool.
10006 if (used_pools < 32 && adapter->num_rx_pools < 16)
10007 reserved_pools = min_t(u16,
10009 16 - adapter->num_rx_pools);
10010 else if (adapter->num_rx_pools < 32)
10011 reserved_pools = min_t(u16,
10013 32 - adapter->num_rx_pools);
10015 reserved_pools = 64 - used_pools;
10018 if (!reserved_pools)
10019 return ERR_PTR(-EBUSY);
10021 adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
10023 /* Force reinit of ring allocation with VMDQ enabled */
10024 err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
10026 return ERR_PTR(err);
10028 if (pool >= adapter->num_rx_pools)
10029 return ERR_PTR(-ENOMEM);
10032 accel = kzalloc(sizeof(*accel), GFP_KERNEL);
10034 return ERR_PTR(-ENOMEM);
10036 set_bit(pool, adapter->fwd_bitmask);
10037 netdev_set_sb_channel(vdev, pool);
10038 accel->pool = pool;
10039 accel->netdev = vdev;
10041 if (!netif_running(pdev))
10044 err = ixgbe_fwd_ring_up(adapter, accel);
10046 return ERR_PTR(err);
10051 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10053 struct ixgbe_fwd_adapter *accel = priv;
10054 struct ixgbe_adapter *adapter = netdev_priv(pdev);
10055 unsigned int rxbase = accel->rx_base_queue;
10058 /* delete unicast filter associated with offloaded interface */
10059 ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10060 VMDQ_P(accel->pool));
10062 /* Allow remaining Rx packets to get flushed out of the
10063 * Rx FIFO before we drop the netdev for the ring.
10065 usleep_range(10000, 20000);
10067 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10068 struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10069 struct ixgbe_q_vector *qv = ring->q_vector;
10071 /* Make sure we aren't processing any packets and clear
10072 * netdev to shut down the ring.
10074 if (netif_running(adapter->netdev))
10075 napi_synchronize(&qv->napi);
10076 ring->netdev = NULL;
10079 /* unbind the queues and drop the subordinate channel config */
10080 netdev_unbind_sb_channel(pdev, accel->netdev);
10081 netdev_set_sb_channel(accel->netdev, 0);
10083 clear_bit(accel->pool, adapter->fwd_bitmask);
10087 #define IXGBE_MAX_MAC_HDR_LEN 127
10088 #define IXGBE_MAX_NETWORK_HDR_LEN 511
10090 static netdev_features_t
10091 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10092 netdev_features_t features)
10094 unsigned int network_hdr_len, mac_hdr_len;
10096 /* Make certain the headers can be described by a context descriptor */
10097 mac_hdr_len = skb_network_header(skb) - skb->data;
10098 if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10099 return features & ~(NETIF_F_HW_CSUM |
10101 NETIF_F_GSO_UDP_L4 |
10102 NETIF_F_HW_VLAN_CTAG_TX |
10106 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10107 if (unlikely(network_hdr_len > IXGBE_MAX_NETWORK_HDR_LEN))
10108 return features & ~(NETIF_F_HW_CSUM |
10110 NETIF_F_GSO_UDP_L4 |
10114 /* We can only support IPV4 TSO in tunnels if we can mangle the
10115 * inner IP ID field, so strip TSO if MANGLEID is not supported.
10116 * IPsec offoad sets skb->encapsulation but still can handle
10117 * the TSO, so it's the exception.
10119 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10120 #ifdef CONFIG_IXGBE_IPSEC
10121 if (!secpath_exists(skb))
10123 features &= ~NETIF_F_TSO;
10129 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10131 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10132 struct ixgbe_adapter *adapter = netdev_priv(dev);
10133 struct bpf_prog *old_prog;
10137 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10140 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10143 /* verify ixgbe ring attributes are sufficient for XDP */
10144 for (i = 0; i < adapter->num_rx_queues; i++) {
10145 struct ixgbe_ring *ring = adapter->rx_ring[i];
10147 if (ring_is_rsc_enabled(ring))
10150 if (frame_size > ixgbe_rx_bufsz(ring))
10154 if (nr_cpu_ids > MAX_XDP_QUEUES)
10157 old_prog = xchg(&adapter->xdp_prog, prog);
10158 need_reset = (!!prog != !!old_prog);
10160 /* If transitioning XDP modes reconfigure rings */
10165 /* Wait until ndo_xsk_wakeup completes. */
10167 err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10170 rcu_assign_pointer(adapter->xdp_prog, old_prog);
10174 for (i = 0; i < adapter->num_rx_queues; i++)
10175 (void)xchg(&adapter->rx_ring[i]->xdp_prog,
10176 adapter->xdp_prog);
10180 bpf_prog_put(old_prog);
10182 /* Kick start the NAPI context if there is an AF_XDP socket open
10183 * on that queue id. This so that receiving will start.
