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 net_prefetch(xdp->data);
2100 /* Note, we get here by enabling legacy-rx via:
2102 * ethtool --set-priv-flags <dev> legacy-rx on
2104 * In this mode, we currently get 0 extra XDP headroom as
2105 * opposed to having legacy-rx off, where we process XDP
2106 * packets going to stack via ixgbe_build_skb(). The latter
2107 * provides us currently with 192 bytes of headroom.
2109 * For ixgbe_construct_skb() mode it means that the
2110 * xdp->data_meta will always point to xdp->data, since
2111 * the helper cannot expand the head. Should this ever
2112 * change in future for legacy-rx mode on, then lets also
2113 * add xdp->data_meta handling here.
2116 /* allocate a skb to store the frags */
2117 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2121 if (size > IXGBE_RX_HDR_SIZE) {
2122 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2123 IXGBE_CB(skb)->dma = rx_buffer->dma;
2125 skb_add_rx_frag(skb, 0, rx_buffer->page,
2126 xdp->data - page_address(rx_buffer->page),
2128 #if (PAGE_SIZE < 8192)
2129 rx_buffer->page_offset ^= truesize;
2131 rx_buffer->page_offset += truesize;
2134 memcpy(__skb_put(skb, size),
2135 xdp->data, ALIGN(size, sizeof(long)));
2136 rx_buffer->pagecnt_bias++;
2142 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2143 struct ixgbe_rx_buffer *rx_buffer,
2144 struct xdp_buff *xdp,
2145 union ixgbe_adv_rx_desc *rx_desc)
2147 unsigned int metasize = xdp->data - xdp->data_meta;
2148 #if (PAGE_SIZE < 8192)
2149 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2151 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2152 SKB_DATA_ALIGN(xdp->data_end -
2153 xdp->data_hard_start);
2155 struct sk_buff *skb;
2157 /* Prefetch first cache line of first page. If xdp->data_meta
2158 * is unused, this points extactly as xdp->data, otherwise we
2159 * likely have a consumer accessing first few bytes of meta
2160 * data, and then actual data.
2162 net_prefetch(xdp->data_meta);
2164 /* build an skb to around the page buffer */
2165 skb = build_skb(xdp->data_hard_start, truesize);
2169 /* update pointers within the skb to store the data */
2170 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2171 __skb_put(skb, xdp->data_end - xdp->data);
2173 skb_metadata_set(skb, metasize);
2175 /* record DMA address if this is the start of a chain of buffers */
2176 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2177 IXGBE_CB(skb)->dma = rx_buffer->dma;
2179 /* update buffer offset */
2180 #if (PAGE_SIZE < 8192)
2181 rx_buffer->page_offset ^= truesize;
2183 rx_buffer->page_offset += truesize;
2189 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2190 struct ixgbe_ring *rx_ring,
2191 struct xdp_buff *xdp)
2193 int err, result = IXGBE_XDP_PASS;
2194 struct bpf_prog *xdp_prog;
2195 struct xdp_frame *xdpf;
2199 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2204 prefetchw(xdp->data_hard_start); /* xdp_frame write */
2206 act = bpf_prog_run_xdp(xdp_prog, xdp);
2211 xdpf = xdp_convert_buff_to_frame(xdp);
2212 if (unlikely(!xdpf)) {
2213 result = IXGBE_XDP_CONSUMED;
2216 result = ixgbe_xmit_xdp_ring(adapter, xdpf);
2219 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2221 result = IXGBE_XDP_REDIR;
2223 result = IXGBE_XDP_CONSUMED;
2226 bpf_warn_invalid_xdp_action(act);
2229 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2230 fallthrough; /* handle aborts by dropping packet */
2232 result = IXGBE_XDP_CONSUMED;
2237 return ERR_PTR(-result);
2240 static unsigned int ixgbe_rx_frame_truesize(struct ixgbe_ring *rx_ring,
2243 unsigned int truesize;
2245 #if (PAGE_SIZE < 8192)
2246 truesize = ixgbe_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
2248 truesize = ring_uses_build_skb(rx_ring) ?
2249 SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) +
2250 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
2251 SKB_DATA_ALIGN(size);
2256 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2257 struct ixgbe_rx_buffer *rx_buffer,
2260 unsigned int truesize = ixgbe_rx_frame_truesize(rx_ring, size);
2261 #if (PAGE_SIZE < 8192)
2262 rx_buffer->page_offset ^= truesize;
2264 rx_buffer->page_offset += truesize;
2269 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2270 * @q_vector: structure containing interrupt and ring information
2271 * @rx_ring: rx descriptor ring to transact packets on
2272 * @budget: Total limit on number of packets to process
2274 * This function provides a "bounce buffer" approach to Rx interrupt
2275 * processing. The advantage to this is that on systems that have
2276 * expensive overhead for IOMMU access this provides a means of avoiding
2277 * it by maintaining the mapping of the page to the syste.
2279 * Returns amount of work completed
2281 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2282 struct ixgbe_ring *rx_ring,
2285 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2286 struct ixgbe_adapter *adapter = q_vector->adapter;
2289 unsigned int mss = 0;
2290 #endif /* IXGBE_FCOE */
2291 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2292 unsigned int xdp_xmit = 0;
2293 struct xdp_buff xdp;
2295 xdp.rxq = &rx_ring->xdp_rxq;
2297 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
2298 #if (PAGE_SIZE < 8192)
2299 xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, 0);
2302 while (likely(total_rx_packets < budget)) {
2303 union ixgbe_adv_rx_desc *rx_desc;
2304 struct ixgbe_rx_buffer *rx_buffer;
2305 struct sk_buff *skb;
2308 /* return some buffers to hardware, one at a time is too slow */
2309 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2310 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2314 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2315 size = le16_to_cpu(rx_desc->wb.upper.length);
2319 /* This memory barrier is needed to keep us from reading
2320 * any other fields out of the rx_desc until we know the
2321 * descriptor has been written back
2325 rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size);
2327 /* retrieve a buffer from the ring */
2329 xdp.data = page_address(rx_buffer->page) +
2330 rx_buffer->page_offset;
2331 xdp.data_meta = xdp.data;
2332 xdp.data_hard_start = xdp.data -
2333 ixgbe_rx_offset(rx_ring);
2334 xdp.data_end = xdp.data + size;
2335 #if (PAGE_SIZE > 4096)
2336 /* At larger PAGE_SIZE, frame_sz depend on len size */
2337 xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, size);
2339 skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2343 unsigned int xdp_res = -PTR_ERR(skb);
2345 if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2346 xdp_xmit |= xdp_res;
2347 ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2349 rx_buffer->pagecnt_bias++;
2352 total_rx_bytes += size;
2354 ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2355 } else if (ring_uses_build_skb(rx_ring)) {
2356 skb = ixgbe_build_skb(rx_ring, rx_buffer,
2359 skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2363 /* exit if we failed to retrieve a buffer */
2365 rx_ring->rx_stats.alloc_rx_buff_failed++;
2366 rx_buffer->pagecnt_bias++;
2370 ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb);
2373 /* place incomplete frames back on ring for completion */
2374 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2377 /* verify the packet layout is correct */
2378 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2381 /* probably a little skewed due to removing CRC */
2382 total_rx_bytes += skb->len;
2384 /* populate checksum, timestamp, VLAN, and protocol */
2385 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2388 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2389 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2390 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2391 /* include DDPed FCoE data */
2392 if (ddp_bytes > 0) {
2394 mss = rx_ring->netdev->mtu -
2395 sizeof(struct fcoe_hdr) -
2396 sizeof(struct fc_frame_header) -
2397 sizeof(struct fcoe_crc_eof);
2401 total_rx_bytes += ddp_bytes;
2402 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2406 dev_kfree_skb_any(skb);
2411 #endif /* IXGBE_FCOE */
2412 ixgbe_rx_skb(q_vector, skb);
2414 /* update budget accounting */
2418 if (xdp_xmit & IXGBE_XDP_REDIR)
2421 if (xdp_xmit & IXGBE_XDP_TX) {
2422 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2424 /* Force memory writes to complete before letting h/w
2425 * know there are new descriptors to fetch.
2428 writel(ring->next_to_use, ring->tail);
2431 u64_stats_update_begin(&rx_ring->syncp);
2432 rx_ring->stats.packets += total_rx_packets;
2433 rx_ring->stats.bytes += total_rx_bytes;
2434 u64_stats_update_end(&rx_ring->syncp);
2435 q_vector->rx.total_packets += total_rx_packets;
2436 q_vector->rx.total_bytes += total_rx_bytes;
2438 return total_rx_packets;
2442 * ixgbe_configure_msix - Configure MSI-X hardware
2443 * @adapter: board private structure
2445 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2448 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2450 struct ixgbe_q_vector *q_vector;
2454 /* Populate MSIX to EITR Select */
2455 if (adapter->num_vfs > 32) {
2456 u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2457 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2461 * Populate the IVAR table and set the ITR values to the
2462 * corresponding register.
2464 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2465 struct ixgbe_ring *ring;
2466 q_vector = adapter->q_vector[v_idx];
2468 ixgbe_for_each_ring(ring, q_vector->rx)
2469 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2471 ixgbe_for_each_ring(ring, q_vector->tx)
2472 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2474 ixgbe_write_eitr(q_vector);
2477 switch (adapter->hw.mac.type) {
2478 case ixgbe_mac_82598EB:
2479 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2482 case ixgbe_mac_82599EB:
2483 case ixgbe_mac_X540:
2484 case ixgbe_mac_X550:
2485 case ixgbe_mac_X550EM_x:
2486 case ixgbe_mac_x550em_a:
2487 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2492 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2494 /* set up to autoclear timer, and the vectors */
2495 mask = IXGBE_EIMS_ENABLE_MASK;
2496 mask &= ~(IXGBE_EIMS_OTHER |
2497 IXGBE_EIMS_MAILBOX |
2500 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2504 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2505 * @q_vector: structure containing interrupt and ring information
2506 * @ring_container: structure containing ring performance data
2508 * Stores a new ITR value based on packets and byte
2509 * counts during the last interrupt. The advantage of per interrupt
2510 * computation is faster updates and more accurate ITR for the current
2511 * traffic pattern. Constants in this function were computed
2512 * based on theoretical maximum wire speed and thresholds were set based
2513 * on testing data as well as attempting to minimize response time
2514 * while increasing bulk throughput.
2516 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2517 struct ixgbe_ring_container *ring_container)
2519 unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2520 IXGBE_ITR_ADAPTIVE_LATENCY;
2521 unsigned int avg_wire_size, packets, bytes;
2522 unsigned long next_update = jiffies;
2524 /* If we don't have any rings just leave ourselves set for maximum
2525 * possible latency so we take ourselves out of the equation.
2527 if (!ring_container->ring)
2530 /* If we didn't update within up to 1 - 2 jiffies we can assume
2531 * that either packets are coming in so slow there hasn't been
2532 * any work, or that there is so much work that NAPI is dealing
2533 * with interrupt moderation and we don't need to do anything.
2535 if (time_after(next_update, ring_container->next_update))
2538 packets = ring_container->total_packets;
2540 /* We have no packets to actually measure against. This means
2541 * either one of the other queues on this vector is active or
2542 * we are a Tx queue doing TSO with too high of an interrupt rate.
2544 * When this occurs just tick up our delay by the minimum value
2545 * and hope that this extra delay will prevent us from being called
2546 * without any work on our queue.
2549 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2550 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2551 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2552 itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2556 bytes = ring_container->total_bytes;
2558 /* If packets are less than 4 or bytes are less than 9000 assume
2559 * insufficient data to use bulk rate limiting approach. We are
2560 * likely latency driven.
2562 if (packets < 4 && bytes < 9000) {
2563 itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2564 goto adjust_by_size;
2567 /* Between 4 and 48 we can assume that our current interrupt delay
2568 * is only slightly too low. As such we should increase it by a small
2572 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2573 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2574 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2578 /* Between 48 and 96 is our "goldilocks" zone where we are working
2579 * out "just right". Just report that our current ITR is good for us.
2582 itr = q_vector->itr >> 2;
2586 /* If packet count is 96 or greater we are likely looking at a slight
2587 * overrun of the delay we want. Try halving our delay to see if that
2588 * will cut the number of packets in half per interrupt.
2590 if (packets < 256) {
2591 itr = q_vector->itr >> 3;
2592 if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2593 itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2597 /* The paths below assume we are dealing with a bulk ITR since number
2598 * of packets is 256 or greater. We are just going to have to compute
2599 * a value and try to bring the count under control, though for smaller
2600 * packet sizes there isn't much we can do as NAPI polling will likely
2601 * be kicking in sooner rather than later.
2603 itr = IXGBE_ITR_ADAPTIVE_BULK;
2606 /* If packet counts are 256 or greater we can assume we have a gross
2607 * overestimation of what the rate should be. Instead of trying to fine
2608 * tune it just use the formula below to try and dial in an exact value
2609 * give the current packet size of the frame.
2611 avg_wire_size = bytes / packets;
2613 /* The following is a crude approximation of:
2614 * wmem_default / (size + overhead) = desired_pkts_per_int
2615 * rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2616 * (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2618 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2619 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2622 * (170 * (size + 24)) / (size + 640) = ITR
2624 * We first do some math on the packet size and then finally bitshift
2625 * by 8 after rounding up. We also have to account for PCIe link speed
2626 * difference as ITR scales based on this.
2628 if (avg_wire_size <= 60) {
2629 /* Start at 50k ints/sec */
2630 avg_wire_size = 5120;
2631 } else if (avg_wire_size <= 316) {
2632 /* 50K ints/sec to 16K ints/sec */
2633 avg_wire_size *= 40;
2634 avg_wire_size += 2720;
2635 } else if (avg_wire_size <= 1084) {
2636 /* 16K ints/sec to 9.2K ints/sec */
2637 avg_wire_size *= 15;
2638 avg_wire_size += 11452;
2639 } else if (avg_wire_size < 1968) {
2640 /* 9.2K ints/sec to 8K ints/sec */
2642 avg_wire_size += 22420;
2644 /* plateau at a limit of 8K ints/sec */
2645 avg_wire_size = 32256;
2648 /* If we are in low latency mode half our delay which doubles the rate
2649 * to somewhere between 100K to 16K ints/sec
2651 if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2652 avg_wire_size >>= 1;
2654 /* Resultant value is 256 times larger than it needs to be. This
2655 * gives us room to adjust the value as needed to either increase
2656 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2658 * Use addition as we have already recorded the new latency flag
2659 * for the ITR value.
