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 offset = (unsigned long)(skb->data) & ~PAGE_MASK;
1830 dma_sync_single_range_for_cpu(rx_ring->dev,
1836 skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1838 dma_sync_single_range_for_cpu(rx_ring->dev,
1841 skb_frag_size(frag),
1845 /* If the page was released, just unmap it. */
1846 if (unlikely(IXGBE_CB(skb)->page_released)) {
1847 dma_unmap_page_attrs(rx_ring->dev, IXGBE_CB(skb)->dma,
1848 ixgbe_rx_pg_size(rx_ring),
1855 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1856 * @rx_ring: rx descriptor ring packet is being transacted on
1857 * @rx_desc: pointer to the EOP Rx descriptor
1858 * @skb: pointer to current skb being fixed
1860 * Check if the skb is valid in the XDP case it will be an error pointer.
1861 * Return true in this case to abort processing and advance to next
1864 * Check for corrupted packet headers caused by senders on the local L2
1865 * embedded NIC switch not setting up their Tx Descriptors right. These
1866 * should be very rare.
1868 * Also address the case where we are pulling data in on pages only
1869 * and as such no data is present in the skb header.
1871 * In addition if skb is not at least 60 bytes we need to pad it so that
1872 * it is large enough to qualify as a valid Ethernet frame.
1874 * Returns true if an error was encountered and skb was freed.
1876 bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1877 union ixgbe_adv_rx_desc *rx_desc,
1878 struct sk_buff *skb)
1880 struct net_device *netdev = rx_ring->netdev;
1882 /* XDP packets use error pointer so abort at this point */
1886 /* Verify netdev is present, and that packet does not have any
1887 * errors that would be unacceptable to the netdev.
1890 (unlikely(ixgbe_test_staterr(rx_desc,
1891 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1892 !(netdev->features & NETIF_F_RXALL)))) {
1893 dev_kfree_skb_any(skb);
1897 /* place header in linear portion of buffer */
1898 if (!skb_headlen(skb))
1899 ixgbe_pull_tail(rx_ring, skb);
1902 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1903 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1907 /* if eth_skb_pad returns an error the skb was freed */
1908 if (eth_skb_pad(skb))
1915 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1916 * @rx_ring: rx descriptor ring to store buffers on
1917 * @old_buff: donor buffer to have page reused
1919 * Synchronizes page for reuse by the adapter
1921 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1922 struct ixgbe_rx_buffer *old_buff)
1924 struct ixgbe_rx_buffer *new_buff;
1925 u16 nta = rx_ring->next_to_alloc;
1927 new_buff = &rx_ring->rx_buffer_info[nta];
1929 /* update, and store next to alloc */
1931 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1933 /* Transfer page from old buffer to new buffer.
1934 * Move each member individually to avoid possible store
1935 * forwarding stalls and unnecessary copy of skb.
1937 new_buff->dma = old_buff->dma;
1938 new_buff->page = old_buff->page;
1939 new_buff->page_offset = old_buff->page_offset;
1940 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
1943 static inline bool ixgbe_page_is_reserved(struct page *page)
1945 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1948 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer)
1950 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1951 struct page *page = rx_buffer->page;
1953 /* avoid re-using remote pages */
1954 if (unlikely(ixgbe_page_is_reserved(page)))
1957 #if (PAGE_SIZE < 8192)
1958 /* if we are only owner of page we can reuse it */
1959 if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
1962 /* The last offset is a bit aggressive in that we assume the
1963 * worst case of FCoE being enabled and using a 3K buffer.
1964 * However this should have minimal impact as the 1K extra is
1965 * still less than one buffer in size.
1967 #define IXGBE_LAST_OFFSET \
1968 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1969 if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1973 /* If we have drained the page fragment pool we need to update
1974 * the pagecnt_bias and page count so that we fully restock the
1975 * number of references the driver holds.
1977 if (unlikely(pagecnt_bias == 1)) {
1978 page_ref_add(page, USHRT_MAX - 1);
1979 rx_buffer->pagecnt_bias = USHRT_MAX;
1986 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1987 * @rx_ring: rx descriptor ring to transact packets on
1988 * @rx_buffer: buffer containing page to add
1989 * @skb: sk_buff to place the data into
1990 * @size: size of data in rx_buffer
1992 * This function will add the data contained in rx_buffer->page to the skb.
1993 * This is done either through a direct copy if the data in the buffer is
1994 * less than the skb header size, otherwise it will just attach the page as
1995 * a frag to the skb.
1997 * The function will then update the page offset if necessary and return
1998 * true if the buffer can be reused by the adapter.
2000 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
2001 struct ixgbe_rx_buffer *rx_buffer,
2002 struct sk_buff *skb,
2005 #if (PAGE_SIZE < 8192)
2006 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2008 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2009 SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2010 SKB_DATA_ALIGN(size);
2012 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2013 rx_buffer->page_offset, size, truesize);
2014 #if (PAGE_SIZE < 8192)
2015 rx_buffer->page_offset ^= truesize;
2017 rx_buffer->page_offset += truesize;
2021 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2022 union ixgbe_adv_rx_desc *rx_desc,
2023 struct sk_buff **skb,
2024 const unsigned int size)
2026 struct ixgbe_rx_buffer *rx_buffer;
2028 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2029 prefetchw(rx_buffer->page);
2030 *skb = rx_buffer->skb;
2032 /* Delay unmapping of the first packet. It carries the header
2033 * information, HW may still access the header after the writeback.
2034 * Only unmap it when EOP is reached
2036 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2041 ixgbe_dma_sync_frag(rx_ring, *skb);
2044 /* we are reusing so sync this buffer for CPU use */
2045 dma_sync_single_range_for_cpu(rx_ring->dev,
2047 rx_buffer->page_offset,
2051 rx_buffer->pagecnt_bias--;
2056 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2057 struct ixgbe_rx_buffer *rx_buffer,
2058 struct sk_buff *skb)
2060 if (ixgbe_can_reuse_rx_page(rx_buffer)) {
2061 /* hand second half of page back to the ring */
2062 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2064 if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2065 /* the page has been released from the ring */
2066 IXGBE_CB(skb)->page_released = true;
2068 /* we are not reusing the buffer so unmap it */
2069 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2070 ixgbe_rx_pg_size(rx_ring),
2074 __page_frag_cache_drain(rx_buffer->page,
2075 rx_buffer->pagecnt_bias);
2078 /* clear contents of rx_buffer */
2079 rx_buffer->page = NULL;
2080 rx_buffer->skb = NULL;
2083 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2084 struct ixgbe_rx_buffer *rx_buffer,
2085 struct xdp_buff *xdp,
2086 union ixgbe_adv_rx_desc *rx_desc)
2088 unsigned int size = xdp->data_end - xdp->data;
2089 #if (PAGE_SIZE < 8192)
2090 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2092 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2093 xdp->data_hard_start);
2095 struct sk_buff *skb;
2097 /* prefetch first cache line of first page */
2098 prefetch(xdp->data);
2099 #if L1_CACHE_BYTES < 128
2100 prefetch(xdp->data + L1_CACHE_BYTES);
2102 /* Note, we get here by enabling legacy-rx via:
2104 * ethtool --set-priv-flags <dev> legacy-rx on
2106 * In this mode, we currently get 0 extra XDP headroom as
2107 * opposed to having legacy-rx off, where we process XDP
2108 * packets going to stack via ixgbe_build_skb(). The latter
2109 * provides us currently with 192 bytes of headroom.
2111 * For ixgbe_construct_skb() mode it means that the
2112 * xdp->data_meta will always point to xdp->data, since
2113 * the helper cannot expand the head. Should this ever
2114 * change in future for legacy-rx mode on, then lets also
2115 * add xdp->data_meta handling here.
2118 /* allocate a skb to store the frags */
2119 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2123 if (size > IXGBE_RX_HDR_SIZE) {
2124 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2125 IXGBE_CB(skb)->dma = rx_buffer->dma;
2127 skb_add_rx_frag(skb, 0, rx_buffer->page,
2128 xdp->data - page_address(rx_buffer->page),
2130 #if (PAGE_SIZE < 8192)
2131 rx_buffer->page_offset ^= truesize;
2133 rx_buffer->page_offset += truesize;
2136 memcpy(__skb_put(skb, size),
2137 xdp->data, ALIGN(size, sizeof(long)));
2138 rx_buffer->pagecnt_bias++;
2144 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2145 struct ixgbe_rx_buffer *rx_buffer,
2146 struct xdp_buff *xdp,
2147 union ixgbe_adv_rx_desc *rx_desc)
2149 unsigned int metasize = xdp->data - xdp->data_meta;
2150 #if (PAGE_SIZE < 8192)
2151 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2153 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2154 SKB_DATA_ALIGN(xdp->data_end -
2155 xdp->data_hard_start);
2157 struct sk_buff *skb;
2159 /* Prefetch first cache line of first page. If xdp->data_meta
2160 * is unused, this points extactly as xdp->data, otherwise we
2161 * likely have a consumer accessing first few bytes of meta
2162 * data, and then actual data.
2164 prefetch(xdp->data_meta);
2165 #if L1_CACHE_BYTES < 128
2166 prefetch(xdp->data_meta + L1_CACHE_BYTES);
2169 /* build an skb to around the page buffer */
2170 skb = build_skb(xdp->data_hard_start, truesize);
2174 /* update pointers within the skb to store the data */
2175 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2176 __skb_put(skb, xdp->data_end - xdp->data);
2178 skb_metadata_set(skb, metasize);
2180 /* record DMA address if this is the start of a chain of buffers */
2181 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2182 IXGBE_CB(skb)->dma = rx_buffer->dma;
2184 /* update buffer offset */
2185 #if (PAGE_SIZE < 8192)
2186 rx_buffer->page_offset ^= truesize;
2188 rx_buffer->page_offset += truesize;
2194 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2195 struct ixgbe_ring *rx_ring,
2196 struct xdp_buff *xdp)
2198 int err, result = IXGBE_XDP_PASS;
2199 struct bpf_prog *xdp_prog;
2200 struct xdp_frame *xdpf;
2204 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2209 prefetchw(xdp->data_hard_start); /* xdp_frame write */
2211 act = bpf_prog_run_xdp(xdp_prog, xdp);
2216 xdpf = xdp_convert_buff_to_frame(xdp);
2217 if (unlikely(!xdpf)) {
2218 result = IXGBE_XDP_CONSUMED;
2221 result = ixgbe_xmit_xdp_ring(adapter, xdpf);
2224 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2226 result = IXGBE_XDP_REDIR;
2228 result = IXGBE_XDP_CONSUMED;
2231 bpf_warn_invalid_xdp_action(act);
2234 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2235 fallthrough; /* handle aborts by dropping packet */
2237 result = IXGBE_XDP_CONSUMED;
2242 return ERR_PTR(-result);
2245 static unsigned int ixgbe_rx_frame_truesize(struct ixgbe_ring *rx_ring,
2248 unsigned int truesize;
2250 #if (PAGE_SIZE < 8192)
2251 truesize = ixgbe_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
2253 truesize = ring_uses_build_skb(rx_ring) ?
2254 SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) +
2255 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
2256 SKB_DATA_ALIGN(size);
2261 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2262 struct ixgbe_rx_buffer *rx_buffer,
2265 unsigned int truesize = ixgbe_rx_frame_truesize(rx_ring, size);
2266 #if (PAGE_SIZE < 8192)
2267 rx_buffer->page_offset ^= truesize;
2269 rx_buffer->page_offset += truesize;
2274 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2275 * @q_vector: structure containing interrupt and ring information
2276 * @rx_ring: rx descriptor ring to transact packets on
2277 * @budget: Total limit on number of packets to process
2279 * This function provides a "bounce buffer" approach to Rx interrupt
2280 * processing. The advantage to this is that on systems that have
2281 * expensive overhead for IOMMU access this provides a means of avoiding
2282 * it by maintaining the mapping of the page to the syste.
2284 * Returns amount of work completed
2286 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2287 struct ixgbe_ring *rx_ring,
2290 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2291 struct ixgbe_adapter *adapter = q_vector->adapter;
2294 unsigned int mss = 0;
2295 #endif /* IXGBE_FCOE */
2296 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2297 unsigned int xdp_xmit = 0;
2298 struct xdp_buff xdp;
2300 xdp.rxq = &rx_ring->xdp_rxq;
2302 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
2303 #if (PAGE_SIZE < 8192)
2304 xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, 0);
2307 while (likely(total_rx_packets < budget)) {
2308 union ixgbe_adv_rx_desc *rx_desc;
2309 struct ixgbe_rx_buffer *rx_buffer;
2310 struct sk_buff *skb;
2313 /* return some buffers to hardware, one at a time is too slow */
2314 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2315 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2319 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2320 size = le16_to_cpu(rx_desc->wb.upper.length);
2324 /* This memory barrier is needed to keep us from reading
2325 * any other fields out of the rx_desc until we know the
2326 * descriptor has been written back
2330 rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size);
2332 /* retrieve a buffer from the ring */
2334 xdp.data = page_address(rx_buffer->page) +
2335 rx_buffer->page_offset;
2336 xdp.data_meta = xdp.data;
2337 xdp.data_hard_start = xdp.data -
2338 ixgbe_rx_offset(rx_ring);
2339 xdp.data_end = xdp.data + size;
2340 #if (PAGE_SIZE > 4096)
2341 /* At larger PAGE_SIZE, frame_sz depend on len size */
2342 xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, size);
2344 skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2348 unsigned int xdp_res = -PTR_ERR(skb);
2350 if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2351 xdp_xmit |= xdp_res;
2352 ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2354 rx_buffer->pagecnt_bias++;
2357 total_rx_bytes += size;
2359 ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2360 } else if (ring_uses_build_skb(rx_ring)) {
2361 skb = ixgbe_build_skb(rx_ring, rx_buffer,
2364 skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2368 /* exit if we failed to retrieve a buffer */
2370 rx_ring->rx_stats.alloc_rx_buff_failed++;
2371 rx_buffer->pagecnt_bias++;
2375 ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb);
2378 /* place incomplete frames back on ring for completion */
2379 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2382 /* verify the packet layout is correct */
2383 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2386 /* probably a little skewed due to removing CRC */
2387 total_rx_bytes += skb->len;
2389 /* populate checksum, timestamp, VLAN, and protocol */
2390 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2393 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2394 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2395 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2396 /* include DDPed FCoE data */
2397 if (ddp_bytes > 0) {
2399 mss = rx_ring->netdev->mtu -
2400 sizeof(struct fcoe_hdr) -
2401 sizeof(struct fc_frame_header) -
2402 sizeof(struct fcoe_crc_eof);
2406 total_rx_bytes += ddp_bytes;
2407 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2411 dev_kfree_skb_any(skb);
2416 #endif /* IXGBE_FCOE */
2417 ixgbe_rx_skb(q_vector, skb);
2419 /* update budget accounting */
2423 if (xdp_xmit & IXGBE_XDP_REDIR)
2426 if (xdp_xmit & IXGBE_XDP_TX) {
2427 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2429 /* Force memory writes to complete before letting h/w
2430 * know there are new descriptors to fetch.
2433 writel(ring->next_to_use, ring->tail);
2436 u64_stats_update_begin(&rx_ring->syncp);
2437 rx_ring->stats.packets += total_rx_packets;
2438 rx_ring->stats.bytes += total_rx_bytes;
2439 u64_stats_update_end(&rx_ring->syncp);
2440 q_vector->rx.total_packets += total_rx_packets;
2441 q_vector->rx.total_bytes += total_rx_bytes;
2443 return total_rx_packets;
2447 * ixgbe_configure_msix - Configure MSI-X hardware
2448 * @adapter: board private structure
2450 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2453 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2455 struct ixgbe_q_vector *q_vector;
2459 /* Populate MSIX to EITR Select */
2460 if (adapter->num_vfs > 32) {
2461 u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2462 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2466 * Populate the IVAR table and set the ITR values to the
2467 * corresponding register.
2469 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2470 struct ixgbe_ring *ring;
2471 q_vector = adapter->q_vector[v_idx];
2473 ixgbe_for_each_ring(ring, q_vector->rx)
2474 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2476 ixgbe_for_each_ring(ring, q_vector->tx)
2477 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2479 ixgbe_write_eitr(q_vector);
2482 switch (adapter->hw.mac.type) {
2483 case ixgbe_mac_82598EB:
2484 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2487 case ixgbe_mac_82599EB:
2488 case ixgbe_mac_X540:
2489 case ixgbe_mac_X550:
2490 case ixgbe_mac_X550EM_x:
2491 case ixgbe_mac_x550em_a:
2492 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2497 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2499 /* set up to autoclear timer, and the vectors */
2500 mask = IXGBE_EIMS_ENABLE_MASK;
2501 mask &= ~(IXGBE_EIMS_OTHER |
2502 IXGBE_EIMS_MAILBOX |
2505 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2509 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2510 * @q_vector: structure containing interrupt and ring information
2511 * @ring_container: structure containing ring performance data
2513 * Stores a new ITR value based on packets and byte
2514 * counts during the last interrupt. The advantage of per interrupt
2515 * computation is faster updates and more accurate ITR for the current
2516 * traffic pattern. Constants in this function were computed
2517 * based on theoretical maximum wire speed and thresholds were set based
2518 * on testing data as well as attempting to minimize response time
2519 * while increasing bulk throughput.
2521 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2522 struct ixgbe_ring_container *ring_container)
2524 unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2525 IXGBE_ITR_ADAPTIVE_LATENCY;
2526 unsigned int avg_wire_size, packets, bytes;
2527 unsigned long next_update = jiffies;
2529 /* If we don't have any rings just leave ourselves set for maximum
2530 * possible latency so we take ourselves out of the equation.
2532 if (!ring_container->ring)
2535 /* If we didn't update within up to 1 - 2 jiffies we can assume
2536 * that either packets are coming in so slow there hasn't been
2537 * any work, or that there is so much work that NAPI is dealing
2538 * with interrupt moderation and we don't need to do anything.
2540 if (time_after(next_update, ring_container->next_update))
2543 packets = ring_container->total_packets;
2545 /* We have no packets to actually measure against. This means
2546 * either one of the other queues on this vector is active or
2547 * we are a Tx queue doing TSO with too high of an interrupt rate.
2549 * When this occurs just tick up our delay by the minimum value
2550 * and hope that this extra delay will prevent us from being called
2551 * without any work on our queue.
2554 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2555 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2556 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2557 itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2561 bytes = ring_container->total_bytes;
2563 /* If packets are less than 4 or bytes are less than 9000 assume
2564 * insufficient data to use bulk rate limiting approach. We are
2565 * likely latency driven.
2567 if (packets < 4 && bytes < 9000) {
2568 itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2569 goto adjust_by_size;
2572 /* Between 4 and 48 we can assume that our current interrupt delay
2573 * is only slightly too low. As such we should increase it by a small
2577 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2578 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2579 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2583 /* Between 48 and 96 is our "goldilocks" zone where we are working
2584 * out "just right". Just report that our current ITR is good for us.
2587 itr = q_vector->itr >> 2;
2591 /* If packet count is 96 or greater we are likely looking at a slight
2592 * overrun of the delay we want. Try halving our delay to see if that
2593 * will cut the number of packets in half per interrupt.
2595 if (packets < 256) {
2596 itr = q_vector->itr >> 3;
2597 if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2598 itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2602 /* The paths below assume we are dealing with a bulk ITR since number
2603 * of packets is 256 or greater. We are just going to have to compute
2604 * a value and try to bring the count under control, though for smaller
2605 * packet sizes there isn't much we can do as NAPI polling will likely
2606 * be kicking in sooner rather than later.
2608 itr = IXGBE_ITR_ADAPTIVE_BULK;
2611 /* If packet counts are 256 or greater we can assume we have a gross
2612 * overestimation of what the rate should be. Instead of trying to fine
2613 * tune it just use the formula below to try and dial in an exact value
2614 * give the current packet size of the frame.
2616 avg_wire_size = bytes / packets;
2618 /* The following is a crude approximation of:
2619 * wmem_default / (size + overhead) = desired_pkts_per_int
2620 * rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2621 * (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2623 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2624 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2627 * (170 * (size + 24)) / (size + 640) = ITR
2629 * We first do some math on the packet size and then finally bitshift
2630 * by 8 after rounding up. We also have to account for PCIe link speed
2631 * difference as ITR scales based on this.
2633 if (avg_wire_size <= 60) {
2634 /* Start at 50k ints/sec */
2635 avg_wire_size = 5120;
2636 } else if (avg_wire_size <= 316) {
2637 /* 50K ints/sec to 16K ints/sec */
2638 avg_wire_size *= 40;
2639 avg_wire_size += 2720;
2640 } else if (avg_wire_size <= 1084) {
2641 /* 16K ints/sec to 9.2K ints/sec */
2642 avg_wire_size *= 15;
2643 avg_wire_size += 11452;
2644 } else if (avg_wire_size < 1968) {
2645 /* 9.2K ints/sec to 8K ints/sec */
2647 avg_wire_size += 22420;
2649 /* plateau at a limit of 8K ints/sec */
2650 avg_wire_size = 32256;
2653 /* If we are in low latency mode half our delay which doubles the rate
2654 * to somewhere between 100K to 16K ints/sec
2656 if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2657 avg_wire_size >>= 1;
2659 /* Resultant value is 256 times larger than it needs to be. This
2660 * gives us room to adjust the value as needed to either increase
2661 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2663 * Use addition as we have already recorded the new latency flag
2664 * for the ITR value.
2666 switch (q_vector->adapter->link_speed) {
2667 case IXGBE_LINK_SPEED_10GB_FULL:
2668 case IXGBE_LINK_SPEED_100_FULL:
2670 itr += DIV_ROUND_UP(avg_wire_size,
2671 IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2672 IXGBE_ITR_ADAPTIVE_MIN_INC;
2674 case IXGBE_LINK_SPEED_2_5GB_FULL:
2675 case IXGBE_LINK_SPEED_1GB_FULL:
2676 case IXGBE_LINK_SPEED_10_FULL:
2677 if (avg_wire_size > 8064)
2678 avg_wire_size = 8064;
2679 itr += DIV_ROUND_UP(avg_wire_size,
2680 IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2681 IXGBE_ITR_ADAPTIVE_MIN_INC;
2686 /* write back value */
2687 ring_container->itr = itr;
2689 /* next update should occur within next jiffy */
2690 ring_container->next_update = next_update + 1;
2692 ring_container->total_bytes = 0;
2693 ring_container->total_packets = 0;
2697 * ixgbe_write_eitr - write EITR register in hardware specific way
2698 * @q_vector: structure containing interrupt and ring information
2700 * This function is made to be called by ethtool and by the driver
2701 * when it needs to update EITR registers at runtime. Hardware
2702 * specific quirks/differences are taken care of here.
2704 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2706 struct ixgbe_adapter *adapter = q_vector->adapter;
2707 struct ixgbe_hw *hw = &adapter->hw;
2708 int v_idx = q_vector->v_idx;
2709 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2711 switch (adapter->hw.mac.type) {
2712 case ixgbe_mac_82598EB:
2713 /* must write high and low 16 bits to reset counter */
2714 itr_reg |= (itr_reg << 16);
2716 case ixgbe_mac_82599EB:
2717 case ixgbe_mac_X540:
2718 case ixgbe_mac_X550:
2719 case ixgbe_mac_X550EM_x:
2720 case ixgbe_mac_x550em_a:
2722 * set the WDIS bit to not clear the timer bits and cause an
2723 * immediate assertion of the interrupt
2725 itr_reg |= IXGBE_EITR_CNT_WDIS;
2730 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2733 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2737 ixgbe_update_itr(q_vector, &q_vector->tx);
2738 ixgbe_update_itr(q_vector, &q_vector->rx);
2740 /* use the smallest value of new ITR delay calculations */
2741 new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2743 /* Clear latency flag if set, shift into correct position */
2744 new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2747 if (new_itr != q_vector->itr) {
2748 /* save the algorithm value here */
2749 q_vector->itr = new_itr;
2751 ixgbe_write_eitr(q_vector);
2756 * ixgbe_check_overtemp_subtask - check for over temperature
2757 * @adapter: pointer to adapter
2759 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2761 struct ixgbe_hw *hw = &adapter->hw;
2762 u32 eicr = adapter->interrupt_event;
2765 if (test_bit(__IXGBE_DOWN, &adapter->state))
2768 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2771 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2773 switch (hw->device_id) {
2774 case IXGBE_DEV_ID_82599_T3_LOM:
2776 * Since the warning interrupt is for both ports
2777 * we don't have to check if:
2778 * - This interrupt wasn't for our port.
2779 * - We may have missed the interrupt so always have to
2780 * check if we got a LSC
2782 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2783 !(eicr & IXGBE_EICR_LSC))
2786 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2788 bool link_up = false;
2790 hw->mac.ops.check_link(hw, &speed, &link_up, false);
2796 /* Check if this is not due to overtemp */
2797 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2801 case IXGBE_DEV_ID_X550EM_A_1G_T:
2802 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2803 rc = hw->phy.ops.check_overtemp(hw);
2804 if (rc != IXGBE_ERR_OVERTEMP)
2808 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2810 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2814 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2816 adapter->interrupt_event = 0;
2819 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2821 struct ixgbe_hw *hw = &adapter->hw;
2823 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2824 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2825 e_crit(probe, "Fan has stopped, replace the adapter\n");
2826 /* write to clear the interrupt */
2827 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2831 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2833 struct ixgbe_hw *hw = &adapter->hw;
2835 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2838 switch (adapter->hw.mac.type) {
2839 case ixgbe_mac_82599EB:
2841 * Need to check link state so complete overtemp check
2844 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2845 (eicr & IXGBE_EICR_LSC)) &&
2846 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2847 adapter->interrupt_event = eicr;
2848 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2849 ixgbe_service_event_schedule(adapter);
2853 case ixgbe_mac_x550em_a:
2854 if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2855 adapter->interrupt_event = eicr;
2856 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2857 ixgbe_service_event_schedule(adapter);
2858 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2859 IXGBE_EICR_GPI_SDP0_X550EM_a);
2860 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2861 IXGBE_EICR_GPI_SDP0_X550EM_a);
2864 case ixgbe_mac_X550:
2865 case ixgbe_mac_X540:
2866 if (!(eicr & IXGBE_EICR_TS))
2873 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2876 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2878 switch (hw->mac.type) {
2879 case ixgbe_mac_82598EB:
2880 if (hw->phy.type == ixgbe_phy_nl)
2883 case ixgbe_mac_82599EB:
2884 case ixgbe_mac_X550EM_x:
2885 case ixgbe_mac_x550em_a:
2886 switch (hw->mac.ops.get_media_type(hw)) {
2887 case ixgbe_media_type_fiber:
2888 case ixgbe_media_type_fiber_qsfp:
2898 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2900 struct ixgbe_hw *hw = &adapter->hw;
2901 u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2903 if (!ixgbe_is_sfp(hw))
2906 /* Later MAC's use different SDP */
2907 if (hw->mac.type >= ixgbe_mac_X540)
2908 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2910 if (eicr & eicr_mask) {
2911 /* Clear the interrupt */
2912 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2913 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2914 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2915 adapter->sfp_poll_time = 0;
2916 ixgbe_service_event_schedule(adapter);
2920 if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2921 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2922 /* Clear the interrupt */
2923 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2924 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2925 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2926 ixgbe_service_event_schedule(adapter);
2931 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2933 struct ixgbe_hw *hw = &adapter->hw;
2936 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2937 adapter->link_check_timeout = jiffies;
2938 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2939 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2940 IXGBE_WRITE_FLUSH(hw);
2941 ixgbe_service_event_schedule(adapter);
2945 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2949 struct ixgbe_hw *hw = &adapter->hw;
2951 switch (hw->mac.type) {
2952 case ixgbe_mac_82598EB:
2953 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2954 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2956 case ixgbe_mac_82599EB:
2957 case ixgbe_mac_X540:
2958 case ixgbe_mac_X550:
2959 case ixgbe_mac_X550EM_x:
2960 case ixgbe_mac_x550em_a:
2961 mask = (qmask & 0xFFFFFFFF);
2963 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2964 mask = (qmask >> 32);
2966 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2971 /* skip the flush */
2975 * ixgbe_irq_enable - Enable default interrupt generation settings
2976 * @adapter: board private structure
2977 * @queues: enable irqs for queues
2978 * @flush: flush register write
2980 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2983 struct ixgbe_hw *hw = &adapter->hw;
2984 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2986 /* don't reenable LSC while waiting for link */
2987 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2988 mask &= ~IXGBE_EIMS_LSC;
2990 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2991 switch (adapter->hw.mac.type) {
2992 case ixgbe_mac_82599EB:
2993 mask |= IXGBE_EIMS_GPI_SDP0(hw);
2995 case ixgbe_mac_X540:
2996 case ixgbe_mac_X550:
2997 case ixgbe_mac_X550EM_x:
2998 case ixgbe_mac_x550em_a:
2999 mask |= IXGBE_EIMS_TS;
3004 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3005 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3006 switch (adapter->hw.mac.type) {
3007 case ixgbe_mac_82599EB:
3008 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3009 mask |= IXGBE_EIMS_GPI_SDP2(hw);
3011 case ixgbe_mac_X540:
3012 case ixgbe_mac_X550:
3013 case ixgbe_mac_X550EM_x:
3014 case ixgbe_mac_x550em_a:
3015 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3016 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3017 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3018 mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3019 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3020 mask |= IXGBE_EICR_GPI_SDP0_X540;
3021 mask |= IXGBE_EIMS_ECC;
3022 mask |= IXGBE_EIMS_MAILBOX;
3028 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3029 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3030 mask |= IXGBE_EIMS_FLOW_DIR;
3032 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3034 ixgbe_irq_enable_queues(adapter, ~0);
3036 IXGBE_WRITE_FLUSH(&adapter->hw);
3039 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3041 struct ixgbe_adapter *adapter = data;
3042 struct ixgbe_hw *hw = &adapter->hw;
3046 * Workaround for Silicon errata. Use clear-by-write instead
3047 * of clear-by-read. Reading with EICS will return the
3048 * interrupt causes without clearing, which later be done
3049 * with the write to EICR.