10185 if (need_reset && prog) {
10186 num_queues = min_t(int, adapter->num_rx_queues,
10187 adapter->num_xdp_queues);
10188 for (i = 0; i < num_queues; i++)
10189 if (adapter->xdp_ring[i]->xsk_pool)
10190 (void)ixgbe_xsk_wakeup(adapter->netdev, i,
10197 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10199 struct ixgbe_adapter *adapter = netdev_priv(dev);
10201 switch (xdp->command) {
10202 case XDP_SETUP_PROG:
10203 return ixgbe_xdp_setup(dev, xdp->prog);
10204 case XDP_SETUP_XSK_POOL:
10205 return ixgbe_xsk_pool_setup(adapter, xdp->xsk.pool,
10206 xdp->xsk.queue_id);
10213 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10215 /* Force memory writes to complete before letting h/w know there
10216 * are new descriptors to fetch.
10219 writel(ring->next_to_use, ring->tail);
10222 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10223 struct xdp_frame **frames, u32 flags)
10225 struct ixgbe_adapter *adapter = netdev_priv(dev);
10226 struct ixgbe_ring *ring;
10230 if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10233 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10236 /* During program transitions its possible adapter->xdp_prog is assigned
10237 * but ring has not been configured yet. In this case simply abort xmit.
10239 ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10240 if (unlikely(!ring))
10243 if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10246 for (i = 0; i < n; i++) {
10247 struct xdp_frame *xdpf = frames[i];
10250 err = ixgbe_xmit_xdp_ring(adapter, xdpf);
10251 if (err != IXGBE_XDP_TX) {
10252 xdp_return_frame_rx_napi(xdpf);
10257 if (unlikely(flags & XDP_XMIT_FLUSH))
10258 ixgbe_xdp_ring_update_tail(ring);
10263 static const struct net_device_ops ixgbe_netdev_ops = {
10264 .ndo_open = ixgbe_open,
10265 .ndo_stop = ixgbe_close,
10266 .ndo_start_xmit = ixgbe_xmit_frame,
10267 .ndo_set_rx_mode = ixgbe_set_rx_mode,
10268 .ndo_validate_addr = eth_validate_addr,
10269 .ndo_set_mac_address = ixgbe_set_mac,
10270 .ndo_change_mtu = ixgbe_change_mtu,
10271 .ndo_tx_timeout = ixgbe_tx_timeout,
10272 .ndo_set_tx_maxrate = ixgbe_tx_maxrate,
10273 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
10274 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
10275 .ndo_do_ioctl = ixgbe_ioctl,
10276 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
10277 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
10278 .ndo_set_vf_rate = ixgbe_ndo_set_vf_bw,
10279 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
10280 .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10281 .ndo_set_vf_trust = ixgbe_ndo_set_vf_trust,
10282 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
10283 .ndo_get_stats64 = ixgbe_get_stats64,
10284 .ndo_setup_tc = __ixgbe_setup_tc,
10286 .ndo_select_queue = ixgbe_select_queue,
10287 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10288 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10289 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10290 .ndo_fcoe_enable = ixgbe_fcoe_enable,
10291 .ndo_fcoe_disable = ixgbe_fcoe_disable,
10292 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10293 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10294 #endif /* IXGBE_FCOE */
10295 .ndo_set_features = ixgbe_set_features,
10296 .ndo_fix_features = ixgbe_fix_features,
10297 .ndo_fdb_add = ixgbe_ndo_fdb_add,
10298 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
10299 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
10300 .ndo_dfwd_add_station = ixgbe_fwd_add,
10301 .ndo_dfwd_del_station = ixgbe_fwd_del,
10302 .ndo_udp_tunnel_add = udp_tunnel_nic_add_port,
10303 .ndo_udp_tunnel_del = udp_tunnel_nic_del_port,
10304 .ndo_features_check = ixgbe_features_check,
10305 .ndo_bpf = ixgbe_xdp,
10306 .ndo_xdp_xmit = ixgbe_xdp_xmit,
10307 .ndo_xsk_wakeup = ixgbe_xsk_wakeup,
10310 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10311 struct ixgbe_ring *tx_ring)
10313 unsigned long wait_delay, delay_interval;
10314 struct ixgbe_hw *hw = &adapter->hw;
10315 u8 reg_idx = tx_ring->reg_idx;
10319 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10321 /* delay mechanism from ixgbe_disable_tx */
10322 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10324 wait_loop = IXGBE_MAX_RX_DESC_POLL;
10325 wait_delay = delay_interval;
10327 while (wait_loop--) {
10328 usleep_range(wait_delay, wait_delay + 10);
10329 wait_delay += delay_interval * 2;
10330 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10332 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10336 e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10339 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10340 struct ixgbe_ring *tx_ring)
10342 set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10343 ixgbe_disable_txr_hw(adapter, tx_ring);
10346 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10347 struct ixgbe_ring *rx_ring)
10349 unsigned long wait_delay, delay_interval;
10350 struct ixgbe_hw *hw = &adapter->hw;
10351 u8 reg_idx = rx_ring->reg_idx;
10355 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10356 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10357 rxdctl |= IXGBE_RXDCTL_SWFLSH;