2661 switch (q_vector->adapter->link_speed) {
2662 case IXGBE_LINK_SPEED_10GB_FULL:
2663 case IXGBE_LINK_SPEED_100_FULL:
2665 itr += DIV_ROUND_UP(avg_wire_size,
2666 IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2667 IXGBE_ITR_ADAPTIVE_MIN_INC;
2669 case IXGBE_LINK_SPEED_2_5GB_FULL:
2670 case IXGBE_LINK_SPEED_1GB_FULL:
2671 case IXGBE_LINK_SPEED_10_FULL:
2672 if (avg_wire_size > 8064)
2673 avg_wire_size = 8064;
2674 itr += DIV_ROUND_UP(avg_wire_size,
2675 IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2676 IXGBE_ITR_ADAPTIVE_MIN_INC;
2681 /* write back value */
2682 ring_container->itr = itr;
2684 /* next update should occur within next jiffy */
2685 ring_container->next_update = next_update + 1;
2687 ring_container->total_bytes = 0;
2688 ring_container->total_packets = 0;
2692 * ixgbe_write_eitr - write EITR register in hardware specific way
2693 * @q_vector: structure containing interrupt and ring information
2695 * This function is made to be called by ethtool and by the driver
2696 * when it needs to update EITR registers at runtime. Hardware
2697 * specific quirks/differences are taken care of here.
2699 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2701 struct ixgbe_adapter *adapter = q_vector->adapter;
2702 struct ixgbe_hw *hw = &adapter->hw;
2703 int v_idx = q_vector->v_idx;
2704 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2706 switch (adapter->hw.mac.type) {
2707 case ixgbe_mac_82598EB:
2708 /* must write high and low 16 bits to reset counter */
2709 itr_reg |= (itr_reg << 16);
2711 case ixgbe_mac_82599EB:
2712 case ixgbe_mac_X540:
2713 case ixgbe_mac_X550:
2714 case ixgbe_mac_X550EM_x:
2715 case ixgbe_mac_x550em_a:
2717 * set the WDIS bit to not clear the timer bits and cause an
2718 * immediate assertion of the interrupt
2720 itr_reg |= IXGBE_EITR_CNT_WDIS;
2725 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2728 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2732 ixgbe_update_itr(q_vector, &q_vector->tx);
2733 ixgbe_update_itr(q_vector, &q_vector->rx);
2735 /* use the smallest value of new ITR delay calculations */
2736 new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2738 /* Clear latency flag if set, shift into correct position */
2739 new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2742 if (new_itr != q_vector->itr) {
2743 /* save the algorithm value here */
2744 q_vector->itr = new_itr;
2746 ixgbe_write_eitr(q_vector);
2751 * ixgbe_check_overtemp_subtask - check for over temperature
2752 * @adapter: pointer to adapter
2754 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2756 struct ixgbe_hw *hw = &adapter->hw;
2757 u32 eicr = adapter->interrupt_event;
2760 if (test_bit(__IXGBE_DOWN, &adapter->state))
2763 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2766 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2768 switch (hw->device_id) {
2769 case IXGBE_DEV_ID_82599_T3_LOM:
2771 * Since the warning interrupt is for both ports
2772 * we don't have to check if:
2773 * - This interrupt wasn't for our port.
2774 * - We may have missed the interrupt so always have to
2775 * check if we got a LSC
2777 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2778 !(eicr & IXGBE_EICR_LSC))
2781 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2783 bool link_up = false;
2785 hw->mac.ops.check_link(hw, &speed, &link_up, false);
2791 /* Check if this is not due to overtemp */
2792 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2796 case IXGBE_DEV_ID_X550EM_A_1G_T:
2797 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2798 rc = hw->phy.ops.check_overtemp(hw);
2799 if (rc != IXGBE_ERR_OVERTEMP)
2803 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2805 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2809 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2811 adapter->interrupt_event = 0;
2814 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2816 struct ixgbe_hw *hw = &adapter->hw;
2818 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2819 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2820 e_crit(probe, "Fan has stopped, replace the adapter\n");
2821 /* write to clear the interrupt */
2822 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2826 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2828 struct ixgbe_hw *hw = &adapter->hw;
2830 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2833 switch (adapter->hw.mac.type) {
2834 case ixgbe_mac_82599EB:
2836 * Need to check link state so complete overtemp check
2839 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2840 (eicr & IXGBE_EICR_LSC)) &&
2841 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2842 adapter->interrupt_event = eicr;
2843 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2844 ixgbe_service_event_schedule(adapter);
2848 case ixgbe_mac_x550em_a:
2849 if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2850 adapter->interrupt_event = eicr;
2851 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2852 ixgbe_service_event_schedule(adapter);
2853 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2854 IXGBE_EICR_GPI_SDP0_X550EM_a);
2855 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2856 IXGBE_EICR_GPI_SDP0_X550EM_a);
2859 case ixgbe_mac_X550:
2860 case ixgbe_mac_X540:
2861 if (!(eicr & IXGBE_EICR_TS))
2868 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2871 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2873 switch (hw->mac.type) {
2874 case ixgbe_mac_82598EB:
2875 if (hw->phy.type == ixgbe_phy_nl)
2878 case ixgbe_mac_82599EB:
2879 case ixgbe_mac_X550EM_x:
2880 case ixgbe_mac_x550em_a:
2881 switch (hw->mac.ops.get_media_type(hw)) {
2882 case ixgbe_media_type_fiber:
2883 case ixgbe_media_type_fiber_qsfp:
2893 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2895 struct ixgbe_hw *hw = &adapter->hw;
2896 u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2898 if (!ixgbe_is_sfp(hw))
2901 /* Later MAC's use different SDP */
2902 if (hw->mac.type >= ixgbe_mac_X540)
2903 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2905 if (eicr & eicr_mask) {
2906 /* Clear the interrupt */
2907 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2908 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2909 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2910 adapter->sfp_poll_time = 0;
2911 ixgbe_service_event_schedule(adapter);
2915 if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2916 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2917 /* Clear the interrupt */
2918 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2919 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2920 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2921 ixgbe_service_event_schedule(adapter);
2926 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2928 struct ixgbe_hw *hw = &adapter->hw;
2931 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2932 adapter->link_check_timeout = jiffies;
2933 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2934 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2935 IXGBE_WRITE_FLUSH(hw);
2936 ixgbe_service_event_schedule(adapter);
2940 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2944 struct ixgbe_hw *hw = &adapter->hw;
2946 switch (hw->mac.type) {
2947 case ixgbe_mac_82598EB:
2948 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2949 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2951 case ixgbe_mac_82599EB:
2952 case ixgbe_mac_X540:
2953 case ixgbe_mac_X550:
2954 case ixgbe_mac_X550EM_x:
2955 case ixgbe_mac_x550em_a:
2956 mask = (qmask & 0xFFFFFFFF);
2958 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2959 mask = (qmask >> 32);
2961 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2966 /* skip the flush */
2970 * ixgbe_irq_enable - Enable default interrupt generation settings
2971 * @adapter: board private structure
2972 * @queues: enable irqs for queues
2973 * @flush: flush register write
2975 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2978 struct ixgbe_hw *hw = &adapter->hw;
2979 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2981 /* don't reenable LSC while waiting for link */
2982 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2983 mask &= ~IXGBE_EIMS_LSC;
2985 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2986 switch (adapter->hw.mac.type) {
2987 case ixgbe_mac_82599EB:
2988 mask |= IXGBE_EIMS_GPI_SDP0(hw);
2990 case ixgbe_mac_X540:
2991 case ixgbe_mac_X550:
2992 case ixgbe_mac_X550EM_x:
2993 case ixgbe_mac_x550em_a:
2994 mask |= IXGBE_EIMS_TS;
2999 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3000 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3001 switch (adapter->hw.mac.type) {
3002 case ixgbe_mac_82599EB:
3003 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3004 mask |= IXGBE_EIMS_GPI_SDP2(hw);
3006 case ixgbe_mac_X540:
3007 case ixgbe_mac_X550:
3008 case ixgbe_mac_X550EM_x:
3009 case ixgbe_mac_x550em_a:
3010 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3011 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3012 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3013 mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3014 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3015 mask |= IXGBE_EICR_GPI_SDP0_X540;
3016 mask |= IXGBE_EIMS_ECC;
3017 mask |= IXGBE_EIMS_MAILBOX;
3023 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3024 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3025 mask |= IXGBE_EIMS_FLOW_DIR;
3027 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3029 ixgbe_irq_enable_queues(adapter, ~0);
3031 IXGBE_WRITE_FLUSH(&adapter->hw);
3034 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3036 struct ixgbe_adapter *adapter = data;
3037 struct ixgbe_hw *hw = &adapter->hw;
3041 * Workaround for Silicon errata. Use clear-by-write instead
3042 * of clear-by-read. Reading with EICS will return the
3043 * interrupt causes without clearing, which later be done
3044 * with the write to EICR.
3046 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3048 /* The lower 16bits of the EICR register are for the queue interrupts
3049 * which should be masked here in order to not accidentally clear them if
3050 * the bits are high when ixgbe_msix_other is called. There is a race
3051 * condition otherwise which results in possible performance loss
3052 * especially if the ixgbe_msix_other interrupt is triggering
3053 * consistently (as it would when PPS is turned on for the X540 device)
3057 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3059 if (eicr & IXGBE_EICR_LSC)
3060 ixgbe_check_lsc(adapter);
3062 if (eicr & IXGBE_EICR_MAILBOX)
3063 ixgbe_msg_task(adapter);
3065 switch (hw->mac.type) {
3066 case ixgbe_mac_82599EB:
3067 case ixgbe_mac_X540:
3068 case ixgbe_mac_X550:
3069 case ixgbe_mac_X550EM_x:
3070 case ixgbe_mac_x550em_a:
3071 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3072 (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3073 adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3074 ixgbe_service_event_schedule(adapter);
3075 IXGBE_WRITE_REG(hw, IXGBE_EICR,
3076 IXGBE_EICR_GPI_SDP0_X540);
3078 if (eicr & IXGBE_EICR_ECC) {
3079 e_info(link, "Received ECC Err, initiating reset\n");
3080 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3081 ixgbe_service_event_schedule(adapter);
3082 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3084 /* Handle Flow Director Full threshold interrupt */
3085 if (eicr & IXGBE_EICR_FLOW_DIR) {
3086 int reinit_count = 0;
3088 for (i = 0; i < adapter->num_tx_queues; i++) {
3089 struct ixgbe_ring *ring = adapter->tx_ring[i];
3090 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3095 /* no more flow director interrupts until after init */
3096 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3097 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3098 ixgbe_service_event_schedule(adapter);
3101 ixgbe_check_sfp_event(adapter, eicr);
3102 ixgbe_check_overtemp_event(adapter, eicr);
3108 ixgbe_check_fan_failure(adapter, eicr);
3110 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3111 ixgbe_ptp_check_pps_event(adapter);
3113 /* re-enable the original interrupt state, no lsc, no queues */
3114 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3115 ixgbe_irq_enable(adapter, false, false);
3120 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3122 struct ixgbe_q_vector *q_vector = data;
3124 /* EIAM disabled interrupts (on this vector) for us */
3126 if (q_vector->rx.ring || q_vector->tx.ring)
3127 napi_schedule_irqoff(&q_vector->napi);
3133 * ixgbe_poll - NAPI Rx polling callback
3134 * @napi: structure for representing this polling device
3135 * @budget: how many packets driver is allowed to clean
3137 * This function is used for legacy and MSI, NAPI mode
3139 int ixgbe_poll(struct napi_struct *napi, int budget)
3141 struct ixgbe_q_vector *q_vector =
3142 container_of(napi, struct ixgbe_q_vector, napi);
3143 struct ixgbe_adapter *adapter = q_vector->adapter;
3144 struct ixgbe_ring *ring;
3145 int per_ring_budget, work_done = 0;
3146 bool clean_complete = true;
3148 #ifdef CONFIG_IXGBE_DCA
3149 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3150 ixgbe_update_dca(q_vector);
3153 ixgbe_for_each_ring(ring, q_vector->tx) {
3154 bool wd = ring->xsk_pool ?
3155 ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3156 ixgbe_clean_tx_irq(q_vector, ring, budget);
3159 clean_complete = false;
3162 /* Exit if we are called by netpoll */
3166 /* attempt to distribute budget to each queue fairly, but don't allow
3167 * the budget to go below 1 because we'll exit polling */
3168 if (q_vector->rx.count > 1)
3169 per_ring_budget = max(budget/q_vector->rx.count, 1);
3171 per_ring_budget = budget;
3173 ixgbe_for_each_ring(ring, q_vector->rx) {
3174 int cleaned = ring->xsk_pool ?
3175 ixgbe_clean_rx_irq_zc(q_vector, ring,
3177 ixgbe_clean_rx_irq(q_vector, ring,
3180 work_done += cleaned;
3181 if (cleaned >= per_ring_budget)
3182 clean_complete = false;
3185 /* If all work not completed, return budget and keep polling */
3186 if (!clean_complete)
3189 /* all work done, exit the polling mode */
3190 if (likely(napi_complete_done(napi, work_done))) {
3191 if (adapter->rx_itr_setting & 1)
3192 ixgbe_set_itr(q_vector);
3193 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3194 ixgbe_irq_enable_queues(adapter,
3195 BIT_ULL(q_vector->v_idx));
3198 return min(work_done, budget - 1);
3202 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3203 * @adapter: board private structure
3205 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3206 * interrupts from the kernel.
3208 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3210 struct net_device *netdev = adapter->netdev;
3211 unsigned int ri = 0, ti = 0;
3214 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3215 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3216 struct msix_entry *entry = &adapter->msix_entries[vector];
3218 if (q_vector->tx.ring && q_vector->rx.ring) {
3219 snprintf(q_vector->name, sizeof(q_vector->name),
3220 "%s-TxRx-%u", netdev->name, ri++);
3222 } else if (q_vector->rx.ring) {
3223 snprintf(q_vector->name, sizeof(q_vector->name),
3224 "%s-rx-%u", netdev->name, ri++);
3225 } else if (q_vector->tx.ring) {
3226 snprintf(q_vector->name, sizeof(q_vector->name),
3227 "%s-tx-%u", netdev->name, ti++);
3229 /* skip this unused q_vector */
3232 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3233 q_vector->name, q_vector);
3235 e_err(probe, "request_irq failed for MSIX interrupt "
3236 "Error: %d\n", err);
3237 goto free_queue_irqs;
3239 /* If Flow Director is enabled, set interrupt affinity */
3240 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3241 /* assign the mask for this irq */
3242 irq_set_affinity_hint(entry->vector,
3243 &q_vector->affinity_mask);
3247 err = request_irq(adapter->msix_entries[vector].vector,
3248 ixgbe_msix_other, 0, netdev->name, adapter);
3250 e_err(probe, "request_irq for msix_other failed: %d\n", err);
3251 goto free_queue_irqs;
3259 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3261 free_irq(adapter->msix_entries[vector].vector,
3262 adapter->q_vector[vector]);
3264 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3265 pci_disable_msix(adapter->pdev);
3266 kfree(adapter->msix_entries);
3267 adapter->msix_entries = NULL;
3272 * ixgbe_intr - legacy mode Interrupt Handler
3273 * @irq: interrupt number
3274 * @data: pointer to a network interface device structure
3276 static irqreturn_t ixgbe_intr(int irq, void *data)
3278 struct ixgbe_adapter *adapter = data;
3279 struct ixgbe_hw *hw = &adapter->hw;
3280 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3284 * Workaround for silicon errata #26 on 82598. Mask the interrupt
3285 * before the read of EICR.