3051 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3053 /* The lower 16bits of the EICR register are for the queue interrupts
3054 * which should be masked here in order to not accidentally clear them if
3055 * the bits are high when ixgbe_msix_other is called. There is a race
3056 * condition otherwise which results in possible performance loss
3057 * especially if the ixgbe_msix_other interrupt is triggering
3058 * consistently (as it would when PPS is turned on for the X540 device)
3062 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3064 if (eicr & IXGBE_EICR_LSC)
3065 ixgbe_check_lsc(adapter);
3067 if (eicr & IXGBE_EICR_MAILBOX)
3068 ixgbe_msg_task(adapter);
3070 switch (hw->mac.type) {
3071 case ixgbe_mac_82599EB:
3072 case ixgbe_mac_X540:
3073 case ixgbe_mac_X550:
3074 case ixgbe_mac_X550EM_x:
3075 case ixgbe_mac_x550em_a:
3076 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3077 (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3078 adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3079 ixgbe_service_event_schedule(adapter);
3080 IXGBE_WRITE_REG(hw, IXGBE_EICR,
3081 IXGBE_EICR_GPI_SDP0_X540);
3083 if (eicr & IXGBE_EICR_ECC) {
3084 e_info(link, "Received ECC Err, initiating reset\n");
3085 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3086 ixgbe_service_event_schedule(adapter);
3087 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3089 /* Handle Flow Director Full threshold interrupt */
3090 if (eicr & IXGBE_EICR_FLOW_DIR) {
3091 int reinit_count = 0;
3093 for (i = 0; i < adapter->num_tx_queues; i++) {
3094 struct ixgbe_ring *ring = adapter->tx_ring[i];
3095 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3100 /* no more flow director interrupts until after init */
3101 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3102 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3103 ixgbe_service_event_schedule(adapter);
3106 ixgbe_check_sfp_event(adapter, eicr);
3107 ixgbe_check_overtemp_event(adapter, eicr);
3113 ixgbe_check_fan_failure(adapter, eicr);
3115 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3116 ixgbe_ptp_check_pps_event(adapter);
3118 /* re-enable the original interrupt state, no lsc, no queues */
3119 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3120 ixgbe_irq_enable(adapter, false, false);
3125 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3127 struct ixgbe_q_vector *q_vector = data;
3129 /* EIAM disabled interrupts (on this vector) for us */
3131 if (q_vector->rx.ring || q_vector->tx.ring)
3132 napi_schedule_irqoff(&q_vector->napi);
3138 * ixgbe_poll - NAPI Rx polling callback
3139 * @napi: structure for representing this polling device
3140 * @budget: how many packets driver is allowed to clean
3142 * This function is used for legacy and MSI, NAPI mode
3144 int ixgbe_poll(struct napi_struct *napi, int budget)
3146 struct ixgbe_q_vector *q_vector =
3147 container_of(napi, struct ixgbe_q_vector, napi);
3148 struct ixgbe_adapter *adapter = q_vector->adapter;
3149 struct ixgbe_ring *ring;
3150 int per_ring_budget, work_done = 0;
3151 bool clean_complete = true;
3153 #ifdef CONFIG_IXGBE_DCA
3154 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3155 ixgbe_update_dca(q_vector);
3158 ixgbe_for_each_ring(ring, q_vector->tx) {
3159 bool wd = ring->xsk_umem ?
3160 ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3161 ixgbe_clean_tx_irq(q_vector, ring, budget);
3164 clean_complete = false;
3167 /* Exit if we are called by netpoll */
3171 /* attempt to distribute budget to each queue fairly, but don't allow
3172 * the budget to go below 1 because we'll exit polling */
3173 if (q_vector->rx.count > 1)
3174 per_ring_budget = max(budget/q_vector->rx.count, 1);
3176 per_ring_budget = budget;
3178 ixgbe_for_each_ring(ring, q_vector->rx) {
3179 int cleaned = ring->xsk_umem ?
3180 ixgbe_clean_rx_irq_zc(q_vector, ring,
3182 ixgbe_clean_rx_irq(q_vector, ring,
3185 work_done += cleaned;
3186 if (cleaned >= per_ring_budget)
3187 clean_complete = false;
3190 /* If all work not completed, return budget and keep polling */
3191 if (!clean_complete)
3194 /* all work done, exit the polling mode */
3195 if (likely(napi_complete_done(napi, work_done))) {
3196 if (adapter->rx_itr_setting & 1)
3197 ixgbe_set_itr(q_vector);
3198 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3199 ixgbe_irq_enable_queues(adapter,
3200 BIT_ULL(q_vector->v_idx));
3203 return min(work_done, budget - 1);
3207 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3208 * @adapter: board private structure
3210 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3211 * interrupts from the kernel.
3213 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3215 struct net_device *netdev = adapter->netdev;
3216 unsigned int ri = 0, ti = 0;
3219 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3220 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3221 struct msix_entry *entry = &adapter->msix_entries[vector];
3223 if (q_vector->tx.ring && q_vector->rx.ring) {
3224 snprintf(q_vector->name, sizeof(q_vector->name),
3225 "%s-TxRx-%u", netdev->name, ri++);
3227 } else if (q_vector->rx.ring) {
3228 snprintf(q_vector->name, sizeof(q_vector->name),
3229 "%s-rx-%u", netdev->name, ri++);
3230 } else if (q_vector->tx.ring) {
3231 snprintf(q_vector->name, sizeof(q_vector->name),
3232 "%s-tx-%u", netdev->name, ti++);
3234 /* skip this unused q_vector */
3237 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3238 q_vector->name, q_vector);
3240 e_err(probe, "request_irq failed for MSIX interrupt "
3241 "Error: %d\n", err);
3242 goto free_queue_irqs;
3244 /* If Flow Director is enabled, set interrupt affinity */
3245 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3246 /* assign the mask for this irq */
3247 irq_set_affinity_hint(entry->vector,
3248 &q_vector->affinity_mask);
3252 err = request_irq(adapter->msix_entries[vector].vector,
3253 ixgbe_msix_other, 0, netdev->name, adapter);
3255 e_err(probe, "request_irq for msix_other failed: %d\n", err);
3256 goto free_queue_irqs;
3264 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3266 free_irq(adapter->msix_entries[vector].vector,
3267 adapter->q_vector[vector]);
3269 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3270 pci_disable_msix(adapter->pdev);
3271 kfree(adapter->msix_entries);
3272 adapter->msix_entries = NULL;
3277 * ixgbe_intr - legacy mode Interrupt Handler
3278 * @irq: interrupt number
3279 * @data: pointer to a network interface device structure
3281 static irqreturn_t ixgbe_intr(int irq, void *data)
3283 struct ixgbe_adapter *adapter = data;
3284 struct ixgbe_hw *hw = &adapter->hw;
3285 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3289 * Workaround for silicon errata #26 on 82598. Mask the interrupt
3290 * before the read of EICR.
3292 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3294 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3295 * therefore no explicit interrupt disable is necessary */
3296 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3299 * shared interrupt alert!
3300 * make sure interrupts are enabled because the read will
3301 * have disabled interrupts due to EIAM
3302 * finish the workaround of silicon errata on 82598. Unmask
3303 * the interrupt that we masked before the EICR read.
3305 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3306 ixgbe_irq_enable(adapter, true, true);
3307 return IRQ_NONE; /* Not our interrupt */
3310 if (eicr & IXGBE_EICR_LSC)
3311 ixgbe_check_lsc(adapter);
3313 switch (hw->mac.type) {
3314 case ixgbe_mac_82599EB:
3315 ixgbe_check_sfp_event(adapter, eicr);
3317 case ixgbe_mac_X540:
3318 case ixgbe_mac_X550:
3319 case ixgbe_mac_X550EM_x:
3320 case ixgbe_mac_x550em_a:
3321 if (eicr & IXGBE_EICR_ECC) {
3322 e_info(link, "Received ECC Err, initiating reset\n");
3323 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3324 ixgbe_service_event_schedule(adapter);
3325 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3327 ixgbe_check_overtemp_event(adapter, eicr);
3333 ixgbe_check_fan_failure(adapter, eicr);
3334 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3335 ixgbe_ptp_check_pps_event(adapter);
3337 /* would disable interrupts here but EIAM disabled it */
3338 napi_schedule_irqoff(&q_vector->napi);
3341 * re-enable link(maybe) and non-queue interrupts, no flush.
3342 * ixgbe_poll will re-enable the queue interrupts
3344 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3345 ixgbe_irq_enable(adapter, false, false);
3351 * ixgbe_request_irq - initialize interrupts
3352 * @adapter: board private structure
3354 * Attempts to configure interrupts using the best available
3355 * capabilities of the hardware and kernel.
3357 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3359 struct net_device *netdev = adapter->netdev;
3362 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3363 err = ixgbe_request_msix_irqs(adapter);
3364 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3365 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3366 netdev->name, adapter);
3368 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3369 netdev->name, adapter);
3372 e_err(probe, "request_irq failed, Error %d\n", err);
3377 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3381 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3382 free_irq(adapter->pdev->irq, adapter);
3386 if (!adapter->msix_entries)
3389 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3390 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3391 struct msix_entry *entry = &adapter->msix_entries[vector];
3393 /* free only the irqs that were actually requested */
3394 if (!q_vector->rx.ring && !q_vector->tx.ring)
3397 /* clear the affinity_mask in the IRQ descriptor */
3398 irq_set_affinity_hint(entry->vector, NULL);
3400 free_irq(entry->vector, q_vector);
3403 free_irq(adapter->msix_entries[vector].vector, adapter);
3407 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3408 * @adapter: board private structure
3410 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3412 switch (adapter->hw.mac.type) {
3413 case ixgbe_mac_82598EB:
3414 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3416 case ixgbe_mac_82599EB:
3417 case ixgbe_mac_X540:
3418 case ixgbe_mac_X550:
3419 case ixgbe_mac_X550EM_x:
3420 case ixgbe_mac_x550em_a:
3421 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3422 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3423 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3428 IXGBE_WRITE_FLUSH(&adapter->hw);
3429 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3432 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3433 synchronize_irq(adapter->msix_entries[vector].vector);
3435 synchronize_irq(adapter->msix_entries[vector++].vector);
3437 synchronize_irq(adapter->pdev->irq);
3442 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3443 * @adapter: board private structure
3446 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3448 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3450 ixgbe_write_eitr(q_vector);
3452 ixgbe_set_ivar(adapter, 0, 0, 0);
3453 ixgbe_set_ivar(adapter, 1, 0, 0);
3455 e_info(hw, "Legacy interrupt IVAR setup done\n");
3459 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3460 * @adapter: board private structure
3461 * @ring: structure containing ring specific data
3463 * Configure the Tx descriptor ring after a reset.
3465 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3466 struct ixgbe_ring *ring)
3468 struct ixgbe_hw *hw = &adapter->hw;
3469 u64 tdba = ring->dma;
3471 u32 txdctl = IXGBE_TXDCTL_ENABLE;
3472 u8 reg_idx = ring->reg_idx;
3474 ring->xsk_umem = NULL;
3475 if (ring_is_xdp(ring))
3476 ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
3478 /* disable queue to avoid issues while updating state */
3479 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3480 IXGBE_WRITE_FLUSH(hw);
3482 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3483 (tdba & DMA_BIT_MASK(32)));
3484 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3485 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3486 ring->count * sizeof(union ixgbe_adv_tx_desc));
3487 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3488 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3489 ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3492 * set WTHRESH to encourage burst writeback, it should not be set
3493 * higher than 1 when:
3494 * - ITR is 0 as it could cause false TX hangs
3495 * - ITR is set to > 100k int/sec and BQL is enabled
3497 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3498 * to or less than the number of on chip descriptors, which is
3501 if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3502 txdctl |= 1u << 16; /* WTHRESH = 1 */
3504 txdctl |= 8u << 16; /* WTHRESH = 8 */
3507 * Setting PTHRESH to 32 both improves performance
3508 * and avoids a TX hang with DFP enabled
3510 txdctl |= (1u << 8) | /* HTHRESH = 1 */
3511 32; /* PTHRESH = 32 */
3513 /* reinitialize flowdirector state */
3514 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3515 ring->atr_sample_rate = adapter->atr_sample_rate;
3516 ring->atr_count = 0;
3517 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3519 ring->atr_sample_rate = 0;
3522 /* initialize XPS */
3523 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3524 struct ixgbe_q_vector *q_vector = ring->q_vector;
3527 netif_set_xps_queue(ring->netdev,
3528 &q_vector->affinity_mask,
3532 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3534 /* reinitialize tx_buffer_info */
3535 memset(ring->tx_buffer_info, 0,
3536 sizeof(struct ixgbe_tx_buffer) * ring->count);
3539 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3541 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3542 if (hw->mac.type == ixgbe_mac_82598EB &&
3543 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3546 /* poll to verify queue is enabled */
3548 usleep_range(1000, 2000);
3549 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3550 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3552 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3555 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3557 struct ixgbe_hw *hw = &adapter->hw;
3559 u8 tcs = adapter->hw_tcs;
3561 if (hw->mac.type == ixgbe_mac_82598EB)
3564 /* disable the arbiter while setting MTQC */
3565 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3566 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3567 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3569 /* set transmit pool layout */
3570 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3571 mtqc = IXGBE_MTQC_VT_ENA;
3573 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3575 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3576 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3577 IXGBE_82599_VMDQ_4Q_MASK)
3578 mtqc |= IXGBE_MTQC_32VF;
3580 mtqc |= IXGBE_MTQC_64VF;
3583 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3584 } else if (tcs > 1) {
3585 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3587 u8 max_txq = adapter->num_tx_queues +
3588 adapter->num_xdp_queues;
3590 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3592 mtqc = IXGBE_MTQC_64Q_1PB;
3596 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3598 /* Enable Security TX Buffer IFG for multiple pb */
3600 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3601 sectx |= IXGBE_SECTX_DCB;
3602 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3605 /* re-enable the arbiter */
3606 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3607 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3611 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3612 * @adapter: board private structure
3614 * Configure the Tx unit of the MAC after a reset.
3616 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3618 struct ixgbe_hw *hw = &adapter->hw;
3622 ixgbe_setup_mtqc(adapter);
3624 if (hw->mac.type != ixgbe_mac_82598EB) {
3625 /* DMATXCTL.EN must be before Tx queues are enabled */
3626 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3627 dmatxctl |= IXGBE_DMATXCTL_TE;
3628 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3631 /* Setup the HW Tx Head and Tail descriptor pointers */
3632 for (i = 0; i < adapter->num_tx_queues; i++)
3633 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3634 for (i = 0; i < adapter->num_xdp_queues; i++)
3635 ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3638 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3639 struct ixgbe_ring *ring)
3641 struct ixgbe_hw *hw = &adapter->hw;
3642 u8 reg_idx = ring->reg_idx;
3643 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3645 srrctl |= IXGBE_SRRCTL_DROP_EN;
3647 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3650 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3651 struct ixgbe_ring *ring)
3653 struct ixgbe_hw *hw = &adapter->hw;
3654 u8 reg_idx = ring->reg_idx;
3655 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3657 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3659 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3662 #ifdef CONFIG_IXGBE_DCB
3663 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3665 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3669 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3671 if (adapter->ixgbe_ieee_pfc)
3672 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3675 * We should set the drop enable bit if:
3678 * Number of Rx queues > 1 and flow control is disabled
3680 * This allows us to avoid head of line blocking for security
3681 * and performance reasons.
3683 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3684 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3685 for (i = 0; i < adapter->num_rx_queues; i++)
3686 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3688 for (i = 0; i < adapter->num_rx_queues; i++)
3689 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3693 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3695 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3696 struct ixgbe_ring *rx_ring)
3698 struct ixgbe_hw *hw = &adapter->hw;
3700 u8 reg_idx = rx_ring->reg_idx;
3702 if (hw->mac.type == ixgbe_mac_82598EB) {
3703 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3706 * if VMDq is not active we must program one srrctl register
3707 * per RSS queue since we have enabled RDRXCTL.MVMEN
3712 /* configure header buffer length, needed for RSC */
3713 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3715 /* configure the packet buffer length */
3716 if (rx_ring->xsk_umem) {
3717 u32 xsk_buf_len = xsk_umem_get_rx_frame_size(rx_ring->xsk_umem);
3719 /* If the MAC support setting RXDCTL.RLPML, the
3720 * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3721 * RXDCTL.RLPML is set to the actual UMEM buffer
3722 * size. If not, then we are stuck with a 1k buffer
3723 * size resolution. In this case frames larger than
3724 * the UMEM buffer size viewed in a 1k resolution will
3727 if (hw->mac.type != ixgbe_mac_82599EB)
3728 srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3730 srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3731 } else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3732 srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3734 srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3737 /* configure descriptor type */
3738 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3740 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3744 * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3745 * @adapter: device handle
3747 * - 82598/82599/X540: 128
3748 * - X550(non-SRIOV mode): 512
3749 * - X550(SRIOV mode): 64
3751 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3753 if (adapter->hw.mac.type < ixgbe_mac_X550)
3755 else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3762 * ixgbe_store_key - Write the RSS key to HW
3763 * @adapter: device handle
3765 * Write the RSS key stored in adapter.rss_key to HW.
3767 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3769 struct ixgbe_hw *hw = &adapter->hw;
3772 for (i = 0; i < 10; i++)
3773 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3777 * ixgbe_init_rss_key - Initialize adapter RSS key
3778 * @adapter: device handle
3780 * Allocates and initializes the RSS key if it is not allocated.
3782 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3786 if (!adapter->rss_key) {
3787 rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3788 if (unlikely(!rss_key))
3791 netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3792 adapter->rss_key = rss_key;
3799 * ixgbe_store_reta - Write the RETA table to HW
3800 * @adapter: device handle
3802 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3804 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3806 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3807 struct ixgbe_hw *hw = &adapter->hw;
3810 u8 *indir_tbl = adapter->rss_indir_tbl;
3812 /* Fill out the redirection table as follows:
3813 * - 82598: 8 bit wide entries containing pair of 4 bit RSS
3815 * - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3816 * - X550: 8 bit wide entries containing 6 bit RSS index
3818 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3819 indices_multi = 0x11;
3821 indices_multi = 0x1;
3823 /* Write redirection table to HW */
3824 for (i = 0; i < reta_entries; i++) {
3825 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3828 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3830 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3838 * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3839 * @adapter: device handle
3841 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3843 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3845 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3846 struct ixgbe_hw *hw = &adapter->hw;
3849 /* Write redirection table to HW */
3850 for (i = 0; i < reta_entries; i++) {
3851 u16 pool = adapter->num_rx_pools;
3853 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3859 IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3865 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3868 u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3869 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3871 /* Program table for at least 4 queues w/ SR-IOV so that VFs can
3872 * make full use of any rings they may have. We will use the
3873 * PSRTYPE register to control how many rings we use within the PF.
3875 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3878 /* Fill out hash function seeds */
3879 ixgbe_store_key(adapter);
3881 /* Fill out redirection table */
3882 memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3884 for (i = 0, j = 0; i < reta_entries; i++, j++) {
3888 adapter->rss_indir_tbl[i] = j;
3891 ixgbe_store_reta(adapter);
3894 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3896 struct ixgbe_hw *hw = &adapter->hw;
3897 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3900 /* Fill out hash function seeds */
3901 for (i = 0; i < 10; i++) {
3902 u16 pool = adapter->num_rx_pools;
3906 IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3907 *(adapter->rss_key + i));
3910 /* Fill out the redirection table */
3911 for (i = 0, j = 0; i < 64; i++, j++) {
3915 adapter->rss_indir_tbl[i] = j;
3918 ixgbe_store_vfreta(adapter);
3921 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3923 struct ixgbe_hw *hw = &adapter->hw;
3924 u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3927 /* Disable indicating checksum in descriptor, enables RSS hash */
3928 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3929 rxcsum |= IXGBE_RXCSUM_PCSD;
3930 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3932 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3933 if (adapter->ring_feature[RING_F_RSS].mask)
3934 mrqc = IXGBE_MRQC_RSSEN;
3936 u8 tcs = adapter->hw_tcs;
3938 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3940 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3942 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3943 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3944 IXGBE_82599_VMDQ_4Q_MASK)
3945 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3947 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3949 /* Enable L3/L4 for Tx Switched packets only for X550,
3950 * older devices do not support this feature
3952 if (hw->mac.type >= ixgbe_mac_X550)
3953 mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3956 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3958 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3960 mrqc = IXGBE_MRQC_RSSEN;
3964 /* Perform hash on these packet types */
3965 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3966 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3967 IXGBE_MRQC_RSS_FIELD_IPV6 |
3968 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3970 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3971 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3972 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3973 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3975 if ((hw->mac.type >= ixgbe_mac_X550) &&
3976 (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3977 u16 pool = adapter->num_rx_pools;
3979 /* Enable VF RSS mode */
3980 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3981 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3983 /* Setup RSS through the VF registers */
3984 ixgbe_setup_vfreta(adapter);
3985 vfmrqc = IXGBE_MRQC_RSSEN;
3986 vfmrqc |= rss_field;
3990 IXGBE_PFVFMRQC(VMDQ_P(pool)),
3993 ixgbe_setup_reta(adapter);
3995 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4000 * ixgbe_configure_rscctl - enable RSC for the indicated ring
4001 * @adapter: address of board private structure
4002 * @ring: structure containing ring specific data
4004 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4005 struct ixgbe_ring *ring)
4007 struct ixgbe_hw *hw = &adapter->hw;
4009 u8 reg_idx = ring->reg_idx;
4011 if (!ring_is_rsc_enabled(ring))
4014 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4015 rscctrl |= IXGBE_RSCCTL_RSCEN;
4017 * we must limit the number of descriptors so that the
4018 * total size of max desc * buf_len is not greater
4021 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4022 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4025 #define IXGBE_MAX_RX_DESC_POLL 10
4026 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4027 struct ixgbe_ring *ring)
4029 struct ixgbe_hw *hw = &adapter->hw;
4030 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4032 u8 reg_idx = ring->reg_idx;
4034 if (ixgbe_removed(hw->hw_addr))
4036 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4037 if (hw->mac.type == ixgbe_mac_82598EB &&
4038 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4042 usleep_range(1000, 2000);
4043 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4044 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4047 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4048 "the polling period\n", reg_idx);
4052 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4053 struct ixgbe_ring *ring)
4055 struct ixgbe_hw *hw = &adapter->hw;
4056 union ixgbe_adv_rx_desc *rx_desc;
4057 u64 rdba = ring->dma;
4059 u8 reg_idx = ring->reg_idx;
4061 xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4062 ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
4063 if (ring->xsk_umem) {
4064 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4065 MEM_TYPE_XSK_BUFF_POOL,
4067 xsk_buff_set_rxq_info(ring->xsk_umem, &ring->xdp_rxq);
4069 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4070 MEM_TYPE_PAGE_SHARED, NULL));
4073 /* disable queue to avoid use of these values while updating state */
4074 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4075 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4077 /* write value back with RXDCTL.ENABLE bit cleared */
4078 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4079 IXGBE_WRITE_FLUSH(hw);
4081 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4082 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4083 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4084 ring->count * sizeof(union ixgbe_adv_rx_desc));
4085 /* Force flushing of IXGBE_RDLEN to prevent MDD */
4086 IXGBE_WRITE_FLUSH(hw);
4088 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4089 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4090 ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4092 ixgbe_configure_srrctl(adapter, ring);
4093 ixgbe_configure_rscctl(adapter, ring);
4095 if (hw->mac.type == ixgbe_mac_82598EB) {
4097 * enable cache line friendly hardware writes:
4098 * PTHRESH=32 descriptors (half the internal cache),
4099 * this also removes ugly rx_no_buffer_count increment
4100 * HTHRESH=4 descriptors (to minimize latency on fetch)
4101 * WTHRESH=8 burst writeback up to two cache lines
4103 rxdctl &= ~0x3FFFFF;
4105 #if (PAGE_SIZE < 8192)
4106 /* RXDCTL.RLPML does not work on 82599 */
4107 } else if (hw->mac.type != ixgbe_mac_82599EB) {
4108 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4109 IXGBE_RXDCTL_RLPML_EN);
4111 /* Limit the maximum frame size so we don't overrun the skb.
4112 * This can happen in SRIOV mode when the MTU of the VF is
4113 * higher than the MTU of the PF.
4115 if (ring_uses_build_skb(ring) &&
4116 !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4117 rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4118 IXGBE_RXDCTL_RLPML_EN;
4122 if (ring->xsk_umem && hw->mac.type != ixgbe_mac_82599EB) {
4123 u32 xsk_buf_len = xsk_umem_get_rx_frame_size(ring->xsk_umem);
4125 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4126 IXGBE_RXDCTL_RLPML_EN);
4127 rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4129 ring->rx_buf_len = xsk_buf_len;
4132 /* initialize rx_buffer_info */
4133 memset(ring->rx_buffer_info, 0,
4134 sizeof(struct ixgbe_rx_buffer) * ring->count);
4136 /* initialize Rx descriptor 0 */
4137 rx_desc = IXGBE_RX_DESC(ring, 0);
4138 rx_desc->wb.upper.length = 0;
4140 /* enable receive descriptor ring */
4141 rxdctl |= IXGBE_RXDCTL_ENABLE;
4142 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4144 ixgbe_rx_desc_queue_enable(adapter, ring);
4146 ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4148 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4151 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4153 struct ixgbe_hw *hw = &adapter->hw;
4154 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4155 u16 pool = adapter->num_rx_pools;
4157 /* PSRTYPE must be initialized in non 82598 adapters */
4158 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4159 IXGBE_PSRTYPE_UDPHDR |
4160 IXGBE_PSRTYPE_IPV4HDR |
4161 IXGBE_PSRTYPE_L2HDR |
4162 IXGBE_PSRTYPE_IPV6HDR;
4164 if (hw->mac.type == ixgbe_mac_82598EB)
4168 psrtype |= 2u << 29;
4170 psrtype |= 1u << 29;
4173 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4176 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4178 struct ixgbe_hw *hw = &adapter->hw;
4179 u16 pool = adapter->num_rx_pools;
4180 u32 reg_offset, vf_shift, vmolr;
4181 u32 gcr_ext, vmdctl;
4184 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4187 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4188 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4189 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4190 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4191 vmdctl |= IXGBE_VT_CTL_REPLEN;
4192 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4194 /* accept untagged packets until a vlan tag is
4195 * specifically set for the VMDQ queue/pool
4197 vmolr = IXGBE_VMOLR_AUPE;
4199 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4201 vf_shift = VMDQ_P(0) % 32;
4202 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4204 /* Enable only the PF's pool for Tx/Rx */
4205 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4206 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4207 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4208 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4209 if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4210 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4212 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4213 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4215 /* clear VLAN promisc flag so VFTA will be updated if necessary */
4216 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4219 * Set up VF register offsets for selected VT Mode,
4220 * i.e. 32 or 64 VFs for SR-IOV
4222 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4223 case IXGBE_82599_VMDQ_8Q_MASK:
4224 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4226 case IXGBE_82599_VMDQ_4Q_MASK:
4227 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4230 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4234 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4236 for (i = 0; i < adapter->num_vfs; i++) {
4237 /* configure spoof checking */
4238 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4239 adapter->vfinfo[i].spoofchk_enabled);
4241 /* Enable/Disable RSS query feature */
4242 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4243 adapter->vfinfo[i].rss_query_enabled);
4247 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4249 struct ixgbe_hw *hw = &adapter->hw;
4250 struct net_device *netdev = adapter->netdev;
4251 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4252 struct ixgbe_ring *rx_ring;
4257 /* adjust max frame to be able to do baby jumbo for FCoE */
4258 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4259 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4260 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4262 #endif /* IXGBE_FCOE */
4264 /* adjust max frame to be at least the size of a standard frame */
4265 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4266 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4268 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4269 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4270 mhadd &= ~IXGBE_MHADD_MFS_MASK;
4271 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4273 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4276 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4277 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4278 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4279 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4282 * Setup the HW Rx Head and Tail Descriptor Pointers and
4283 * the Base and Length of the Rx Descriptor Ring
4285 for (i = 0; i < adapter->num_rx_queues; i++) {
4286 rx_ring = adapter->rx_ring[i];
4288 clear_ring_rsc_enabled(rx_ring);
4289 clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4290 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4292 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4293 set_ring_rsc_enabled(rx_ring);
4295 if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4296 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4298 if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4301 set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4303 #if (PAGE_SIZE < 8192)
4304 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4305 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4307 if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4308 (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4309 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4314 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4316 struct ixgbe_hw *hw = &adapter->hw;
4317 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4319 switch (hw->mac.type) {
4320 case ixgbe_mac_82598EB:
4322 * For VMDq support of different descriptor types or
4323 * buffer sizes through the use of multiple SRRCTL
4324 * registers, RDRXCTL.MVMEN must be set to 1
4326 * also, the manual doesn't mention it clearly but DCA hints
4327 * will only use queue 0's tags unless this bit is set. Side
4328 * effects of setting this bit are only that SRRCTL must be
4329 * fully programmed [0..15]
4331 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4333 case ixgbe_mac_X550:
4334 case ixgbe_mac_X550EM_x:
4335 case ixgbe_mac_x550em_a:
4336 if (adapter->num_vfs)
4337 rdrxctl |= IXGBE_RDRXCTL_PSP;
4339 case ixgbe_mac_82599EB:
4340 case ixgbe_mac_X540:
4341 /* Disable RSC for ACK packets */
4342 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4343 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4344 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4345 /* hardware requires some bits to be set by default */
4346 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4347 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4350 /* We should do nothing since we don't know this hardware */
4354 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4358 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4359 * @adapter: board private structure
4361 * Configure the Rx unit of the MAC after a reset.