10359 /* write value back with RXDCTL.ENABLE bit cleared */
10360 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10362 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10363 if (hw->mac.type == ixgbe_mac_82598EB &&
10364 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10367 /* delay mechanism from ixgbe_disable_rx */
10368 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10370 wait_loop = IXGBE_MAX_RX_DESC_POLL;
10371 wait_delay = delay_interval;
10373 while (wait_loop--) {
10374 usleep_range(wait_delay, wait_delay + 10);
10375 wait_delay += delay_interval * 2;
10376 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10378 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10382 e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10385 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10387 memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10388 memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10391 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10393 memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10394 memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10398 * ixgbe_irq_disable_single - Disable single IRQ vector
10399 * @adapter: adapter structure
10400 * @ring: ring index
10402 static void ixgbe_irq_disable_single(struct ixgbe_adapter *adapter, u32 ring)
10404 struct ixgbe_hw *hw = &adapter->hw;
10405 u64 qmask = BIT_ULL(ring);
10408 switch (adapter->hw.mac.type) {
10409 case ixgbe_mac_82598EB:
10410 mask = qmask & IXGBE_EIMC_RTX_QUEUE;
10411 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
10413 case ixgbe_mac_82599EB:
10414 case ixgbe_mac_X540:
10415 case ixgbe_mac_X550:
10416 case ixgbe_mac_X550EM_x:
10417 case ixgbe_mac_x550em_a:
10418 mask = (qmask & 0xFFFFFFFF);
10420 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
10421 mask = (qmask >> 32);
10423 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
10428 IXGBE_WRITE_FLUSH(&adapter->hw);
10429 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
10430 synchronize_irq(adapter->msix_entries[ring].vector);
10432 synchronize_irq(adapter->pdev->irq);
10436 * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10437 * @adapter: adapter structure
10438 * @ring: ring index
10440 * This function disables a certain Rx/Tx/XDP Tx ring. The function
10441 * assumes that the netdev is running.
10443 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10445 struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10447 rx_ring = adapter->rx_ring[ring];
10448 tx_ring = adapter->tx_ring[ring];
10449 xdp_ring = adapter->xdp_ring[ring];
10451 ixgbe_irq_disable_single(adapter, ring);
10453 /* Rx/Tx/XDP Tx share the same napi context. */
10454 napi_disable(&rx_ring->q_vector->napi);
10456 ixgbe_disable_txr(adapter, tx_ring);
10458 ixgbe_disable_txr(adapter, xdp_ring);
10459 ixgbe_disable_rxr_hw(adapter, rx_ring);
10464 ixgbe_clean_tx_ring(tx_ring);
10466 ixgbe_clean_tx_ring(xdp_ring);
10467 ixgbe_clean_rx_ring(rx_ring);
10469 ixgbe_reset_txr_stats(tx_ring);
10471 ixgbe_reset_txr_stats(xdp_ring);
10472 ixgbe_reset_rxr_stats(rx_ring);
10476 * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10477 * @adapter: adapter structure
10478 * @ring: ring index
10480 * This function enables a certain Rx/Tx/XDP Tx ring. The function
10481 * assumes that the netdev is running.
10483 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10485 struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10487 rx_ring = adapter->rx_ring[ring];
10488 tx_ring = adapter->tx_ring[ring];
10489 xdp_ring = adapter->xdp_ring[ring];
10491 ixgbe_configure_tx_ring(adapter, tx_ring);
10493 ixgbe_configure_tx_ring(adapter, xdp_ring);
10494 ixgbe_configure_rx_ring(adapter, rx_ring);
10496 clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10498 clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10500 /* Rx/Tx/XDP Tx share the same napi context. */
10501 napi_enable(&rx_ring->q_vector->napi);
10502 ixgbe_irq_enable_queues(adapter, BIT_ULL(ring));
10503 IXGBE_WRITE_FLUSH(&adapter->hw);
10507 * ixgbe_enumerate_functions - Get the number of ports this device has
10508 * @adapter: adapter structure
10510 * This function enumerates the phsyical functions co-located on a single slot,
10511 * in order to determine how many ports a device has. This is most useful in
10512 * determining the required GT/s of PCIe bandwidth necessary for optimal
10515 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10517 struct pci_dev *entry, *pdev = adapter->pdev;
10520 /* Some cards can not use the generic count PCIe functions method,
10521 * because they are behind a parent switch, so we hardcode these with
10522 * the correct number of functions.
10524 if (ixgbe_pcie_from_parent(&adapter->hw))
10527 list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10528 /* don't count virtual functions */
10529 if (entry->is_virtfn)
10532 /* When the devices on the bus don't all match our device ID,
10533 * we can't reliably determine the correct number of
10534 * functions. This can occur if a function has been direct
10535 * attached to a virtual machine using VT-d, for example. In
10536 * this case, simply return -1 to indicate this.