3287 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3289 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3290 * therefore no explicit interrupt disable is necessary */
3291 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3294 * shared interrupt alert!
3295 * make sure interrupts are enabled because the read will
3296 * have disabled interrupts due to EIAM
3297 * finish the workaround of silicon errata on 82598. Unmask
3298 * the interrupt that we masked before the EICR read.
3300 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3301 ixgbe_irq_enable(adapter, true, true);
3302 return IRQ_NONE; /* Not our interrupt */
3305 if (eicr & IXGBE_EICR_LSC)
3306 ixgbe_check_lsc(adapter);
3308 switch (hw->mac.type) {
3309 case ixgbe_mac_82599EB:
3310 ixgbe_check_sfp_event(adapter, eicr);
3312 case ixgbe_mac_X540:
3313 case ixgbe_mac_X550:
3314 case ixgbe_mac_X550EM_x:
3315 case ixgbe_mac_x550em_a:
3316 if (eicr & IXGBE_EICR_ECC) {
3317 e_info(link, "Received ECC Err, initiating reset\n");
3318 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3319 ixgbe_service_event_schedule(adapter);
3320 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3322 ixgbe_check_overtemp_event(adapter, eicr);
3328 ixgbe_check_fan_failure(adapter, eicr);
3329 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3330 ixgbe_ptp_check_pps_event(adapter);
3332 /* would disable interrupts here but EIAM disabled it */
3333 napi_schedule_irqoff(&q_vector->napi);
3336 * re-enable link(maybe) and non-queue interrupts, no flush.
3337 * ixgbe_poll will re-enable the queue interrupts
3339 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3340 ixgbe_irq_enable(adapter, false, false);
3346 * ixgbe_request_irq - initialize interrupts
3347 * @adapter: board private structure
3349 * Attempts to configure interrupts using the best available
3350 * capabilities of the hardware and kernel.
3352 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3354 struct net_device *netdev = adapter->netdev;
3357 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3358 err = ixgbe_request_msix_irqs(adapter);
3359 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3360 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3361 netdev->name, adapter);
3363 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3364 netdev->name, adapter);
3367 e_err(probe, "request_irq failed, Error %d\n", err);
3372 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3376 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3377 free_irq(adapter->pdev->irq, adapter);
3381 if (!adapter->msix_entries)
3384 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3385 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3386 struct msix_entry *entry = &adapter->msix_entries[vector];
3388 /* free only the irqs that were actually requested */
3389 if (!q_vector->rx.ring && !q_vector->tx.ring)
3392 /* clear the affinity_mask in the IRQ descriptor */
3393 irq_set_affinity_hint(entry->vector, NULL);
3395 free_irq(entry->vector, q_vector);
3398 free_irq(adapter->msix_entries[vector].vector, adapter);
3402 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3403 * @adapter: board private structure
3405 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3407 switch (adapter->hw.mac.type) {
3408 case ixgbe_mac_82598EB:
3409 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3411 case ixgbe_mac_82599EB:
3412 case ixgbe_mac_X540:
3413 case ixgbe_mac_X550:
3414 case ixgbe_mac_X550EM_x:
3415 case ixgbe_mac_x550em_a:
3416 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3417 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3418 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3423 IXGBE_WRITE_FLUSH(&adapter->hw);
3424 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3427 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3428 synchronize_irq(adapter->msix_entries[vector].vector);
3430 synchronize_irq(adapter->msix_entries[vector++].vector);
3432 synchronize_irq(adapter->pdev->irq);
3437 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3438 * @adapter: board private structure
3441 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3443 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3445 ixgbe_write_eitr(q_vector);
3447 ixgbe_set_ivar(adapter, 0, 0, 0);
3448 ixgbe_set_ivar(adapter, 1, 0, 0);
3450 e_info(hw, "Legacy interrupt IVAR setup done\n");
3454 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3455 * @adapter: board private structure
3456 * @ring: structure containing ring specific data
3458 * Configure the Tx descriptor ring after a reset.
3460 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3461 struct ixgbe_ring *ring)
3463 struct ixgbe_hw *hw = &adapter->hw;
3464 u64 tdba = ring->dma;
3466 u32 txdctl = IXGBE_TXDCTL_ENABLE;
3467 u8 reg_idx = ring->reg_idx;
3469 ring->xsk_pool = NULL;
3470 if (ring_is_xdp(ring))
3471 ring->xsk_pool = ixgbe_xsk_pool(adapter, ring);
3473 /* disable queue to avoid issues while updating state */
3474 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3475 IXGBE_WRITE_FLUSH(hw);
3477 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3478 (tdba & DMA_BIT_MASK(32)));
3479 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3480 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3481 ring->count * sizeof(union ixgbe_adv_tx_desc));
3482 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3483 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3484 ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3487 * set WTHRESH to encourage burst writeback, it should not be set
3488 * higher than 1 when:
3489 * - ITR is 0 as it could cause false TX hangs
3490 * - ITR is set to > 100k int/sec and BQL is enabled
3492 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3493 * to or less than the number of on chip descriptors, which is
3496 if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3497 txdctl |= 1u << 16; /* WTHRESH = 1 */
3499 txdctl |= 8u << 16; /* WTHRESH = 8 */
3502 * Setting PTHRESH to 32 both improves performance
3503 * and avoids a TX hang with DFP enabled
3505 txdctl |= (1u << 8) | /* HTHRESH = 1 */
3506 32; /* PTHRESH = 32 */
3508 /* reinitialize flowdirector state */
3509 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3510 ring->atr_sample_rate = adapter->atr_sample_rate;
3511 ring->atr_count = 0;
3512 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3514 ring->atr_sample_rate = 0;
3517 /* initialize XPS */
3518 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3519 struct ixgbe_q_vector *q_vector = ring->q_vector;
3522 netif_set_xps_queue(ring->netdev,
3523 &q_vector->affinity_mask,
3527 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3529 /* reinitialize tx_buffer_info */
3530 memset(ring->tx_buffer_info, 0,
3531 sizeof(struct ixgbe_tx_buffer) * ring->count);
3534 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3536 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3537 if (hw->mac.type == ixgbe_mac_82598EB &&
3538 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3541 /* poll to verify queue is enabled */
3543 usleep_range(1000, 2000);
3544 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3545 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3547 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3550 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3552 struct ixgbe_hw *hw = &adapter->hw;
3554 u8 tcs = adapter->hw_tcs;
3556 if (hw->mac.type == ixgbe_mac_82598EB)
3559 /* disable the arbiter while setting MTQC */
3560 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3561 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3562 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3564 /* set transmit pool layout */
3565 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3566 mtqc = IXGBE_MTQC_VT_ENA;
3568 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3570 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3571 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3572 IXGBE_82599_VMDQ_4Q_MASK)
3573 mtqc |= IXGBE_MTQC_32VF;
3575 mtqc |= IXGBE_MTQC_64VF;
3578 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3579 } else if (tcs > 1) {
3580 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3582 u8 max_txq = adapter->num_tx_queues +
3583 adapter->num_xdp_queues;
3585 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3587 mtqc = IXGBE_MTQC_64Q_1PB;
3591 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3593 /* Enable Security TX Buffer IFG for multiple pb */
3595 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3596 sectx |= IXGBE_SECTX_DCB;
3597 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3600 /* re-enable the arbiter */
3601 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3602 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3606 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3607 * @adapter: board private structure
3609 * Configure the Tx unit of the MAC after a reset.
3611 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3613 struct ixgbe_hw *hw = &adapter->hw;
3617 ixgbe_setup_mtqc(adapter);
3619 if (hw->mac.type != ixgbe_mac_82598EB) {
3620 /* DMATXCTL.EN must be before Tx queues are enabled */
3621 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3622 dmatxctl |= IXGBE_DMATXCTL_TE;
3623 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3626 /* Setup the HW Tx Head and Tail descriptor pointers */
3627 for (i = 0; i < adapter->num_tx_queues; i++)
3628 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3629 for (i = 0; i < adapter->num_xdp_queues; i++)
3630 ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3633 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3634 struct ixgbe_ring *ring)
3636 struct ixgbe_hw *hw = &adapter->hw;
3637 u8 reg_idx = ring->reg_idx;
3638 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3640 srrctl |= IXGBE_SRRCTL_DROP_EN;
3642 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3645 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3646 struct ixgbe_ring *ring)
3648 struct ixgbe_hw *hw = &adapter->hw;
3649 u8 reg_idx = ring->reg_idx;
3650 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3652 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3654 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3657 #ifdef CONFIG_IXGBE_DCB
3658 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3660 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3664 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3666 if (adapter->ixgbe_ieee_pfc)
3667 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3670 * We should set the drop enable bit if:
3673 * Number of Rx queues > 1 and flow control is disabled
3675 * This allows us to avoid head of line blocking for security
3676 * and performance reasons.
3678 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3679 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3680 for (i = 0; i < adapter->num_rx_queues; i++)
3681 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3683 for (i = 0; i < adapter->num_rx_queues; i++)
3684 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3688 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3690 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3691 struct ixgbe_ring *rx_ring)
3693 struct ixgbe_hw *hw = &adapter->hw;
3695 u8 reg_idx = rx_ring->reg_idx;
3697 if (hw->mac.type == ixgbe_mac_82598EB) {
3698 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3701 * if VMDq is not active we must program one srrctl register
3702 * per RSS queue since we have enabled RDRXCTL.MVMEN
3707 /* configure header buffer length, needed for RSC */
3708 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3710 /* configure the packet buffer length */
3711 if (rx_ring->xsk_pool) {
3712 u32 xsk_buf_len = xsk_pool_get_rx_frame_size(rx_ring->xsk_pool);
3714 /* If the MAC support setting RXDCTL.RLPML, the
3715 * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3716 * RXDCTL.RLPML is set to the actual UMEM buffer
3717 * size. If not, then we are stuck with a 1k buffer
3718 * size resolution. In this case frames larger than
3719 * the UMEM buffer size viewed in a 1k resolution will
3722 if (hw->mac.type != ixgbe_mac_82599EB)
3723 srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3725 srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3726 } else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3727 srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3729 srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3732 /* configure descriptor type */
3733 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3735 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3739 * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3740 * @adapter: device handle
3742 * - 82598/82599/X540: 128
3743 * - X550(non-SRIOV mode): 512
3744 * - X550(SRIOV mode): 64
3746 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3748 if (adapter->hw.mac.type < ixgbe_mac_X550)
3750 else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3757 * ixgbe_store_key - Write the RSS key to HW
3758 * @adapter: device handle
3760 * Write the RSS key stored in adapter.rss_key to HW.
3762 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3764 struct ixgbe_hw *hw = &adapter->hw;
3767 for (i = 0; i < 10; i++)
3768 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3772 * ixgbe_init_rss_key - Initialize adapter RSS key
3773 * @adapter: device handle
3775 * Allocates and initializes the RSS key if it is not allocated.
3777 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3781 if (!adapter->rss_key) {
3782 rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3783 if (unlikely(!rss_key))
3786 netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3787 adapter->rss_key = rss_key;
3794 * ixgbe_store_reta - Write the RETA table to HW
3795 * @adapter: device handle
3797 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3799 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3801 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3802 struct ixgbe_hw *hw = &adapter->hw;
3805 u8 *indir_tbl = adapter->rss_indir_tbl;
3807 /* Fill out the redirection table as follows:
3808 * - 82598: 8 bit wide entries containing pair of 4 bit RSS
3810 * - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3811 * - X550: 8 bit wide entries containing 6 bit RSS index
3813 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3814 indices_multi = 0x11;
3816 indices_multi = 0x1;
3818 /* Write redirection table to HW */
3819 for (i = 0; i < reta_entries; i++) {
3820 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3823 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3825 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3833 * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3834 * @adapter: device handle
3836 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3838 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3840 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3841 struct ixgbe_hw *hw = &adapter->hw;
3844 /* Write redirection table to HW */
3845 for (i = 0; i < reta_entries; i++) {
3846 u16 pool = adapter->num_rx_pools;
3848 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3854 IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3860 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3863 u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3864 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3866 /* Program table for at least 4 queues w/ SR-IOV so that VFs can
3867 * make full use of any rings they may have. We will use the
3868 * PSRTYPE register to control how many rings we use within the PF.