4363 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4365 struct ixgbe_hw *hw = &adapter->hw;
4369 /* disable receives while setting up the descriptors */
4370 hw->mac.ops.disable_rx(hw);
4372 ixgbe_setup_psrtype(adapter);
4373 ixgbe_setup_rdrxctl(adapter);
4376 rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4377 rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4378 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4379 rfctl |= IXGBE_RFCTL_RSC_DIS;
4381 /* disable NFS filtering */
4382 rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4383 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4385 /* Program registers for the distribution of queues */
4386 ixgbe_setup_mrqc(adapter);
4388 /* set_rx_buffer_len must be called before ring initialization */
4389 ixgbe_set_rx_buffer_len(adapter);
4392 * Setup the HW Rx Head and Tail Descriptor Pointers and
4393 * the Base and Length of the Rx Descriptor Ring
4395 for (i = 0; i < adapter->num_rx_queues; i++)
4396 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4398 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4399 /* disable drop enable for 82598 parts */
4400 if (hw->mac.type == ixgbe_mac_82598EB)
4401 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4403 /* enable all receives */
4404 rxctrl |= IXGBE_RXCTRL_RXEN;
4405 hw->mac.ops.enable_rx_dma(hw, rxctrl);
4408 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4409 __be16 proto, u16 vid)
4411 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4412 struct ixgbe_hw *hw = &adapter->hw;
4414 /* add VID to filter table */
4415 if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4416 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4418 set_bit(vid, adapter->active_vlans);
4423 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4428 /* short cut the special case */
4432 /* Search for the vlan id in the VLVF entries */
4433 for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4434 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4435 if ((vlvf & VLAN_VID_MASK) == vlan)
4442 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4444 struct ixgbe_hw *hw = &adapter->hw;
4448 idx = ixgbe_find_vlvf_entry(hw, vid);
4452 /* See if any other pools are set for this VLAN filter
4453 * entry other than the PF.
4455 word = idx * 2 + (VMDQ_P(0) / 32);
4456 bits = ~BIT(VMDQ_P(0) % 32);
4457 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4459 /* Disable the filter so this falls into the default pool. */
4460 if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4461 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4462 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4463 IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4467 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4468 __be16 proto, u16 vid)
4470 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4471 struct ixgbe_hw *hw = &adapter->hw;
4473 /* remove VID from filter table */
4474 if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4475 hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4477 clear_bit(vid, adapter->active_vlans);
4483 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4484 * @adapter: driver data
4486 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4488 struct ixgbe_hw *hw = &adapter->hw;
4492 switch (hw->mac.type) {
4493 case ixgbe_mac_82598EB:
4494 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4495 vlnctrl &= ~IXGBE_VLNCTRL_VME;
4496 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4498 case ixgbe_mac_82599EB:
4499 case ixgbe_mac_X540:
4500 case ixgbe_mac_X550:
4501 case ixgbe_mac_X550EM_x:
4502 case ixgbe_mac_x550em_a:
4503 for (i = 0; i < adapter->num_rx_queues; i++) {
4504 struct ixgbe_ring *ring = adapter->rx_ring[i];
4506 if (!netif_is_ixgbe(ring->netdev))
4510 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4511 vlnctrl &= ~IXGBE_RXDCTL_VME;
4512 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4521 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4522 * @adapter: driver data
4524 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4526 struct ixgbe_hw *hw = &adapter->hw;
4530 switch (hw->mac.type) {
4531 case ixgbe_mac_82598EB:
4532 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4533 vlnctrl |= IXGBE_VLNCTRL_VME;
4534 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4536 case ixgbe_mac_82599EB:
4537 case ixgbe_mac_X540:
4538 case ixgbe_mac_X550:
4539 case ixgbe_mac_X550EM_x:
4540 case ixgbe_mac_x550em_a:
4541 for (i = 0; i < adapter->num_rx_queues; i++) {
4542 struct ixgbe_ring *ring = adapter->rx_ring[i];
4544 if (!netif_is_ixgbe(ring->netdev))
4548 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4549 vlnctrl |= IXGBE_RXDCTL_VME;
4550 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4558 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4560 struct ixgbe_hw *hw = &adapter->hw;
4563 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4565 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4566 /* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4567 vlnctrl |= IXGBE_VLNCTRL_VFE;
4568 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4570 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4571 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4575 /* Nothing to do for 82598 */
4576 if (hw->mac.type == ixgbe_mac_82598EB)
4579 /* We are already in VLAN promisc, nothing to do */
4580 if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4583 /* Set flag so we don't redo unnecessary work */
4584 adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4586 /* Add PF to all active pools */
4587 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4588 u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4589 u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4591 vlvfb |= BIT(VMDQ_P(0) % 32);
4592 IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4595 /* Set all bits in the VLAN filter table array */
4596 for (i = hw->mac.vft_size; i--;)
4597 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4600 #define VFTA_BLOCK_SIZE 8
4601 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4603 struct ixgbe_hw *hw = &adapter->hw;
4604 u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4605 u32 vid_start = vfta_offset * 32;
4606 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4607 u32 i, vid, word, bits;
4609 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4610 u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4612 /* pull VLAN ID from VLVF */
4613 vid = vlvf & VLAN_VID_MASK;
4615 /* only concern outselves with a certain range */
4616 if (vid < vid_start || vid >= vid_end)
4620 /* record VLAN ID in VFTA */
4621 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4623 /* if PF is part of this then continue */
4624 if (test_bit(vid, adapter->active_vlans))
4628 /* remove PF from the pool */
4629 word = i * 2 + VMDQ_P(0) / 32;
4630 bits = ~BIT(VMDQ_P(0) % 32);
4631 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4632 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4635 /* extract values from active_vlans and write back to VFTA */
4636 for (i = VFTA_BLOCK_SIZE; i--;) {
4637 vid = (vfta_offset + i) * 32;
4638 word = vid / BITS_PER_LONG;
4639 bits = vid % BITS_PER_LONG;
4641 vfta[i] |= adapter->active_vlans[word] >> bits;
4643 IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4647 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4649 struct ixgbe_hw *hw = &adapter->hw;
4652 /* Set VLAN filtering to enabled */
4653 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4654 vlnctrl |= IXGBE_VLNCTRL_VFE;
4655 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4657 if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4658 hw->mac.type == ixgbe_mac_82598EB)
4661 /* We are not in VLAN promisc, nothing to do */
4662 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4665 /* Set flag so we don't redo unnecessary work */
4666 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4668 for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4669 ixgbe_scrub_vfta(adapter, i);
4672 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4676 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4678 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4679 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4683 * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4684 * @netdev: network interface device structure
4686 * Writes multicast address list to the MTA hash table.
4687 * Returns: -ENOMEM on failure
4688 * 0 on no addresses written
4689 * X on writing X addresses to MTA
4691 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4693 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4694 struct ixgbe_hw *hw = &adapter->hw;
4696 if (!netif_running(netdev))
4699 if (hw->mac.ops.update_mc_addr_list)
4700 hw->mac.ops.update_mc_addr_list(hw, netdev);
4704 #ifdef CONFIG_PCI_IOV
4705 ixgbe_restore_vf_multicasts(adapter);
4708 return netdev_mc_count(netdev);
4711 #ifdef CONFIG_PCI_IOV
4712 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4714 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4715 struct ixgbe_hw *hw = &adapter->hw;
4718 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4719 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4721 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4722 hw->mac.ops.set_rar(hw, i,
4727 hw->mac.ops.clear_rar(hw, i);
4732 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4734 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4735 struct ixgbe_hw *hw = &adapter->hw;
4738 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4739 if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4742 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4744 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4745 hw->mac.ops.set_rar(hw, i,
4750 hw->mac.ops.clear_rar(hw, i);
4754 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4756 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4757 struct ixgbe_hw *hw = &adapter->hw;
4760 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4761 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4762 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4765 ixgbe_sync_mac_table(adapter);
4768 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4770 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4771 struct ixgbe_hw *hw = &adapter->hw;
4774 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4775 /* do not count default RAR as available */
4776 if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4779 /* only count unused and addresses that belong to us */
4780 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4781 if (mac_table->pool != pool)
4791 /* this function destroys the first RAR entry */
4792 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4794 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4795 struct ixgbe_hw *hw = &adapter->hw;
4797 memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4798 mac_table->pool = VMDQ_P(0);
4800 mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4802 hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4806 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4807 const u8 *addr, u16 pool)
4809 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4810 struct ixgbe_hw *hw = &adapter->hw;
4813 if (is_zero_ether_addr(addr))
4816 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4817 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4820 ether_addr_copy(mac_table->addr, addr);
4821 mac_table->pool = pool;
4823 mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4824 IXGBE_MAC_STATE_IN_USE;
4826 ixgbe_sync_mac_table(adapter);
4834 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4835 const u8 *addr, u16 pool)
4837 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4838 struct ixgbe_hw *hw = &adapter->hw;
4841 if (is_zero_ether_addr(addr))
4844 /* search table for addr, if found clear IN_USE flag and sync */
4845 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4846 /* we can only delete an entry if it is in use */
4847 if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4849 /* we only care about entries that belong to the given pool */
4850 if (mac_table->pool != pool)
4852 /* we only care about a specific MAC address */
4853 if (!ether_addr_equal(addr, mac_table->addr))
4856 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4857 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4859 ixgbe_sync_mac_table(adapter);
4867 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4869 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4872 ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4874 return min_t(int, ret, 0);
4877 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4879 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4881 ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4887 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4888 * @netdev: network interface device structure
4890 * The set_rx_method entry point is called whenever the unicast/multicast
4891 * address list or the network interface flags are updated. This routine is
4892 * responsible for configuring the hardware for proper unicast, multicast and
4895 void ixgbe_set_rx_mode(struct net_device *netdev)
4897 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4898 struct ixgbe_hw *hw = &adapter->hw;
4899 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4900 netdev_features_t features = netdev->features;
4903 /* Check for Promiscuous and All Multicast modes */
4904 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4906 /* set all bits that we expect to always be set */
4907 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4908 fctrl |= IXGBE_FCTRL_BAM;
4909 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4910 fctrl |= IXGBE_FCTRL_PMCF;
4912 /* clear the bits we are changing the status of */
4913 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4914 if (netdev->flags & IFF_PROMISC) {
4915 hw->addr_ctrl.user_set_promisc = true;
4916 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4917 vmolr |= IXGBE_VMOLR_MPE;
4918 features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4920 if (netdev->flags & IFF_ALLMULTI) {
4921 fctrl |= IXGBE_FCTRL_MPE;
4922 vmolr |= IXGBE_VMOLR_MPE;
4924 hw->addr_ctrl.user_set_promisc = false;
4928 * Write addresses to available RAR registers, if there is not
4929 * sufficient space to store all the addresses then enable
4930 * unicast promiscuous mode
4932 if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4933 fctrl |= IXGBE_FCTRL_UPE;
4934 vmolr |= IXGBE_VMOLR_ROPE;
4937 /* Write addresses to the MTA, if the attempt fails
4938 * then we should just turn on promiscuous mode so
4939 * that we can at least receive multicast traffic
4941 count = ixgbe_write_mc_addr_list(netdev);
4943 fctrl |= IXGBE_FCTRL_MPE;
4944 vmolr |= IXGBE_VMOLR_MPE;
4946 vmolr |= IXGBE_VMOLR_ROMPE;
4949 if (hw->mac.type != ixgbe_mac_82598EB) {
4950 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4951 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4953 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4956 /* This is useful for sniffing bad packets. */
4957 if (features & NETIF_F_RXALL) {
4958 /* UPE and MPE will be handled by normal PROMISC logic
4959 * in e1000e_set_rx_mode */
4960 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4961 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4962 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4964 fctrl &= ~(IXGBE_FCTRL_DPF);
4965 /* NOTE: VLAN filtering is disabled by setting PROMISC */
4968 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4970 if (features & NETIF_F_HW_VLAN_CTAG_RX)
4971 ixgbe_vlan_strip_enable(adapter);
4973 ixgbe_vlan_strip_disable(adapter);
4975 if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4976 ixgbe_vlan_promisc_disable(adapter);
4978 ixgbe_vlan_promisc_enable(adapter);
4981 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4985 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4986 napi_enable(&adapter->q_vector[q_idx]->napi);
4989 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4993 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4994 napi_disable(&adapter->q_vector[q_idx]->napi);
4997 static void ixgbe_clear_udp_tunnel_port(struct ixgbe_adapter *adapter, u32 mask)
4999 struct ixgbe_hw *hw = &adapter->hw;
5002 if (!(adapter->flags & (IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE |
5003 IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)))
5006 vxlanctrl = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) & ~mask;
5007 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, vxlanctrl);
5009 if (mask & IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK)
5010 adapter->vxlan_port = 0;
5012 if (mask & IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK)
5013 adapter->geneve_port = 0;
5016 #ifdef CONFIG_IXGBE_DCB
5018 * ixgbe_configure_dcb - Configure DCB hardware
5019 * @adapter: ixgbe adapter struct
5021 * This is called by the driver on open to configure the DCB hardware.
5022 * This is also called by the gennetlink interface when reconfiguring
5025 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5027 struct ixgbe_hw *hw = &adapter->hw;
5028 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5030 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5031 if (hw->mac.type == ixgbe_mac_82598EB)
5032 netif_set_gso_max_size(adapter->netdev, 65536);
5036 if (hw->mac.type == ixgbe_mac_82598EB)
5037 netif_set_gso_max_size(adapter->netdev, 32768);
5040 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5041 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5044 /* reconfigure the hardware */
5045 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5046 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5048 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5050 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5051 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5052 ixgbe_dcb_hw_ets(&adapter->hw,
5053 adapter->ixgbe_ieee_ets,
5055 ixgbe_dcb_hw_pfc_config(&adapter->hw,
5056 adapter->ixgbe_ieee_pfc->pfc_en,
5057 adapter->ixgbe_ieee_ets->prio_tc);
5060 /* Enable RSS Hash per TC */
5061 if (hw->mac.type != ixgbe_mac_82598EB) {
5063 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5070 /* write msb to all 8 TCs in one write */
5071 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5076 /* Additional bittime to account for IXGBE framing */
5077 #define IXGBE_ETH_FRAMING 20
5080 * ixgbe_hpbthresh - calculate high water mark for flow control
5082 * @adapter: board private structure to calculate for
5083 * @pb: packet buffer to calculate
5085 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5087 struct ixgbe_hw *hw = &adapter->hw;
5088 struct net_device *dev = adapter->netdev;
5089 int link, tc, kb, marker;
5092 /* Calculate max LAN frame size */
5093 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5096 /* FCoE traffic class uses FCOE jumbo frames */
5097 if ((dev->features & NETIF_F_FCOE_MTU) &&
5098 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5099 (pb == ixgbe_fcoe_get_tc(adapter)))
5100 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5103 /* Calculate delay value for device */
5104 switch (hw->mac.type) {
5105 case ixgbe_mac_X540:
5106 case ixgbe_mac_X550:
5107 case ixgbe_mac_X550EM_x:
5108 case ixgbe_mac_x550em_a:
5109 dv_id = IXGBE_DV_X540(link, tc);
5112 dv_id = IXGBE_DV(link, tc);
5116 /* Loopback switch introduces additional latency */
5117 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5118 dv_id += IXGBE_B2BT(tc);
5120 /* Delay value is calculated in bit times convert to KB */
5121 kb = IXGBE_BT2KB(dv_id);
5122 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5124 marker = rx_pba - kb;
5126 /* It is possible that the packet buffer is not large enough
5127 * to provide required headroom. In this case throw an error
5128 * to user and a do the best we can.
5131 e_warn(drv, "Packet Buffer(%i) can not provide enough"
5132 "headroom to support flow control."
5133 "Decrease MTU or number of traffic classes\n", pb);
5141 * ixgbe_lpbthresh - calculate low water mark for for flow control
5143 * @adapter: board private structure to calculate for
5144 * @pb: packet buffer to calculate
5146 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5148 struct ixgbe_hw *hw = &adapter->hw;
5149 struct net_device *dev = adapter->netdev;
5153 /* Calculate max LAN frame size */
5154 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5157 /* FCoE traffic class uses FCOE jumbo frames */
5158 if ((dev->features & NETIF_F_FCOE_MTU) &&
5159 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5160 (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5161 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5164 /* Calculate delay value for device */
5165 switch (hw->mac.type) {
5166 case ixgbe_mac_X540:
5167 case ixgbe_mac_X550:
5168 case ixgbe_mac_X550EM_x:
5169 case ixgbe_mac_x550em_a:
5170 dv_id = IXGBE_LOW_DV_X540(tc);
5173 dv_id = IXGBE_LOW_DV(tc);
5177 /* Delay value is calculated in bit times convert to KB */
5178 return IXGBE_BT2KB(dv_id);
5182 * ixgbe_pbthresh_setup - calculate and setup high low water marks
5184 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5186 struct ixgbe_hw *hw = &adapter->hw;
5187 int num_tc = adapter->hw_tcs;
5193 for (i = 0; i < num_tc; i++) {
5194 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5195 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5197 /* Low water marks must not be larger than high water marks */
5198 if (hw->fc.low_water[i] > hw->fc.high_water[i])
5199 hw->fc.low_water[i] = 0;
5202 for (; i < MAX_TRAFFIC_CLASS; i++)
5203 hw->fc.high_water[i] = 0;
5206 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5208 struct ixgbe_hw *hw = &adapter->hw;
5210 u8 tc = adapter->hw_tcs;
5212 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5213 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5214 hdrm = 32 << adapter->fdir_pballoc;
5218 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5219 ixgbe_pbthresh_setup(adapter);
5222 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5224 struct ixgbe_hw *hw = &adapter->hw;
5225 struct hlist_node *node2;
5226 struct ixgbe_fdir_filter *filter;
5229 spin_lock(&adapter->fdir_perfect_lock);
5231 if (!hlist_empty(&adapter->fdir_filter_list))
5232 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5234 hlist_for_each_entry_safe(filter, node2,
5235 &adapter->fdir_filter_list, fdir_node) {
5236 if (filter->action == IXGBE_FDIR_DROP_QUEUE) {
5237 queue = IXGBE_FDIR_DROP_QUEUE;
5239 u32 ring = ethtool_get_flow_spec_ring(filter->action);
5240 u8 vf = ethtool_get_flow_spec_ring_vf(filter->action);
5242 if (!vf && (ring >= adapter->num_rx_queues)) {
5243 e_err(drv, "FDIR restore failed without VF, ring: %u\n",
5247 ((vf > adapter->num_vfs) ||
5248 ring >= adapter->num_rx_queues_per_pool)) {
5249 e_err(drv, "FDIR restore failed with VF, vf: %hhu, ring: %u\n",
5254 /* Map the ring onto the absolute queue index */
5256 queue = adapter->rx_ring[ring]->reg_idx;
5259 adapter->num_rx_queues_per_pool) + ring;
5262 ixgbe_fdir_write_perfect_filter_82599(hw,
5263 &filter->filter, filter->sw_idx, queue);
5266 spin_unlock(&adapter->fdir_perfect_lock);
5270 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5271 * @rx_ring: ring to free buffers from
5273 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5275 u16 i = rx_ring->next_to_clean;
5276 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5278 if (rx_ring->xsk_umem) {
5279 ixgbe_xsk_clean_rx_ring(rx_ring);
5283 /* Free all the Rx ring sk_buffs */
5284 while (i != rx_ring->next_to_alloc) {
5285 if (rx_buffer->skb) {
5286 struct sk_buff *skb = rx_buffer->skb;
5287 if (IXGBE_CB(skb)->page_released)
5288 dma_unmap_page_attrs(rx_ring->dev,
5290 ixgbe_rx_pg_size(rx_ring),
5296 /* Invalidate cache lines that may have been written to by
5297 * device so that we avoid corrupting memory.
5299 dma_sync_single_range_for_cpu(rx_ring->dev,
5301 rx_buffer->page_offset,
5302 ixgbe_rx_bufsz(rx_ring),
5305 /* free resources associated with mapping */
5306 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5307 ixgbe_rx_pg_size(rx_ring),
5310 __page_frag_cache_drain(rx_buffer->page,
5311 rx_buffer->pagecnt_bias);
5315 if (i == rx_ring->count) {
5317 rx_buffer = rx_ring->rx_buffer_info;
5322 rx_ring->next_to_alloc = 0;
5323 rx_ring->next_to_clean = 0;
5324 rx_ring->next_to_use = 0;
5327 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5328 struct ixgbe_fwd_adapter *accel)
5330 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5331 int num_tc = netdev_get_num_tc(adapter->netdev);
5332 struct net_device *vdev = accel->netdev;
5335 baseq = accel->pool * adapter->num_rx_queues_per_pool;
5336 netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5337 accel->pool, adapter->num_rx_pools,
5338 baseq, baseq + adapter->num_rx_queues_per_pool);
5340 accel->rx_base_queue = baseq;
5341 accel->tx_base_queue = baseq;
5343 /* record configuration for macvlan interface in vdev */
5344 for (i = 0; i < num_tc; i++)
5345 netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5346 i, rss_i, baseq + (rss_i * i));
5348 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5349 adapter->rx_ring[baseq + i]->netdev = vdev;
5351 /* Guarantee all rings are updated before we update the
5352 * MAC address filter.
5356 /* ixgbe_add_mac_filter will return an index if it succeeds, so we
5357 * need to only treat it as an error value if it is negative.
5359 err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5360 VMDQ_P(accel->pool));
5364 /* if we cannot add the MAC rule then disable the offload */
5365 macvlan_release_l2fw_offload(vdev);
5367 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5368 adapter->rx_ring[baseq + i]->netdev = NULL;
5370 netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5372 /* unbind the queues and drop the subordinate channel config */
5373 netdev_unbind_sb_channel(adapter->netdev, vdev);
5374 netdev_set_sb_channel(vdev, 0);
5376 clear_bit(accel->pool, adapter->fwd_bitmask);
5382 static int ixgbe_macvlan_up(struct net_device *vdev, void *data)
5384 struct ixgbe_adapter *adapter = data;
5385 struct ixgbe_fwd_adapter *accel;
5387 if (!netif_is_macvlan(vdev))
5390 accel = macvlan_accel_priv(vdev);
5394 ixgbe_fwd_ring_up(adapter, accel);
5399 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5401 netdev_walk_all_upper_dev_rcu(adapter->netdev,
5402 ixgbe_macvlan_up, adapter);
5405 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5407 struct ixgbe_hw *hw = &adapter->hw;
5409 ixgbe_configure_pb(adapter);
5410 #ifdef CONFIG_IXGBE_DCB
5411 ixgbe_configure_dcb(adapter);
5414 * We must restore virtualization before VLANs or else
5415 * the VLVF registers will not be populated
5417 ixgbe_configure_virtualization(adapter);
5419 ixgbe_set_rx_mode(adapter->netdev);
5420 ixgbe_restore_vlan(adapter);
5421 ixgbe_ipsec_restore(adapter);
5423 switch (hw->mac.type) {
5424 case ixgbe_mac_82599EB:
5425 case ixgbe_mac_X540:
5426 hw->mac.ops.disable_rx_buff(hw);
5432 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5433 ixgbe_init_fdir_signature_82599(&adapter->hw,
5434 adapter->fdir_pballoc);
5435 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5436 ixgbe_init_fdir_perfect_82599(&adapter->hw,
5437 adapter->fdir_pballoc);
5438 ixgbe_fdir_filter_restore(adapter);
5441 switch (hw->mac.type) {
5442 case ixgbe_mac_82599EB:
5443 case ixgbe_mac_X540:
5444 hw->mac.ops.enable_rx_buff(hw);
5450 #ifdef CONFIG_IXGBE_DCA
5452 if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5453 ixgbe_setup_dca(adapter);
5454 #endif /* CONFIG_IXGBE_DCA */
5457 /* configure FCoE L2 filters, redirection table, and Rx control */
5458 ixgbe_configure_fcoe(adapter);
5460 #endif /* IXGBE_FCOE */
5461 ixgbe_configure_tx(adapter);
5462 ixgbe_configure_rx(adapter);
5463 ixgbe_configure_dfwd(adapter);
5467 * ixgbe_sfp_link_config - set up SFP+ link
5468 * @adapter: pointer to private adapter struct
5470 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5473 * We are assuming the worst case scenario here, and that
5474 * is that an SFP was inserted/removed after the reset
5475 * but before SFP detection was enabled. As such the best
5476 * solution is to just start searching as soon as we start
5478 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5479 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5481 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5482 adapter->sfp_poll_time = 0;
5486 * ixgbe_non_sfp_link_config - set up non-SFP+ link
5487 * @hw: pointer to private hardware struct
5489 * Returns 0 on success, negative on failure
5491 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5494 bool autoneg, link_up = false;
5495 int ret = IXGBE_ERR_LINK_SETUP;
5497 if (hw->mac.ops.check_link)
5498 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5503 speed = hw->phy.autoneg_advertised;
5504 if (!speed && hw->mac.ops.get_link_capabilities) {
5505 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5507 speed &= ~(IXGBE_LINK_SPEED_5GB_FULL |
5508 IXGBE_LINK_SPEED_2_5GB_FULL);
5514 if (hw->mac.ops.setup_link)
5515 ret = hw->mac.ops.setup_link(hw, speed, link_up);
5520 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5522 struct ixgbe_hw *hw = &adapter->hw;
5525 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5526 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5528 gpie |= IXGBE_GPIE_EIAME;
5530 * use EIAM to auto-mask when MSI-X interrupt is asserted
5531 * this saves a register write for every interrupt
5533 switch (hw->mac.type) {
5534 case ixgbe_mac_82598EB:
5535 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5537 case ixgbe_mac_82599EB:
5538 case ixgbe_mac_X540:
5539 case ixgbe_mac_X550:
5540 case ixgbe_mac_X550EM_x:
5541 case ixgbe_mac_x550em_a:
5543 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5544 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5548 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
5549 * specifically only auto mask tx and rx interrupts */
5550 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5553 /* XXX: to interrupt immediately for EICS writes, enable this */
5554 /* gpie |= IXGBE_GPIE_EIMEN; */
5556 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5557 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5559 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5560 case IXGBE_82599_VMDQ_8Q_MASK:
5561 gpie |= IXGBE_GPIE_VTMODE_16;
5563 case IXGBE_82599_VMDQ_4Q_MASK:
5564 gpie |= IXGBE_GPIE_VTMODE_32;
5567 gpie |= IXGBE_GPIE_VTMODE_64;
5572 /* Enable Thermal over heat sensor interrupt */
5573 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5574 switch (adapter->hw.mac.type) {
5575 case ixgbe_mac_82599EB:
5576 gpie |= IXGBE_SDP0_GPIEN_8259X;
5583 /* Enable fan failure interrupt */
5584 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5585 gpie |= IXGBE_SDP1_GPIEN(hw);
5587 switch (hw->mac.type) {
5588 case ixgbe_mac_82599EB:
5589 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5591 case ixgbe_mac_X550EM_x:
5592 case ixgbe_mac_x550em_a:
5593 gpie |= IXGBE_SDP0_GPIEN_X540;
5599 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5602 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5604 struct ixgbe_hw *hw = &adapter->hw;
5608 ixgbe_get_hw_control(adapter);
5609 ixgbe_setup_gpie(adapter);
5611 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5612 ixgbe_configure_msix(adapter);
5614 ixgbe_configure_msi_and_legacy(adapter);
5616 /* enable the optics for 82599 SFP+ fiber */
5617 if (hw->mac.ops.enable_tx_laser)
5618 hw->mac.ops.enable_tx_laser(hw);
5620 if (hw->phy.ops.set_phy_power)
5621 hw->phy.ops.set_phy_power(hw, true);
5623 smp_mb__before_atomic();
5624 clear_bit(__IXGBE_DOWN, &adapter->state);
5625 ixgbe_napi_enable_all(adapter);
5627 if (ixgbe_is_sfp(hw)) {
5628 ixgbe_sfp_link_config(adapter);
5630 err = ixgbe_non_sfp_link_config(hw);
5632 e_err(probe, "link_config FAILED %d\n", err);
5635 /* clear any pending interrupts, may auto mask */
5636 IXGBE_READ_REG(hw, IXGBE_EICR);
5637 ixgbe_irq_enable(adapter, true, true);
5640 * If this adapter has a fan, check to see if we had a failure
5641 * before we enabled the interrupt.