10538 if ((entry->vendor != pdev->vendor) ||
10539 (entry->device != pdev->device))
10549 * ixgbe_wol_supported - Check whether device supports WoL
10550 * @adapter: the adapter private structure
10551 * @device_id: the device ID
10552 * @subdevice_id: the subsystem device ID
10554 * This function is used by probe and ethtool to determine
10555 * which devices have WoL support
10558 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10561 struct ixgbe_hw *hw = &adapter->hw;
10562 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10564 /* WOL not supported on 82598 */
10565 if (hw->mac.type == ixgbe_mac_82598EB)
10568 /* check eeprom to see if WOL is enabled for X540 and newer */
10569 if (hw->mac.type >= ixgbe_mac_X540) {
10570 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10571 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10572 (hw->bus.func == 0)))
10576 /* WOL is determined based on device IDs for 82599 MACs */
10577 switch (device_id) {
10578 case IXGBE_DEV_ID_82599_SFP:
10579 /* Only these subdevices could supports WOL */
10580 switch (subdevice_id) {
10581 case IXGBE_SUBDEV_ID_82599_560FLR:
10582 case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10583 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10584 case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10585 /* only support first port */
10586 if (hw->bus.func != 0)
10589 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10590 case IXGBE_SUBDEV_ID_82599_SFP:
10591 case IXGBE_SUBDEV_ID_82599_RNDC:
10592 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10593 case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10594 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10595 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10599 case IXGBE_DEV_ID_82599EN_SFP:
10600 /* Only these subdevices support WOL */
10601 switch (subdevice_id) {
10602 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10606 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10607 /* All except this subdevice support WOL */
10608 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10611 case IXGBE_DEV_ID_82599_KX4:
10621 * ixgbe_set_fw_version - Set FW version
10622 * @adapter: the adapter private structure
10624 * This function is used by probe and ethtool to determine the FW version to
10625 * format to display. The FW version is taken from the EEPROM/NVM.
10627 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10629 struct ixgbe_hw *hw = &adapter->hw;
10630 struct ixgbe_nvm_version nvm_ver;
10632 ixgbe_get_oem_prod_version(hw, &nvm_ver);
10633 if (nvm_ver.oem_valid) {
10634 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10635 "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10636 nvm_ver.oem_release);
10640 ixgbe_get_etk_id(hw, &nvm_ver);
10641 ixgbe_get_orom_version(hw, &nvm_ver);
10643 if (nvm_ver.or_valid) {
10644 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10645 "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10646 nvm_ver.or_build, nvm_ver.or_patch);
10650 /* Set ETrack ID format */
10651 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10652 "0x%08x", nvm_ver.etk_id);
10656 * ixgbe_probe - Device Initialization Routine
10657 * @pdev: PCI device information struct
10658 * @ent: entry in ixgbe_pci_tbl
10660 * Returns 0 on success, negative on failure
10662 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10663 * The OS initialization, configuring of the adapter private structure,
10664 * and a hardware reset occur.
10666 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10668 struct net_device *netdev;
10669 struct ixgbe_adapter *adapter = NULL;
10670 struct ixgbe_hw *hw;
10671 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10672 int i, err, pci_using_dac, expected_gts;
10673 unsigned int indices = MAX_TX_QUEUES;
10674 u8 part_str[IXGBE_PBANUM_LENGTH];
10675 bool disable_dev = false;
10681 /* Catch broken hardware that put the wrong VF device ID in
10682 * the PCIe SR-IOV capability.
10684 if (pdev->is_virtfn) {
10685 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10686 pci_name(pdev), pdev->vendor, pdev->device);
10690 err = pci_enable_device_mem(pdev);
10694 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10697 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10699 dev_err(&pdev->dev,
10700 "No usable DMA configuration, aborting\n");
10706 err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10708 dev_err(&pdev->dev,
10709 "pci_request_selected_regions failed 0x%x\n", err);
10713 pci_enable_pcie_error_reporting(pdev);
10715 pci_set_master(pdev);
10716 pci_save_state(pdev);
10718 if (ii->mac == ixgbe_mac_82598EB) {
10719 #ifdef CONFIG_IXGBE_DCB
10720 /* 8 TC w/ 4 queues per TC */
10721 indices = 4 * MAX_TRAFFIC_CLASS;
10723 indices = IXGBE_MAX_RSS_INDICES;
10727 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10730 goto err_alloc_etherdev;
10733 SET_NETDEV_DEV(netdev, &pdev->dev);
10735 adapter = netdev_priv(netdev);
10737 adapter->netdev = netdev;
10738 adapter->pdev = pdev;
10740 hw->back = adapter;
10741 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10743 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10744 pci_resource_len(pdev, 0));
10745 adapter->io_addr = hw->hw_addr;
10746 if (!hw->hw_addr) {