3870 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3873 /* Fill out hash function seeds */
3874 ixgbe_store_key(adapter);
3876 /* Fill out redirection table */
3877 memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3879 for (i = 0, j = 0; i < reta_entries; i++, j++) {
3883 adapter->rss_indir_tbl[i] = j;
3886 ixgbe_store_reta(adapter);
3889 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3891 struct ixgbe_hw *hw = &adapter->hw;
3892 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3895 /* Fill out hash function seeds */
3896 for (i = 0; i < 10; i++) {
3897 u16 pool = adapter->num_rx_pools;
3901 IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3902 *(adapter->rss_key + i));
3905 /* Fill out the redirection table */
3906 for (i = 0, j = 0; i < 64; i++, j++) {
3910 adapter->rss_indir_tbl[i] = j;
3913 ixgbe_store_vfreta(adapter);
3916 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3918 struct ixgbe_hw *hw = &adapter->hw;
3919 u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3922 /* Disable indicating checksum in descriptor, enables RSS hash */
3923 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3924 rxcsum |= IXGBE_RXCSUM_PCSD;
3925 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3927 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3928 if (adapter->ring_feature[RING_F_RSS].mask)
3929 mrqc = IXGBE_MRQC_RSSEN;
3931 u8 tcs = adapter->hw_tcs;
3933 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3935 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3937 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3938 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3939 IXGBE_82599_VMDQ_4Q_MASK)
3940 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3942 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3944 /* Enable L3/L4 for Tx Switched packets only for X550,
3945 * older devices do not support this feature
3947 if (hw->mac.type >= ixgbe_mac_X550)
3948 mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3951 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3953 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3955 mrqc = IXGBE_MRQC_RSSEN;
3959 /* Perform hash on these packet types */
3960 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3961 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3962 IXGBE_MRQC_RSS_FIELD_IPV6 |
3963 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3965 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3966 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3967 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3968 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3970 if ((hw->mac.type >= ixgbe_mac_X550) &&
3971 (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3972 u16 pool = adapter->num_rx_pools;
3974 /* Enable VF RSS mode */
3975 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3976 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3978 /* Setup RSS through the VF registers */
3979 ixgbe_setup_vfreta(adapter);
3980 vfmrqc = IXGBE_MRQC_RSSEN;
3981 vfmrqc |= rss_field;
3985 IXGBE_PFVFMRQC(VMDQ_P(pool)),
3988 ixgbe_setup_reta(adapter);
3990 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3995 * ixgbe_configure_rscctl - enable RSC for the indicated ring
3996 * @adapter: address of board private structure
3997 * @ring: structure containing ring specific data
3999 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4000 struct ixgbe_ring *ring)
4002 struct ixgbe_hw *hw = &adapter->hw;
4004 u8 reg_idx = ring->reg_idx;
4006 if (!ring_is_rsc_enabled(ring))
4009 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4010 rscctrl |= IXGBE_RSCCTL_RSCEN;
4012 * we must limit the number of descriptors so that the
4013 * total size of max desc * buf_len is not greater
4016 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4017 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4020 #define IXGBE_MAX_RX_DESC_POLL 10
4021 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4022 struct ixgbe_ring *ring)
4024 struct ixgbe_hw *hw = &adapter->hw;
4025 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4027 u8 reg_idx = ring->reg_idx;
4029 if (ixgbe_removed(hw->hw_addr))
4031 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4032 if (hw->mac.type == ixgbe_mac_82598EB &&
4033 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4037 usleep_range(1000, 2000);
4038 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4039 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4042 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4043 "the polling period\n", reg_idx);
4047 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4048 struct ixgbe_ring *ring)
4050 struct ixgbe_hw *hw = &adapter->hw;
4051 union ixgbe_adv_rx_desc *rx_desc;
4052 u64 rdba = ring->dma;
4054 u8 reg_idx = ring->reg_idx;
4056 xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4057 ring->xsk_pool = ixgbe_xsk_pool(adapter, ring);
4058 if (ring->xsk_pool) {
4059 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4060 MEM_TYPE_XSK_BUFF_POOL,
4062 xsk_pool_set_rxq_info(ring->xsk_pool, &ring->xdp_rxq);
4064 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4065 MEM_TYPE_PAGE_SHARED, NULL));
4068 /* disable queue to avoid use of these values while updating state */
4069 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4070 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4072 /* write value back with RXDCTL.ENABLE bit cleared */
4073 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4074 IXGBE_WRITE_FLUSH(hw);
4076 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4077 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4078 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4079 ring->count * sizeof(union ixgbe_adv_rx_desc));
4080 /* Force flushing of IXGBE_RDLEN to prevent MDD */
4081 IXGBE_WRITE_FLUSH(hw);
4083 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4084 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4085 ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4087 ixgbe_configure_srrctl(adapter, ring);
4088 ixgbe_configure_rscctl(adapter, ring);
4090 if (hw->mac.type == ixgbe_mac_82598EB) {
4092 * enable cache line friendly hardware writes:
4093 * PTHRESH=32 descriptors (half the internal cache),
4094 * this also removes ugly rx_no_buffer_count increment
4095 * HTHRESH=4 descriptors (to minimize latency on fetch)
4096 * WTHRESH=8 burst writeback up to two cache lines
4098 rxdctl &= ~0x3FFFFF;
4100 #if (PAGE_SIZE < 8192)
4101 /* RXDCTL.RLPML does not work on 82599 */
4102 } else if (hw->mac.type != ixgbe_mac_82599EB) {
4103 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4104 IXGBE_RXDCTL_RLPML_EN);
4106 /* Limit the maximum frame size so we don't overrun the skb.
4107 * This can happen in SRIOV mode when the MTU of the VF is
4108 * higher than the MTU of the PF.
4110 if (ring_uses_build_skb(ring) &&
4111 !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4112 rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4113 IXGBE_RXDCTL_RLPML_EN;
4117 if (ring->xsk_pool && hw->mac.type != ixgbe_mac_82599EB) {
4118 u32 xsk_buf_len = xsk_pool_get_rx_frame_size(ring->xsk_pool);
4120 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4121 IXGBE_RXDCTL_RLPML_EN);
4122 rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4124 ring->rx_buf_len = xsk_buf_len;
4127 /* initialize rx_buffer_info */
4128 memset(ring->rx_buffer_info, 0,
4129 sizeof(struct ixgbe_rx_buffer) * ring->count);
4131 /* initialize Rx descriptor 0 */
4132 rx_desc = IXGBE_RX_DESC(ring, 0);
4133 rx_desc->wb.upper.length = 0;
4135 /* enable receive descriptor ring */
4136 rxdctl |= IXGBE_RXDCTL_ENABLE;
4137 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4139 ixgbe_rx_desc_queue_enable(adapter, ring);
4141 ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4143 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4146 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4148 struct ixgbe_hw *hw = &adapter->hw;
4149 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4150 u16 pool = adapter->num_rx_pools;
4152 /* PSRTYPE must be initialized in non 82598 adapters */
4153 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4154 IXGBE_PSRTYPE_UDPHDR |
4155 IXGBE_PSRTYPE_IPV4HDR |
4156 IXGBE_PSRTYPE_L2HDR |
4157 IXGBE_PSRTYPE_IPV6HDR;
4159 if (hw->mac.type == ixgbe_mac_82598EB)
4163 psrtype |= 2u << 29;
4165 psrtype |= 1u << 29;
4168 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4171 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4173 struct ixgbe_hw *hw = &adapter->hw;
4174 u16 pool = adapter->num_rx_pools;
4175 u32 reg_offset, vf_shift, vmolr;
4176 u32 gcr_ext, vmdctl;
4179 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4182 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4183 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4184 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4185 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4186 vmdctl |= IXGBE_VT_CTL_REPLEN;
4187 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4189 /* accept untagged packets until a vlan tag is
4190 * specifically set for the VMDQ queue/pool
4192 vmolr = IXGBE_VMOLR_AUPE;
4194 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4196 vf_shift = VMDQ_P(0) % 32;
4197 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4199 /* Enable only the PF's pool for Tx/Rx */
4200 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4201 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4202 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4203 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4204 if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4205 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4207 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4208 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4210 /* clear VLAN promisc flag so VFTA will be updated if necessary */
4211 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4214 * Set up VF register offsets for selected VT Mode,
4215 * i.e. 32 or 64 VFs for SR-IOV
4217 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4218 case IXGBE_82599_VMDQ_8Q_MASK:
4219 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4221 case IXGBE_82599_VMDQ_4Q_MASK:
4222 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4225 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4229 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4231 for (i = 0; i < adapter->num_vfs; i++) {
4232 /* configure spoof checking */
4233 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4234 adapter->vfinfo[i].spoofchk_enabled);
4236 /* Enable/Disable RSS query feature */
4237 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4238 adapter->vfinfo[i].rss_query_enabled);
4242 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4244 struct ixgbe_hw *hw = &adapter->hw;
4245 struct net_device *netdev = adapter->netdev;
4246 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4247 struct ixgbe_ring *rx_ring;
4252 /* adjust max frame to be able to do baby jumbo for FCoE */
4253 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4254 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4255 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4257 #endif /* IXGBE_FCOE */
4259 /* adjust max frame to be at least the size of a standard frame */
4260 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4261 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4263 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4264 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4265 mhadd &= ~IXGBE_MHADD_MFS_MASK;
4266 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4268 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4271 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4272 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4273 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4274 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4277 * Setup the HW Rx Head and Tail Descriptor Pointers and
4278 * the Base and Length of the Rx Descriptor Ring
4280 for (i = 0; i < adapter->num_rx_queues; i++) {
4281 rx_ring = adapter->rx_ring[i];
4283 clear_ring_rsc_enabled(rx_ring);
4284 clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4285 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4287 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4288 set_ring_rsc_enabled(rx_ring);
4290 if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4291 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4293 if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4296 set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4298 #if (PAGE_SIZE < 8192)
4299 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4300 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4302 if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4303 (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4304 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4309 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4311 struct ixgbe_hw *hw = &adapter->hw;
4312 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4314 switch (hw->mac.type) {
4315 case ixgbe_mac_82598EB:
4317 * For VMDq support of different descriptor types or
4318 * buffer sizes through the use of multiple SRRCTL
4319 * registers, RDRXCTL.MVMEN must be set to 1
4321 * also, the manual doesn't mention it clearly but DCA hints
4322 * will only use queue 0's tags unless this bit is set. Side
4323 * effects of setting this bit are only that SRRCTL must be
4324 * fully programmed [0..15]
4326 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4328 case ixgbe_mac_X550:
4329 case ixgbe_mac_X550EM_x:
4330 case ixgbe_mac_x550em_a:
4331 if (adapter->num_vfs)
4332 rdrxctl |= IXGBE_RDRXCTL_PSP;
4334 case ixgbe_mac_82599EB:
4335 case ixgbe_mac_X540:
4336 /* Disable RSC for ACK packets */
4337 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4338 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4339 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4340 /* hardware requires some bits to be set by default */
4341 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4342 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4345 /* We should do nothing since we don't know this hardware */
4349 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4353 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4354 * @adapter: board private structure
4356 * Configure the Rx unit of the MAC after a reset.
4358 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4360 struct ixgbe_hw *hw = &adapter->hw;
4364 /* disable receives while setting up the descriptors */
4365 hw->mac.ops.disable_rx(hw);
4367 ixgbe_setup_psrtype(adapter);
4368 ixgbe_setup_rdrxctl(adapter);
4371 rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4372 rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4373 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4374 rfctl |= IXGBE_RFCTL_RSC_DIS;
4376 /* disable NFS filtering */
4377 rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4378 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4380 /* Program registers for the distribution of queues */
4381 ixgbe_setup_mrqc(adapter);
4383 /* set_rx_buffer_len must be called before ring initialization */
4384 ixgbe_set_rx_buffer_len(adapter);
4387 * Setup the HW Rx Head and Tail Descriptor Pointers and
4388 * the Base and Length of the Rx Descriptor Ring
4390 for (i = 0; i < adapter->num_rx_queues; i++)
4391 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4393 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4394 /* disable drop enable for 82598 parts */
4395 if (hw->mac.type == ixgbe_mac_82598EB)
4396 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4398 /* enable all receives */
4399 rxctrl |= IXGBE_RXCTRL_RXEN;
4400 hw->mac.ops.enable_rx_dma(hw, rxctrl);
4403 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4404 __be16 proto, u16 vid)
4406 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4407 struct ixgbe_hw *hw = &adapter->hw;
4409 /* add VID to filter table */
4410 if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4411 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4413 set_bit(vid, adapter->active_vlans);
4418 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4423 /* short cut the special case */
4427 /* Search for the vlan id in the VLVF entries */
4428 for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4429 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4430 if ((vlvf & VLAN_VID_MASK) == vlan)
4437 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4439 struct ixgbe_hw *hw = &adapter->hw;
4443 idx = ixgbe_find_vlvf_entry(hw, vid);
4447 /* See if any other pools are set for this VLAN filter
4448 * entry other than the PF.
4450 word = idx * 2 + (VMDQ_P(0) / 32);
4451 bits = ~BIT(VMDQ_P(0) % 32);
4452 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4454 /* Disable the filter so this falls into the default pool. */
4455 if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4456 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4457 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4458 IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4462 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4463 __be16 proto, u16 vid)
4465 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4466 struct ixgbe_hw *hw = &adapter->hw;
4468 /* remove VID from filter table */
4469 if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4470 hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4472 clear_bit(vid, adapter->active_vlans);
4478 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4479 * @adapter: driver data
4481 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4483 struct ixgbe_hw *hw = &adapter->hw;
4487 switch (hw->mac.type) {
4488 case ixgbe_mac_82598EB:
4489 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4490 vlnctrl &= ~IXGBE_VLNCTRL_VME;
4491 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4493 case ixgbe_mac_82599EB:
4494 case ixgbe_mac_X540:
4495 case ixgbe_mac_X550:
4496 case ixgbe_mac_X550EM_x:
4497 case ixgbe_mac_x550em_a:
4498 for (i = 0; i < adapter->num_rx_queues; i++) {
4499 struct ixgbe_ring *ring = adapter->rx_ring[i];
4501 if (!netif_is_ixgbe(ring->netdev))
4505 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4506 vlnctrl &= ~IXGBE_RXDCTL_VME;
4507 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4516 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4517 * @adapter: driver data
4519 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4521 struct ixgbe_hw *hw = &adapter->hw;
4525 switch (hw->mac.type) {
4526 case ixgbe_mac_82598EB:
4527 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4528 vlnctrl |= IXGBE_VLNCTRL_VME;
4529 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4531 case ixgbe_mac_82599EB:
4532 case ixgbe_mac_X540:
4533 case ixgbe_mac_X550:
4534 case ixgbe_mac_X550EM_x:
4535 case ixgbe_mac_x550em_a:
4536 for (i = 0; i < adapter->num_rx_queues; i++) {
4537 struct ixgbe_ring *ring = adapter->rx_ring[i];
4539 if (!netif_is_ixgbe(ring->netdev))
4543 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4544 vlnctrl |= IXGBE_RXDCTL_VME;
4545 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4553 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4555 struct ixgbe_hw *hw = &adapter->hw;
4558 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4560 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4561 /* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4562 vlnctrl |= IXGBE_VLNCTRL_VFE;
4563 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4565 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4566 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4570 /* Nothing to do for 82598 */
4571 if (hw->mac.type == ixgbe_mac_82598EB)
4574 /* We are already in VLAN promisc, nothing to do */
4575 if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4578 /* Set flag so we don't redo unnecessary work */
4579 adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4581 /* Add PF to all active pools */
4582 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4583 u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4584 u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4586 vlvfb |= BIT(VMDQ_P(0) % 32);
4587 IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4590 /* Set all bits in the VLAN filter table array */
4591 for (i = hw->mac.vft_size; i--;)
4592 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4595 #define VFTA_BLOCK_SIZE 8
4596 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4598 struct ixgbe_hw *hw = &adapter->hw;
4599 u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4600 u32 vid_start = vfta_offset * 32;
4601 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4602 u32 i, vid, word, bits;
4604 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4605 u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4607 /* pull VLAN ID from VLVF */
4608 vid = vlvf & VLAN_VID_MASK;
4610 /* only concern outselves with a certain range */
4611 if (vid < vid_start || vid >= vid_end)
4615 /* record VLAN ID in VFTA */
4616 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4618 /* if PF is part of this then continue */
4619 if (test_bit(vid, adapter->active_vlans))
4623 /* remove PF from the pool */
4624 word = i * 2 + VMDQ_P(0) / 32;
4625 bits = ~BIT(VMDQ_P(0) % 32);
4626 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4627 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4630 /* extract values from active_vlans and write back to VFTA */
4631 for (i = VFTA_BLOCK_SIZE; i--;) {
4632 vid = (vfta_offset + i) * 32;
4633 word = vid / BITS_PER_LONG;
4634 bits = vid % BITS_PER_LONG;
4636 vfta[i] |= adapter->active_vlans[word] >> bits;
4638 IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4642 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4644 struct ixgbe_hw *hw = &adapter->hw;
4647 /* Set VLAN filtering to enabled */
4648 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4649 vlnctrl |= IXGBE_VLNCTRL_VFE;
4650 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4652 if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4653 hw->mac.type == ixgbe_mac_82598EB)
4656 /* We are not in VLAN promisc, nothing to do */
4657 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4660 /* Set flag so we don't redo unnecessary work */
4661 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4663 for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4664 ixgbe_scrub_vfta(adapter, i);
4667 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4671 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4673 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4674 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4678 * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4679 * @netdev: network interface device structure
4681 * Writes multicast address list to the MTA hash table.