5643 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5644 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5645 if (esdp & IXGBE_ESDP_SDP1)
5646 e_crit(drv, "Fan has stopped, replace the adapter\n");
5649 /* bring the link up in the watchdog, this could race with our first
5650 * link up interrupt but shouldn't be a problem */
5651 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5652 adapter->link_check_timeout = jiffies;
5653 mod_timer(&adapter->service_timer, jiffies);
5655 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5656 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5657 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5658 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5661 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5663 WARN_ON(in_interrupt());
5664 /* put off any impending NetWatchDogTimeout */
5665 netif_trans_update(adapter->netdev);
5667 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5668 usleep_range(1000, 2000);
5669 if (adapter->hw.phy.type == ixgbe_phy_fw)
5670 ixgbe_watchdog_link_is_down(adapter);
5671 ixgbe_down(adapter);
5673 * If SR-IOV enabled then wait a bit before bringing the adapter
5674 * back up to give the VFs time to respond to the reset. The
5675 * two second wait is based upon the watchdog timer cycle in
5678 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5681 clear_bit(__IXGBE_RESETTING, &adapter->state);
5684 void ixgbe_up(struct ixgbe_adapter *adapter)
5686 /* hardware has been reset, we need to reload some things */
5687 ixgbe_configure(adapter);
5689 ixgbe_up_complete(adapter);
5692 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5696 pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5698 switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5699 case IXGBE_PCIDEVCTRL2_17_34s:
5700 case IXGBE_PCIDEVCTRL2_4_8s:
5701 /* For now we cap the upper limit on delay to 2 seconds
5702 * as we end up going up to 34 seconds of delay in worst
5703 * case timeout value.
5705 case IXGBE_PCIDEVCTRL2_1_2s:
5706 return 2000000ul; /* 2.0 s */
5707 case IXGBE_PCIDEVCTRL2_260_520ms:
5708 return 520000ul; /* 520 ms */
5709 case IXGBE_PCIDEVCTRL2_65_130ms:
5710 return 130000ul; /* 130 ms */
5711 case IXGBE_PCIDEVCTRL2_16_32ms:
5712 return 32000ul; /* 32 ms */
5713 case IXGBE_PCIDEVCTRL2_1_2ms:
5714 return 2000ul; /* 2 ms */
5715 case IXGBE_PCIDEVCTRL2_50_100us:
5716 return 100ul; /* 100 us */
5717 case IXGBE_PCIDEVCTRL2_16_32ms_def:
5718 return 32000ul; /* 32 ms */
5723 /* We shouldn't need to hit this path, but just in case default as
5724 * though completion timeout is not supported and support 32ms.
5729 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5731 unsigned long wait_delay, delay_interval;
5732 struct ixgbe_hw *hw = &adapter->hw;
5736 /* disable receives */
5737 hw->mac.ops.disable_rx(hw);
5739 if (ixgbe_removed(hw->hw_addr))
5742 /* disable all enabled Rx queues */
5743 for (i = 0; i < adapter->num_rx_queues; i++) {
5744 struct ixgbe_ring *ring = adapter->rx_ring[i];
5745 u8 reg_idx = ring->reg_idx;
5747 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5748 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5749 rxdctl |= IXGBE_RXDCTL_SWFLSH;
5751 /* write value back with RXDCTL.ENABLE bit cleared */
5752 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5755 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5756 if (hw->mac.type == ixgbe_mac_82598EB &&
5757 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5760 /* Determine our minimum delay interval. We will increase this value
5761 * with each subsequent test. This way if the device returns quickly
5762 * we should spend as little time as possible waiting, however as
5763 * the time increases we will wait for larger periods of time.
5765 * The trick here is that we increase the interval using the
5766 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5767 * of that wait is that it totals up to 100x whatever interval we
5768 * choose. Since our minimum wait is 100us we can just divide the
5769 * total timeout by 100 to get our minimum delay interval.
5771 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5773 wait_loop = IXGBE_MAX_RX_DESC_POLL;
5774 wait_delay = delay_interval;
5776 while (wait_loop--) {
5777 usleep_range(wait_delay, wait_delay + 10);
5778 wait_delay += delay_interval * 2;
5781 /* OR together the reading of all the active RXDCTL registers,
5782 * and then test the result. We need the disable to complete
5783 * before we start freeing the memory and invalidating the
5786 for (i = 0; i < adapter->num_rx_queues; i++) {
5787 struct ixgbe_ring *ring = adapter->rx_ring[i];
5788 u8 reg_idx = ring->reg_idx;
5790 rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5793 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5798 "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5801 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5803 unsigned long wait_delay, delay_interval;
5804 struct ixgbe_hw *hw = &adapter->hw;
5808 if (ixgbe_removed(hw->hw_addr))
5811 /* disable all enabled Tx queues */
5812 for (i = 0; i < adapter->num_tx_queues; i++) {
5813 struct ixgbe_ring *ring = adapter->tx_ring[i];
5814 u8 reg_idx = ring->reg_idx;
5816 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5819 /* disable all enabled XDP Tx queues */
5820 for (i = 0; i < adapter->num_xdp_queues; i++) {
5821 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5822 u8 reg_idx = ring->reg_idx;
5824 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5827 /* If the link is not up there shouldn't be much in the way of
5828 * pending transactions. Those that are left will be flushed out
5829 * when the reset logic goes through the flush sequence to clean out
5830 * the pending Tx transactions.
5832 if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5833 goto dma_engine_disable;
5835 /* Determine our minimum delay interval. We will increase this value
5836 * with each subsequent test. This way if the device returns quickly
5837 * we should spend as little time as possible waiting, however as
5838 * the time increases we will wait for larger periods of time.
5840 * The trick here is that we increase the interval using the
5841 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5842 * of that wait is that it totals up to 100x whatever interval we
5843 * choose. Since our minimum wait is 100us we can just divide the
5844 * total timeout by 100 to get our minimum delay interval.
5846 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5848 wait_loop = IXGBE_MAX_RX_DESC_POLL;
5849 wait_delay = delay_interval;
5851 while (wait_loop--) {
5852 usleep_range(wait_delay, wait_delay + 10);
5853 wait_delay += delay_interval * 2;
5856 /* OR together the reading of all the active TXDCTL registers,
5857 * and then test the result. We need the disable to complete
5858 * before we start freeing the memory and invalidating the
5861 for (i = 0; i < adapter->num_tx_queues; i++) {
5862 struct ixgbe_ring *ring = adapter->tx_ring[i];
5863 u8 reg_idx = ring->reg_idx;
5865 txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5867 for (i = 0; i < adapter->num_xdp_queues; i++) {
5868 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5869 u8 reg_idx = ring->reg_idx;
5871 txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5874 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5875 goto dma_engine_disable;
5879 "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5882 /* Disable the Tx DMA engine on 82599 and later MAC */
5883 switch (hw->mac.type) {
5884 case ixgbe_mac_82599EB:
5885 case ixgbe_mac_X540:
5886 case ixgbe_mac_X550:
5887 case ixgbe_mac_X550EM_x:
5888 case ixgbe_mac_x550em_a:
5889 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5890 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5891 ~IXGBE_DMATXCTL_TE));
5898 void ixgbe_reset(struct ixgbe_adapter *adapter)
5900 struct ixgbe_hw *hw = &adapter->hw;
5901 struct net_device *netdev = adapter->netdev;
5904 if (ixgbe_removed(hw->hw_addr))
5906 /* lock SFP init bit to prevent race conditions with the watchdog */
5907 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5908 usleep_range(1000, 2000);
5910 /* clear all SFP and link config related flags while holding SFP_INIT */
5911 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5912 IXGBE_FLAG2_SFP_NEEDS_RESET);
5913 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5915 err = hw->mac.ops.init_hw(hw);
5918 case IXGBE_ERR_SFP_NOT_PRESENT:
5919 case IXGBE_ERR_SFP_NOT_SUPPORTED:
5921 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5922 e_dev_err("master disable timed out\n");
5924 case IXGBE_ERR_EEPROM_VERSION:
5925 /* We are running on a pre-production device, log a warning */
5926 e_dev_warn("This device is a pre-production adapter/LOM. "
5927 "Please be aware there may be issues associated with "
5928 "your hardware. If you are experiencing problems "
5929 "please contact your Intel or hardware "
5930 "representative who provided you with this "
5934 e_dev_err("Hardware Error: %d\n", err);
5937 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5939 /* flush entries out of MAC table */
5940 ixgbe_flush_sw_mac_table(adapter);
5941 __dev_uc_unsync(netdev, NULL);
5943 /* do not flush user set addresses */
5944 ixgbe_mac_set_default_filter(adapter);
5946 /* update SAN MAC vmdq pool selection */
5947 if (hw->mac.san_mac_rar_index)
5948 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5950 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5951 ixgbe_ptp_reset(adapter);
5953 if (hw->phy.ops.set_phy_power) {
5954 if (!netif_running(adapter->netdev) && !adapter->wol)
5955 hw->phy.ops.set_phy_power(hw, false);
5957 hw->phy.ops.set_phy_power(hw, true);
5962 * ixgbe_clean_tx_ring - Free Tx Buffers
5963 * @tx_ring: ring to be cleaned
5965 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5967 u16 i = tx_ring->next_to_clean;
5968 struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
5970 if (tx_ring->xsk_umem) {
5971 ixgbe_xsk_clean_tx_ring(tx_ring);
5975 while (i != tx_ring->next_to_use) {
5976 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
5978 /* Free all the Tx ring sk_buffs */
5979 if (ring_is_xdp(tx_ring))
5980 xdp_return_frame(tx_buffer->xdpf);
5982 dev_kfree_skb_any(tx_buffer->skb);
5984 /* unmap skb header data */
5985 dma_unmap_single(tx_ring->dev,
5986 dma_unmap_addr(tx_buffer, dma),
5987 dma_unmap_len(tx_buffer, len),
5990 /* check for eop_desc to determine the end of the packet */
5991 eop_desc = tx_buffer->next_to_watch;
5992 tx_desc = IXGBE_TX_DESC(tx_ring, i);
5994 /* unmap remaining buffers */
5995 while (tx_desc != eop_desc) {
5999 if (unlikely(i == tx_ring->count)) {
6001 tx_buffer = tx_ring->tx_buffer_info;
6002 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6005 /* unmap any remaining paged data */
6006 if (dma_unmap_len(tx_buffer, len))
6007 dma_unmap_page(tx_ring->dev,
6008 dma_unmap_addr(tx_buffer, dma),
6009 dma_unmap_len(tx_buffer, len),
6013 /* move us one more past the eop_desc for start of next pkt */
6016 if (unlikely(i == tx_ring->count)) {
6018 tx_buffer = tx_ring->tx_buffer_info;
6022 /* reset BQL for queue */
6023 if (!ring_is_xdp(tx_ring))
6024 netdev_tx_reset_queue(txring_txq(tx_ring));
6027 /* reset next_to_use and next_to_clean */
6028 tx_ring->next_to_use = 0;
6029 tx_ring->next_to_clean = 0;
6033 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6034 * @adapter: board private structure
6036 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6040 for (i = 0; i < adapter->num_rx_queues; i++)
6041 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6045 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6046 * @adapter: board private structure
6048 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6052 for (i = 0; i < adapter->num_tx_queues; i++)
6053 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6054 for (i = 0; i < adapter->num_xdp_queues; i++)
6055 ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6058 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6060 struct hlist_node *node2;
6061 struct ixgbe_fdir_filter *filter;
6063 spin_lock(&adapter->fdir_perfect_lock);
6065 hlist_for_each_entry_safe(filter, node2,
6066 &adapter->fdir_filter_list, fdir_node) {
6067 hlist_del(&filter->fdir_node);
6070 adapter->fdir_filter_count = 0;
6072 spin_unlock(&adapter->fdir_perfect_lock);
6075 void ixgbe_down(struct ixgbe_adapter *adapter)
6077 struct net_device *netdev = adapter->netdev;
6078 struct ixgbe_hw *hw = &adapter->hw;
6081 /* signal that we are down to the interrupt handler */
6082 if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6083 return; /* do nothing if already down */
6085 /* Shut off incoming Tx traffic */
6086 netif_tx_stop_all_queues(netdev);
6088 /* call carrier off first to avoid false dev_watchdog timeouts */
6089 netif_carrier_off(netdev);
6090 netif_tx_disable(netdev);
6093 ixgbe_disable_rx(adapter);
6095 /* synchronize_rcu() needed for pending XDP buffers to drain */
6096 if (adapter->xdp_ring[0])
6099 ixgbe_irq_disable(adapter);
6101 ixgbe_napi_disable_all(adapter);
6103 clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6104 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6105 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6107 del_timer_sync(&adapter->service_timer);
6109 if (adapter->num_vfs) {
6110 /* Clear EITR Select mapping */
6111 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6113 /* Mark all the VFs as inactive */
6114 for (i = 0 ; i < adapter->num_vfs; i++)
6115 adapter->vfinfo[i].clear_to_send = false;
6117 /* ping all the active vfs to let them know we are going down */
6118 ixgbe_ping_all_vfs(adapter);
6120 /* Disable all VFTE/VFRE TX/RX */
6121 ixgbe_disable_tx_rx(adapter);
6124 /* disable transmits in the hardware now that interrupts are off */
6125 ixgbe_disable_tx(adapter);
6127 if (!pci_channel_offline(adapter->pdev))
6128 ixgbe_reset(adapter);
6130 /* power down the optics for 82599 SFP+ fiber */
6131 if (hw->mac.ops.disable_tx_laser)
6132 hw->mac.ops.disable_tx_laser(hw);
6134 ixgbe_clean_all_tx_rings(adapter);
6135 ixgbe_clean_all_rx_rings(adapter);
6139 * ixgbe_eee_capable - helper function to determine EEE support on X550
6140 * @adapter: board private structure
6142 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6144 struct ixgbe_hw *hw = &adapter->hw;
6146 switch (hw->device_id) {
6147 case IXGBE_DEV_ID_X550EM_A_1G_T:
6148 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6149 if (!hw->phy.eee_speeds_supported)
6151 adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6152 if (!hw->phy.eee_speeds_advertised)
6154 adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6157 adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6158 adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6164 * ixgbe_tx_timeout - Respond to a Tx Hang
6165 * @netdev: network interface device structure
6167 static void ixgbe_tx_timeout(struct net_device *netdev, unsigned int txqueue)
6169 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6171 /* Do the reset outside of interrupt context */
6172 ixgbe_tx_timeout_reset(adapter);
6175 #ifdef CONFIG_IXGBE_DCB
6176 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6178 struct ixgbe_hw *hw = &adapter->hw;
6179 struct tc_configuration *tc;
6182 switch (hw->mac.type) {
6183 case ixgbe_mac_82598EB:
6184 case ixgbe_mac_82599EB:
6185 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6186 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6188 case ixgbe_mac_X540:
6189 case ixgbe_mac_X550:
6190 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6191 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6193 case ixgbe_mac_X550EM_x:
6194 case ixgbe_mac_x550em_a:
6196 adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6197 adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6201 /* Configure DCB traffic classes */
6202 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6203 tc = &adapter->dcb_cfg.tc_config[j];
6204 tc->path[DCB_TX_CONFIG].bwg_id = 0;
6205 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6206 tc->path[DCB_RX_CONFIG].bwg_id = 0;
6207 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6208 tc->dcb_pfc = pfc_disabled;
6211 /* Initialize default user to priority mapping, UPx->TC0 */
6212 tc = &adapter->dcb_cfg.tc_config[0];
6213 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6214 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6216 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6217 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6218 adapter->dcb_cfg.pfc_mode_enable = false;
6219 adapter->dcb_set_bitmap = 0x00;
6220 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6221 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6222 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6223 sizeof(adapter->temp_dcb_cfg));
6228 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6229 * @adapter: board private structure to initialize
6230 * @ii: pointer to ixgbe_info for device
6232 * ixgbe_sw_init initializes the Adapter private data structure.
6233 * Fields are initialized based on PCI device information and
6234 * OS network device settings (MTU size).
6236 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6237 const struct ixgbe_info *ii)
6239 struct ixgbe_hw *hw = &adapter->hw;
6240 struct pci_dev *pdev = adapter->pdev;
6241 unsigned int rss, fdir;
6245 /* PCI config space info */
6247 hw->vendor_id = pdev->vendor;
6248 hw->device_id = pdev->device;
6249 hw->revision_id = pdev->revision;
6250 hw->subsystem_vendor_id = pdev->subsystem_vendor;
6251 hw->subsystem_device_id = pdev->subsystem_device;
6253 /* get_invariants needs the device IDs */
6254 ii->get_invariants(hw);
6256 /* Set common capability flags and settings */
6257 rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6258 adapter->ring_feature[RING_F_RSS].limit = rss;
6259 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6260 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6261 adapter->atr_sample_rate = 20;
6262 fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6263 adapter->ring_feature[RING_F_FDIR].limit = fdir;
6264 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6265 adapter->ring_feature[RING_F_VMDQ].limit = 1;
6266 #ifdef CONFIG_IXGBE_DCA
6267 adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6269 #ifdef CONFIG_IXGBE_DCB
6270 adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6271 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6274 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6275 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6276 #ifdef CONFIG_IXGBE_DCB
6277 /* Default traffic class to use for FCoE */
6278 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6279 #endif /* CONFIG_IXGBE_DCB */
6280 #endif /* IXGBE_FCOE */
6282 /* initialize static ixgbe jump table entries */
6283 adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6285 if (!adapter->jump_tables[0])
6287 adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6289 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6290 adapter->jump_tables[i] = NULL;
6292 adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6293 sizeof(struct ixgbe_mac_addr),
6295 if (!adapter->mac_table)
6298 if (ixgbe_init_rss_key(adapter))
6301 adapter->af_xdp_zc_qps = bitmap_zalloc(MAX_XDP_QUEUES, GFP_KERNEL);
6302 if (!adapter->af_xdp_zc_qps)
6305 /* Set MAC specific capability flags and exceptions */
6306 switch (hw->mac.type) {
6307 case ixgbe_mac_82598EB:
6308 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6310 if (hw->device_id == IXGBE_DEV_ID_82598AT)
6311 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6313 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6314 adapter->ring_feature[RING_F_FDIR].limit = 0;
6315 adapter->atr_sample_rate = 0;
6316 adapter->fdir_pballoc = 0;
6318 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6319 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6320 #ifdef CONFIG_IXGBE_DCB
6321 adapter->fcoe.up = 0;
6322 #endif /* IXGBE_DCB */
6323 #endif /* IXGBE_FCOE */
6325 case ixgbe_mac_82599EB:
6326 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6327 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6329 case ixgbe_mac_X540:
6330 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6331 if (fwsm & IXGBE_FWSM_TS_ENABLED)
6332 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6334 case ixgbe_mac_x550em_a:
6335 adapter->flags |= IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE;
6336 switch (hw->device_id) {
6337 case IXGBE_DEV_ID_X550EM_A_1G_T:
6338 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6339 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6345 case ixgbe_mac_X550EM_x:
6346 #ifdef CONFIG_IXGBE_DCB
6347 adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6350 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6351 #ifdef CONFIG_IXGBE_DCB
6352 adapter->fcoe.up = 0;
6353 #endif /* IXGBE_DCB */
6354 #endif /* IXGBE_FCOE */
6356 case ixgbe_mac_X550:
6357 if (hw->mac.type == ixgbe_mac_X550)
6358 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6359 #ifdef CONFIG_IXGBE_DCA
6360 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6362 adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
6369 /* FCoE support exists, always init the FCoE lock */
6370 spin_lock_init(&adapter->fcoe.lock);
6373 /* n-tuple support exists, always init our spinlock */
6374 spin_lock_init(&adapter->fdir_perfect_lock);
6376 #ifdef CONFIG_IXGBE_DCB
6377 ixgbe_init_dcb(adapter);
6379 ixgbe_init_ipsec_offload(adapter);
6381 /* default flow control settings */
6382 hw->fc.requested_mode = ixgbe_fc_full;
6383 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
6384 ixgbe_pbthresh_setup(adapter);
6385 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6386 hw->fc.send_xon = true;
6387 hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6389 #ifdef CONFIG_PCI_IOV
6391 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6393 /* assign number of SR-IOV VFs */
6394 if (hw->mac.type != ixgbe_mac_82598EB) {
6395 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6397 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6400 #endif /* CONFIG_PCI_IOV */
6402 /* enable itr by default in dynamic mode */
6403 adapter->rx_itr_setting = 1;
6404 adapter->tx_itr_setting = 1;
6406 /* set default ring sizes */
6407 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6408 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6410 /* set default work limits */
6411 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6413 /* initialize eeprom parameters */
6414 if (ixgbe_init_eeprom_params_generic(hw)) {
6415 e_dev_err("EEPROM initialization failed\n");
6419 /* PF holds first pool slot */
6420 set_bit(0, adapter->fwd_bitmask);
6421 set_bit(__IXGBE_DOWN, &adapter->state);
6427 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6428 * @tx_ring: tx descriptor ring (for a specific queue) to setup
6430 * Return 0 on success, negative on failure
6432 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6434 struct device *dev = tx_ring->dev;
6435 int orig_node = dev_to_node(dev);
6436 int ring_node = NUMA_NO_NODE;
6439 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6441 if (tx_ring->q_vector)
6442 ring_node = tx_ring->q_vector->numa_node;
6444 tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6445 if (!tx_ring->tx_buffer_info)
6446 tx_ring->tx_buffer_info = vmalloc(size);
6447 if (!tx_ring->tx_buffer_info)
6450 /* round up to nearest 4K */
6451 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6452 tx_ring->size = ALIGN(tx_ring->size, 4096);
6454 set_dev_node(dev, ring_node);
6455 tx_ring->desc = dma_alloc_coherent(dev,
6459 set_dev_node(dev, orig_node);
6461 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6462 &tx_ring->dma, GFP_KERNEL);
6466 tx_ring->next_to_use = 0;
6467 tx_ring->next_to_clean = 0;
6471 vfree(tx_ring->tx_buffer_info);
6472 tx_ring->tx_buffer_info = NULL;
6473 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6478 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6479 * @adapter: board private structure
6481 * If this function returns with an error, then it's possible one or
6482 * more of the rings is populated (while the rest are not). It is the
6483 * callers duty to clean those orphaned rings.
6485 * Return 0 on success, negative on failure
6487 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6489 int i, j = 0, err = 0;
6491 for (i = 0; i < adapter->num_tx_queues; i++) {
6492 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6496 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6499 for (j = 0; j < adapter->num_xdp_queues; j++) {
6500 err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6504 e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6510 /* rewind the index freeing the rings as we go */
6512 ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6514 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6519 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6520 * @adapter: pointer to ixgbe_adapter
6521 * @rx_ring: rx descriptor ring (for a specific queue) to setup
6523 * Returns 0 on success, negative on failure
6525 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6526 struct ixgbe_ring *rx_ring)
6528 struct device *dev = rx_ring->dev;
6529 int orig_node = dev_to_node(dev);
6530 int ring_node = NUMA_NO_NODE;
6533 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6535 if (rx_ring->q_vector)
6536 ring_node = rx_ring->q_vector->numa_node;
6538 rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6539 if (!rx_ring->rx_buffer_info)
6540 rx_ring->rx_buffer_info = vmalloc(size);
6541 if (!rx_ring->rx_buffer_info)
6544 /* Round up to nearest 4K */
6545 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6546 rx_ring->size = ALIGN(rx_ring->size, 4096);
6548 set_dev_node(dev, ring_node);
6549 rx_ring->desc = dma_alloc_coherent(dev,
6553 set_dev_node(dev, orig_node);
6555 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6556 &rx_ring->dma, GFP_KERNEL);
6560 rx_ring->next_to_clean = 0;
6561 rx_ring->next_to_use = 0;
6563 /* XDP RX-queue info */
6564 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6565 rx_ring->queue_index) < 0)
6568 rx_ring->xdp_prog = adapter->xdp_prog;
6572 vfree(rx_ring->rx_buffer_info);
6573 rx_ring->rx_buffer_info = NULL;
6574 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6579 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6580 * @adapter: board private structure
6582 * If this function returns with an error, then it's possible one or
6583 * more of the rings is populated (while the rest are not). It is the
6584 * callers duty to clean those orphaned rings.
6586 * Return 0 on success, negative on failure
6588 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6592 for (i = 0; i < adapter->num_rx_queues; i++) {
6593 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6597 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6602 err = ixgbe_setup_fcoe_ddp_resources(adapter);
6607 /* rewind the index freeing the rings as we go */
6609 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6614 * ixgbe_free_tx_resources - Free Tx Resources per Queue
6615 * @tx_ring: Tx descriptor ring for a specific queue
6617 * Free all transmit software resources
6619 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6621 ixgbe_clean_tx_ring(tx_ring);
6623 vfree(tx_ring->tx_buffer_info);
6624 tx_ring->tx_buffer_info = NULL;
6626 /* if not set, then don't free */
6630 dma_free_coherent(tx_ring->dev, tx_ring->size,
6631 tx_ring->desc, tx_ring->dma);
6633 tx_ring->desc = NULL;
6637 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6638 * @adapter: board private structure
6640 * Free all transmit software resources
6642 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6646 for (i = 0; i < adapter->num_tx_queues; i++)
6647 if (adapter->tx_ring[i]->desc)
6648 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6649 for (i = 0; i < adapter->num_xdp_queues; i++)
6650 if (adapter->xdp_ring[i]->desc)
6651 ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6655 * ixgbe_free_rx_resources - Free Rx Resources
6656 * @rx_ring: ring to clean the resources from
6658 * Free all receive software resources
6660 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6662 ixgbe_clean_rx_ring(rx_ring);
6664 rx_ring->xdp_prog = NULL;
6665 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6666 vfree(rx_ring->rx_buffer_info);
6667 rx_ring->rx_buffer_info = NULL;
6669 /* if not set, then don't free */
6673 dma_free_coherent(rx_ring->dev, rx_ring->size,
6674 rx_ring->desc, rx_ring->dma);
6676 rx_ring->desc = NULL;
6680 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6681 * @adapter: board private structure
6683 * Free all receive software resources
6685 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6690 ixgbe_free_fcoe_ddp_resources(adapter);
6693 for (i = 0; i < adapter->num_rx_queues; i++)
6694 if (adapter->rx_ring[i]->desc)
6695 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6699 * ixgbe_change_mtu - Change the Maximum Transfer Unit
6700 * @netdev: network interface device structure
6701 * @new_mtu: new value for maximum frame size
6703 * Returns 0 on success, negative on failure
6705 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6707 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6709 if (adapter->xdp_prog) {
6710 int new_frame_size = new_mtu + ETH_HLEN + ETH_FCS_LEN +
6714 for (i = 0; i < adapter->num_rx_queues; i++) {
6715 struct ixgbe_ring *ring = adapter->rx_ring[i];
6717 if (new_frame_size > ixgbe_rx_bufsz(ring)) {
6718 e_warn(probe, "Requested MTU size is not supported with XDP\n");
6725 * For 82599EB we cannot allow legacy VFs to enable their receive
6726 * paths when MTU greater than 1500 is configured. So display a
6727 * warning that legacy VFs will be disabled.