10751 netdev->netdev_ops = &ixgbe_netdev_ops;
10752 ixgbe_set_ethtool_ops(netdev);
10753 netdev->watchdog_timeo = 5 * HZ;
10754 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10757 hw->mac.ops = *ii->mac_ops;
10758 hw->mac.type = ii->mac;
10759 hw->mvals = ii->mvals;
10761 hw->link.ops = *ii->link_ops;
10764 hw->eeprom.ops = *ii->eeprom_ops;
10765 eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10766 if (ixgbe_removed(hw->hw_addr)) {
10770 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10771 if (!(eec & BIT(8)))
10772 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10775 hw->phy.ops = *ii->phy_ops;
10776 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10777 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
10778 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10779 hw->phy.mdio.mmds = 0;
10780 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10781 hw->phy.mdio.dev = netdev;
10782 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10783 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10785 /* setup the private structure */
10786 err = ixgbe_sw_init(adapter, ii);
10790 switch (adapter->hw.mac.type) {
10791 case ixgbe_mac_X550:
10792 case ixgbe_mac_X550EM_x:
10793 netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550;
10795 case ixgbe_mac_x550em_a:
10796 netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550em_a;
10802 /* Make sure the SWFW semaphore is in a valid state */
10803 if (hw->mac.ops.init_swfw_sync)
10804 hw->mac.ops.init_swfw_sync(hw);
10806 /* Make it possible the adapter to be woken up via WOL */
10807 switch (adapter->hw.mac.type) {
10808 case ixgbe_mac_82599EB:
10809 case ixgbe_mac_X540:
10810 case ixgbe_mac_X550:
10811 case ixgbe_mac_X550EM_x:
10812 case ixgbe_mac_x550em_a:
10813 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10820 * If there is a fan on this device and it has failed log the
10823 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10824 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10825 if (esdp & IXGBE_ESDP_SDP1)
10826 e_crit(probe, "Fan has stopped, replace the adapter\n");
10829 if (allow_unsupported_sfp)
10830 hw->allow_unsupported_sfp = allow_unsupported_sfp;
10832 /* reset_hw fills in the perm_addr as well */
10833 hw->phy.reset_if_overtemp = true;
10834 err = hw->mac.ops.reset_hw(hw);
10835 hw->phy.reset_if_overtemp = false;
10836 ixgbe_set_eee_capable(adapter);
10837 if (err == -ENOENT) {
10839 } else if (err == -EOPNOTSUPP) {
10840 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10841 e_dev_err("Reload the driver after installing a supported module.\n");
10844 e_dev_err("HW Init failed: %d\n", err);
10848 #ifdef CONFIG_PCI_IOV
10849 /* SR-IOV not supported on the 82598 */
10850 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10853 ixgbe_init_mbx_params_pf(hw);
10854 hw->mbx.ops = ii->mbx_ops;
10855 pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10856 ixgbe_enable_sriov(adapter, max_vfs);
10860 netdev->features = NETIF_F_SG |
10867 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10868 NETIF_F_GSO_GRE_CSUM | \
10869 NETIF_F_GSO_IPXIP4 | \
10870 NETIF_F_GSO_IPXIP6 | \
10871 NETIF_F_GSO_UDP_TUNNEL | \
10872 NETIF_F_GSO_UDP_TUNNEL_CSUM)
10874 netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10875 netdev->features |= NETIF_F_GSO_PARTIAL |
10876 IXGBE_GSO_PARTIAL_FEATURES;
10878 if (hw->mac.type >= ixgbe_mac_82599EB)
10879 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
10881 #ifdef CONFIG_IXGBE_IPSEC
10882 #define IXGBE_ESP_FEATURES (NETIF_F_HW_ESP | \
10883 NETIF_F_HW_ESP_TX_CSUM | \
10886 if (adapter->ipsec)
10887 netdev->features |= IXGBE_ESP_FEATURES;
10889 /* copy netdev features into list of user selectable features */
10890 netdev->hw_features |= netdev->features |
10891 NETIF_F_HW_VLAN_CTAG_FILTER |
10892 NETIF_F_HW_VLAN_CTAG_RX |
10893 NETIF_F_HW_VLAN_CTAG_TX |
10895 NETIF_F_HW_L2FW_DOFFLOAD;
10897 if (hw->mac.type >= ixgbe_mac_82599EB)
10898 netdev->hw_features |= NETIF_F_NTUPLE |
10902 netdev->features |= NETIF_F_HIGHDMA;
10904 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10905 netdev->hw_enc_features |= netdev->vlan_features;
10906 netdev->mpls_features |= NETIF_F_SG |
10910 netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10912 /* set this bit last since it cannot be part of vlan_features */
10913 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10914 NETIF_F_HW_VLAN_CTAG_RX |
10915 NETIF_F_HW_VLAN_CTAG_TX;
10917 netdev->priv_flags |= IFF_UNICAST_FLT;
10918 netdev->priv_flags |= IFF_SUPP_NOFCS;
10920 /* MTU range: 68 - 9710 */
10921 netdev->min_mtu = ETH_MIN_MTU;
10922 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
10924 #ifdef CONFIG_IXGBE_DCB
10925 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
10926 netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
10930 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
10931 unsigned int fcoe_l;
10933 if (hw->mac.ops.get_device_caps) {
10934 hw->mac.ops.get_device_caps(hw, &device_caps);
10935 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
10936 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
10940 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
10941 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
10943 netdev->features |= NETIF_F_FSO |