4682 * Returns: -ENOMEM on failure
4683 * 0 on no addresses written
4684 * X on writing X addresses to MTA
4686 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4688 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4689 struct ixgbe_hw *hw = &adapter->hw;
4691 if (!netif_running(netdev))
4694 if (hw->mac.ops.update_mc_addr_list)
4695 hw->mac.ops.update_mc_addr_list(hw, netdev);
4699 #ifdef CONFIG_PCI_IOV
4700 ixgbe_restore_vf_multicasts(adapter);
4703 return netdev_mc_count(netdev);
4706 #ifdef CONFIG_PCI_IOV
4707 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4709 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4710 struct ixgbe_hw *hw = &adapter->hw;
4713 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4714 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4716 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4717 hw->mac.ops.set_rar(hw, i,
4722 hw->mac.ops.clear_rar(hw, i);
4727 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4729 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4730 struct ixgbe_hw *hw = &adapter->hw;
4733 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4734 if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4737 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4739 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4740 hw->mac.ops.set_rar(hw, i,
4745 hw->mac.ops.clear_rar(hw, i);
4749 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4751 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4752 struct ixgbe_hw *hw = &adapter->hw;
4755 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4756 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4757 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4760 ixgbe_sync_mac_table(adapter);
4763 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4765 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4766 struct ixgbe_hw *hw = &adapter->hw;
4769 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4770 /* do not count default RAR as available */
4771 if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4774 /* only count unused and addresses that belong to us */
4775 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4776 if (mac_table->pool != pool)
4786 /* this function destroys the first RAR entry */
4787 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4789 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4790 struct ixgbe_hw *hw = &adapter->hw;
4792 memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4793 mac_table->pool = VMDQ_P(0);
4795 mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4797 hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4801 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4802 const u8 *addr, u16 pool)
4804 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4805 struct ixgbe_hw *hw = &adapter->hw;
4808 if (is_zero_ether_addr(addr))
4811 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4812 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4815 ether_addr_copy(mac_table->addr, addr);
4816 mac_table->pool = pool;
4818 mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4819 IXGBE_MAC_STATE_IN_USE;
4821 ixgbe_sync_mac_table(adapter);
4829 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4830 const u8 *addr, u16 pool)
4832 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4833 struct ixgbe_hw *hw = &adapter->hw;
4836 if (is_zero_ether_addr(addr))
4839 /* search table for addr, if found clear IN_USE flag and sync */
4840 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4841 /* we can only delete an entry if it is in use */
4842 if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4844 /* we only care about entries that belong to the given pool */
4845 if (mac_table->pool != pool)
4847 /* we only care about a specific MAC address */
4848 if (!ether_addr_equal(addr, mac_table->addr))
4851 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4852 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4854 ixgbe_sync_mac_table(adapter);
4862 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4864 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4867 ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4869 return min_t(int, ret, 0);
4872 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4874 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4876 ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4882 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4883 * @netdev: network interface device structure
4885 * The set_rx_method entry point is called whenever the unicast/multicast
4886 * address list or the network interface flags are updated. This routine is
4887 * responsible for configuring the hardware for proper unicast, multicast and
4890 void ixgbe_set_rx_mode(struct net_device *netdev)
4892 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4893 struct ixgbe_hw *hw = &adapter->hw;
4894 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4895 netdev_features_t features = netdev->features;
4898 /* Check for Promiscuous and All Multicast modes */
4899 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4901 /* set all bits that we expect to always be set */
4902 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4903 fctrl |= IXGBE_FCTRL_BAM;
4904 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4905 fctrl |= IXGBE_FCTRL_PMCF;
4907 /* clear the bits we are changing the status of */
4908 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4909 if (netdev->flags & IFF_PROMISC) {
4910 hw->addr_ctrl.user_set_promisc = true;
4911 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4912 vmolr |= IXGBE_VMOLR_MPE;
4913 features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4915 if (netdev->flags & IFF_ALLMULTI) {
4916 fctrl |= IXGBE_FCTRL_MPE;
4917 vmolr |= IXGBE_VMOLR_MPE;
4919 hw->addr_ctrl.user_set_promisc = false;
4923 * Write addresses to available RAR registers, if there is not
4924 * sufficient space to store all the addresses then enable
4925 * unicast promiscuous mode
4927 if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4928 fctrl |= IXGBE_FCTRL_UPE;
4929 vmolr |= IXGBE_VMOLR_ROPE;
4932 /* Write addresses to the MTA, if the attempt fails
4933 * then we should just turn on promiscuous mode so
4934 * that we can at least receive multicast traffic
4936 count = ixgbe_write_mc_addr_list(netdev);
4938 fctrl |= IXGBE_FCTRL_MPE;
4939 vmolr |= IXGBE_VMOLR_MPE;
4941 vmolr |= IXGBE_VMOLR_ROMPE;
4944 if (hw->mac.type != ixgbe_mac_82598EB) {
4945 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4946 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4948 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4951 /* This is useful for sniffing bad packets. */
4952 if (features & NETIF_F_RXALL) {
4953 /* UPE and MPE will be handled by normal PROMISC logic
4954 * in e1000e_set_rx_mode */
4955 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4956 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4957 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4959 fctrl &= ~(IXGBE_FCTRL_DPF);
4960 /* NOTE: VLAN filtering is disabled by setting PROMISC */
4963 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4965 if (features & NETIF_F_HW_VLAN_CTAG_RX)
4966 ixgbe_vlan_strip_enable(adapter);
4968 ixgbe_vlan_strip_disable(adapter);
4970 if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4971 ixgbe_vlan_promisc_disable(adapter);
4973 ixgbe_vlan_promisc_enable(adapter);
4976 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4980 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4981 napi_enable(&adapter->q_vector[q_idx]->napi);
4984 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4988 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4989 napi_disable(&adapter->q_vector[q_idx]->napi);
4992 static int ixgbe_udp_tunnel_sync(struct net_device *dev, unsigned int table)
4994 struct ixgbe_adapter *adapter = netdev_priv(dev);
4995 struct ixgbe_hw *hw = &adapter->hw;
4996 struct udp_tunnel_info ti;
4998 udp_tunnel_nic_get_port(dev, table, 0, &ti);
4999 if (ti.type == UDP_TUNNEL_TYPE_VXLAN)
5000 adapter->vxlan_port = ti.port;
5002 adapter->geneve_port = ti.port;
5004 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL,
5005 ntohs(adapter->vxlan_port) |
5006 ntohs(adapter->geneve_port) <<
5007 IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT);
5011 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550 = {
5012 .sync_table = ixgbe_udp_tunnel_sync,
5013 .flags = UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5015 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
5019 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550em_a = {
5020 .sync_table = ixgbe_udp_tunnel_sync,
5021 .flags = UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5023 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
5024 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_GENEVE, },
5028 #ifdef CONFIG_IXGBE_DCB
5030 * ixgbe_configure_dcb - Configure DCB hardware
5031 * @adapter: ixgbe adapter struct
5033 * This is called by the driver on open to configure the DCB hardware.
5034 * This is also called by the gennetlink interface when reconfiguring
5037 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5039 struct ixgbe_hw *hw = &adapter->hw;
5040 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5042 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5043 if (hw->mac.type == ixgbe_mac_82598EB)
5044 netif_set_gso_max_size(adapter->netdev, 65536);
5048 if (hw->mac.type == ixgbe_mac_82598EB)
5049 netif_set_gso_max_size(adapter->netdev, 32768);
5052 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5053 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5056 /* reconfigure the hardware */
5057 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5058 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5060 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5062 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5063 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5064 ixgbe_dcb_hw_ets(&adapter->hw,
5065 adapter->ixgbe_ieee_ets,
5067 ixgbe_dcb_hw_pfc_config(&adapter->hw,
5068 adapter->ixgbe_ieee_pfc->pfc_en,
5069 adapter->ixgbe_ieee_ets->prio_tc);
5072 /* Enable RSS Hash per TC */
5073 if (hw->mac.type != ixgbe_mac_82598EB) {
5075 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5082 /* write msb to all 8 TCs in one write */
5083 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5088 /* Additional bittime to account for IXGBE framing */
5089 #define IXGBE_ETH_FRAMING 20
5092 * ixgbe_hpbthresh - calculate high water mark for flow control
5094 * @adapter: board private structure to calculate for
5095 * @pb: packet buffer to calculate
5097 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5099 struct ixgbe_hw *hw = &adapter->hw;
5100 struct net_device *dev = adapter->netdev;
5101 int link, tc, kb, marker;
5104 /* Calculate max LAN frame size */
5105 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5108 /* FCoE traffic class uses FCOE jumbo frames */
5109 if ((dev->features & NETIF_F_FCOE_MTU) &&
5110 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5111 (pb == ixgbe_fcoe_get_tc(adapter)))
5112 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5115 /* Calculate delay value for device */
5116 switch (hw->mac.type) {
5117 case ixgbe_mac_X540:
5118 case ixgbe_mac_X550:
5119 case ixgbe_mac_X550EM_x:
5120 case ixgbe_mac_x550em_a:
5121 dv_id = IXGBE_DV_X540(link, tc);
5124 dv_id = IXGBE_DV(link, tc);
5128 /* Loopback switch introduces additional latency */
5129 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5130 dv_id += IXGBE_B2BT(tc);
5132 /* Delay value is calculated in bit times convert to KB */
5133 kb = IXGBE_BT2KB(dv_id);
5134 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5136 marker = rx_pba - kb;
5138 /* It is possible that the packet buffer is not large enough
5139 * to provide required headroom. In this case throw an error
5140 * to user and a do the best we can.
5143 e_warn(drv, "Packet Buffer(%i) can not provide enough"
5144 "headroom to support flow control."
5145 "Decrease MTU or number of traffic classes\n", pb);
5153 * ixgbe_lpbthresh - calculate low water mark for for flow control
5155 * @adapter: board private structure to calculate for
5156 * @pb: packet buffer to calculate
5158 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5160 struct ixgbe_hw *hw = &adapter->hw;
5161 struct net_device *dev = adapter->netdev;
5165 /* Calculate max LAN frame size */
5166 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5169 /* FCoE traffic class uses FCOE jumbo frames */
5170 if ((dev->features & NETIF_F_FCOE_MTU) &&
5171 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5172 (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5173 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5176 /* Calculate delay value for device */
5177 switch (hw->mac.type) {
5178 case ixgbe_mac_X540:
5179 case ixgbe_mac_X550:
5180 case ixgbe_mac_X550EM_x:
5181 case ixgbe_mac_x550em_a:
5182 dv_id = IXGBE_LOW_DV_X540(tc);
5185 dv_id = IXGBE_LOW_DV(tc);
5189 /* Delay value is calculated in bit times convert to KB */
5190 return IXGBE_BT2KB(dv_id);
5194 * ixgbe_pbthresh_setup - calculate and setup high low water marks
5196 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5198 struct ixgbe_hw *hw = &adapter->hw;
5199 int num_tc = adapter->hw_tcs;
5205 for (i = 0; i < num_tc; i++) {
5206 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5207 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5209 /* Low water marks must not be larger than high water marks */
5210 if (hw->fc.low_water[i] > hw->fc.high_water[i])
5211 hw->fc.low_water[i] = 0;
5214 for (; i < MAX_TRAFFIC_CLASS; i++)
5215 hw->fc.high_water[i] = 0;
5218 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5220 struct ixgbe_hw *hw = &adapter->hw;
5222 u8 tc = adapter->hw_tcs;
5224 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5225 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5226 hdrm = 32 << adapter->fdir_pballoc;
5230 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5231 ixgbe_pbthresh_setup(adapter);
5234 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5236 struct ixgbe_hw *hw = &adapter->hw;
5237 struct hlist_node *node2;
5238 struct ixgbe_fdir_filter *filter;
5241 spin_lock(&adapter->fdir_perfect_lock);
5243 if (!hlist_empty(&adapter->fdir_filter_list))
5244 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5246 hlist_for_each_entry_safe(filter, node2,
5247 &adapter->fdir_filter_list, fdir_node) {
5248 if (filter->action == IXGBE_FDIR_DROP_QUEUE) {
5249 queue = IXGBE_FDIR_DROP_QUEUE;
5251 u32 ring = ethtool_get_flow_spec_ring(filter->action);
5252 u8 vf = ethtool_get_flow_spec_ring_vf(filter->action);
5254 if (!vf && (ring >= adapter->num_rx_queues)) {
5255 e_err(drv, "FDIR restore failed without VF, ring: %u\n",
5259 ((vf > adapter->num_vfs) ||
5260 ring >= adapter->num_rx_queues_per_pool)) {
5261 e_err(drv, "FDIR restore failed with VF, vf: %hhu, ring: %u\n",
5266 /* Map the ring onto the absolute queue index */
5268 queue = adapter->rx_ring[ring]->reg_idx;
5271 adapter->num_rx_queues_per_pool) + ring;
5274 ixgbe_fdir_write_perfect_filter_82599(hw,
5275 &filter->filter, filter->sw_idx, queue);
5278 spin_unlock(&adapter->fdir_perfect_lock);
5282 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5283 * @rx_ring: ring to free buffers from
5285 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5287 u16 i = rx_ring->next_to_clean;
5288 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5290 if (rx_ring->xsk_pool) {
5291 ixgbe_xsk_clean_rx_ring(rx_ring);
5295 /* Free all the Rx ring sk_buffs */
5296 while (i != rx_ring->next_to_alloc) {
5297 if (rx_buffer->skb) {
5298 struct sk_buff *skb = rx_buffer->skb;
5299 if (IXGBE_CB(skb)->page_released)
5300 dma_unmap_page_attrs(rx_ring->dev,
5302 ixgbe_rx_pg_size(rx_ring),
5308 /* Invalidate cache lines that may have been written to by
5309 * device so that we avoid corrupting memory.