6729 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6730 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6731 (new_mtu > ETH_DATA_LEN))
6732 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6734 netdev_dbg(netdev, "changing MTU from %d to %d\n",
6735 netdev->mtu, new_mtu);
6737 /* must set new MTU before calling down or up */
6738 netdev->mtu = new_mtu;
6740 if (netif_running(netdev))
6741 ixgbe_reinit_locked(adapter);
6747 * ixgbe_open - Called when a network interface is made active
6748 * @netdev: network interface device structure
6750 * Returns 0 on success, negative value on failure
6752 * The open entry point is called when a network interface is made
6753 * active by the system (IFF_UP). At this point all resources needed
6754 * for transmit and receive operations are allocated, the interrupt
6755 * handler is registered with the OS, the watchdog timer is started,
6756 * and the stack is notified that the interface is ready.
6758 int ixgbe_open(struct net_device *netdev)
6760 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6761 struct ixgbe_hw *hw = &adapter->hw;
6764 /* disallow open during test */
6765 if (test_bit(__IXGBE_TESTING, &adapter->state))
6768 netif_carrier_off(netdev);
6770 /* allocate transmit descriptors */
6771 err = ixgbe_setup_all_tx_resources(adapter);
6775 /* allocate receive descriptors */
6776 err = ixgbe_setup_all_rx_resources(adapter);
6780 ixgbe_configure(adapter);
6782 err = ixgbe_request_irq(adapter);
6786 /* Notify the stack of the actual queue counts. */
6787 queues = adapter->num_tx_queues;
6788 err = netif_set_real_num_tx_queues(netdev, queues);
6790 goto err_set_queues;
6792 queues = adapter->num_rx_queues;
6793 err = netif_set_real_num_rx_queues(netdev, queues);
6795 goto err_set_queues;
6797 ixgbe_ptp_init(adapter);
6799 ixgbe_up_complete(adapter);
6801 ixgbe_clear_udp_tunnel_port(adapter, IXGBE_VXLANCTRL_ALL_UDPPORT_MASK);
6802 udp_tunnel_get_rx_info(netdev);
6807 ixgbe_free_irq(adapter);
6809 ixgbe_free_all_rx_resources(adapter);
6810 if (hw->phy.ops.set_phy_power && !adapter->wol)
6811 hw->phy.ops.set_phy_power(&adapter->hw, false);
6813 ixgbe_free_all_tx_resources(adapter);
6815 ixgbe_reset(adapter);
6820 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6822 ixgbe_ptp_suspend(adapter);
6824 if (adapter->hw.phy.ops.enter_lplu) {
6825 adapter->hw.phy.reset_disable = true;
6826 ixgbe_down(adapter);
6827 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6828 adapter->hw.phy.reset_disable = false;
6830 ixgbe_down(adapter);
6833 ixgbe_free_irq(adapter);
6835 ixgbe_free_all_tx_resources(adapter);
6836 ixgbe_free_all_rx_resources(adapter);
6840 * ixgbe_close - Disables a network interface
6841 * @netdev: network interface device structure
6843 * Returns 0, this is not allowed to fail
6845 * The close entry point is called when an interface is de-activated
6846 * by the OS. The hardware is still under the drivers control, but
6847 * needs to be disabled. A global MAC reset is issued to stop the
6848 * hardware, and all transmit and receive resources are freed.
6850 int ixgbe_close(struct net_device *netdev)
6852 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6854 ixgbe_ptp_stop(adapter);
6856 if (netif_device_present(netdev))
6857 ixgbe_close_suspend(adapter);
6859 ixgbe_fdir_filter_exit(adapter);
6861 ixgbe_release_hw_control(adapter);
6867 static int ixgbe_resume(struct pci_dev *pdev)
6869 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6870 struct net_device *netdev = adapter->netdev;
6873 adapter->hw.hw_addr = adapter->io_addr;
6874 pci_set_power_state(pdev, PCI_D0);
6875 pci_restore_state(pdev);
6877 * pci_restore_state clears dev->state_saved so call
6878 * pci_save_state to restore it.
6880 pci_save_state(pdev);
6882 err = pci_enable_device_mem(pdev);
6884 e_dev_err("Cannot enable PCI device from suspend\n");
6887 smp_mb__before_atomic();
6888 clear_bit(__IXGBE_DISABLED, &adapter->state);
6889 pci_set_master(pdev);
6891 pci_wake_from_d3(pdev, false);
6893 ixgbe_reset(adapter);
6895 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6898 err = ixgbe_init_interrupt_scheme(adapter);
6899 if (!err && netif_running(netdev))
6900 err = ixgbe_open(netdev);
6904 netif_device_attach(netdev);
6909 #endif /* CONFIG_PM */
6911 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6913 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6914 struct net_device *netdev = adapter->netdev;
6915 struct ixgbe_hw *hw = &adapter->hw;
6917 u32 wufc = adapter->wol;
6923 netif_device_detach(netdev);
6925 if (netif_running(netdev))
6926 ixgbe_close_suspend(adapter);
6928 ixgbe_clear_interrupt_scheme(adapter);
6932 retval = pci_save_state(pdev);
6937 if (hw->mac.ops.stop_link_on_d3)
6938 hw->mac.ops.stop_link_on_d3(hw);
6943 ixgbe_set_rx_mode(netdev);
6945 /* enable the optics for 82599 SFP+ fiber as we can WoL */
6946 if (hw->mac.ops.enable_tx_laser)
6947 hw->mac.ops.enable_tx_laser(hw);
6949 /* enable the reception of multicast packets */
6950 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6951 fctrl |= IXGBE_FCTRL_MPE;
6952 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6954 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6955 ctrl |= IXGBE_CTRL_GIO_DIS;
6956 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6958 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6960 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6961 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6964 switch (hw->mac.type) {
6965 case ixgbe_mac_82598EB:
6966 pci_wake_from_d3(pdev, false);
6968 case ixgbe_mac_82599EB:
6969 case ixgbe_mac_X540:
6970 case ixgbe_mac_X550:
6971 case ixgbe_mac_X550EM_x:
6972 case ixgbe_mac_x550em_a:
6973 pci_wake_from_d3(pdev, !!wufc);
6979 *enable_wake = !!wufc;
6980 if (hw->phy.ops.set_phy_power && !*enable_wake)
6981 hw->phy.ops.set_phy_power(hw, false);
6983 ixgbe_release_hw_control(adapter);
6985 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
6986 pci_disable_device(pdev);
6992 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
6997 retval = __ixgbe_shutdown(pdev, &wake);
7002 pci_prepare_to_sleep(pdev);
7004 pci_wake_from_d3(pdev, false);
7005 pci_set_power_state(pdev, PCI_D3hot);
7010 #endif /* CONFIG_PM */
7012 static void ixgbe_shutdown(struct pci_dev *pdev)
7016 __ixgbe_shutdown(pdev, &wake);
7018 if (system_state == SYSTEM_POWER_OFF) {
7019 pci_wake_from_d3(pdev, wake);
7020 pci_set_power_state(pdev, PCI_D3hot);
7025 * ixgbe_update_stats - Update the board statistics counters.
7026 * @adapter: board private structure
7028 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7030 struct net_device *netdev = adapter->netdev;
7031 struct ixgbe_hw *hw = &adapter->hw;
7032 struct ixgbe_hw_stats *hwstats = &adapter->stats;
7034 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7035 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7036 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7037 u64 alloc_rx_page = 0;
7038 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7040 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7041 test_bit(__IXGBE_RESETTING, &adapter->state))
7044 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7047 for (i = 0; i < adapter->num_rx_queues; i++) {
7048 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7049 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7051 adapter->rsc_total_count = rsc_count;
7052 adapter->rsc_total_flush = rsc_flush;
7055 for (i = 0; i < adapter->num_rx_queues; i++) {
7056 struct ixgbe_ring *rx_ring = READ_ONCE(adapter->rx_ring[i]);
7060 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7061 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7062 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7063 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7064 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7065 bytes += rx_ring->stats.bytes;
7066 packets += rx_ring->stats.packets;
7068 adapter->non_eop_descs = non_eop_descs;
7069 adapter->alloc_rx_page = alloc_rx_page;
7070 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7071 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7072 adapter->hw_csum_rx_error = hw_csum_rx_error;
7073 netdev->stats.rx_bytes = bytes;
7074 netdev->stats.rx_packets = packets;
7078 /* gather some stats to the adapter struct that are per queue */
7079 for (i = 0; i < adapter->num_tx_queues; i++) {
7080 struct ixgbe_ring *tx_ring = READ_ONCE(adapter->tx_ring[i]);
7084 restart_queue += tx_ring->tx_stats.restart_queue;
7085 tx_busy += tx_ring->tx_stats.tx_busy;
7086 bytes += tx_ring->stats.bytes;
7087 packets += tx_ring->stats.packets;
7089 for (i = 0; i < adapter->num_xdp_queues; i++) {
7090 struct ixgbe_ring *xdp_ring = READ_ONCE(adapter->xdp_ring[i]);
7094 restart_queue += xdp_ring->tx_stats.restart_queue;
7095 tx_busy += xdp_ring->tx_stats.tx_busy;
7096 bytes += xdp_ring->stats.bytes;
7097 packets += xdp_ring->stats.packets;
7099 adapter->restart_queue = restart_queue;
7100 adapter->tx_busy = tx_busy;
7101 netdev->stats.tx_bytes = bytes;
7102 netdev->stats.tx_packets = packets;
7104 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7106 /* 8 register reads */
7107 for (i = 0; i < 8; i++) {
7108 /* for packet buffers not used, the register should read 0 */
7109 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7111 hwstats->mpc[i] += mpc;
7112 total_mpc += hwstats->mpc[i];
7113 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7114 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7115 switch (hw->mac.type) {
7116 case ixgbe_mac_82598EB:
7117 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7118 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7119 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7120 hwstats->pxonrxc[i] +=
7121 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7123 case ixgbe_mac_82599EB:
7124 case ixgbe_mac_X540:
7125 case ixgbe_mac_X550:
7126 case ixgbe_mac_X550EM_x:
7127 case ixgbe_mac_x550em_a:
7128 hwstats->pxonrxc[i] +=
7129 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7136 /*16 register reads */
7137 for (i = 0; i < 16; i++) {
7138 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7139 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7140 if ((hw->mac.type == ixgbe_mac_82599EB) ||
7141 (hw->mac.type == ixgbe_mac_X540) ||
7142 (hw->mac.type == ixgbe_mac_X550) ||
7143 (hw->mac.type == ixgbe_mac_X550EM_x) ||
7144 (hw->mac.type == ixgbe_mac_x550em_a)) {
7145 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7146 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7147 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7148 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7152 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7153 /* work around hardware counting issue */
7154 hwstats->gprc -= missed_rx;
7156 ixgbe_update_xoff_received(adapter);
7158 /* 82598 hardware only has a 32 bit counter in the high register */
7159 switch (hw->mac.type) {
7160 case ixgbe_mac_82598EB:
7161 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7162 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7163 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7164 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7166 case ixgbe_mac_X540:
7167 case ixgbe_mac_X550:
7168 case ixgbe_mac_X550EM_x:
7169 case ixgbe_mac_x550em_a:
7170 /* OS2BMC stats are X540 and later */
7171 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7172 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7173 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7174 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7176 case ixgbe_mac_82599EB:
7177 for (i = 0; i < 16; i++)
7178 adapter->hw_rx_no_dma_resources +=
7179 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7180 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7181 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7182 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7183 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7184 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7185 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7186 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7187 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7188 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7190 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7191 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7192 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7193 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7194 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7195 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7196 /* Add up per cpu counters for total ddp aloc fail */
7197 if (adapter->fcoe.ddp_pool) {
7198 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7199 struct ixgbe_fcoe_ddp_pool *ddp_pool;
7201 u64 noddp = 0, noddp_ext_buff = 0;
7202 for_each_possible_cpu(cpu) {
7203 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7204 noddp += ddp_pool->noddp;
7205 noddp_ext_buff += ddp_pool->noddp_ext_buff;
7207 hwstats->fcoe_noddp = noddp;
7208 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7210 #endif /* IXGBE_FCOE */
7215 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7216 hwstats->bprc += bprc;
7217 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7218 if (hw->mac.type == ixgbe_mac_82598EB)
7219 hwstats->mprc -= bprc;
7220 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7221 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7222 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7223 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7224 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7225 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7226 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7227 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7228 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7229 hwstats->lxontxc += lxon;
7230 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7231 hwstats->lxofftxc += lxoff;
7232 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7233 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7235 * 82598 errata - tx of flow control packets is included in tx counters
7237 xon_off_tot = lxon + lxoff;
7238 hwstats->gptc -= xon_off_tot;
7239 hwstats->mptc -= xon_off_tot;
7240 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7241 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7242 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7243 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7244 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7245 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7246 hwstats->ptc64 -= xon_off_tot;
7247 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7248 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7249 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7250 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7251 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7252 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7254 /* Fill out the OS statistics structure */
7255 netdev->stats.multicast = hwstats->mprc;
7258 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7259 netdev->stats.rx_dropped = 0;
7260 netdev->stats.rx_length_errors = hwstats->rlec;
7261 netdev->stats.rx_crc_errors = hwstats->crcerrs;
7262 netdev->stats.rx_missed_errors = total_mpc;
7266 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7267 * @adapter: pointer to the device adapter structure
7269 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7271 struct ixgbe_hw *hw = &adapter->hw;
7274 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7277 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7279 /* if interface is down do nothing */
7280 if (test_bit(__IXGBE_DOWN, &adapter->state))
7283 /* do nothing if we are not using signature filters */
7284 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7287 adapter->fdir_overflow++;
7289 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7290 for (i = 0; i < adapter->num_tx_queues; i++)
7291 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7292 &(adapter->tx_ring[i]->state));
7293 for (i = 0; i < adapter->num_xdp_queues; i++)
7294 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7295 &adapter->xdp_ring[i]->state);
7296 /* re-enable flow director interrupts */
7297 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7299 e_err(probe, "failed to finish FDIR re-initialization, "
7300 "ignored adding FDIR ATR filters\n");
7305 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7306 * @adapter: pointer to the device adapter structure
7308 * This function serves two purposes. First it strobes the interrupt lines
7309 * in order to make certain interrupts are occurring. Secondly it sets the
7310 * bits needed to check for TX hangs. As a result we should immediately
7311 * determine if a hang has occurred.
7313 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7315 struct ixgbe_hw *hw = &adapter->hw;
7319 /* If we're down, removing or resetting, just bail */
7320 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7321 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7322 test_bit(__IXGBE_RESETTING, &adapter->state))
7325 /* Force detection of hung controller */
7326 if (netif_carrier_ok(adapter->netdev)) {
7327 for (i = 0; i < adapter->num_tx_queues; i++)
7328 set_check_for_tx_hang(adapter->tx_ring[i]);
7329 for (i = 0; i < adapter->num_xdp_queues; i++)
7330 set_check_for_tx_hang(adapter->xdp_ring[i]);
7333 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7335 * for legacy and MSI interrupts don't set any bits
7336 * that are enabled for EIAM, because this operation
7337 * would set *both* EIMS and EICS for any bit in EIAM
7339 IXGBE_WRITE_REG(hw, IXGBE_EICS,
7340 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7342 /* get one bit for every active tx/rx interrupt vector */
7343 for (i = 0; i < adapter->num_q_vectors; i++) {
7344 struct ixgbe_q_vector *qv = adapter->q_vector[i];
7345 if (qv->rx.ring || qv->tx.ring)
7350 /* Cause software interrupt to ensure rings are cleaned */
7351 ixgbe_irq_rearm_queues(adapter, eics);
7355 * ixgbe_watchdog_update_link - update the link status
7356 * @adapter: pointer to the device adapter structure
7358 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7360 struct ixgbe_hw *hw = &adapter->hw;
7361 u32 link_speed = adapter->link_speed;
7362 bool link_up = adapter->link_up;
7363 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7365 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7368 if (hw->mac.ops.check_link) {
7369 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7371 /* always assume link is up, if no check link function */
7372 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7376 if (adapter->ixgbe_ieee_pfc)
7377 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7379 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7380 hw->mac.ops.fc_enable(hw);
7381 ixgbe_set_rx_drop_en(adapter);
7385 time_after(jiffies, (adapter->link_check_timeout +
7386 IXGBE_TRY_LINK_TIMEOUT))) {
7387 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7388 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7389 IXGBE_WRITE_FLUSH(hw);
7392 adapter->link_up = link_up;
7393 adapter->link_speed = link_speed;
7396 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7398 #ifdef CONFIG_IXGBE_DCB
7399 struct net_device *netdev = adapter->netdev;
7400 struct dcb_app app = {
7401 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7406 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7407 up = dcb_ieee_getapp_mask(netdev, &app);
7409 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7414 * ixgbe_watchdog_link_is_up - update netif_carrier status and
7415 * print link up message
7416 * @adapter: pointer to the device adapter structure
7418 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7420 struct net_device *netdev = adapter->netdev;
7421 struct ixgbe_hw *hw = &adapter->hw;
7422 u32 link_speed = adapter->link_speed;
7423 const char *speed_str;
7424 bool flow_rx, flow_tx;
7426 /* only continue if link was previously down */
7427 if (netif_carrier_ok(netdev))
7430 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7432 switch (hw->mac.type) {
7433 case ixgbe_mac_82598EB: {
7434 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7435 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7436 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7437 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7440 case ixgbe_mac_X540:
7441 case ixgbe_mac_X550:
7442 case ixgbe_mac_X550EM_x:
7443 case ixgbe_mac_x550em_a:
7444 case ixgbe_mac_82599EB: {
7445 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7446 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7447 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7448 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7457 adapter->last_rx_ptp_check = jiffies;
7459 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7460 ixgbe_ptp_start_cyclecounter(adapter);
7462 switch (link_speed) {
7463 case IXGBE_LINK_SPEED_10GB_FULL:
7464 speed_str = "10 Gbps";
7466 case IXGBE_LINK_SPEED_5GB_FULL:
7467 speed_str = "5 Gbps";
7469 case IXGBE_LINK_SPEED_2_5GB_FULL:
7470 speed_str = "2.5 Gbps";
7472 case IXGBE_LINK_SPEED_1GB_FULL:
7473 speed_str = "1 Gbps";
7475 case IXGBE_LINK_SPEED_100_FULL:
7476 speed_str = "100 Mbps";
7478 case IXGBE_LINK_SPEED_10_FULL:
7479 speed_str = "10 Mbps";
7482 speed_str = "unknown speed";
7485 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7486 ((flow_rx && flow_tx) ? "RX/TX" :
7488 (flow_tx ? "TX" : "None"))));
7490 netif_carrier_on(netdev);
7491 ixgbe_check_vf_rate_limit(adapter);
7493 /* enable transmits */
7494 netif_tx_wake_all_queues(adapter->netdev);
7496 /* update the default user priority for VFs */
7497 ixgbe_update_default_up(adapter);
7499 /* ping all the active vfs to let them know link has changed */
7500 ixgbe_ping_all_vfs(adapter);
7504 * ixgbe_watchdog_link_is_down - update netif_carrier status and
7505 * print link down message
7506 * @adapter: pointer to the adapter structure
7508 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7510 struct net_device *netdev = adapter->netdev;
7511 struct ixgbe_hw *hw = &adapter->hw;
7513 adapter->link_up = false;
7514 adapter->link_speed = 0;
7516 /* only continue if link was up previously */
7517 if (!netif_carrier_ok(netdev))
7520 /* poll for SFP+ cable when link is down */
7521 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7522 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7524 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7525 ixgbe_ptp_start_cyclecounter(adapter);
7527 e_info(drv, "NIC Link is Down\n");
7528 netif_carrier_off(netdev);
7530 /* ping all the active vfs to let them know link has changed */
7531 ixgbe_ping_all_vfs(adapter);
7534 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7538 for (i = 0; i < adapter->num_tx_queues; i++) {
7539 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7541 if (tx_ring->next_to_use != tx_ring->next_to_clean)
7545 for (i = 0; i < adapter->num_xdp_queues; i++) {
7546 struct ixgbe_ring *ring = adapter->xdp_ring[i];
7548 if (ring->next_to_use != ring->next_to_clean)
7555 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7557 struct ixgbe_hw *hw = &adapter->hw;
7558 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7559 u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7563 if (!adapter->num_vfs)
7566 /* resetting the PF is only needed for MAC before X550 */
7567 if (hw->mac.type >= ixgbe_mac_X550)
7570 for (i = 0; i < adapter->num_vfs; i++) {
7571 for (j = 0; j < q_per_pool; j++) {
7574 h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7575 t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7586 * ixgbe_watchdog_flush_tx - flush queues on link down
7587 * @adapter: pointer to the device adapter structure
7589 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7591 if (!netif_carrier_ok(adapter->netdev)) {
7592 if (ixgbe_ring_tx_pending(adapter) ||
7593 ixgbe_vf_tx_pending(adapter)) {
7594 /* We've lost link, so the controller stops DMA,
7595 * but we've got queued Tx work that's never going
7596 * to get done, so reset controller to flush Tx.
7597 * (Do the reset outside of interrupt context).
7599 e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7600 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7605 #ifdef CONFIG_PCI_IOV
7606 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7608 struct ixgbe_hw *hw = &adapter->hw;
7609 struct pci_dev *pdev = adapter->pdev;
7613 if (!(netif_carrier_ok(adapter->netdev)))
7616 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7617 if (gpc) /* If incrementing then no need for the check below */
7619 /* Check to see if a bad DMA write target from an errant or
7620 * malicious VF has caused a PCIe error. If so then we can
7621 * issue a VFLR to the offending VF(s) and then resume without
7622 * requesting a full slot reset.
7628 /* check status reg for all VFs owned by this PF */
7629 for (vf = 0; vf < adapter->num_vfs; ++vf) {
7630 struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7635 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7636 if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7637 status_reg & PCI_STATUS_REC_MASTER_ABORT)
7642 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7646 /* Do not perform spoof check for 82598 or if not in IOV mode */
7647 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7648 adapter->num_vfs == 0)
7651 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7654 * ssvpc register is cleared on read, if zero then no
7655 * spoofed packets in the last interval.
7660 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7663 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7668 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7671 #endif /* CONFIG_PCI_IOV */
7675 * ixgbe_watchdog_subtask - check and bring link up
7676 * @adapter: pointer to the device adapter structure
7678 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7680 /* if interface is down, removing or resetting, do nothing */
7681 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7682 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7683 test_bit(__IXGBE_RESETTING, &adapter->state))
7686 ixgbe_watchdog_update_link(adapter);
7688 if (adapter->link_up)
7689 ixgbe_watchdog_link_is_up(adapter);
7691 ixgbe_watchdog_link_is_down(adapter);
7693 ixgbe_check_for_bad_vf(adapter);
7694 ixgbe_spoof_check(adapter);
7695 ixgbe_update_stats(adapter);
7697 ixgbe_watchdog_flush_tx(adapter);
7701 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7702 * @adapter: the ixgbe adapter structure
7704 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7706 struct ixgbe_hw *hw = &adapter->hw;
7709 /* not searching for SFP so there is nothing to do here */
7710 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7711 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7714 if (adapter->sfp_poll_time &&
7715 time_after(adapter->sfp_poll_time, jiffies))
7716 return; /* If not yet time to poll for SFP */
7718 /* someone else is in init, wait until next service event */
7719 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7722 adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7724 err = hw->phy.ops.identify_sfp(hw);
7725 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7728 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7729 /* If no cable is present, then we need to reset
7730 * the next time we find a good cable. */
7731 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7738 /* exit if reset not needed */
7739 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7742 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7745 * A module may be identified correctly, but the EEPROM may not have
7746 * support for that module. setup_sfp() will fail in that case, so
7747 * we should not allow that module to load.
7749 if (hw->mac.type == ixgbe_mac_82598EB)
7750 err = hw->phy.ops.reset(hw);
7752 err = hw->mac.ops.setup_sfp(hw);
7754 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7757 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7758 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7761 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7763 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7764 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7765 e_dev_err("failed to initialize because an unsupported "
7766 "SFP+ module type was detected.\n");
7767 e_dev_err("Reload the driver after installing a "
7768 "supported module.\n");
7769 unregister_netdev(adapter->netdev);
7774 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7775 * @adapter: the ixgbe adapter structure
7777 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7779 struct ixgbe_hw *hw = &adapter->hw;
7782 bool autoneg = false;
7784 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7787 /* someone else is in init, wait until next service event */
7788 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7791 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7793 hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7795 /* advertise highest capable link speed */
7796 if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7797 speed = IXGBE_LINK_SPEED_10GB_FULL;
7799 speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7800 IXGBE_LINK_SPEED_1GB_FULL);
7802 if (hw->mac.ops.setup_link)
7803 hw->mac.ops.setup_link(hw, speed, true);
7805 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7806 adapter->link_check_timeout = jiffies;
7807 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7811 * ixgbe_service_timer - Timer Call-back
7812 * @t: pointer to timer_list structure
7814 static void ixgbe_service_timer(struct timer_list *t)
7816 struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7817 unsigned long next_event_offset;
7819 /* poll faster when waiting for link */
7820 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7821 next_event_offset = HZ / 10;
7823 next_event_offset = HZ * 2;
7825 /* Reset the timer */
7826 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7828 ixgbe_service_event_schedule(adapter);
7831 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7833 struct ixgbe_hw *hw = &adapter->hw;
7836 if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7839 adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7841 if (!hw->phy.ops.handle_lasi)
7844 status = hw->phy.ops.handle_lasi(&adapter->hw);
7845 if (status != IXGBE_ERR_OVERTEMP)
7848 e_crit(drv, "%s\n", ixgbe_overheat_msg);
7851 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7853 if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7857 /* If we're already down, removing or resetting, just bail */
7858 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7859 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7860 test_bit(__IXGBE_RESETTING, &adapter->state)) {
7865 ixgbe_dump(adapter);
7866 netdev_err(adapter->netdev, "Reset adapter\n");
7867 adapter->tx_timeout_count++;
7869 ixgbe_reinit_locked(adapter);
7874 * ixgbe_check_fw_error - Check firmware for errors
7875 * @adapter: the adapter private structure
7877 * Check firmware errors in register FWSM
7879 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
7881 struct ixgbe_hw *hw = &adapter->hw;
7884 /* read fwsm.ext_err_ind register and log errors */
7885 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
7887 if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
7888 !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
7889 e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7892 if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
7893 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");
7901 * ixgbe_service_task - manages and runs subtasks
7902 * @work: pointer to work_struct containing our data
7904 static void ixgbe_service_task(struct work_struct *work)
7906 struct ixgbe_adapter *adapter = container_of(work,
7907 struct ixgbe_adapter,
7909 if (ixgbe_removed(adapter->hw.hw_addr)) {
7910 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7912 ixgbe_down(adapter);
7915 ixgbe_service_event_complete(adapter);
7918 if (ixgbe_check_fw_error(adapter)) {
7919 if (!test_bit(__IXGBE_DOWN, &adapter->state))
7920 unregister_netdev(adapter->netdev);
7921 ixgbe_service_event_complete(adapter);
7924 if (adapter->flags2 & IXGBE_FLAG2_UDP_TUN_REREG_NEEDED) {
7926 adapter->flags2 &= ~IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
7927 udp_tunnel_get_rx_info(adapter->netdev);
7930 ixgbe_reset_subtask(adapter);
7931 ixgbe_phy_interrupt_subtask(adapter);
7932 ixgbe_sfp_detection_subtask(adapter);
7933 ixgbe_sfp_link_config_subtask(adapter);
7934 ixgbe_check_overtemp_subtask(adapter);
7935 ixgbe_watchdog_subtask(adapter);
7936 ixgbe_fdir_reinit_subtask(adapter);
7937 ixgbe_check_hang_subtask(adapter);
7939 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7940 ixgbe_ptp_overflow_check(adapter);
7941 if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
7942 ixgbe_ptp_rx_hang(adapter);
7943 ixgbe_ptp_tx_hang(adapter);
7946 ixgbe_service_event_complete(adapter);
7949 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7950 struct ixgbe_tx_buffer *first,
7952 struct ixgbe_ipsec_tx_data *itd)
7954 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7955 struct sk_buff *skb = first->skb;
7966 u32 paylen, l4_offset;
7967 u32 fceof_saidx = 0;
7970 if (skb->ip_summed != CHECKSUM_PARTIAL)
7973 if (!skb_is_gso(skb))
7976 err = skb_cow_head(skb, 0);
7980 if (eth_p_mpls(first->protocol))
7981 ip.hdr = skb_inner_network_header(skb);
7983 ip.hdr = skb_network_header(skb);
7984 l4.hdr = skb_checksum_start(skb);
7986 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7987 type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
7988 IXGBE_ADVTXD_TUCMD_L4T_UDP : IXGBE_ADVTXD_TUCMD_L4T_TCP;
7990 /* initialize outer IP header fields */
7991 if (ip.v4->version == 4) {
7992 unsigned char *csum_start = skb_checksum_start(skb);
7993 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
7994 int len = csum_start - trans_start;
7996 /* IP header will have to cancel out any data that
7997 * is not a part of the outer IP header, so set to
7998 * a reverse csum if needed, else init check to 0.