10946 netdev->vlan_features |= NETIF_F_FSO |
10950 #endif /* IXGBE_FCOE */
10951 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
10952 netdev->hw_features |= NETIF_F_LRO;
10953 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
10954 netdev->features |= NETIF_F_LRO;
10956 if (ixgbe_check_fw_error(adapter)) {
10961 /* make sure the EEPROM is good */
10962 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
10963 e_dev_err("The EEPROM Checksum Is Not Valid\n");
10968 eth_platform_get_mac_address(&adapter->pdev->dev,
10969 adapter->hw.mac.perm_addr);
10971 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
10973 if (!is_valid_ether_addr(netdev->dev_addr)) {
10974 e_dev_err("invalid MAC address\n");
10979 /* Set hw->mac.addr to permanent MAC address */
10980 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
10981 ixgbe_mac_set_default_filter(adapter);
10983 timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
10985 if (ixgbe_removed(hw->hw_addr)) {
10989 INIT_WORK(&adapter->service_task, ixgbe_service_task);
10990 set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
10991 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
10993 err = ixgbe_init_interrupt_scheme(adapter);
10997 for (i = 0; i < adapter->num_rx_queues; i++)
10998 u64_stats_init(&adapter->rx_ring[i]->syncp);
10999 for (i = 0; i < adapter->num_tx_queues; i++)
11000 u64_stats_init(&adapter->tx_ring[i]->syncp);
11001 for (i = 0; i < adapter->num_xdp_queues; i++)
11002 u64_stats_init(&adapter->xdp_ring[i]->syncp);
11004 /* WOL not supported for all devices */
11006 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
11007 hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
11008 pdev->subsystem_device);
11009 if (hw->wol_enabled)
11010 adapter->wol = IXGBE_WUFC_MAG;
11012 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
11014 /* save off EEPROM version number */
11015 ixgbe_set_fw_version(adapter);
11017 /* pick up the PCI bus settings for reporting later */
11018 if (ixgbe_pcie_from_parent(hw))
11019 ixgbe_get_parent_bus_info(adapter);
11021 hw->mac.ops.get_bus_info(hw);
11023 /* calculate the expected PCIe bandwidth required for optimal
11024 * performance. Note that some older parts will never have enough
11025 * bandwidth due to being older generation PCIe parts. We clamp these
11026 * parts to ensure no warning is displayed if it can't be fixed.
11028 switch (hw->mac.type) {
11029 case ixgbe_mac_82598EB:
11030 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
11033 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
11037 /* don't check link if we failed to enumerate functions */
11038 if (expected_gts > 0)
11039 ixgbe_check_minimum_link(adapter, expected_gts);
11041 err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
11043 strlcpy(part_str, "Unknown", sizeof(part_str));
11044 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
11045 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
11046 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
11049 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
11050 hw->mac.type, hw->phy.type, part_str);
11052 e_dev_info("%pM\n", netdev->dev_addr);
11054 /* reset the hardware with the new settings */
11055 err = hw->mac.ops.start_hw(hw);
11056 if (err == -EACCES) {
11057 /* We are running on a pre-production device, log a warning */
11058 e_dev_warn("This device is a pre-production adapter/LOM. "
11059 "Please be aware there may be issues associated "
11060 "with your hardware. If you are experiencing "
11061 "problems please contact your Intel or hardware "
11062 "representative who provided you with this "
11065 strcpy(netdev->name, "eth%d");
11066 pci_set_drvdata(pdev, adapter);
11067 err = register_netdev(netdev);
11072 /* power down the optics for 82599 SFP+ fiber */
11073 if (hw->mac.ops.disable_tx_laser)
11074 hw->mac.ops.disable_tx_laser(hw);
11076 /* carrier off reporting is important to ethtool even BEFORE open */
11077 netif_carrier_off(netdev);
11079 #ifdef CONFIG_IXGBE_DCA
11080 if (dca_add_requester(&pdev->dev) == 0) {
11081 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11082 ixgbe_setup_dca(adapter);
11085 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11086 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11087 for (i = 0; i < adapter->num_vfs; i++)
11088 ixgbe_vf_configuration(pdev, (i | 0x10000000));
11091 /* firmware requires driver version to be 0xFFFFFFFF
11092 * since os does not support feature
11094 if (hw->mac.ops.set_fw_drv_ver)
11095 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11096 sizeof(UTS_RELEASE) - 1,
11099 /* add san mac addr to netdev */
11100 ixgbe_add_sanmac_netdev(netdev);
11102 e_dev_info("%s\n", ixgbe_default_device_descr);
11104 #ifdef CONFIG_IXGBE_HWMON
11105 if (ixgbe_sysfs_init(adapter))
11106 e_err(probe, "failed to allocate sysfs resources\n");
11107 #endif /* CONFIG_IXGBE_HWMON */
11109 ixgbe_dbg_adapter_init(adapter);
11111 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11112 if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11113 hw->mac.ops.setup_link(hw,
11114 IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11117 err = ixgbe_mii_bus_init(hw);
11124 unregister_netdev(netdev);
11126 ixgbe_release_hw_control(adapter);