5311 dma_sync_single_range_for_cpu(rx_ring->dev,
5313 rx_buffer->page_offset,
5314 ixgbe_rx_bufsz(rx_ring),
5317 /* free resources associated with mapping */
5318 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5319 ixgbe_rx_pg_size(rx_ring),
5322 __page_frag_cache_drain(rx_buffer->page,
5323 rx_buffer->pagecnt_bias);
5327 if (i == rx_ring->count) {
5329 rx_buffer = rx_ring->rx_buffer_info;
5334 rx_ring->next_to_alloc = 0;
5335 rx_ring->next_to_clean = 0;
5336 rx_ring->next_to_use = 0;
5339 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5340 struct ixgbe_fwd_adapter *accel)
5342 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5343 int num_tc = netdev_get_num_tc(adapter->netdev);
5344 struct net_device *vdev = accel->netdev;
5347 baseq = accel->pool * adapter->num_rx_queues_per_pool;
5348 netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5349 accel->pool, adapter->num_rx_pools,
5350 baseq, baseq + adapter->num_rx_queues_per_pool);
5352 accel->rx_base_queue = baseq;
5353 accel->tx_base_queue = baseq;
5355 /* record configuration for macvlan interface in vdev */
5356 for (i = 0; i < num_tc; i++)
5357 netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5358 i, rss_i, baseq + (rss_i * i));
5360 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5361 adapter->rx_ring[baseq + i]->netdev = vdev;
5363 /* Guarantee all rings are updated before we update the
5364 * MAC address filter.
5368 /* ixgbe_add_mac_filter will return an index if it succeeds, so we
5369 * need to only treat it as an error value if it is negative.
5371 err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5372 VMDQ_P(accel->pool));
5376 /* if we cannot add the MAC rule then disable the offload */
5377 macvlan_release_l2fw_offload(vdev);
5379 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5380 adapter->rx_ring[baseq + i]->netdev = NULL;
5382 netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5384 /* unbind the queues and drop the subordinate channel config */
5385 netdev_unbind_sb_channel(adapter->netdev, vdev);
5386 netdev_set_sb_channel(vdev, 0);
5388 clear_bit(accel->pool, adapter->fwd_bitmask);
5394 static int ixgbe_macvlan_up(struct net_device *vdev, void *data)
5396 struct ixgbe_adapter *adapter = data;
5397 struct ixgbe_fwd_adapter *accel;
5399 if (!netif_is_macvlan(vdev))
5402 accel = macvlan_accel_priv(vdev);
5406 ixgbe_fwd_ring_up(adapter, accel);
5411 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5413 netdev_walk_all_upper_dev_rcu(adapter->netdev,
5414 ixgbe_macvlan_up, adapter);
5417 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5419 struct ixgbe_hw *hw = &adapter->hw;
5421 ixgbe_configure_pb(adapter);
5422 #ifdef CONFIG_IXGBE_DCB
5423 ixgbe_configure_dcb(adapter);
5426 * We must restore virtualization before VLANs or else
5427 * the VLVF registers will not be populated
5429 ixgbe_configure_virtualization(adapter);
5431 ixgbe_set_rx_mode(adapter->netdev);
5432 ixgbe_restore_vlan(adapter);
5433 ixgbe_ipsec_restore(adapter);
5435 switch (hw->mac.type) {
5436 case ixgbe_mac_82599EB:
5437 case ixgbe_mac_X540:
5438 hw->mac.ops.disable_rx_buff(hw);
5444 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5445 ixgbe_init_fdir_signature_82599(&adapter->hw,
5446 adapter->fdir_pballoc);
5447 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5448 ixgbe_init_fdir_perfect_82599(&adapter->hw,
5449 adapter->fdir_pballoc);
5450 ixgbe_fdir_filter_restore(adapter);
5453 switch (hw->mac.type) {
5454 case ixgbe_mac_82599EB:
5455 case ixgbe_mac_X540:
5456 hw->mac.ops.enable_rx_buff(hw);
5462 #ifdef CONFIG_IXGBE_DCA
5464 if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5465 ixgbe_setup_dca(adapter);
5466 #endif /* CONFIG_IXGBE_DCA */
5469 /* configure FCoE L2 filters, redirection table, and Rx control */
5470 ixgbe_configure_fcoe(adapter);
5472 #endif /* IXGBE_FCOE */
5473 ixgbe_configure_tx(adapter);
5474 ixgbe_configure_rx(adapter);
5475 ixgbe_configure_dfwd(adapter);
5479 * ixgbe_sfp_link_config - set up SFP+ link
5480 * @adapter: pointer to private adapter struct
5482 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5485 * We are assuming the worst case scenario here, and that
5486 * is that an SFP was inserted/removed after the reset
5487 * but before SFP detection was enabled. As such the best
5488 * solution is to just start searching as soon as we start
5490 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5491 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5493 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5494 adapter->sfp_poll_time = 0;
5498 * ixgbe_non_sfp_link_config - set up non-SFP+ link
5499 * @hw: pointer to private hardware struct
5501 * Returns 0 on success, negative on failure
5503 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5506 bool autoneg, link_up = false;
5507 int ret = IXGBE_ERR_LINK_SETUP;
5509 if (hw->mac.ops.check_link)
5510 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5515 speed = hw->phy.autoneg_advertised;
5516 if (!speed && hw->mac.ops.get_link_capabilities) {
5517 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5519 speed &= ~(IXGBE_LINK_SPEED_5GB_FULL |
5520 IXGBE_LINK_SPEED_2_5GB_FULL);
5526 if (hw->mac.ops.setup_link)
5527 ret = hw->mac.ops.setup_link(hw, speed, link_up);
5532 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5534 struct ixgbe_hw *hw = &adapter->hw;
5537 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5538 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5540 gpie |= IXGBE_GPIE_EIAME;
5542 * use EIAM to auto-mask when MSI-X interrupt is asserted
5543 * this saves a register write for every interrupt
5545 switch (hw->mac.type) {
5546 case ixgbe_mac_82598EB:
5547 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5549 case ixgbe_mac_82599EB:
5550 case ixgbe_mac_X540:
5551 case ixgbe_mac_X550:
5552 case ixgbe_mac_X550EM_x:
5553 case ixgbe_mac_x550em_a:
5555 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5556 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5560 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
5561 * specifically only auto mask tx and rx interrupts */
5562 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5565 /* XXX: to interrupt immediately for EICS writes, enable this */
5566 /* gpie |= IXGBE_GPIE_EIMEN; */
5568 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5569 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5571 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5572 case IXGBE_82599_VMDQ_8Q_MASK:
5573 gpie |= IXGBE_GPIE_VTMODE_16;
5575 case IXGBE_82599_VMDQ_4Q_MASK:
5576 gpie |= IXGBE_GPIE_VTMODE_32;
5579 gpie |= IXGBE_GPIE_VTMODE_64;
5584 /* Enable Thermal over heat sensor interrupt */
5585 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5586 switch (adapter->hw.mac.type) {
5587 case ixgbe_mac_82599EB:
5588 gpie |= IXGBE_SDP0_GPIEN_8259X;
5595 /* Enable fan failure interrupt */
5596 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5597 gpie |= IXGBE_SDP1_GPIEN(hw);
5599 switch (hw->mac.type) {
5600 case ixgbe_mac_82599EB:
5601 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5603 case ixgbe_mac_X550EM_x:
5604 case ixgbe_mac_x550em_a:
5605 gpie |= IXGBE_SDP0_GPIEN_X540;
5611 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5614 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5616 struct ixgbe_hw *hw = &adapter->hw;
5620 ixgbe_get_hw_control(adapter);
5621 ixgbe_setup_gpie(adapter);
5623 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5624 ixgbe_configure_msix(adapter);
5626 ixgbe_configure_msi_and_legacy(adapter);
5628 /* enable the optics for 82599 SFP+ fiber */
5629 if (hw->mac.ops.enable_tx_laser)
5630 hw->mac.ops.enable_tx_laser(hw);
5632 if (hw->phy.ops.set_phy_power)
5633 hw->phy.ops.set_phy_power(hw, true);
5635 smp_mb__before_atomic();
5636 clear_bit(__IXGBE_DOWN, &adapter->state);
5637 ixgbe_napi_enable_all(adapter);
5639 if (ixgbe_is_sfp(hw)) {
5640 ixgbe_sfp_link_config(adapter);
5642 err = ixgbe_non_sfp_link_config(hw);
5644 e_err(probe, "link_config FAILED %d\n", err);
5647 /* clear any pending interrupts, may auto mask */
5648 IXGBE_READ_REG(hw, IXGBE_EICR);
5649 ixgbe_irq_enable(adapter, true, true);
5652 * If this adapter has a fan, check to see if we had a failure
5653 * before we enabled the interrupt.
5655 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5656 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5657 if (esdp & IXGBE_ESDP_SDP1)
5658 e_crit(drv, "Fan has stopped, replace the adapter\n");
5661 /* bring the link up in the watchdog, this could race with our first
5662 * link up interrupt but shouldn't be a problem */
5663 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5664 adapter->link_check_timeout = jiffies;
5665 mod_timer(&adapter->service_timer, jiffies);
5667 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5668 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5669 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5670 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5673 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5675 WARN_ON(in_interrupt());
5676 /* put off any impending NetWatchDogTimeout */
5677 netif_trans_update(adapter->netdev);
5679 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5680 usleep_range(1000, 2000);
5681 if (adapter->hw.phy.type == ixgbe_phy_fw)
5682 ixgbe_watchdog_link_is_down(adapter);
5683 ixgbe_down(adapter);
5685 * If SR-IOV enabled then wait a bit before bringing the adapter
5686 * back up to give the VFs time to respond to the reset. The
5687 * two second wait is based upon the watchdog timer cycle in
5690 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5693 clear_bit(__IXGBE_RESETTING, &adapter->state);
5696 void ixgbe_up(struct ixgbe_adapter *adapter)
5698 /* hardware has been reset, we need to reload some things */
5699 ixgbe_configure(adapter);
5701 ixgbe_up_complete(adapter);
5704 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5708 pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5710 switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5711 case IXGBE_PCIDEVCTRL2_17_34s:
5712 case IXGBE_PCIDEVCTRL2_4_8s:
5713 /* For now we cap the upper limit on delay to 2 seconds
5714 * as we end up going up to 34 seconds of delay in worst
5715 * case timeout value.
5717 case IXGBE_PCIDEVCTRL2_1_2s:
5718 return 2000000ul; /* 2.0 s */
5719 case IXGBE_PCIDEVCTRL2_260_520ms:
5720 return 520000ul; /* 520 ms */
5721 case IXGBE_PCIDEVCTRL2_65_130ms:
5722 return 130000ul; /* 130 ms */
5723 case IXGBE_PCIDEVCTRL2_16_32ms:
5724 return 32000ul; /* 32 ms */
5725 case IXGBE_PCIDEVCTRL2_1_2ms:
5726 return 2000ul; /* 2 ms */
5727 case IXGBE_PCIDEVCTRL2_50_100us:
5728 return 100ul; /* 100 us */
5729 case IXGBE_PCIDEVCTRL2_16_32ms_def:
5730 return 32000ul; /* 32 ms */
5735 /* We shouldn't need to hit this path, but just in case default as
5736 * though completion timeout is not supported and support 32ms.
5741 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5743 unsigned long wait_delay, delay_interval;
5744 struct ixgbe_hw *hw = &adapter->hw;
5748 /* disable receives */
5749 hw->mac.ops.disable_rx(hw);
5751 if (ixgbe_removed(hw->hw_addr))
5754 /* disable all enabled Rx queues */
5755 for (i = 0; i < adapter->num_rx_queues; i++) {
5756 struct ixgbe_ring *ring = adapter->rx_ring[i];
5757 u8 reg_idx = ring->reg_idx;
5759 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5760 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5761 rxdctl |= IXGBE_RXDCTL_SWFLSH;
5763 /* write value back with RXDCTL.ENABLE bit cleared */
5764 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5767 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5768 if (hw->mac.type == ixgbe_mac_82598EB &&
5769 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5772 /* Determine our minimum delay interval. We will increase this value
5773 * with each subsequent test. This way if the device returns quickly
5774 * we should spend as little time as possible waiting, however as
5775 * the time increases we will wait for larger periods of time.
5777 * The trick here is that we increase the interval using the
5778 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5779 * of that wait is that it totals up to 100x whatever interval we
5780 * choose. Since our minimum wait is 100us we can just divide the
5781 * total timeout by 100 to get our minimum delay interval.
5783 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5785 wait_loop = IXGBE_MAX_RX_DESC_POLL;
5786 wait_delay = delay_interval;
5788 while (wait_loop--) {
5789 usleep_range(wait_delay, wait_delay + 10);
5790 wait_delay += delay_interval * 2;
5793 /* OR together the reading of all the active RXDCTL registers,
5794 * and then test the result. We need the disable to complete
5795 * before we start freeing the memory and invalidating the
5798 for (i = 0; i < adapter->num_rx_queues; i++) {
5799 struct ixgbe_ring *ring = adapter->rx_ring[i];
5800 u8 reg_idx = ring->reg_idx;
5802 rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5805 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5810 "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5813 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5815 unsigned long wait_delay, delay_interval;
5816 struct ixgbe_hw *hw = &adapter->hw;
5820 if (ixgbe_removed(hw->hw_addr))
5823 /* disable all enabled Tx queues */
5824 for (i = 0; i < adapter->num_tx_queues; i++) {
5825 struct ixgbe_ring *ring = adapter->tx_ring[i];
5826 u8 reg_idx = ring->reg_idx;
5828 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5831 /* disable all enabled XDP Tx queues */
5832 for (i = 0; i < adapter->num_xdp_queues; i++) {
5833 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5834 u8 reg_idx = ring->reg_idx;
5836 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5839 /* If the link is not up there shouldn't be much in the way of
5840 * pending transactions. Those that are left will be flushed out
5841 * when the reset logic goes through the flush sequence to clean out
5842 * the pending Tx transactions.