8000 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
8001 csum_fold(csum_partial(trans_start,
8003 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
8006 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8007 IXGBE_TX_FLAGS_CSUM |
8008 IXGBE_TX_FLAGS_IPV4;
8010 ip.v6->payload_len = 0;
8011 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8012 IXGBE_TX_FLAGS_CSUM;
8015 /* determine offset of inner transport header */
8016 l4_offset = l4.hdr - skb->data;
8018 /* remove payload length from inner checksum */
8019 paylen = skb->len - l4_offset;
8021 if (type_tucmd & IXGBE_ADVTXD_TUCMD_L4T_TCP) {
8022 /* compute length of segmentation header */
8023 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
8024 csum_replace_by_diff(&l4.tcp->check,
8025 (__force __wsum)htonl(paylen));
8027 /* compute length of segmentation header */
8028 *hdr_len = sizeof(*l4.udp) + l4_offset;
8029 csum_replace_by_diff(&l4.udp->check,
8030 (__force __wsum)htonl(paylen));
8033 /* update gso size and bytecount with header size */
8034 first->gso_segs = skb_shinfo(skb)->gso_segs;
8035 first->bytecount += (first->gso_segs - 1) * *hdr_len;
8037 /* mss_l4len_id: use 0 as index for TSO */
8038 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8039 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8041 fceof_saidx |= itd->sa_idx;
8042 type_tucmd |= itd->flags | itd->trailer_len;
8044 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8045 vlan_macip_lens = l4.hdr - ip.hdr;
8046 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8047 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8049 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8055 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb)
8057 unsigned int offset = 0;
8059 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
8061 return offset == skb_checksum_start_offset(skb);
8064 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8065 struct ixgbe_tx_buffer *first,
8066 struct ixgbe_ipsec_tx_data *itd)
8068 struct sk_buff *skb = first->skb;
8069 u32 vlan_macip_lens = 0;
8070 u32 fceof_saidx = 0;
8073 if (skb->ip_summed != CHECKSUM_PARTIAL) {
8075 if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8076 IXGBE_TX_FLAGS_CC)))
8081 switch (skb->csum_offset) {
8082 case offsetof(struct tcphdr, check):
8083 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8085 case offsetof(struct udphdr, check):
8087 case offsetof(struct sctphdr, checksum):
8088 /* validate that this is actually an SCTP request */
8089 if (((first->protocol == htons(ETH_P_IP)) &&
8090 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
8091 ((first->protocol == htons(ETH_P_IPV6)) &&
8092 ixgbe_ipv6_csum_is_sctp(skb))) {
8093 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8098 skb_checksum_help(skb);
8102 /* update TX checksum flag */
8103 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8104 vlan_macip_lens = skb_checksum_start_offset(skb) -
8105 skb_network_offset(skb);
8107 /* vlan_macip_lens: MACLEN, VLAN tag */
8108 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8109 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8111 fceof_saidx |= itd->sa_idx;
8112 type_tucmd |= itd->flags | itd->trailer_len;
8114 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8117 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8118 ((_flag <= _result) ? \
8119 ((u32)(_input & _flag) * (_result / _flag)) : \
8120 ((u32)(_input & _flag) / (_flag / _result)))
8122 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8124 /* set type for advanced descriptor with frame checksum insertion */
8125 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8126 IXGBE_ADVTXD_DCMD_DEXT |
8127 IXGBE_ADVTXD_DCMD_IFCS;
8129 /* set HW vlan bit if vlan is present */
8130 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8131 IXGBE_ADVTXD_DCMD_VLE);
8133 /* set segmentation enable bits for TSO/FSO */
8134 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8135 IXGBE_ADVTXD_DCMD_TSE);
8137 /* set timestamp bit if present */
8138 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8139 IXGBE_ADVTXD_MAC_TSTAMP);
8141 /* insert frame checksum */
8142 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8147 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8148 u32 tx_flags, unsigned int paylen)
8150 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8152 /* enable L4 checksum for TSO and TX checksum offload */
8153 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8154 IXGBE_TX_FLAGS_CSUM,
8155 IXGBE_ADVTXD_POPTS_TXSM);
8157 /* enable IPv4 checksum for TSO */
8158 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8159 IXGBE_TX_FLAGS_IPV4,
8160 IXGBE_ADVTXD_POPTS_IXSM);
8163 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8164 IXGBE_TX_FLAGS_IPSEC,
8165 IXGBE_ADVTXD_POPTS_IPSEC);
8168 * Check Context must be set if Tx switch is enabled, which it
8169 * always is for case where virtual functions are running
8171 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8175 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8178 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8180 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8182 /* Herbert's original patch had:
8183 * smp_mb__after_netif_stop_queue();
8184 * but since that doesn't exist yet, just open code it.
8188 /* We need to check again in a case another CPU has just
8189 * made room available.
8191 if (likely(ixgbe_desc_unused(tx_ring) < size))
8194 /* A reprieve! - use start_queue because it doesn't call schedule */
8195 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8196 ++tx_ring->tx_stats.restart_queue;
8200 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8202 if (likely(ixgbe_desc_unused(tx_ring) >= size))
8205 return __ixgbe_maybe_stop_tx(tx_ring, size);
8208 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8209 struct ixgbe_tx_buffer *first,
8212 struct sk_buff *skb = first->skb;
8213 struct ixgbe_tx_buffer *tx_buffer;
8214 union ixgbe_adv_tx_desc *tx_desc;
8217 unsigned int data_len, size;
8218 u32 tx_flags = first->tx_flags;
8219 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8220 u16 i = tx_ring->next_to_use;
8222 tx_desc = IXGBE_TX_DESC(tx_ring, i);
8224 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8226 size = skb_headlen(skb);
8227 data_len = skb->data_len;
8230 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8231 if (data_len < sizeof(struct fcoe_crc_eof)) {
8232 size -= sizeof(struct fcoe_crc_eof) - data_len;
8235 data_len -= sizeof(struct fcoe_crc_eof);
8240 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8244 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8245 if (dma_mapping_error(tx_ring->dev, dma))
8248 /* record length, and DMA address */
8249 dma_unmap_len_set(tx_buffer, len, size);
8250 dma_unmap_addr_set(tx_buffer, dma, dma);
8252 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8254 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8255 tx_desc->read.cmd_type_len =
8256 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8260 if (i == tx_ring->count) {
8261 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8264 tx_desc->read.olinfo_status = 0;
8266 dma += IXGBE_MAX_DATA_PER_TXD;
8267 size -= IXGBE_MAX_DATA_PER_TXD;
8269 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8272 if (likely(!data_len))
8275 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8279 if (i == tx_ring->count) {
8280 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8283 tx_desc->read.olinfo_status = 0;
8286 size = min_t(unsigned int, data_len, skb_frag_size(frag));
8288 size = skb_frag_size(frag);
8292 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8295 tx_buffer = &tx_ring->tx_buffer_info[i];
8298 /* write last descriptor with RS and EOP bits */
8299 cmd_type |= size | IXGBE_TXD_CMD;
8300 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8302 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8304 /* set the timestamp */
8305 first->time_stamp = jiffies;
8307 skb_tx_timestamp(skb);
8310 * Force memory writes to complete before letting h/w know there
8311 * are new descriptors to fetch. (Only applicable for weak-ordered
8312 * memory model archs, such as IA-64).
8314 * We also need this memory barrier to make certain all of the
8315 * status bits have been updated before next_to_watch is written.
8319 /* set next_to_watch value indicating a packet is present */
8320 first->next_to_watch = tx_desc;
8323 if (i == tx_ring->count)
8326 tx_ring->next_to_use = i;
8328 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8330 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
8331 writel(i, tx_ring->tail);
8336 dev_err(tx_ring->dev, "TX DMA map failed\n");
8338 /* clear dma mappings for failed tx_buffer_info map */
8340 tx_buffer = &tx_ring->tx_buffer_info[i];
8341 if (dma_unmap_len(tx_buffer, len))
8342 dma_unmap_page(tx_ring->dev,
8343 dma_unmap_addr(tx_buffer, dma),
8344 dma_unmap_len(tx_buffer, len),
8346 dma_unmap_len_set(tx_buffer, len, 0);
8347 if (tx_buffer == first)
8350 i += tx_ring->count;
8354 dev_kfree_skb_any(first->skb);
8357 tx_ring->next_to_use = i;
8362 static void ixgbe_atr(struct ixgbe_ring *ring,
8363 struct ixgbe_tx_buffer *first)
8365 struct ixgbe_q_vector *q_vector = ring->q_vector;
8366 union ixgbe_atr_hash_dword input = { .dword = 0 };
8367 union ixgbe_atr_hash_dword common = { .dword = 0 };
8369 unsigned char *network;
8371 struct ipv6hdr *ipv6;
8375 struct sk_buff *skb;
8379 /* if ring doesn't have a interrupt vector, cannot perform ATR */
8383 /* do nothing if sampling is disabled */
8384 if (!ring->atr_sample_rate)
8389 /* currently only IPv4/IPv6 with TCP is supported */
8390 if ((first->protocol != htons(ETH_P_IP)) &&
8391 (first->protocol != htons(ETH_P_IPV6)))
8394 /* snag network header to get L4 type and address */
8396 hdr.network = skb_network_header(skb);
8397 if (unlikely(hdr.network <= skb->data))
8399 if (skb->encapsulation &&
8400 first->protocol == htons(ETH_P_IP) &&
8401 hdr.ipv4->protocol == IPPROTO_UDP) {
8402 struct ixgbe_adapter *adapter = q_vector->adapter;
8404 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8408 /* verify the port is recognized as VXLAN */
8409 if (adapter->vxlan_port &&
8410 udp_hdr(skb)->dest == adapter->vxlan_port)
8411 hdr.network = skb_inner_network_header(skb);
8413 if (adapter->geneve_port &&
8414 udp_hdr(skb)->dest == adapter->geneve_port)
8415 hdr.network = skb_inner_network_header(skb);
8418 /* Make sure we have at least [minimum IPv4 header + TCP]
8419 * or [IPv6 header] bytes
8421 if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8424 /* Currently only IPv4/IPv6 with TCP is supported */
8425 switch (hdr.ipv4->version) {
8427 /* access ihl as u8 to avoid unaligned access on ia64 */
8428 hlen = (hdr.network[0] & 0x0F) << 2;
8429 l4_proto = hdr.ipv4->protocol;
8432 hlen = hdr.network - skb->data;
8433 l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8434 hlen -= hdr.network - skb->data;
8440 if (l4_proto != IPPROTO_TCP)
8443 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8444 hlen + sizeof(struct tcphdr)))
8447 th = (struct tcphdr *)(hdr.network + hlen);
8449 /* skip this packet since the socket is closing */
8453 /* sample on all syn packets or once every atr sample count */
8454 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8457 /* reset sample count */
8458 ring->atr_count = 0;
8460 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8463 * src and dst are inverted, think how the receiver sees them
8465 * The input is broken into two sections, a non-compressed section
8466 * containing vm_pool, vlan_id, and flow_type. The rest of the data
8467 * is XORed together and stored in the compressed dword.
8469 input.formatted.vlan_id = vlan_id;
8472 * since src port and flex bytes occupy the same word XOR them together
8473 * and write the value to source port portion of compressed dword
8475 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8476 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8478 common.port.src ^= th->dest ^ first->protocol;
8479 common.port.dst ^= th->source;
8481 switch (hdr.ipv4->version) {
8483 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8484 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8487 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8488 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8489 hdr.ipv6->saddr.s6_addr32[1] ^
8490 hdr.ipv6->saddr.s6_addr32[2] ^
8491 hdr.ipv6->saddr.s6_addr32[3] ^
8492 hdr.ipv6->daddr.s6_addr32[0] ^
8493 hdr.ipv6->daddr.s6_addr32[1] ^
8494 hdr.ipv6->daddr.s6_addr32[2] ^
8495 hdr.ipv6->daddr.s6_addr32[3];
8501 if (hdr.network != skb_network_header(skb))
8502 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8504 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
8505 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8506 input, common, ring->queue_index);
8510 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8511 struct net_device *sb_dev)
8513 struct ixgbe_adapter *adapter;
8514 struct ixgbe_ring_feature *f;
8518 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8519 struct net_device *vdev = sb_dev;
8521 txq = vdev->tc_to_txq[tc].offset;
8522 txq += reciprocal_scale(skb_get_hash(skb),
8523 vdev->tc_to_txq[tc].count);
8529 * only execute the code below if protocol is FCoE
8530 * or FIP and we have FCoE enabled on the adapter
8532 switch (vlan_get_protocol(skb)) {
8533 case htons(ETH_P_FCOE):
8534 case htons(ETH_P_FIP):
8535 adapter = netdev_priv(dev);
8537 if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8541 return netdev_pick_tx(dev, skb, sb_dev);
8544 f = &adapter->ring_feature[RING_F_FCOE];
8546 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8549 while (txq >= f->indices)
8552 return txq + f->offset;
8556 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8557 struct xdp_frame *xdpf)
8559 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8560 struct ixgbe_tx_buffer *tx_buffer;
8561 union ixgbe_adv_tx_desc *tx_desc;
8568 if (unlikely(!ixgbe_desc_unused(ring)))
8569 return IXGBE_XDP_CONSUMED;
8571 dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
8572 if (dma_mapping_error(ring->dev, dma))
8573 return IXGBE_XDP_CONSUMED;
8575 /* record the location of the first descriptor for this packet */
8576 tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8577 tx_buffer->bytecount = len;
8578 tx_buffer->gso_segs = 1;
8579 tx_buffer->protocol = 0;
8581 i = ring->next_to_use;
8582 tx_desc = IXGBE_TX_DESC(ring, i);
8584 dma_unmap_len_set(tx_buffer, len, len);
8585 dma_unmap_addr_set(tx_buffer, dma, dma);
8586 tx_buffer->xdpf = xdpf;
8588 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8590 /* put descriptor type bits */
8591 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8592 IXGBE_ADVTXD_DCMD_DEXT |
8593 IXGBE_ADVTXD_DCMD_IFCS;
8594 cmd_type |= len | IXGBE_TXD_CMD;
8595 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8596 tx_desc->read.olinfo_status =
8597 cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8599 /* Avoid any potential race with xdp_xmit and cleanup */
8602 /* set next_to_watch value indicating a packet is present */
8604 if (i == ring->count)
8607 tx_buffer->next_to_watch = tx_desc;
8608 ring->next_to_use = i;
8610 return IXGBE_XDP_TX;
8613 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8614 struct ixgbe_adapter *adapter,
8615 struct ixgbe_ring *tx_ring)
8617 struct ixgbe_tx_buffer *first;
8621 u16 count = TXD_USE_COUNT(skb_headlen(skb));
8622 struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8623 __be16 protocol = skb->protocol;
8627 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8628 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8629 * + 2 desc gap to keep tail from touching head,
8630 * + 1 desc for context descriptor,
8631 * otherwise try next time
8633 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8634 count += TXD_USE_COUNT(skb_frag_size(
8635 &skb_shinfo(skb)->frags[f]));
8637 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8638 tx_ring->tx_stats.tx_busy++;
8639 return NETDEV_TX_BUSY;
8642 /* record the location of the first descriptor for this packet */
8643 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8645 first->bytecount = skb->len;
8646 first->gso_segs = 1;
8648 /* if we have a HW VLAN tag being added default to the HW one */
8649 if (skb_vlan_tag_present(skb)) {
8650 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8651 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8652 /* else if it is a SW VLAN check the next protocol and store the tag */
8653 } else if (protocol == htons(ETH_P_8021Q)) {
8654 struct vlan_hdr *vhdr, _vhdr;
8655 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8659 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8660 IXGBE_TX_FLAGS_VLAN_SHIFT;
8661 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8663 protocol = vlan_get_protocol(skb);
8665 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8666 adapter->ptp_clock) {
8667 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
8668 !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8670 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8671 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8673 /* schedule check for Tx timestamp */
8674 adapter->ptp_tx_skb = skb_get(skb);
8675 adapter->ptp_tx_start = jiffies;
8676 schedule_work(&adapter->ptp_tx_work);
8678 adapter->tx_hwtstamp_skipped++;
8682 #ifdef CONFIG_PCI_IOV
8684 * Use the l2switch_enable flag - would be false if the DMA
8685 * Tx switch had been disabled.
8687 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8688 tx_flags |= IXGBE_TX_FLAGS_CC;
8691 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8692 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8693 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8694 (skb->priority != TC_PRIO_CONTROL))) {
8695 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8696 tx_flags |= (skb->priority & 0x7) <<
8697 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8698 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8699 struct vlan_ethhdr *vhdr;
8701 if (skb_cow_head(skb, 0))
8703 vhdr = (struct vlan_ethhdr *)skb->data;
8704 vhdr->h_vlan_TCI = htons(tx_flags >>
8705 IXGBE_TX_FLAGS_VLAN_SHIFT);
8707 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8711 /* record initial flags and protocol */
8712 first->tx_flags = tx_flags;
8713 first->protocol = protocol;
8716 /* setup tx offload for FCoE */
8717 if ((protocol == htons(ETH_P_FCOE)) &&
8718 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8719 tso = ixgbe_fso(tx_ring, first, &hdr_len);
8726 #endif /* IXGBE_FCOE */
8728 #ifdef CONFIG_IXGBE_IPSEC
8729 if (xfrm_offload(skb) &&
8730 !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8733 tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8737 ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8739 /* add the ATR filter if ATR is on */
8740 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8741 ixgbe_atr(tx_ring, first);
8745 #endif /* IXGBE_FCOE */
8746 if (ixgbe_tx_map(tx_ring, first, hdr_len))
8747 goto cleanup_tx_timestamp;
8749 return NETDEV_TX_OK;
8752 dev_kfree_skb_any(first->skb);
8754 cleanup_tx_timestamp:
8755 if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8756 dev_kfree_skb_any(adapter->ptp_tx_skb);
8757 adapter->ptp_tx_skb = NULL;
8758 cancel_work_sync(&adapter->ptp_tx_work);
8759 clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8762 return NETDEV_TX_OK;
8765 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8766 struct net_device *netdev,
8767 struct ixgbe_ring *ring)
8769 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8770 struct ixgbe_ring *tx_ring;
8773 * The minimum packet size for olinfo paylen is 17 so pad the skb
8774 * in order to meet this minimum size requirement.
8776 if (skb_put_padto(skb, 17))
8777 return NETDEV_TX_OK;
8779 tx_ring = ring ? ring : adapter->tx_ring[skb_get_queue_mapping(skb)];
8780 if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8781 return NETDEV_TX_BUSY;
8783 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8786 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8787 struct net_device *netdev)
8789 return __ixgbe_xmit_frame(skb, netdev, NULL);
8793 * ixgbe_set_mac - Change the Ethernet Address of the NIC
8794 * @netdev: network interface device structure
8795 * @p: pointer to an address structure
8797 * Returns 0 on success, negative on failure
8799 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8801 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8802 struct ixgbe_hw *hw = &adapter->hw;
8803 struct sockaddr *addr = p;
8805 if (!is_valid_ether_addr(addr->sa_data))
8806 return -EADDRNOTAVAIL;
8808 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8809 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8811 ixgbe_mac_set_default_filter(adapter);
8817 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8819 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8820 struct ixgbe_hw *hw = &adapter->hw;
8824 if (adapter->mii_bus) {
8827 if (devad != MDIO_DEVAD_NONE)
8828 regnum |= (devad << 16) | MII_ADDR_C45;
8830 return mdiobus_read(adapter->mii_bus, prtad, regnum);
8833 if (prtad != hw->phy.mdio.prtad)
8835 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8841 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8842 u16 addr, u16 value)
8844 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8845 struct ixgbe_hw *hw = &adapter->hw;
8847 if (adapter->mii_bus) {
8850 if (devad != MDIO_DEVAD_NONE)
8851 regnum |= (devad << 16) | MII_ADDR_C45;
8853 return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8856 if (prtad != hw->phy.mdio.prtad)
8858 return hw->phy.ops.write_reg(hw, addr, devad, value);
8861 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8863 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8867 return ixgbe_ptp_set_ts_config(adapter, req);
8869 return ixgbe_ptp_get_ts_config(adapter, req);
8871 if (!adapter->hw.phy.ops.read_reg)
8875 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8880 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8882 * @dev: network interface device structure
8884 * Returns non-zero on failure
8886 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8889 struct ixgbe_adapter *adapter = netdev_priv(dev);
8890 struct ixgbe_hw *hw = &adapter->hw;
8892 if (is_valid_ether_addr(hw->mac.san_addr)) {
8894 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8897 /* update SAN MAC vmdq pool selection */
8898 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8904 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8906 * @dev: network interface device structure
8908 * Returns non-zero on failure
8910 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8913 struct ixgbe_adapter *adapter = netdev_priv(dev);
8914 struct ixgbe_mac_info *mac = &adapter->hw.mac;
8916 if (is_valid_ether_addr(mac->san_addr)) {
8918 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8924 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8925 struct ixgbe_ring *ring)
8932 start = u64_stats_fetch_begin_irq(&ring->syncp);
8933 packets = ring->stats.packets;
8934 bytes = ring->stats.bytes;
8935 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8936 stats->tx_packets += packets;
8937 stats->tx_bytes += bytes;
8941 static void ixgbe_get_stats64(struct net_device *netdev,
8942 struct rtnl_link_stats64 *stats)
8944 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8948 for (i = 0; i < adapter->num_rx_queues; i++) {
8949 struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8955 start = u64_stats_fetch_begin_irq(&ring->syncp);
8956 packets = ring->stats.packets;
8957 bytes = ring->stats.bytes;
8958 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8959 stats->rx_packets += packets;
8960 stats->rx_bytes += bytes;
8964 for (i = 0; i < adapter->num_tx_queues; i++) {
8965 struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8967 ixgbe_get_ring_stats64(stats, ring);
8969 for (i = 0; i < adapter->num_xdp_queues; i++) {
8970 struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8972 ixgbe_get_ring_stats64(stats, ring);
8976 /* following stats updated by ixgbe_watchdog_task() */
8977 stats->multicast = netdev->stats.multicast;
8978 stats->rx_errors = netdev->stats.rx_errors;
8979 stats->rx_length_errors = netdev->stats.rx_length_errors;
8980 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
8981 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
8984 #ifdef CONFIG_IXGBE_DCB
8986 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8987 * @adapter: pointer to ixgbe_adapter
8988 * @tc: number of traffic classes currently enabled
8990 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8991 * 802.1Q priority maps to a packet buffer that exists.
8993 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8995 struct ixgbe_hw *hw = &adapter->hw;
8999 /* 82598 have a static priority to TC mapping that can not
9000 * be changed so no validation is needed.
9002 if (hw->mac.type == ixgbe_mac_82598EB)
9005 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
9008 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
9009 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
9011 /* If up2tc is out of bounds default to zero */
9013 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
9017 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
9023 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
9024 * @adapter: Pointer to adapter struct
9026 * Populate the netdev user priority to tc map
9028 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
9030 struct net_device *dev = adapter->netdev;
9031 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9032 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9035 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9038 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9039 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9041 tc = ets->prio_tc[prio];
9043 netdev_set_prio_tc_map(dev, prio, tc);
9047 #endif /* CONFIG_IXGBE_DCB */
9048 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev, void *data)
9050 struct ixgbe_adapter *adapter = data;
9051 struct ixgbe_fwd_adapter *accel;
9054 /* we only care about macvlans... */
9055 if (!netif_is_macvlan(vdev))
9058 /* that have hardware offload enabled... */
9059 accel = macvlan_accel_priv(vdev);
9063 /* If we can relocate to a different bit do so */
9064 pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9065 if (pool < adapter->num_rx_pools) {
9066 set_bit(pool, adapter->fwd_bitmask);
9071 /* if we cannot find a free pool then disable the offload */
9072 netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9073 macvlan_release_l2fw_offload(vdev);
9075 /* unbind the queues and drop the subordinate channel config */
9076 netdev_unbind_sb_channel(adapter->netdev, vdev);
9077 netdev_set_sb_channel(vdev, 0);
9084 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9086 struct ixgbe_adapter *adapter = netdev_priv(dev);
9088 /* flush any stale bits out of the fwd bitmask */
9089 bitmap_clear(adapter->fwd_bitmask, 1, 63);
9091 /* walk through upper devices reassigning pools */
9092 netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9097 * ixgbe_setup_tc - configure net_device for multiple traffic classes
9099 * @dev: net device to configure
9100 * @tc: number of traffic classes to enable
9102 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9104 struct ixgbe_adapter *adapter = netdev_priv(dev);
9105 struct ixgbe_hw *hw = &adapter->hw;
9107 /* Hardware supports up to 8 traffic classes */
9108 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9111 if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9114 /* Hardware has to reinitialize queues and interrupts to
9115 * match packet buffer alignment. Unfortunately, the
9116 * hardware is not flexible enough to do this dynamically.
9118 if (netif_running(dev))
9121 ixgbe_reset(adapter);
9123 ixgbe_clear_interrupt_scheme(adapter);
9125 #ifdef CONFIG_IXGBE_DCB
9127 if (adapter->xdp_prog) {
9128 e_warn(probe, "DCB is not supported with XDP\n");
9130 ixgbe_init_interrupt_scheme(adapter);
9131 if (netif_running(dev))
9136 netdev_set_num_tc(dev, tc);
9137 ixgbe_set_prio_tc_map(adapter);
9139 adapter->hw_tcs = tc;
9140 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9142 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9143 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9144 adapter->hw.fc.requested_mode = ixgbe_fc_none;
9147 netdev_reset_tc(dev);
9149 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9150 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9152 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9153 adapter->hw_tcs = tc;
9155 adapter->temp_dcb_cfg.pfc_mode_enable = false;
9156 adapter->dcb_cfg.pfc_mode_enable = false;
9159 ixgbe_validate_rtr(adapter, tc);
9161 #endif /* CONFIG_IXGBE_DCB */
9162 ixgbe_init_interrupt_scheme(adapter);
9164 ixgbe_defrag_macvlan_pools(dev);
9166 if (netif_running(dev))
9167 return ixgbe_open(dev);
9172 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9173 struct tc_cls_u32_offload *cls)
9175 u32 hdl = cls->knode.handle;
9176 u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9177 u32 loc = cls->knode.handle & 0xfffff;
9179 struct ixgbe_jump_table *jump = NULL;
9181 if (loc > IXGBE_MAX_HW_ENTRIES)
9184 if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9187 /* Clear this filter in the link data it is associated with */
9188 if (uhtid != 0x800) {
9189 jump = adapter->jump_tables[uhtid];
9192 if (!test_bit(loc - 1, jump->child_loc_map))
9194 clear_bit(loc - 1, jump->child_loc_map);
9197 /* Check if the filter being deleted is a link */
9198 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9199 jump = adapter->jump_tables[i];
9200 if (jump && jump->link_hdl == hdl) {
9201 /* Delete filters in the hardware in the child hash
9202 * table associated with this link
9204 for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9205 if (!test_bit(j, jump->child_loc_map))
9207 spin_lock(&adapter->fdir_perfect_lock);
9208 err = ixgbe_update_ethtool_fdir_entry(adapter,
9211 spin_unlock(&adapter->fdir_perfect_lock);
9212 clear_bit(j, jump->child_loc_map);
9214 /* Remove resources for this link */
9218 adapter->jump_tables[i] = NULL;
9223 spin_lock(&adapter->fdir_perfect_lock);
9224 err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9225 spin_unlock(&adapter->fdir_perfect_lock);
9229 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9230 struct tc_cls_u32_offload *cls)
9232 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9234 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9237 /* This ixgbe devices do not support hash tables at the moment
9238 * so abort when given hash tables.