11127 ixgbe_clear_interrupt_scheme(adapter);
11129 ixgbe_disable_sriov(adapter);
11130 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11131 iounmap(adapter->io_addr);
11132 kfree(adapter->jump_tables[0]);
11133 kfree(adapter->mac_table);
11134 kfree(adapter->rss_key);
11135 bitmap_free(adapter->af_xdp_zc_qps);
11137 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11138 free_netdev(netdev);
11139 err_alloc_etherdev:
11140 pci_disable_pcie_error_reporting(pdev);
11141 pci_release_mem_regions(pdev);
11144 if (!adapter || disable_dev)
11145 pci_disable_device(pdev);
11150 * ixgbe_remove - Device Removal Routine
11151 * @pdev: PCI device information struct
11153 * ixgbe_remove is called by the PCI subsystem to alert the driver
11154 * that it should release a PCI device. The could be caused by a
11155 * Hot-Plug event, or because the driver is going to be removed from
11158 static void ixgbe_remove(struct pci_dev *pdev)
11160 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11161 struct net_device *netdev;
11165 /* if !adapter then we already cleaned up in probe */
11169 netdev = adapter->netdev;
11170 ixgbe_dbg_adapter_exit(adapter);
11172 set_bit(__IXGBE_REMOVING, &adapter->state);
11173 cancel_work_sync(&adapter->service_task);
11175 if (adapter->mii_bus)
11176 mdiobus_unregister(adapter->mii_bus);
11178 #ifdef CONFIG_IXGBE_DCA
11179 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11180 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11181 dca_remove_requester(&pdev->dev);
11182 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11183 IXGBE_DCA_CTRL_DCA_DISABLE);
11187 #ifdef CONFIG_IXGBE_HWMON
11188 ixgbe_sysfs_exit(adapter);
11189 #endif /* CONFIG_IXGBE_HWMON */
11191 /* remove the added san mac */
11192 ixgbe_del_sanmac_netdev(netdev);
11194 #ifdef CONFIG_PCI_IOV
11195 ixgbe_disable_sriov(adapter);
11197 if (netdev->reg_state == NETREG_REGISTERED)
11198 unregister_netdev(netdev);
11200 ixgbe_stop_ipsec_offload(adapter);
11201 ixgbe_clear_interrupt_scheme(adapter);
11203 ixgbe_release_hw_control(adapter);
11206 kfree(adapter->ixgbe_ieee_pfc);
11207 kfree(adapter->ixgbe_ieee_ets);
11210 iounmap(adapter->io_addr);
11211 pci_release_mem_regions(pdev);
11213 e_dev_info("complete\n");
11215 for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11216 if (adapter->jump_tables[i]) {
11217 kfree(adapter->jump_tables[i]->input);
11218 kfree(adapter->jump_tables[i]->mask);
11220 kfree(adapter->jump_tables[i]);
11223 kfree(adapter->mac_table);
11224 kfree(adapter->rss_key);
11225 bitmap_free(adapter->af_xdp_zc_qps);
11226 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11227 free_netdev(netdev);
11229 pci_disable_pcie_error_reporting(pdev);
11232 pci_disable_device(pdev);
11236 * ixgbe_io_error_detected - called when PCI error is detected
11237 * @pdev: Pointer to PCI device
11238 * @state: The current pci connection state
11240 * This function is called after a PCI bus error affecting
11241 * this device has been detected.
11243 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11244 pci_channel_state_t state)
11246 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11247 struct net_device *netdev = adapter->netdev;
11249 #ifdef CONFIG_PCI_IOV
11250 struct ixgbe_hw *hw = &adapter->hw;
11251 struct pci_dev *bdev, *vfdev;
11252 u32 dw0, dw1, dw2, dw3;
11254 u16 req_id, pf_func;
11256 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11257 adapter->num_vfs == 0)
11258 goto skip_bad_vf_detection;
11260 bdev = pdev->bus->self;
11261 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11262 bdev = bdev->bus->self;
11265 goto skip_bad_vf_detection;
11267 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11269 goto skip_bad_vf_detection;
11271 dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11272 dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11273 dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11274 dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11275 if (ixgbe_removed(hw->hw_addr))
11276 goto skip_bad_vf_detection;
11278 req_id = dw1 >> 16;
11279 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11280 if (!(req_id & 0x0080))
11281 goto skip_bad_vf_detection;
11283 pf_func = req_id & 0x01;
11284 if ((pf_func & 1) == (pdev->devfn & 1)) {
11285 unsigned int device_id;
11287 vf = (req_id & 0x7F) >> 1;
11288 e_dev_err("VF %d has caused a PCIe error\n", vf);
11289 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11290 "%8.8x\tdw3: %8.8x\n",
11291 dw0, dw1, dw2, dw3);
11292 switch (adapter->hw.mac.type) {
11293 case ixgbe_mac_82599EB:
11294 device_id = IXGBE_82599_VF_DEVICE_ID;
11296 case ixgbe_mac_X540:
11297 device_id = IXGBE_X540_VF_DEVICE_ID;
11299 case ixgbe_mac_X550:
11300 device_id = IXGBE_DEV_ID_X550_VF;
11302 case ixgbe_mac_X550EM_x:
11303 device_id = IXGBE_DEV_ID_X550EM_X_VF;
11305 case ixgbe_mac_x550em_a:
11306 device_id = IXGBE_DEV_ID_X550EM_A_VF;
11313 /* Find the pci device of the offending VF */
11314 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11316 if (vfdev->devfn == (req_id & 0xFF))
11318 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11322 * There's a slim chance the VF could have been hot plugged,
11323 * so if it is no longer present we don't need to issue the
11324 * VFLR. Just clean up the AER in that case.