5844 if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5845 goto dma_engine_disable;
5847 /* Determine our minimum delay interval. We will increase this value
5848 * with each subsequent test. This way if the device returns quickly
5849 * we should spend as little time as possible waiting, however as
5850 * the time increases we will wait for larger periods of time.
5852 * The trick here is that we increase the interval using the
5853 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5854 * of that wait is that it totals up to 100x whatever interval we
5855 * choose. Since our minimum wait is 100us we can just divide the
5856 * total timeout by 100 to get our minimum delay interval.
5858 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5860 wait_loop = IXGBE_MAX_RX_DESC_POLL;
5861 wait_delay = delay_interval;
5863 while (wait_loop--) {
5864 usleep_range(wait_delay, wait_delay + 10);
5865 wait_delay += delay_interval * 2;
5868 /* OR together the reading of all the active TXDCTL registers,
5869 * and then test the result. We need the disable to complete
5870 * before we start freeing the memory and invalidating the
5873 for (i = 0; i < adapter->num_tx_queues; i++) {
5874 struct ixgbe_ring *ring = adapter->tx_ring[i];
5875 u8 reg_idx = ring->reg_idx;
5877 txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5879 for (i = 0; i < adapter->num_xdp_queues; i++) {
5880 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5881 u8 reg_idx = ring->reg_idx;
5883 txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5886 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5887 goto dma_engine_disable;
5891 "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5894 /* Disable the Tx DMA engine on 82599 and later MAC */
5895 switch (hw->mac.type) {
5896 case ixgbe_mac_82599EB:
5897 case ixgbe_mac_X540:
5898 case ixgbe_mac_X550:
5899 case ixgbe_mac_X550EM_x:
5900 case ixgbe_mac_x550em_a:
5901 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5902 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5903 ~IXGBE_DMATXCTL_TE));
5910 void ixgbe_reset(struct ixgbe_adapter *adapter)
5912 struct ixgbe_hw *hw = &adapter->hw;
5913 struct net_device *netdev = adapter->netdev;
5916 if (ixgbe_removed(hw->hw_addr))
5918 /* lock SFP init bit to prevent race conditions with the watchdog */
5919 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5920 usleep_range(1000, 2000);
5922 /* clear all SFP and link config related flags while holding SFP_INIT */
5923 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5924 IXGBE_FLAG2_SFP_NEEDS_RESET);
5925 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5927 err = hw->mac.ops.init_hw(hw);
5930 case IXGBE_ERR_SFP_NOT_PRESENT:
5931 case IXGBE_ERR_SFP_NOT_SUPPORTED:
5933 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5934 e_dev_err("master disable timed out\n");
5936 case IXGBE_ERR_EEPROM_VERSION:
5937 /* We are running on a pre-production device, log a warning */
5938 e_dev_warn("This device is a pre-production adapter/LOM. "
5939 "Please be aware there may be issues associated with "
5940 "your hardware. If you are experiencing problems "
5941 "please contact your Intel or hardware "
5942 "representative who provided you with this "
5946 e_dev_err("Hardware Error: %d\n", err);
5949 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5951 /* flush entries out of MAC table */
5952 ixgbe_flush_sw_mac_table(adapter);
5953 __dev_uc_unsync(netdev, NULL);
5955 /* do not flush user set addresses */
5956 ixgbe_mac_set_default_filter(adapter);
5958 /* update SAN MAC vmdq pool selection */
5959 if (hw->mac.san_mac_rar_index)
5960 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5962 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5963 ixgbe_ptp_reset(adapter);
5965 if (hw->phy.ops.set_phy_power) {
5966 if (!netif_running(adapter->netdev) && !adapter->wol)
5967 hw->phy.ops.set_phy_power(hw, false);
5969 hw->phy.ops.set_phy_power(hw, true);
5974 * ixgbe_clean_tx_ring - Free Tx Buffers
5975 * @tx_ring: ring to be cleaned
5977 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5979 u16 i = tx_ring->next_to_clean;
5980 struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
5982 if (tx_ring->xsk_pool) {
5983 ixgbe_xsk_clean_tx_ring(tx_ring);
5987 while (i != tx_ring->next_to_use) {
5988 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
5990 /* Free all the Tx ring sk_buffs */
5991 if (ring_is_xdp(tx_ring))
5992 xdp_return_frame(tx_buffer->xdpf);
5994 dev_kfree_skb_any(tx_buffer->skb);
5996 /* unmap skb header data */
5997 dma_unmap_single(tx_ring->dev,
5998 dma_unmap_addr(tx_buffer, dma),
5999 dma_unmap_len(tx_buffer, len),
6002 /* check for eop_desc to determine the end of the packet */
6003 eop_desc = tx_buffer->next_to_watch;
6004 tx_desc = IXGBE_TX_DESC(tx_ring, i);
6006 /* unmap remaining buffers */
6007 while (tx_desc != eop_desc) {
6011 if (unlikely(i == tx_ring->count)) {
6013 tx_buffer = tx_ring->tx_buffer_info;
6014 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6017 /* unmap any remaining paged data */
6018 if (dma_unmap_len(tx_buffer, len))
6019 dma_unmap_page(tx_ring->dev,
6020 dma_unmap_addr(tx_buffer, dma),
6021 dma_unmap_len(tx_buffer, len),
6025 /* move us one more past the eop_desc for start of next pkt */
6028 if (unlikely(i == tx_ring->count)) {
6030 tx_buffer = tx_ring->tx_buffer_info;
6034 /* reset BQL for queue */
6035 if (!ring_is_xdp(tx_ring))
6036 netdev_tx_reset_queue(txring_txq(tx_ring));
6039 /* reset next_to_use and next_to_clean */
6040 tx_ring->next_to_use = 0;
6041 tx_ring->next_to_clean = 0;
6045 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6046 * @adapter: board private structure
6048 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6052 for (i = 0; i < adapter->num_rx_queues; i++)
6053 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6057 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6058 * @adapter: board private structure
6060 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6064 for (i = 0; i < adapter->num_tx_queues; i++)
6065 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6066 for (i = 0; i < adapter->num_xdp_queues; i++)
6067 ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6070 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6072 struct hlist_node *node2;
6073 struct ixgbe_fdir_filter *filter;
6075 spin_lock(&adapter->fdir_perfect_lock);
6077 hlist_for_each_entry_safe(filter, node2,
6078 &adapter->fdir_filter_list, fdir_node) {
6079 hlist_del(&filter->fdir_node);
6082 adapter->fdir_filter_count = 0;
6084 spin_unlock(&adapter->fdir_perfect_lock);
6087 void ixgbe_down(struct ixgbe_adapter *adapter)
6089 struct net_device *netdev = adapter->netdev;
6090 struct ixgbe_hw *hw = &adapter->hw;
6093 /* signal that we are down to the interrupt handler */
6094 if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6095 return; /* do nothing if already down */
6097 /* Shut off incoming Tx traffic */
6098 netif_tx_stop_all_queues(netdev);
6100 /* call carrier off first to avoid false dev_watchdog timeouts */
6101 netif_carrier_off(netdev);
6102 netif_tx_disable(netdev);
6105 ixgbe_disable_rx(adapter);
6107 /* synchronize_rcu() needed for pending XDP buffers to drain */
6108 if (adapter->xdp_ring[0])
6111 ixgbe_irq_disable(adapter);
6113 ixgbe_napi_disable_all(adapter);
6115 clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6116 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6117 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6119 del_timer_sync(&adapter->service_timer);
6121 if (adapter->num_vfs) {
6122 /* Clear EITR Select mapping */
6123 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6125 /* Mark all the VFs as inactive */
6126 for (i = 0 ; i < adapter->num_vfs; i++)
6127 adapter->vfinfo[i].clear_to_send = false;
6129 /* ping all the active vfs to let them know we are going down */
6130 ixgbe_ping_all_vfs(adapter);
6132 /* Disable all VFTE/VFRE TX/RX */
6133 ixgbe_disable_tx_rx(adapter);
6136 /* disable transmits in the hardware now that interrupts are off */
6137 ixgbe_disable_tx(adapter);
6139 if (!pci_channel_offline(adapter->pdev))
6140 ixgbe_reset(adapter);
6142 /* power down the optics for 82599 SFP+ fiber */
6143 if (hw->mac.ops.disable_tx_laser)
6144 hw->mac.ops.disable_tx_laser(hw);
6146 ixgbe_clean_all_tx_rings(adapter);
6147 ixgbe_clean_all_rx_rings(adapter);
6151 * ixgbe_eee_capable - helper function to determine EEE support on X550
6152 * @adapter: board private structure
6154 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6156 struct ixgbe_hw *hw = &adapter->hw;
6158 switch (hw->device_id) {
6159 case IXGBE_DEV_ID_X550EM_A_1G_T:
6160 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6161 if (!hw->phy.eee_speeds_supported)
6163 adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6164 if (!hw->phy.eee_speeds_advertised)
6166 adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6169 adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6170 adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6176 * ixgbe_tx_timeout - Respond to a Tx Hang
6177 * @netdev: network interface device structure
6179 static void ixgbe_tx_timeout(struct net_device *netdev, unsigned int txqueue)
6181 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6183 /* Do the reset outside of interrupt context */
6184 ixgbe_tx_timeout_reset(adapter);
6187 #ifdef CONFIG_IXGBE_DCB
6188 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6190 struct ixgbe_hw *hw = &adapter->hw;
6191 struct tc_configuration *tc;
6194 switch (hw->mac.type) {
6195 case ixgbe_mac_82598EB:
6196 case ixgbe_mac_82599EB:
6197 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6198 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6200 case ixgbe_mac_X540:
6201 case ixgbe_mac_X550:
6202 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6203 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6205 case ixgbe_mac_X550EM_x:
6206 case ixgbe_mac_x550em_a:
6208 adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6209 adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6213 /* Configure DCB traffic classes */
6214 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6215 tc = &adapter->dcb_cfg.tc_config[j];
6216 tc->path[DCB_TX_CONFIG].bwg_id = 0;
6217 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6218 tc->path[DCB_RX_CONFIG].bwg_id = 0;
6219 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6220 tc->dcb_pfc = pfc_disabled;
6223 /* Initialize default user to priority mapping, UPx->TC0 */
6224 tc = &adapter->dcb_cfg.tc_config[0];
6225 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6226 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6228 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6229 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6230 adapter->dcb_cfg.pfc_mode_enable = false;
6231 adapter->dcb_set_bitmap = 0x00;
6232 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6233 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6234 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6235 sizeof(adapter->temp_dcb_cfg));
6240 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6241 * @adapter: board private structure to initialize
6242 * @ii: pointer to ixgbe_info for device
6244 * ixgbe_sw_init initializes the Adapter private data structure.
6245 * Fields are initialized based on PCI device information and
6246 * OS network device settings (MTU size).
6248 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6249 const struct ixgbe_info *ii)
6251 struct ixgbe_hw *hw = &adapter->hw;
6252 struct pci_dev *pdev = adapter->pdev;
6253 unsigned int rss, fdir;
6257 /* PCI config space info */
6259 hw->vendor_id = pdev->vendor;
6260 hw->device_id = pdev->device;
6261 hw->revision_id = pdev->revision;
6262 hw->subsystem_vendor_id = pdev->subsystem_vendor;
6263 hw->subsystem_device_id = pdev->subsystem_device;
6265 /* get_invariants needs the device IDs */
6266 ii->get_invariants(hw);
6268 /* Set common capability flags and settings */
6269 rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6270 adapter->ring_feature[RING_F_RSS].limit = rss;
6271 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6272 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6273 adapter->atr_sample_rate = 20;
6274 fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6275 adapter->ring_feature[RING_F_FDIR].limit = fdir;
6276 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6277 adapter->ring_feature[RING_F_VMDQ].limit = 1;
6278 #ifdef CONFIG_IXGBE_DCA
6279 adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6281 #ifdef CONFIG_IXGBE_DCB
6282 adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6283 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6286 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6287 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6288 #ifdef CONFIG_IXGBE_DCB
6289 /* Default traffic class to use for FCoE */
6290 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6291 #endif /* CONFIG_IXGBE_DCB */
6292 #endif /* IXGBE_FCOE */
6294 /* initialize static ixgbe jump table entries */
6295 adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6297 if (!adapter->jump_tables[0])
6299 adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6301 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6302 adapter->jump_tables[i] = NULL;
6304 adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6305 sizeof(struct ixgbe_mac_addr),
6307 if (!adapter->mac_table)
6310 if (ixgbe_init_rss_key(adapter))
6313 adapter->af_xdp_zc_qps = bitmap_zalloc(MAX_XDP_QUEUES, GFP_KERNEL);
6314 if (!adapter->af_xdp_zc_qps)
6317 /* Set MAC specific capability flags and exceptions */
6318 switch (hw->mac.type) {
6319 case ixgbe_mac_82598EB:
6320 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6322 if (hw->device_id == IXGBE_DEV_ID_82598AT)
6323 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6325 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6326 adapter->ring_feature[RING_F_FDIR].limit = 0;
6327 adapter->atr_sample_rate = 0;
6328 adapter->fdir_pballoc = 0;
6330 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6331 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6332 #ifdef CONFIG_IXGBE_DCB
6333 adapter->fcoe.up = 0;
6334 #endif /* IXGBE_DCB */
6335 #endif /* IXGBE_FCOE */
6337 case ixgbe_mac_82599EB:
6338 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6339 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6341 case ixgbe_mac_X540:
6342 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6343 if (fwsm & IXGBE_FWSM_TS_ENABLED)
6344 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6346 case ixgbe_mac_x550em_a:
6347 switch (hw->device_id) {
6348 case IXGBE_DEV_ID_X550EM_A_1G_T:
6349 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6350 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6356 case ixgbe_mac_X550EM_x:
6357 #ifdef CONFIG_IXGBE_DCB
6358 adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6361 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6362 #ifdef CONFIG_IXGBE_DCB
6363 adapter->fcoe.up = 0;
6364 #endif /* IXGBE_DCB */
6365 #endif /* IXGBE_FCOE */
6367 case ixgbe_mac_X550:
6368 if (hw->mac.type == ixgbe_mac_X550)
6369 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6370 #ifdef CONFIG_IXGBE_DCA
6371 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6379 /* FCoE support exists, always init the FCoE lock */
6380 spin_lock_init(&adapter->fcoe.lock);
6383 /* n-tuple support exists, always init our spinlock */
6384 spin_lock_init(&adapter->fdir_perfect_lock);
6386 #ifdef CONFIG_IXGBE_DCB
6387 ixgbe_init_dcb(adapter);
6389 ixgbe_init_ipsec_offload(adapter);
6391 /* default flow control settings */
6392 hw->fc.requested_mode = ixgbe_fc_full;
6393 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
6394 ixgbe_pbthresh_setup(adapter);
6395 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6396 hw->fc.send_xon = true;
6397 hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6399 #ifdef CONFIG_PCI_IOV
6401 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6403 /* assign number of SR-IOV VFs */
6404 if (hw->mac.type != ixgbe_mac_82598EB) {
6405 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6407 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6410 #endif /* CONFIG_PCI_IOV */
6412 /* enable itr by default in dynamic mode */
6413 adapter->rx_itr_setting = 1;
6414 adapter->tx_itr_setting = 1;
6416 /* set default ring sizes */
6417 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6418 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6420 /* set default work limits */
6421 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6423 /* initialize eeprom parameters */
6424 if (ixgbe_init_eeprom_params_generic(hw)) {
6425 e_dev_err("EEPROM initialization failed\n");
6429 /* PF holds first pool slot */
6430 set_bit(0, adapter->fwd_bitmask);
6431 set_bit(__IXGBE_DOWN, &adapter->state);
6437 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6438 * @tx_ring: tx descriptor ring (for a specific queue) to setup
6440 * Return 0 on success, negative on failure
6442 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6444 struct device *dev = tx_ring->dev;
6445 int orig_node = dev_to_node(dev);
6446 int ring_node = NUMA_NO_NODE;
6449 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6451 if (tx_ring->q_vector)
6452 ring_node = tx_ring->q_vector->numa_node;
6454 tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6455 if (!tx_ring->tx_buffer_info)
6456 tx_ring->tx_buffer_info = vmalloc(size);
6457 if (!tx_ring->tx_buffer_info)
6460 /* round up to nearest 4K */
6461 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6462 tx_ring->size = ALIGN(tx_ring->size, 4096);
6464 set_dev_node(dev, ring_node);
6465 tx_ring->desc = dma_alloc_coherent(dev,
6469 set_dev_node(dev, orig_node);
6471 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6472 &tx_ring->dma, GFP_KERNEL);
6476 tx_ring->next_to_use = 0;
6477 tx_ring->next_to_clean = 0;
6481 vfree(tx_ring->tx_buffer_info);
6482 tx_ring->tx_buffer_info = NULL;
6483 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6488 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6489 * @adapter: board private structure
6491 * If this function returns with an error, then it's possible one or
6492 * more of the rings is populated (while the rest are not). It is the
6493 * callers duty to clean those orphaned rings.