9240 if (cls->hnode.divisor > 0)
9243 set_bit(uhtid - 1, &adapter->tables);
9247 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9248 struct tc_cls_u32_offload *cls)
9250 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9252 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9255 clear_bit(uhtid - 1, &adapter->tables);
9259 #ifdef CONFIG_NET_CLS_ACT
9260 struct upper_walk_data {
9261 struct ixgbe_adapter *adapter;
9267 static int get_macvlan_queue(struct net_device *upper, void *_data)
9269 if (netif_is_macvlan(upper)) {
9270 struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9271 struct upper_walk_data *data = _data;
9272 struct ixgbe_adapter *adapter = data->adapter;
9273 int ifindex = data->ifindex;
9275 if (vadapter && upper->ifindex == ifindex) {
9276 data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9277 data->action = data->queue;
9285 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9286 u8 *queue, u64 *action)
9288 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9289 unsigned int num_vfs = adapter->num_vfs, vf;
9290 struct upper_walk_data data;
9291 struct net_device *upper;
9293 /* redirect to a SRIOV VF */
9294 for (vf = 0; vf < num_vfs; ++vf) {
9295 upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9296 if (upper->ifindex == ifindex) {
9297 *queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9299 *action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9304 /* redirect to a offloaded macvlan netdev */
9305 data.adapter = adapter;
9306 data.ifindex = ifindex;
9309 if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9310 get_macvlan_queue, &data)) {
9311 *action = data.action;
9312 *queue = data.queue;
9320 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9321 struct tcf_exts *exts, u64 *action, u8 *queue)
9323 const struct tc_action *a;
9326 if (!tcf_exts_has_actions(exts))
9329 tcf_exts_for_each_action(i, a, exts) {
9331 if (is_tcf_gact_shot(a)) {
9332 *action = IXGBE_FDIR_DROP_QUEUE;
9333 *queue = IXGBE_FDIR_DROP_QUEUE;
9337 /* Redirect to a VF or a offloaded macvlan */
9338 if (is_tcf_mirred_egress_redirect(a)) {
9339 struct net_device *dev = tcf_mirred_dev(a);
9343 return handle_redirect_action(adapter, dev->ifindex,
9353 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9354 struct tcf_exts *exts, u64 *action, u8 *queue)
9358 #endif /* CONFIG_NET_CLS_ACT */
9360 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9361 union ixgbe_atr_input *mask,
9362 struct tc_cls_u32_offload *cls,
9363 struct ixgbe_mat_field *field_ptr,
9364 struct ixgbe_nexthdr *nexthdr)
9368 bool found_entry = false, found_jump_field = false;
9370 for (i = 0; i < cls->knode.sel->nkeys; i++) {
9371 off = cls->knode.sel->keys[i].off;
9372 val = cls->knode.sel->keys[i].val;
9373 m = cls->knode.sel->keys[i].mask;
9375 for (j = 0; field_ptr[j].val; j++) {
9376 if (field_ptr[j].off == off) {
9377 field_ptr[j].val(input, mask, (__force u32)val,
9379 input->filter.formatted.flow_type |=
9386 if (nexthdr->off == cls->knode.sel->keys[i].off &&
9388 (__force u32)cls->knode.sel->keys[i].val &&
9390 (__force u32)cls->knode.sel->keys[i].mask)
9391 found_jump_field = true;
9397 if (nexthdr && !found_jump_field)
9403 mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9404 IXGBE_ATR_L4TYPE_MASK;
9406 if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9407 mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9412 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9413 struct tc_cls_u32_offload *cls)
9415 __be16 protocol = cls->common.protocol;
9416 u32 loc = cls->knode.handle & 0xfffff;
9417 struct ixgbe_hw *hw = &adapter->hw;
9418 struct ixgbe_mat_field *field_ptr;
9419 struct ixgbe_fdir_filter *input = NULL;
9420 union ixgbe_atr_input *mask = NULL;
9421 struct ixgbe_jump_table *jump = NULL;
9422 int i, err = -EINVAL;
9424 u32 uhtid, link_uhtid;
9426 uhtid = TC_U32_USERHTID(cls->knode.handle);
9427 link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9429 /* At the moment cls_u32 jumps to network layer and skips past
9430 * L2 headers. The canonical method to match L2 frames is to use
9431 * negative values. However this is error prone at best but really
9432 * just broken because there is no way to "know" what sort of hdr
9433 * is in front of the network layer. Fix cls_u32 to support L2
9434 * headers when needed.
9436 if (protocol != htons(ETH_P_IP))
9439 if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9440 e_err(drv, "Location out of range\n");
9444 /* cls u32 is a graph starting at root node 0x800. The driver tracks
9445 * links and also the fields used to advance the parser across each
9446 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9447 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9448 * To add support for new nodes update ixgbe_model.h parse structures
9449 * this function _should_ be generic try not to hardcode values here.
9451 if (uhtid == 0x800) {
9452 field_ptr = (adapter->jump_tables[0])->mat;
9454 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9456 if (!adapter->jump_tables[uhtid])
9458 field_ptr = (adapter->jump_tables[uhtid])->mat;
9464 /* At this point we know the field_ptr is valid and need to either
9465 * build cls_u32 link or attach filter. Because adding a link to
9466 * a handle that does not exist is invalid and the same for adding
9467 * rules to handles that don't exist.
9471 struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9473 if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9476 if (!test_bit(link_uhtid - 1, &adapter->tables))
9479 /* Multiple filters as links to the same hash table are not
9480 * supported. To add a new filter with the same next header
9481 * but different match/jump conditions, create a new hash table
9484 if (adapter->jump_tables[link_uhtid] &&
9485 (adapter->jump_tables[link_uhtid])->link_hdl) {
9486 e_err(drv, "Link filter exists for link: %x\n",
9491 for (i = 0; nexthdr[i].jump; i++) {
9492 if (nexthdr[i].o != cls->knode.sel->offoff ||
9493 nexthdr[i].s != cls->knode.sel->offshift ||
9495 (__force u32)cls->knode.sel->offmask)
9498 jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9501 input = kzalloc(sizeof(*input), GFP_KERNEL);
9506 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9511 jump->input = input;
9513 jump->link_hdl = cls->knode.handle;
9515 err = ixgbe_clsu32_build_input(input, mask, cls,
9516 field_ptr, &nexthdr[i]);
9518 jump->mat = nexthdr[i].jump;
9519 adapter->jump_tables[link_uhtid] = jump;
9530 input = kzalloc(sizeof(*input), GFP_KERNEL);
9533 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9539 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9540 if ((adapter->jump_tables[uhtid])->input)
9541 memcpy(input, (adapter->jump_tables[uhtid])->input,
9543 if ((adapter->jump_tables[uhtid])->mask)
9544 memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9547 /* Lookup in all child hash tables if this location is already
9548 * filled with a filter
9550 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9551 struct ixgbe_jump_table *link = adapter->jump_tables[i];
9553 if (link && (test_bit(loc - 1, link->child_loc_map))) {
9554 e_err(drv, "Filter exists in location: %x\n",
9561 err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9565 err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9570 input->sw_idx = loc;
9572 spin_lock(&adapter->fdir_perfect_lock);
9574 if (hlist_empty(&adapter->fdir_filter_list)) {
9575 memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9576 err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9578 goto err_out_w_lock;
9579 } else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9581 goto err_out_w_lock;
9584 ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9585 err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9586 input->sw_idx, queue);
9588 ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9589 spin_unlock(&adapter->fdir_perfect_lock);
9591 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9592 set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9597 spin_unlock(&adapter->fdir_perfect_lock);
9607 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9608 struct tc_cls_u32_offload *cls_u32)
9610 switch (cls_u32->command) {
9611 case TC_CLSU32_NEW_KNODE:
9612 case TC_CLSU32_REPLACE_KNODE:
9613 return ixgbe_configure_clsu32(adapter, cls_u32);
9614 case TC_CLSU32_DELETE_KNODE:
9615 return ixgbe_delete_clsu32(adapter, cls_u32);
9616 case TC_CLSU32_NEW_HNODE:
9617 case TC_CLSU32_REPLACE_HNODE:
9618 return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9619 case TC_CLSU32_DELETE_HNODE:
9620 return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9626 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9629 struct ixgbe_adapter *adapter = cb_priv;
9631 if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9635 case TC_SETUP_CLSU32:
9636 return ixgbe_setup_tc_cls_u32(adapter, type_data);
9642 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9643 struct tc_mqprio_qopt *mqprio)
9645 mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9646 return ixgbe_setup_tc(dev, mqprio->num_tc);
9649 static LIST_HEAD(ixgbe_block_cb_list);
9651 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9654 struct ixgbe_adapter *adapter = netdev_priv(dev);
9657 case TC_SETUP_BLOCK:
9658 return flow_block_cb_setup_simple(type_data,
9659 &ixgbe_block_cb_list,
9660 ixgbe_setup_tc_block_cb,
9661 adapter, adapter, true);
9662 case TC_SETUP_QDISC_MQPRIO:
9663 return ixgbe_setup_tc_mqprio(dev, type_data);
9669 #ifdef CONFIG_PCI_IOV
9670 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9672 struct net_device *netdev = adapter->netdev;
9675 ixgbe_setup_tc(netdev, adapter->hw_tcs);
9680 void ixgbe_do_reset(struct net_device *netdev)
9682 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9684 if (netif_running(netdev))
9685 ixgbe_reinit_locked(adapter);
9687 ixgbe_reset(adapter);
9690 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9691 netdev_features_t features)
9693 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9695 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9696 if (!(features & NETIF_F_RXCSUM))
9697 features &= ~NETIF_F_LRO;
9699 /* Turn off LRO if not RSC capable */
9700 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9701 features &= ~NETIF_F_LRO;
9703 if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9704 e_dev_err("LRO is not supported with XDP\n");
9705 features &= ~NETIF_F_LRO;
9711 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9713 int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9716 /* go back to full RSS if we're not running SR-IOV */
9717 if (!adapter->ring_feature[RING_F_VMDQ].offset)
9718 adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9719 IXGBE_FLAG_SRIOV_ENABLED);
9721 adapter->ring_feature[RING_F_RSS].limit = rss;
9722 adapter->ring_feature[RING_F_VMDQ].limit = 1;
9724 ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9727 static int ixgbe_set_features(struct net_device *netdev,
9728 netdev_features_t features)
9730 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9731 netdev_features_t changed = netdev->features ^ features;
9732 bool need_reset = false;
9734 /* Make sure RSC matches LRO, reset if change */
9735 if (!(features & NETIF_F_LRO)) {
9736 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9738 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9739 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9740 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9741 if (adapter->rx_itr_setting == 1 ||
9742 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9743 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9745 } else if ((changed ^ features) & NETIF_F_LRO) {
9746 e_info(probe, "rx-usecs set too low, "
9752 * Check if Flow Director n-tuple support or hw_tc support was
9753 * enabled or disabled. If the state changed, we need to reset.
9755 if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9756 /* turn off ATR, enable perfect filters and reset */
9757 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9760 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9761 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9763 /* turn off perfect filters, enable ATR and reset */
9764 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9767 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9769 /* We cannot enable ATR if SR-IOV is enabled */
9770 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9771 /* We cannot enable ATR if we have 2 or more tcs */
9772 (adapter->hw_tcs > 1) ||
9773 /* We cannot enable ATR if RSS is disabled */
9774 (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9775 /* A sample rate of 0 indicates ATR disabled */
9776 (!adapter->atr_sample_rate))
9777 ; /* do nothing not supported */
9778 else /* otherwise supported and set the flag */
9779 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9782 if (changed & NETIF_F_RXALL)
9785 netdev->features = features;
9787 if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9788 ixgbe_reset_l2fw_offload(adapter);
9789 else if (need_reset)
9790 ixgbe_do_reset(netdev);
9791 else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9792 NETIF_F_HW_VLAN_CTAG_FILTER))
9793 ixgbe_set_rx_mode(netdev);
9799 * ixgbe_add_udp_tunnel_port - Get notifications about adding UDP tunnel ports
9800 * @dev: The port's netdev
9801 * @ti: Tunnel endpoint information
9803 static void ixgbe_add_udp_tunnel_port(struct net_device *dev,
9804 struct udp_tunnel_info *ti)
9806 struct ixgbe_adapter *adapter = netdev_priv(dev);
9807 struct ixgbe_hw *hw = &adapter->hw;
9808 __be16 port = ti->port;
9812 if (ti->sa_family != AF_INET)
9816 case UDP_TUNNEL_TYPE_VXLAN:
9817 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9820 if (adapter->vxlan_port == port)
9823 if (adapter->vxlan_port) {
9825 "VXLAN port %d set, not adding port %d\n",
9826 ntohs(adapter->vxlan_port),
9831 adapter->vxlan_port = port;
9833 case UDP_TUNNEL_TYPE_GENEVE:
9834 if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9837 if (adapter->geneve_port == port)
9840 if (adapter->geneve_port) {
9842 "GENEVE port %d set, not adding port %d\n",
9843 ntohs(adapter->geneve_port),
9848 port_shift = IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT;
9849 adapter->geneve_port = port;
9855 reg = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) | ntohs(port) << port_shift;
9856 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, reg);
9860 * ixgbe_del_udp_tunnel_port - Get notifications about removing UDP tunnel ports
9861 * @dev: The port's netdev
9862 * @ti: Tunnel endpoint information
9864 static void ixgbe_del_udp_tunnel_port(struct net_device *dev,
9865 struct udp_tunnel_info *ti)
9867 struct ixgbe_adapter *adapter = netdev_priv(dev);
9870 if (ti->type != UDP_TUNNEL_TYPE_VXLAN &&
9871 ti->type != UDP_TUNNEL_TYPE_GENEVE)
9874 if (ti->sa_family != AF_INET)
9878 case UDP_TUNNEL_TYPE_VXLAN:
9879 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9882 if (adapter->vxlan_port != ti->port) {
9883 netdev_info(dev, "VXLAN port %d not found\n",
9888 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9890 case UDP_TUNNEL_TYPE_GENEVE:
9891 if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9894 if (adapter->geneve_port != ti->port) {
9895 netdev_info(dev, "GENEVE port %d not found\n",
9900 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9906 ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9907 adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9910 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9911 struct net_device *dev,
9912 const unsigned char *addr, u16 vid,
9914 struct netlink_ext_ack *extack)
9916 /* guarantee we can provide a unique filter for the unicast address */
9917 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9918 struct ixgbe_adapter *adapter = netdev_priv(dev);
9919 u16 pool = VMDQ_P(0);
9921 if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9925 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9929 * ixgbe_configure_bridge_mode - set various bridge modes
9930 * @adapter: the private structure
9931 * @mode: requested bridge mode
9933 * Configure some settings require for various bridge modes.
9935 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9938 struct ixgbe_hw *hw = &adapter->hw;
9939 unsigned int p, num_pools;
9943 case BRIDGE_MODE_VEPA:
9944 /* disable Tx loopback, rely on switch hairpin mode */
9945 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9947 /* must enable Rx switching replication to allow multicast
9948 * packet reception on all VFs, and to enable source address
9951 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9952 vmdctl |= IXGBE_VT_CTL_REPLEN;
9953 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9955 /* enable Rx source address pruning. Note, this requires
9956 * replication to be enabled or else it does nothing.
9958 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9959 for (p = 0; p < num_pools; p++) {
9960 if (hw->mac.ops.set_source_address_pruning)
9961 hw->mac.ops.set_source_address_pruning(hw,
9966 case BRIDGE_MODE_VEB:
9967 /* enable Tx loopback for internal VF/PF communication */
9968 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9969 IXGBE_PFDTXGSWC_VT_LBEN);
9971 /* disable Rx switching replication unless we have SR-IOV
9974 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9975 if (!adapter->num_vfs)
9976 vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9977 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9979 /* disable Rx source address pruning, since we don't expect to
9980 * be receiving external loopback of our transmitted frames.
9982 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9983 for (p = 0; p < num_pools; p++) {
9984 if (hw->mac.ops.set_source_address_pruning)
9985 hw->mac.ops.set_source_address_pruning(hw,
9994 adapter->bridge_mode = mode;
9996 e_info(drv, "enabling bridge mode: %s\n",
9997 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
10002 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
10003 struct nlmsghdr *nlh, u16 flags,
10004 struct netlink_ext_ack *extack)
10006 struct ixgbe_adapter *adapter = netdev_priv(dev);
10007 struct nlattr *attr, *br_spec;
10010 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10011 return -EOPNOTSUPP;
10013 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
10017 nla_for_each_nested(attr, br_spec, rem) {
10021 if (nla_type(attr) != IFLA_BRIDGE_MODE)
10024 if (nla_len(attr) < sizeof(mode))
10027 mode = nla_get_u16(attr);
10028 status = ixgbe_configure_bridge_mode(adapter, mode);
10038 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
10039 struct net_device *dev,
10040 u32 filter_mask, int nlflags)
10042 struct ixgbe_adapter *adapter = netdev_priv(dev);
10044 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10047 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
10048 adapter->bridge_mode, 0, 0, nlflags,
10049 filter_mask, NULL);
10052 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
10054 struct ixgbe_adapter *adapter = netdev_priv(pdev);
10055 struct ixgbe_fwd_adapter *accel;
10056 int tcs = adapter->hw_tcs ? : 1;
10059 if (adapter->xdp_prog) {
10060 e_warn(probe, "L2FW offload is not supported with XDP\n");
10061 return ERR_PTR(-EINVAL);
10064 /* The hardware supported by ixgbe only filters on the destination MAC
10065 * address. In order to avoid issues we only support offloading modes
10066 * where the hardware can actually provide the functionality.
10068 if (!macvlan_supports_dest_filter(vdev))
10069 return ERR_PTR(-EMEDIUMTYPE);
10071 /* We need to lock down the macvlan to be a single queue device so that
10072 * we can reuse the tc_to_txq field in the macvlan netdev to represent
10073 * the queue mapping to our netdev.
10075 if (netif_is_multiqueue(vdev))
10076 return ERR_PTR(-ERANGE);
10078 pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
10079 if (pool == adapter->num_rx_pools) {
10080 u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
10081 u16 reserved_pools;
10083 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
10084 adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
10085 adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
10086 return ERR_PTR(-EBUSY);
10088 /* Hardware has a limited number of available pools. Each VF,
10089 * and the PF require a pool. Check to ensure we don't
10090 * attempt to use more then the available number of pools.
10092 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
10093 return ERR_PTR(-EBUSY);
10095 /* Enable VMDq flag so device will be set in VM mode */
10096 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
10097 IXGBE_FLAG_SRIOV_ENABLED;
10099 /* Try to reserve as many queues per pool as possible,
10100 * we start with the configurations that support 4 queues
10101 * per pools, followed by 2, and then by just 1 per pool.
10103 if (used_pools < 32 && adapter->num_rx_pools < 16)
10104 reserved_pools = min_t(u16,
10106 16 - adapter->num_rx_pools);
10107 else if (adapter->num_rx_pools < 32)
10108 reserved_pools = min_t(u16,
10110 32 - adapter->num_rx_pools);
10112 reserved_pools = 64 - used_pools;
10115 if (!reserved_pools)
10116 return ERR_PTR(-EBUSY);
10118 adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
10120 /* Force reinit of ring allocation with VMDQ enabled */
10121 err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
10123 return ERR_PTR(err);
10125 if (pool >= adapter->num_rx_pools)
10126 return ERR_PTR(-ENOMEM);
10129 accel = kzalloc(sizeof(*accel), GFP_KERNEL);
10131 return ERR_PTR(-ENOMEM);
10133 set_bit(pool, adapter->fwd_bitmask);
10134 netdev_set_sb_channel(vdev, pool);
10135 accel->pool = pool;
10136 accel->netdev = vdev;
10138 if (!netif_running(pdev))
10141 err = ixgbe_fwd_ring_up(adapter, accel);
10143 return ERR_PTR(err);
10148 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10150 struct ixgbe_fwd_adapter *accel = priv;
10151 struct ixgbe_adapter *adapter = netdev_priv(pdev);
10152 unsigned int rxbase = accel->rx_base_queue;
10155 /* delete unicast filter associated with offloaded interface */
10156 ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10157 VMDQ_P(accel->pool));
10159 /* Allow remaining Rx packets to get flushed out of the
10160 * Rx FIFO before we drop the netdev for the ring.
10162 usleep_range(10000, 20000);
10164 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10165 struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10166 struct ixgbe_q_vector *qv = ring->q_vector;
10168 /* Make sure we aren't processing any packets and clear
10169 * netdev to shut down the ring.
10171 if (netif_running(adapter->netdev))
10172 napi_synchronize(&qv->napi);
10173 ring->netdev = NULL;
10176 /* unbind the queues and drop the subordinate channel config */
10177 netdev_unbind_sb_channel(pdev, accel->netdev);
10178 netdev_set_sb_channel(accel->netdev, 0);
10180 clear_bit(accel->pool, adapter->fwd_bitmask);
10184 #define IXGBE_MAX_MAC_HDR_LEN 127
10185 #define IXGBE_MAX_NETWORK_HDR_LEN 511
10187 static netdev_features_t
10188 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10189 netdev_features_t features)
10191 unsigned int network_hdr_len, mac_hdr_len;
10193 /* Make certain the headers can be described by a context descriptor */
10194 mac_hdr_len = skb_network_header(skb) - skb->data;
10195 if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10196 return features & ~(NETIF_F_HW_CSUM |
10198 NETIF_F_GSO_UDP_L4 |
10199 NETIF_F_HW_VLAN_CTAG_TX |
10203 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10204 if (unlikely(network_hdr_len > IXGBE_MAX_NETWORK_HDR_LEN))
10205 return features & ~(NETIF_F_HW_CSUM |
10207 NETIF_F_GSO_UDP_L4 |
10211 /* We can only support IPV4 TSO in tunnels if we can mangle the
10212 * inner IP ID field, so strip TSO if MANGLEID is not supported.
10213 * IPsec offoad sets skb->encapsulation but still can handle
10214 * the TSO, so it's the exception.
10216 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10217 #ifdef CONFIG_IXGBE_IPSEC
10218 if (!secpath_exists(skb))
10220 features &= ~NETIF_F_TSO;
10226 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10228 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10229 struct ixgbe_adapter *adapter = netdev_priv(dev);
10230 struct bpf_prog *old_prog;
10233 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10236 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10239 /* verify ixgbe ring attributes are sufficient for XDP */
10240 for (i = 0; i < adapter->num_rx_queues; i++) {
10241 struct ixgbe_ring *ring = adapter->rx_ring[i];
10243 if (ring_is_rsc_enabled(ring))
10246 if (frame_size > ixgbe_rx_bufsz(ring))
10250 if (nr_cpu_ids > MAX_XDP_QUEUES)
10253 old_prog = xchg(&adapter->xdp_prog, prog);
10254 need_reset = (!!prog != !!old_prog);
10256 /* If transitioning XDP modes reconfigure rings */
10261 /* Wait until ndo_xsk_wakeup completes. */
10263 err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10266 rcu_assign_pointer(adapter->xdp_prog, old_prog);
10270 for (i = 0; i < adapter->num_rx_queues; i++)
10271 (void)xchg(&adapter->rx_ring[i]->xdp_prog,
10272 adapter->xdp_prog);
10276 bpf_prog_put(old_prog);
10278 /* Kick start the NAPI context if there is an AF_XDP socket open
10279 * on that queue id. This so that receiving will start.
10281 if (need_reset && prog)
10282 for (i = 0; i < adapter->num_rx_queues; i++)
10283 if (adapter->xdp_ring[i]->xsk_umem)
10284 (void)ixgbe_xsk_wakeup(adapter->netdev, i,
10290 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10292 struct ixgbe_adapter *adapter = netdev_priv(dev);
10294 switch (xdp->command) {
10295 case XDP_SETUP_PROG:
10296 return ixgbe_xdp_setup(dev, xdp->prog);
10297 case XDP_QUERY_PROG:
10298 xdp->prog_id = adapter->xdp_prog ?
10299 adapter->xdp_prog->aux->id : 0;
10301 case XDP_SETUP_XSK_UMEM:
10302 return ixgbe_xsk_umem_setup(adapter, xdp->xsk.umem,
10303 xdp->xsk.queue_id);
10310 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10312 /* Force memory writes to complete before letting h/w know there
10313 * are new descriptors to fetch.
10316 writel(ring->next_to_use, ring->tail);
10319 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10320 struct xdp_frame **frames, u32 flags)
10322 struct ixgbe_adapter *adapter = netdev_priv(dev);
10323 struct ixgbe_ring *ring;
10327 if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10330 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10333 /* During program transitions its possible adapter->xdp_prog is assigned
10334 * but ring has not been configured yet. In this case simply abort xmit.
10336 ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10337 if (unlikely(!ring))
10340 if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10343 for (i = 0; i < n; i++) {
10344 struct xdp_frame *xdpf = frames[i];
10347 err = ixgbe_xmit_xdp_ring(adapter, xdpf);
10348 if (err != IXGBE_XDP_TX) {
10349 xdp_return_frame_rx_napi(xdpf);
10354 if (unlikely(flags & XDP_XMIT_FLUSH))
10355 ixgbe_xdp_ring_update_tail(ring);
10360 static const struct net_device_ops ixgbe_netdev_ops = {
10361 .ndo_open = ixgbe_open,
10362 .ndo_stop = ixgbe_close,
10363 .ndo_start_xmit = ixgbe_xmit_frame,
10364 .ndo_set_rx_mode = ixgbe_set_rx_mode,
10365 .ndo_validate_addr = eth_validate_addr,
10366 .ndo_set_mac_address = ixgbe_set_mac,
10367 .ndo_change_mtu = ixgbe_change_mtu,
10368 .ndo_tx_timeout = ixgbe_tx_timeout,
10369 .ndo_set_tx_maxrate = ixgbe_tx_maxrate,
10370 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
10371 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
10372 .ndo_do_ioctl = ixgbe_ioctl,
10373 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
10374 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
10375 .ndo_set_vf_rate = ixgbe_ndo_set_vf_bw,
10376 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
10377 .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10378 .ndo_set_vf_trust = ixgbe_ndo_set_vf_trust,
10379 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
10380 .ndo_get_stats64 = ixgbe_get_stats64,
10381 .ndo_setup_tc = __ixgbe_setup_tc,
10383 .ndo_select_queue = ixgbe_select_queue,
10384 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10385 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10386 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10387 .ndo_fcoe_enable = ixgbe_fcoe_enable,
10388 .ndo_fcoe_disable = ixgbe_fcoe_disable,
10389 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10390 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10391 #endif /* IXGBE_FCOE */
10392 .ndo_set_features = ixgbe_set_features,
10393 .ndo_fix_features = ixgbe_fix_features,
10394 .ndo_fdb_add = ixgbe_ndo_fdb_add,
10395 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
10396 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
10397 .ndo_dfwd_add_station = ixgbe_fwd_add,
10398 .ndo_dfwd_del_station = ixgbe_fwd_del,
10399 .ndo_udp_tunnel_add = ixgbe_add_udp_tunnel_port,
10400 .ndo_udp_tunnel_del = ixgbe_del_udp_tunnel_port,
10401 .ndo_features_check = ixgbe_features_check,
10402 .ndo_bpf = ixgbe_xdp,
10403 .ndo_xdp_xmit = ixgbe_xdp_xmit,
10404 .ndo_xsk_wakeup = ixgbe_xsk_wakeup,
10407 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10408 struct ixgbe_ring *tx_ring)
10410 unsigned long wait_delay, delay_interval;
10411 struct ixgbe_hw *hw = &adapter->hw;
10412 u8 reg_idx = tx_ring->reg_idx;
10416 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10418 /* delay mechanism from ixgbe_disable_tx */
10419 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10421 wait_loop = IXGBE_MAX_RX_DESC_POLL;
10422 wait_delay = delay_interval;
10424 while (wait_loop--) {
10425 usleep_range(wait_delay, wait_delay + 10);
10426 wait_delay += delay_interval * 2;
10427 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10429 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10433 e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10436 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10437 struct ixgbe_ring *tx_ring)
10439 set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10440 ixgbe_disable_txr_hw(adapter, tx_ring);
10443 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10444 struct ixgbe_ring *rx_ring)
10446 unsigned long wait_delay, delay_interval;
10447 struct ixgbe_hw *hw = &adapter->hw;
10448 u8 reg_idx = rx_ring->reg_idx;
10452 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10453 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10454 rxdctl |= IXGBE_RXDCTL_SWFLSH;
10456 /* write value back with RXDCTL.ENABLE bit cleared */
10457 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10459 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10460 if (hw->mac.type == ixgbe_mac_82598EB &&
10461 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10464 /* delay mechanism from ixgbe_disable_rx */
10465 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10467 wait_loop = IXGBE_MAX_RX_DESC_POLL;
10468 wait_delay = delay_interval;
10470 while (wait_loop--) {
10471 usleep_range(wait_delay, wait_delay + 10);
10472 wait_delay += delay_interval * 2;
10473 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10475 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10479 e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10482 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10484 memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10485 memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10488 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10490 memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10491 memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10495 * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10496 * @adapter: adapter structure
10497 * @ring: ring index
10499 * This function disables a certain Rx/Tx/XDP Tx ring. The function
10500 * assumes that the netdev is running.