11328 /* Free device reference count */
11329 pci_dev_put(vfdev);
11334 * Even though the error may have occurred on the other port
11335 * we still need to increment the vf error reference count for
11336 * both ports because the I/O resume function will be called
11337 * for both of them.
11339 adapter->vferr_refcount++;
11341 return PCI_ERS_RESULT_RECOVERED;
11343 skip_bad_vf_detection:
11344 #endif /* CONFIG_PCI_IOV */
11345 if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11346 return PCI_ERS_RESULT_DISCONNECT;
11348 if (!netif_device_present(netdev))
11349 return PCI_ERS_RESULT_DISCONNECT;
11352 netif_device_detach(netdev);
11354 if (netif_running(netdev))
11355 ixgbe_close_suspend(adapter);
11357 if (state == pci_channel_io_perm_failure) {
11359 return PCI_ERS_RESULT_DISCONNECT;
11362 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11363 pci_disable_device(pdev);
11366 /* Request a slot reset. */
11367 return PCI_ERS_RESULT_NEED_RESET;
11371 * ixgbe_io_slot_reset - called after the pci bus has been reset.
11372 * @pdev: Pointer to PCI device
11374 * Restart the card from scratch, as if from a cold-boot.
11376 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11378 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11379 pci_ers_result_t result;
11381 if (pci_enable_device_mem(pdev)) {
11382 e_err(probe, "Cannot re-enable PCI device after reset.\n");
11383 result = PCI_ERS_RESULT_DISCONNECT;
11385 smp_mb__before_atomic();
11386 clear_bit(__IXGBE_DISABLED, &adapter->state);
11387 adapter->hw.hw_addr = adapter->io_addr;
11388 pci_set_master(pdev);
11389 pci_restore_state(pdev);
11390 pci_save_state(pdev);
11392 pci_wake_from_d3(pdev, false);
11394 ixgbe_reset(adapter);
11395 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11396 result = PCI_ERS_RESULT_RECOVERED;
11403 * ixgbe_io_resume - called when traffic can start flowing again.
11404 * @pdev: Pointer to PCI device
11406 * This callback is called when the error recovery driver tells us that
11407 * its OK to resume normal operation.
11409 static void ixgbe_io_resume(struct pci_dev *pdev)
11411 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11412 struct net_device *netdev = adapter->netdev;
11414 #ifdef CONFIG_PCI_IOV
11415 if (adapter->vferr_refcount) {
11416 e_info(drv, "Resuming after VF err\n");
11417 adapter->vferr_refcount--;
11423 if (netif_running(netdev))
11424 ixgbe_open(netdev);
11426 netif_device_attach(netdev);
11430 static const struct pci_error_handlers ixgbe_err_handler = {
11431 .error_detected = ixgbe_io_error_detected,
11432 .slot_reset = ixgbe_io_slot_reset,
11433 .resume = ixgbe_io_resume,
11436 static SIMPLE_DEV_PM_OPS(ixgbe_pm_ops, ixgbe_suspend, ixgbe_resume);
11438 static struct pci_driver ixgbe_driver = {
11439 .name = ixgbe_driver_name,
11440 .id_table = ixgbe_pci_tbl,
11441 .probe = ixgbe_probe,
11442 .remove = ixgbe_remove,
11443 .driver.pm = &ixgbe_pm_ops,
11444 .shutdown = ixgbe_shutdown,
11445 .sriov_configure = ixgbe_pci_sriov_configure,
11446 .err_handler = &ixgbe_err_handler
11450 * ixgbe_init_module - Driver Registration Routine
11452 * ixgbe_init_module is the first routine called when the driver is
11453 * loaded. All it does is register with the PCI subsystem.
11455 static int __init ixgbe_init_module(void)
11458 pr_info("%s\n", ixgbe_driver_string);
11459 pr_info("%s\n", ixgbe_copyright);
11461 ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11463 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11469 ret = pci_register_driver(&ixgbe_driver);
11471 destroy_workqueue(ixgbe_wq);
11476 #ifdef CONFIG_IXGBE_DCA
11477 dca_register_notify(&dca_notifier);
11483 module_init(ixgbe_init_module);
11486 * ixgbe_exit_module - Driver Exit Cleanup Routine
11488 * ixgbe_exit_module is called just before the driver is removed
11491 static void __exit ixgbe_exit_module(void)
11493 #ifdef CONFIG_IXGBE_DCA
11494 dca_unregister_notify(&dca_notifier);
11496 pci_unregister_driver(&ixgbe_driver);
11500 destroy_workqueue(ixgbe_wq);
11505 #ifdef CONFIG_IXGBE_DCA
11506 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11511 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11512 __ixgbe_notify_dca);
11514 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11517 #endif /* CONFIG_IXGBE_DCA */
11519 module_exit(ixgbe_exit_module);