10502 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10504 struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10506 rx_ring = adapter->rx_ring[ring];
10507 tx_ring = adapter->tx_ring[ring];
10508 xdp_ring = adapter->xdp_ring[ring];
10510 ixgbe_disable_txr(adapter, tx_ring);
10512 ixgbe_disable_txr(adapter, xdp_ring);
10513 ixgbe_disable_rxr_hw(adapter, rx_ring);
10518 /* Rx/Tx/XDP Tx share the same napi context. */
10519 napi_disable(&rx_ring->q_vector->napi);
10521 ixgbe_clean_tx_ring(tx_ring);
10523 ixgbe_clean_tx_ring(xdp_ring);
10524 ixgbe_clean_rx_ring(rx_ring);
10526 ixgbe_reset_txr_stats(tx_ring);
10528 ixgbe_reset_txr_stats(xdp_ring);
10529 ixgbe_reset_rxr_stats(rx_ring);
10533 * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10534 * @adapter: adapter structure
10535 * @ring: ring index
10537 * This function enables a certain Rx/Tx/XDP Tx ring. The function
10538 * assumes that the netdev is running.
10540 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10542 struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10544 rx_ring = adapter->rx_ring[ring];
10545 tx_ring = adapter->tx_ring[ring];
10546 xdp_ring = adapter->xdp_ring[ring];
10548 /* Rx/Tx/XDP Tx share the same napi context. */
10549 napi_enable(&rx_ring->q_vector->napi);
10551 ixgbe_configure_tx_ring(adapter, tx_ring);
10553 ixgbe_configure_tx_ring(adapter, xdp_ring);
10554 ixgbe_configure_rx_ring(adapter, rx_ring);
10556 clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10558 clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10562 * ixgbe_enumerate_functions - Get the number of ports this device has
10563 * @adapter: adapter structure
10565 * This function enumerates the phsyical functions co-located on a single slot,
10566 * in order to determine how many ports a device has. This is most useful in
10567 * determining the required GT/s of PCIe bandwidth necessary for optimal
10570 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10572 struct pci_dev *entry, *pdev = adapter->pdev;
10575 /* Some cards can not use the generic count PCIe functions method,
10576 * because they are behind a parent switch, so we hardcode these with
10577 * the correct number of functions.
10579 if (ixgbe_pcie_from_parent(&adapter->hw))
10582 list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10583 /* don't count virtual functions */
10584 if (entry->is_virtfn)
10587 /* When the devices on the bus don't all match our device ID,
10588 * we can't reliably determine the correct number of
10589 * functions. This can occur if a function has been direct
10590 * attached to a virtual machine using VT-d, for example. In
10591 * this case, simply return -1 to indicate this.
10593 if ((entry->vendor != pdev->vendor) ||
10594 (entry->device != pdev->device))
10604 * ixgbe_wol_supported - Check whether device supports WoL
10605 * @adapter: the adapter private structure
10606 * @device_id: the device ID
10607 * @subdevice_id: the subsystem device ID
10609 * This function is used by probe and ethtool to determine
10610 * which devices have WoL support
10613 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10616 struct ixgbe_hw *hw = &adapter->hw;
10617 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10619 /* WOL not supported on 82598 */
10620 if (hw->mac.type == ixgbe_mac_82598EB)
10623 /* check eeprom to see if WOL is enabled for X540 and newer */
10624 if (hw->mac.type >= ixgbe_mac_X540) {
10625 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10626 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10627 (hw->bus.func == 0)))
10631 /* WOL is determined based on device IDs for 82599 MACs */
10632 switch (device_id) {
10633 case IXGBE_DEV_ID_82599_SFP:
10634 /* Only these subdevices could supports WOL */
10635 switch (subdevice_id) {
10636 case IXGBE_SUBDEV_ID_82599_560FLR:
10637 case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10638 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10639 case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10640 /* only support first port */
10641 if (hw->bus.func != 0)
10644 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10645 case IXGBE_SUBDEV_ID_82599_SFP:
10646 case IXGBE_SUBDEV_ID_82599_RNDC:
10647 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10648 case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10649 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10650 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10654 case IXGBE_DEV_ID_82599EN_SFP:
10655 /* Only these subdevices support WOL */
10656 switch (subdevice_id) {
10657 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10661 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10662 /* All except this subdevice support WOL */
10663 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10666 case IXGBE_DEV_ID_82599_KX4:
10676 * ixgbe_set_fw_version - Set FW version
10677 * @adapter: the adapter private structure
10679 * This function is used by probe and ethtool to determine the FW version to
10680 * format to display. The FW version is taken from the EEPROM/NVM.
10682 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10684 struct ixgbe_hw *hw = &adapter->hw;
10685 struct ixgbe_nvm_version nvm_ver;
10687 ixgbe_get_oem_prod_version(hw, &nvm_ver);
10688 if (nvm_ver.oem_valid) {
10689 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10690 "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10691 nvm_ver.oem_release);
10695 ixgbe_get_etk_id(hw, &nvm_ver);
10696 ixgbe_get_orom_version(hw, &nvm_ver);
10698 if (nvm_ver.or_valid) {
10699 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10700 "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10701 nvm_ver.or_build, nvm_ver.or_patch);
10705 /* Set ETrack ID format */
10706 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10707 "0x%08x", nvm_ver.etk_id);
10711 * ixgbe_probe - Device Initialization Routine
10712 * @pdev: PCI device information struct
10713 * @ent: entry in ixgbe_pci_tbl
10715 * Returns 0 on success, negative on failure
10717 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10718 * The OS initialization, configuring of the adapter private structure,
10719 * and a hardware reset occur.
10721 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10723 struct net_device *netdev;
10724 struct ixgbe_adapter *adapter = NULL;
10725 struct ixgbe_hw *hw;
10726 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10727 int i, err, pci_using_dac, expected_gts;
10728 unsigned int indices = MAX_TX_QUEUES;
10729 u8 part_str[IXGBE_PBANUM_LENGTH];
10730 bool disable_dev = false;
10736 /* Catch broken hardware that put the wrong VF device ID in
10737 * the PCIe SR-IOV capability.
10739 if (pdev->is_virtfn) {
10740 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10741 pci_name(pdev), pdev->vendor, pdev->device);
10745 err = pci_enable_device_mem(pdev);
10749 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10752 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10754 dev_err(&pdev->dev,
10755 "No usable DMA configuration, aborting\n");
10761 err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10763 dev_err(&pdev->dev,
10764 "pci_request_selected_regions failed 0x%x\n", err);
10768 pci_enable_pcie_error_reporting(pdev);
10770 pci_set_master(pdev);
10771 pci_save_state(pdev);
10773 if (ii->mac == ixgbe_mac_82598EB) {
10774 #ifdef CONFIG_IXGBE_DCB
10775 /* 8 TC w/ 4 queues per TC */
10776 indices = 4 * MAX_TRAFFIC_CLASS;
10778 indices = IXGBE_MAX_RSS_INDICES;
10782 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10785 goto err_alloc_etherdev;
10788 SET_NETDEV_DEV(netdev, &pdev->dev);
10790 adapter = netdev_priv(netdev);
10792 adapter->netdev = netdev;
10793 adapter->pdev = pdev;
10795 hw->back = adapter;
10796 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10798 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10799 pci_resource_len(pdev, 0));
10800 adapter->io_addr = hw->hw_addr;
10801 if (!hw->hw_addr) {
10806 netdev->netdev_ops = &ixgbe_netdev_ops;
10807 ixgbe_set_ethtool_ops(netdev);
10808 netdev->watchdog_timeo = 5 * HZ;
10809 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10812 hw->mac.ops = *ii->mac_ops;
10813 hw->mac.type = ii->mac;
10814 hw->mvals = ii->mvals;
10816 hw->link.ops = *ii->link_ops;
10819 hw->eeprom.ops = *ii->eeprom_ops;
10820 eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10821 if (ixgbe_removed(hw->hw_addr)) {
10825 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10826 if (!(eec & BIT(8)))
10827 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10830 hw->phy.ops = *ii->phy_ops;
10831 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10832 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
10833 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10834 hw->phy.mdio.mmds = 0;
10835 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10836 hw->phy.mdio.dev = netdev;
10837 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10838 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10840 /* setup the private structure */
10841 err = ixgbe_sw_init(adapter, ii);
10845 /* Make sure the SWFW semaphore is in a valid state */
10846 if (hw->mac.ops.init_swfw_sync)
10847 hw->mac.ops.init_swfw_sync(hw);
10849 /* Make it possible the adapter to be woken up via WOL */
10850 switch (adapter->hw.mac.type) {
10851 case ixgbe_mac_82599EB:
10852 case ixgbe_mac_X540:
10853 case ixgbe_mac_X550:
10854 case ixgbe_mac_X550EM_x:
10855 case ixgbe_mac_x550em_a:
10856 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10863 * If there is a fan on this device and it has failed log the
10866 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10867 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10868 if (esdp & IXGBE_ESDP_SDP1)
10869 e_crit(probe, "Fan has stopped, replace the adapter\n");
10872 if (allow_unsupported_sfp)
10873 hw->allow_unsupported_sfp = allow_unsupported_sfp;
10875 /* reset_hw fills in the perm_addr as well */
10876 hw->phy.reset_if_overtemp = true;
10877 err = hw->mac.ops.reset_hw(hw);
10878 hw->phy.reset_if_overtemp = false;
10879 ixgbe_set_eee_capable(adapter);
10880 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10882 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10883 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10884 e_dev_err("Reload the driver after installing a supported module.\n");
10887 e_dev_err("HW Init failed: %d\n", err);
10891 #ifdef CONFIG_PCI_IOV
10892 /* SR-IOV not supported on the 82598 */
10893 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10896 ixgbe_init_mbx_params_pf(hw);
10897 hw->mbx.ops = ii->mbx_ops;
10898 pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10899 ixgbe_enable_sriov(adapter, max_vfs);
10903 netdev->features = NETIF_F_SG |
10910 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10911 NETIF_F_GSO_GRE_CSUM | \
10912 NETIF_F_GSO_IPXIP4 | \
10913 NETIF_F_GSO_IPXIP6 | \
10914 NETIF_F_GSO_UDP_TUNNEL | \
10915 NETIF_F_GSO_UDP_TUNNEL_CSUM)
10917 netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10918 netdev->features |= NETIF_F_GSO_PARTIAL |
10919 IXGBE_GSO_PARTIAL_FEATURES;
10921 if (hw->mac.type >= ixgbe_mac_82599EB)
10922 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
10924 #ifdef CONFIG_IXGBE_IPSEC
10925 #define IXGBE_ESP_FEATURES (NETIF_F_HW_ESP | \
10926 NETIF_F_HW_ESP_TX_CSUM | \
10929 if (adapter->ipsec)
10930 netdev->features |= IXGBE_ESP_FEATURES;
10932 /* copy netdev features into list of user selectable features */
10933 netdev->hw_features |= netdev->features |
10934 NETIF_F_HW_VLAN_CTAG_FILTER |
10935 NETIF_F_HW_VLAN_CTAG_RX |
10936 NETIF_F_HW_VLAN_CTAG_TX |
10938 NETIF_F_HW_L2FW_DOFFLOAD;
10940 if (hw->mac.type >= ixgbe_mac_82599EB)
10941 netdev->hw_features |= NETIF_F_NTUPLE |
10945 netdev->features |= NETIF_F_HIGHDMA;
10947 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10948 netdev->hw_enc_features |= netdev->vlan_features;
10949 netdev->mpls_features |= NETIF_F_SG |
10953 netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10955 /* set this bit last since it cannot be part of vlan_features */
10956 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10957 NETIF_F_HW_VLAN_CTAG_RX |
10958 NETIF_F_HW_VLAN_CTAG_TX;
10960 netdev->priv_flags |= IFF_UNICAST_FLT;
10961 netdev->priv_flags |= IFF_SUPP_NOFCS;
10963 /* MTU range: 68 - 9710 */
10964 netdev->min_mtu = ETH_MIN_MTU;
10965 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
10967 #ifdef CONFIG_IXGBE_DCB
10968 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
10969 netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
10973 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
10974 unsigned int fcoe_l;
10976 if (hw->mac.ops.get_device_caps) {
10977 hw->mac.ops.get_device_caps(hw, &device_caps);
10978 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
10979 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
10983 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
10984 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
10986 netdev->features |= NETIF_F_FSO |
10989 netdev->vlan_features |= NETIF_F_FSO |
10993 #endif /* IXGBE_FCOE */
10994 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
10995 netdev->hw_features |= NETIF_F_LRO;
10996 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
10997 netdev->features |= NETIF_F_LRO;
10999 if (ixgbe_check_fw_error(adapter)) {
11004 /* make sure the EEPROM is good */
11005 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
11006 e_dev_err("The EEPROM Checksum Is Not Valid\n");
11011 eth_platform_get_mac_address(&adapter->pdev->dev,
11012 adapter->hw.mac.perm_addr);
11014 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
11016 if (!is_valid_ether_addr(netdev->dev_addr)) {
11017 e_dev_err("invalid MAC address\n");
11022 /* Set hw->mac.addr to permanent MAC address */
11023 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
11024 ixgbe_mac_set_default_filter(adapter);
11026 timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
11028 if (ixgbe_removed(hw->hw_addr)) {
11032 INIT_WORK(&adapter->service_task, ixgbe_service_task);
11033 set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
11034 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
11036 err = ixgbe_init_interrupt_scheme(adapter);
11040 for (i = 0; i < adapter->num_rx_queues; i++)
11041 u64_stats_init(&adapter->rx_ring[i]->syncp);
11042 for (i = 0; i < adapter->num_tx_queues; i++)
11043 u64_stats_init(&adapter->tx_ring[i]->syncp);
11044 for (i = 0; i < adapter->num_xdp_queues; i++)
11045 u64_stats_init(&adapter->xdp_ring[i]->syncp);
11047 /* WOL not supported for all devices */
11049 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
11050 hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
11051 pdev->subsystem_device);
11052 if (hw->wol_enabled)
11053 adapter->wol = IXGBE_WUFC_MAG;
11055 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
11057 /* save off EEPROM version number */
11058 ixgbe_set_fw_version(adapter);
11060 /* pick up the PCI bus settings for reporting later */
11061 if (ixgbe_pcie_from_parent(hw))
11062 ixgbe_get_parent_bus_info(adapter);
11064 hw->mac.ops.get_bus_info(hw);
11066 /* calculate the expected PCIe bandwidth required for optimal
11067 * performance. Note that some older parts will never have enough
11068 * bandwidth due to being older generation PCIe parts. We clamp these
11069 * parts to ensure no warning is displayed if it can't be fixed.
11071 switch (hw->mac.type) {
11072 case ixgbe_mac_82598EB:
11073 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
11076 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
11080 /* don't check link if we failed to enumerate functions */
11081 if (expected_gts > 0)
11082 ixgbe_check_minimum_link(adapter, expected_gts);
11084 err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
11086 strlcpy(part_str, "Unknown", sizeof(part_str));
11087 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
11088 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
11089 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
11092 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
11093 hw->mac.type, hw->phy.type, part_str);
11095 e_dev_info("%pM\n", netdev->dev_addr);
11097 /* reset the hardware with the new settings */
11098 err = hw->mac.ops.start_hw(hw);
11099 if (err == IXGBE_ERR_EEPROM_VERSION) {
11100 /* We are running on a pre-production device, log a warning */
11101 e_dev_warn("This device is a pre-production adapter/LOM. "
11102 "Please be aware there may be issues associated "
11103 "with your hardware. If you are experiencing "
11104 "problems please contact your Intel or hardware "
11105 "representative who provided you with this "
11108 strcpy(netdev->name, "eth%d");
11109 pci_set_drvdata(pdev, adapter);
11110 err = register_netdev(netdev);
11115 /* power down the optics for 82599 SFP+ fiber */
11116 if (hw->mac.ops.disable_tx_laser)
11117 hw->mac.ops.disable_tx_laser(hw);
11119 /* carrier off reporting is important to ethtool even BEFORE open */
11120 netif_carrier_off(netdev);
11122 #ifdef CONFIG_IXGBE_DCA
11123 if (dca_add_requester(&pdev->dev) == 0) {
11124 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11125 ixgbe_setup_dca(adapter);
11128 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11129 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11130 for (i = 0; i < adapter->num_vfs; i++)
11131 ixgbe_vf_configuration(pdev, (i | 0x10000000));
11134 /* firmware requires driver version to be 0xFFFFFFFF
11135 * since os does not support feature
11137 if (hw->mac.ops.set_fw_drv_ver)
11138 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11139 sizeof(UTS_RELEASE) - 1,
11142 /* add san mac addr to netdev */
11143 ixgbe_add_sanmac_netdev(netdev);
11145 e_dev_info("%s\n", ixgbe_default_device_descr);
11147 #ifdef CONFIG_IXGBE_HWMON
11148 if (ixgbe_sysfs_init(adapter))
11149 e_err(probe, "failed to allocate sysfs resources\n");
11150 #endif /* CONFIG_IXGBE_HWMON */
11152 ixgbe_dbg_adapter_init(adapter);
11154 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11155 if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11156 hw->mac.ops.setup_link(hw,
11157 IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11160 err = ixgbe_mii_bus_init(hw);
11167 unregister_netdev(netdev);
11169 ixgbe_release_hw_control(adapter);
11170 ixgbe_clear_interrupt_scheme(adapter);
11172 ixgbe_disable_sriov(adapter);
11173 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11174 iounmap(adapter->io_addr);
11175 kfree(adapter->jump_tables[0]);
11176 kfree(adapter->mac_table);
11177 kfree(adapter->rss_key);
11178 bitmap_free(adapter->af_xdp_zc_qps);
11180 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11181 free_netdev(netdev);
11182 err_alloc_etherdev:
11183 pci_release_mem_regions(pdev);
11186 if (!adapter || disable_dev)
11187 pci_disable_device(pdev);
11192 * ixgbe_remove - Device Removal Routine
11193 * @pdev: PCI device information struct
11195 * ixgbe_remove is called by the PCI subsystem to alert the driver
11196 * that it should release a PCI device. The could be caused by a
11197 * Hot-Plug event, or because the driver is going to be removed from
11200 static void ixgbe_remove(struct pci_dev *pdev)
11202 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11203 struct net_device *netdev;
11207 /* if !adapter then we already cleaned up in probe */
11211 netdev = adapter->netdev;
11212 ixgbe_dbg_adapter_exit(adapter);
11214 set_bit(__IXGBE_REMOVING, &adapter->state);
11215 cancel_work_sync(&adapter->service_task);
11217 if (adapter->mii_bus)
11218 mdiobus_unregister(adapter->mii_bus);
11220 #ifdef CONFIG_IXGBE_DCA
11221 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11222 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11223 dca_remove_requester(&pdev->dev);
11224 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11225 IXGBE_DCA_CTRL_DCA_DISABLE);
11229 #ifdef CONFIG_IXGBE_HWMON
11230 ixgbe_sysfs_exit(adapter);
11231 #endif /* CONFIG_IXGBE_HWMON */
11233 /* remove the added san mac */
11234 ixgbe_del_sanmac_netdev(netdev);
11236 #ifdef CONFIG_PCI_IOV
11237 ixgbe_disable_sriov(adapter);
11239 if (netdev->reg_state == NETREG_REGISTERED)
11240 unregister_netdev(netdev);
11242 ixgbe_stop_ipsec_offload(adapter);
11243 ixgbe_clear_interrupt_scheme(adapter);
11245 ixgbe_release_hw_control(adapter);
11248 kfree(adapter->ixgbe_ieee_pfc);
11249 kfree(adapter->ixgbe_ieee_ets);
11252 iounmap(adapter->io_addr);
11253 pci_release_mem_regions(pdev);
11255 e_dev_info("complete\n");
11257 for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11258 if (adapter->jump_tables[i]) {
11259 kfree(adapter->jump_tables[i]->input);
11260 kfree(adapter->jump_tables[i]->mask);
11262 kfree(adapter->jump_tables[i]);
11265 kfree(adapter->mac_table);
11266 kfree(adapter->rss_key);
11267 bitmap_free(adapter->af_xdp_zc_qps);
11268 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11269 free_netdev(netdev);
11271 pci_disable_pcie_error_reporting(pdev);
11274 pci_disable_device(pdev);
11278 * ixgbe_io_error_detected - called when PCI error is detected
11279 * @pdev: Pointer to PCI device
11280 * @state: The current pci connection state
11282 * This function is called after a PCI bus error affecting
11283 * this device has been detected.
11285 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11286 pci_channel_state_t state)
11288 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11289 struct net_device *netdev = adapter->netdev;
11291 #ifdef CONFIG_PCI_IOV
11292 struct ixgbe_hw *hw = &adapter->hw;
11293 struct pci_dev *bdev, *vfdev;
11294 u32 dw0, dw1, dw2, dw3;
11296 u16 req_id, pf_func;
11298 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11299 adapter->num_vfs == 0)
11300 goto skip_bad_vf_detection;
11302 bdev = pdev->bus->self;
11303 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11304 bdev = bdev->bus->self;
11307 goto skip_bad_vf_detection;
11309 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11311 goto skip_bad_vf_detection;
11313 dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11314 dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11315 dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11316 dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11317 if (ixgbe_removed(hw->hw_addr))
11318 goto skip_bad_vf_detection;
11320 req_id = dw1 >> 16;
11321 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11322 if (!(req_id & 0x0080))
11323 goto skip_bad_vf_detection;
11325 pf_func = req_id & 0x01;
11326 if ((pf_func & 1) == (pdev->devfn & 1)) {
11327 unsigned int device_id;
11329 vf = (req_id & 0x7F) >> 1;
11330 e_dev_err("VF %d has caused a PCIe error\n", vf);
11331 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11332 "%8.8x\tdw3: %8.8x\n",
11333 dw0, dw1, dw2, dw3);
11334 switch (adapter->hw.mac.type) {
11335 case ixgbe_mac_82599EB:
11336 device_id = IXGBE_82599_VF_DEVICE_ID;
11338 case ixgbe_mac_X540:
11339 device_id = IXGBE_X540_VF_DEVICE_ID;
11341 case ixgbe_mac_X550:
11342 device_id = IXGBE_DEV_ID_X550_VF;
11344 case ixgbe_mac_X550EM_x:
11345 device_id = IXGBE_DEV_ID_X550EM_X_VF;
11347 case ixgbe_mac_x550em_a:
11348 device_id = IXGBE_DEV_ID_X550EM_A_VF;
11355 /* Find the pci device of the offending VF */
11356 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11358 if (vfdev->devfn == (req_id & 0xFF))
11360 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11364 * There's a slim chance the VF could have been hot plugged,
11365 * so if it is no longer present we don't need to issue the
11366 * VFLR. Just clean up the AER in that case.
11370 /* Free device reference count */
11371 pci_dev_put(vfdev);
11376 * Even though the error may have occurred on the other port
11377 * we still need to increment the vf error reference count for
11378 * both ports because the I/O resume function will be called
11379 * for both of them.
11381 adapter->vferr_refcount++;
11383 return PCI_ERS_RESULT_RECOVERED;
11385 skip_bad_vf_detection:
11386 #endif /* CONFIG_PCI_IOV */
11387 if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11388 return PCI_ERS_RESULT_DISCONNECT;
11390 if (!netif_device_present(netdev))
11391 return PCI_ERS_RESULT_DISCONNECT;
11394 netif_device_detach(netdev);
11396 if (netif_running(netdev))
11397 ixgbe_close_suspend(adapter);
11399 if (state == pci_channel_io_perm_failure) {
11401 return PCI_ERS_RESULT_DISCONNECT;
11404 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11405 pci_disable_device(pdev);
11408 /* Request a slot reset. */
11409 return PCI_ERS_RESULT_NEED_RESET;
11413 * ixgbe_io_slot_reset - called after the pci bus has been reset.
11414 * @pdev: Pointer to PCI device
11416 * Restart the card from scratch, as if from a cold-boot.
11418 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11420 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11421 pci_ers_result_t result;
11423 if (pci_enable_device_mem(pdev)) {
11424 e_err(probe, "Cannot re-enable PCI device after reset.\n");
11425 result = PCI_ERS_RESULT_DISCONNECT;
11427 smp_mb__before_atomic();
11428 clear_bit(__IXGBE_DISABLED, &adapter->state);
11429 adapter->hw.hw_addr = adapter->io_addr;
11430 pci_set_master(pdev);
11431 pci_restore_state(pdev);
11432 pci_save_state(pdev);
11434 pci_wake_from_d3(pdev, false);
11436 ixgbe_reset(adapter);
11437 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11438 result = PCI_ERS_RESULT_RECOVERED;
11445 * ixgbe_io_resume - called when traffic can start flowing again.
11446 * @pdev: Pointer to PCI device
11448 * This callback is called when the error recovery driver tells us that
11449 * its OK to resume normal operation.
11451 static void ixgbe_io_resume(struct pci_dev *pdev)
11453 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11454 struct net_device *netdev = adapter->netdev;
11456 #ifdef CONFIG_PCI_IOV
11457 if (adapter->vferr_refcount) {
11458 e_info(drv, "Resuming after VF err\n");
11459 adapter->vferr_refcount--;
11465 if (netif_running(netdev))
11466 ixgbe_open(netdev);
11468 netif_device_attach(netdev);
11472 static const struct pci_error_handlers ixgbe_err_handler = {
11473 .error_detected = ixgbe_io_error_detected,
11474 .slot_reset = ixgbe_io_slot_reset,
11475 .resume = ixgbe_io_resume,
11478 static struct pci_driver ixgbe_driver = {
11479 .name = ixgbe_driver_name,
11480 .id_table = ixgbe_pci_tbl,
11481 .probe = ixgbe_probe,
11482 .remove = ixgbe_remove,
11484 .suspend = ixgbe_suspend,
11485 .resume = ixgbe_resume,
11487 .shutdown = ixgbe_shutdown,
11488 .sriov_configure = ixgbe_pci_sriov_configure,
11489 .err_handler = &ixgbe_err_handler
11493 * ixgbe_init_module - Driver Registration Routine
11495 * ixgbe_init_module is the first routine called when the driver is
11496 * loaded. All it does is register with the PCI subsystem.
11498 static int __init ixgbe_init_module(void)
11501 pr_info("%s\n", ixgbe_driver_string);
11502 pr_info("%s\n", ixgbe_copyright);
11504 ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11506 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11512 ret = pci_register_driver(&ixgbe_driver);
11514 destroy_workqueue(ixgbe_wq);
11519 #ifdef CONFIG_IXGBE_DCA
11520 dca_register_notify(&dca_notifier);
11526 module_init(ixgbe_init_module);
11529 * ixgbe_exit_module - Driver Exit Cleanup Routine
11531 * ixgbe_exit_module is called just before the driver is removed
11534 static void __exit ixgbe_exit_module(void)
11536 #ifdef CONFIG_IXGBE_DCA
11537 dca_unregister_notify(&dca_notifier);
11539 pci_unregister_driver(&ixgbe_driver);
11543 destroy_workqueue(ixgbe_wq);
11548 #ifdef CONFIG_IXGBE_DCA
11549 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11554 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11555 __ixgbe_notify_dca);
11557 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11560 #endif /* CONFIG_IXGBE_DCA */
11562 module_exit(ixgbe_exit_module);