6495 * Return 0 on success, negative on failure
6497 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6499 int i, j = 0, err = 0;
6501 for (i = 0; i < adapter->num_tx_queues; i++) {
6502 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6506 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6509 for (j = 0; j < adapter->num_xdp_queues; j++) {
6510 err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6514 e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6520 /* rewind the index freeing the rings as we go */
6522 ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6524 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6529 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6530 * @adapter: pointer to ixgbe_adapter
6531 * @rx_ring: rx descriptor ring (for a specific queue) to setup
6533 * Returns 0 on success, negative on failure
6535 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6536 struct ixgbe_ring *rx_ring)
6538 struct device *dev = rx_ring->dev;
6539 int orig_node = dev_to_node(dev);
6540 int ring_node = NUMA_NO_NODE;
6543 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6545 if (rx_ring->q_vector)
6546 ring_node = rx_ring->q_vector->numa_node;
6548 rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6549 if (!rx_ring->rx_buffer_info)
6550 rx_ring->rx_buffer_info = vmalloc(size);
6551 if (!rx_ring->rx_buffer_info)
6554 /* Round up to nearest 4K */
6555 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6556 rx_ring->size = ALIGN(rx_ring->size, 4096);
6558 set_dev_node(dev, ring_node);
6559 rx_ring->desc = dma_alloc_coherent(dev,
6563 set_dev_node(dev, orig_node);
6565 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6566 &rx_ring->dma, GFP_KERNEL);
6570 rx_ring->next_to_clean = 0;
6571 rx_ring->next_to_use = 0;
6573 /* XDP RX-queue info */
6574 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6575 rx_ring->queue_index) < 0)
6578 rx_ring->xdp_prog = adapter->xdp_prog;
6582 vfree(rx_ring->rx_buffer_info);
6583 rx_ring->rx_buffer_info = NULL;
6584 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6589 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6590 * @adapter: board private structure
6592 * If this function returns with an error, then it's possible one or
6593 * more of the rings is populated (while the rest are not). It is the
6594 * callers duty to clean those orphaned rings.
6596 * Return 0 on success, negative on failure
6598 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6602 for (i = 0; i < adapter->num_rx_queues; i++) {
6603 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6607 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6612 err = ixgbe_setup_fcoe_ddp_resources(adapter);
6617 /* rewind the index freeing the rings as we go */
6619 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6624 * ixgbe_free_tx_resources - Free Tx Resources per Queue
6625 * @tx_ring: Tx descriptor ring for a specific queue
6627 * Free all transmit software resources
6629 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6631 ixgbe_clean_tx_ring(tx_ring);
6633 vfree(tx_ring->tx_buffer_info);
6634 tx_ring->tx_buffer_info = NULL;
6636 /* if not set, then don't free */
6640 dma_free_coherent(tx_ring->dev, tx_ring->size,
6641 tx_ring->desc, tx_ring->dma);
6643 tx_ring->desc = NULL;
6647 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6648 * @adapter: board private structure
6650 * Free all transmit software resources
6652 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6656 for (i = 0; i < adapter->num_tx_queues; i++)
6657 if (adapter->tx_ring[i]->desc)
6658 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6659 for (i = 0; i < adapter->num_xdp_queues; i++)
6660 if (adapter->xdp_ring[i]->desc)
6661 ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6665 * ixgbe_free_rx_resources - Free Rx Resources
6666 * @rx_ring: ring to clean the resources from
6668 * Free all receive software resources
6670 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6672 ixgbe_clean_rx_ring(rx_ring);
6674 rx_ring->xdp_prog = NULL;
6675 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6676 vfree(rx_ring->rx_buffer_info);
6677 rx_ring->rx_buffer_info = NULL;
6679 /* if not set, then don't free */
6683 dma_free_coherent(rx_ring->dev, rx_ring->size,
6684 rx_ring->desc, rx_ring->dma);
6686 rx_ring->desc = NULL;
6690 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6691 * @adapter: board private structure
6693 * Free all receive software resources
6695 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6700 ixgbe_free_fcoe_ddp_resources(adapter);
6703 for (i = 0; i < adapter->num_rx_queues; i++)
6704 if (adapter->rx_ring[i]->desc)
6705 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6709 * ixgbe_change_mtu - Change the Maximum Transfer Unit
6710 * @netdev: network interface device structure
6711 * @new_mtu: new value for maximum frame size
6713 * Returns 0 on success, negative on failure
6715 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6717 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6719 if (adapter->xdp_prog) {
6720 int new_frame_size = new_mtu + ETH_HLEN + ETH_FCS_LEN +
6724 for (i = 0; i < adapter->num_rx_queues; i++) {
6725 struct ixgbe_ring *ring = adapter->rx_ring[i];
6727 if (new_frame_size > ixgbe_rx_bufsz(ring)) {
6728 e_warn(probe, "Requested MTU size is not supported with XDP\n");
6735 * For 82599EB we cannot allow legacy VFs to enable their receive
6736 * paths when MTU greater than 1500 is configured. So display a
6737 * warning that legacy VFs will be disabled.
6739 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6740 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6741 (new_mtu > ETH_DATA_LEN))
6742 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6744 netdev_dbg(netdev, "changing MTU from %d to %d\n",
6745 netdev->mtu, new_mtu);
6747 /* must set new MTU before calling down or up */
6748 netdev->mtu = new_mtu;
6750 if (netif_running(netdev))
6751 ixgbe_reinit_locked(adapter);
6757 * ixgbe_open - Called when a network interface is made active
6758 * @netdev: network interface device structure
6760 * Returns 0 on success, negative value on failure
6762 * The open entry point is called when a network interface is made
6763 * active by the system (IFF_UP). At this point all resources needed
6764 * for transmit and receive operations are allocated, the interrupt
6765 * handler is registered with the OS, the watchdog timer is started,
6766 * and the stack is notified that the interface is ready.
6768 int ixgbe_open(struct net_device *netdev)
6770 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6771 struct ixgbe_hw *hw = &adapter->hw;
6774 /* disallow open during test */
6775 if (test_bit(__IXGBE_TESTING, &adapter->state))
6778 netif_carrier_off(netdev);
6780 /* allocate transmit descriptors */
6781 err = ixgbe_setup_all_tx_resources(adapter);
6785 /* allocate receive descriptors */
6786 err = ixgbe_setup_all_rx_resources(adapter);
6790 ixgbe_configure(adapter);
6792 err = ixgbe_request_irq(adapter);
6796 /* Notify the stack of the actual queue counts. */
6797 queues = adapter->num_tx_queues;
6798 err = netif_set_real_num_tx_queues(netdev, queues);
6800 goto err_set_queues;
6802 queues = adapter->num_rx_queues;
6803 err = netif_set_real_num_rx_queues(netdev, queues);
6805 goto err_set_queues;
6807 ixgbe_ptp_init(adapter);
6809 ixgbe_up_complete(adapter);
6811 udp_tunnel_nic_reset_ntf(netdev);
6816 ixgbe_free_irq(adapter);
6818 ixgbe_free_all_rx_resources(adapter);
6819 if (hw->phy.ops.set_phy_power && !adapter->wol)
6820 hw->phy.ops.set_phy_power(&adapter->hw, false);
6822 ixgbe_free_all_tx_resources(adapter);
6824 ixgbe_reset(adapter);
6829 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6831 ixgbe_ptp_suspend(adapter);
6833 if (adapter->hw.phy.ops.enter_lplu) {
6834 adapter->hw.phy.reset_disable = true;
6835 ixgbe_down(adapter);
6836 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6837 adapter->hw.phy.reset_disable = false;
6839 ixgbe_down(adapter);
6842 ixgbe_free_irq(adapter);
6844 ixgbe_free_all_tx_resources(adapter);
6845 ixgbe_free_all_rx_resources(adapter);
6849 * ixgbe_close - Disables a network interface
6850 * @netdev: network interface device structure
6852 * Returns 0, this is not allowed to fail
6854 * The close entry point is called when an interface is de-activated
6855 * by the OS. The hardware is still under the drivers control, but
6856 * needs to be disabled. A global MAC reset is issued to stop the
6857 * hardware, and all transmit and receive resources are freed.
6859 int ixgbe_close(struct net_device *netdev)
6861 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6863 ixgbe_ptp_stop(adapter);
6865 if (netif_device_present(netdev))
6866 ixgbe_close_suspend(adapter);
6868 ixgbe_fdir_filter_exit(adapter);
6870 ixgbe_release_hw_control(adapter);
6875 static int __maybe_unused ixgbe_resume(struct device *dev_d)
6877 struct pci_dev *pdev = to_pci_dev(dev_d);
6878 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6879 struct net_device *netdev = adapter->netdev;
6882 adapter->hw.hw_addr = adapter->io_addr;
6884 smp_mb__before_atomic();
6885 clear_bit(__IXGBE_DISABLED, &adapter->state);
6886 pci_set_master(pdev);
6888 device_wakeup_disable(dev_d);
6890 ixgbe_reset(adapter);
6892 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6895 err = ixgbe_init_interrupt_scheme(adapter);
6896 if (!err && netif_running(netdev))
6897 err = ixgbe_open(netdev);
6901 netif_device_attach(netdev);
6907 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6909 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6910 struct net_device *netdev = adapter->netdev;
6911 struct ixgbe_hw *hw = &adapter->hw;
6913 u32 wufc = adapter->wol;
6916 netif_device_detach(netdev);
6918 if (netif_running(netdev))
6919 ixgbe_close_suspend(adapter);
6921 ixgbe_clear_interrupt_scheme(adapter);
6924 if (hw->mac.ops.stop_link_on_d3)
6925 hw->mac.ops.stop_link_on_d3(hw);
6930 ixgbe_set_rx_mode(netdev);
6932 /* enable the optics for 82599 SFP+ fiber as we can WoL */
6933 if (hw->mac.ops.enable_tx_laser)
6934 hw->mac.ops.enable_tx_laser(hw);
6936 /* enable the reception of multicast packets */
6937 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6938 fctrl |= IXGBE_FCTRL_MPE;
6939 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6941 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6942 ctrl |= IXGBE_CTRL_GIO_DIS;
6943 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6945 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6947 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6948 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6951 switch (hw->mac.type) {
6952 case ixgbe_mac_82598EB:
6953 pci_wake_from_d3(pdev, false);
6955 case ixgbe_mac_82599EB:
6956 case ixgbe_mac_X540:
6957 case ixgbe_mac_X550:
6958 case ixgbe_mac_X550EM_x:
6959 case ixgbe_mac_x550em_a:
6960 pci_wake_from_d3(pdev, !!wufc);
6966 *enable_wake = !!wufc;
6967 if (hw->phy.ops.set_phy_power && !*enable_wake)
6968 hw->phy.ops.set_phy_power(hw, false);
6970 ixgbe_release_hw_control(adapter);
6972 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
6973 pci_disable_device(pdev);
6978 static int __maybe_unused ixgbe_suspend(struct device *dev_d)
6980 struct pci_dev *pdev = to_pci_dev(dev_d);
6984 retval = __ixgbe_shutdown(pdev, &wake);
6986 device_set_wakeup_enable(dev_d, wake);
6991 static void ixgbe_shutdown(struct pci_dev *pdev)
6995 __ixgbe_shutdown(pdev, &wake);
6997 if (system_state == SYSTEM_POWER_OFF) {
6998 pci_wake_from_d3(pdev, wake);
6999 pci_set_power_state(pdev, PCI_D3hot);
7004 * ixgbe_update_stats - Update the board statistics counters.
7005 * @adapter: board private structure
7007 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7009 struct net_device *netdev = adapter->netdev;
7010 struct ixgbe_hw *hw = &adapter->hw;
7011 struct ixgbe_hw_stats *hwstats = &adapter->stats;
7013 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7014 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7015 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7016 u64 alloc_rx_page = 0;
7017 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7019 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7020 test_bit(__IXGBE_RESETTING, &adapter->state))
7023 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7026 for (i = 0; i < adapter->num_rx_queues; i++) {
7027 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7028 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7030 adapter->rsc_total_count = rsc_count;
7031 adapter->rsc_total_flush = rsc_flush;
7034 for (i = 0; i < adapter->num_rx_queues; i++) {
7035 struct ixgbe_ring *rx_ring = READ_ONCE(adapter->rx_ring[i]);
7039 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7040 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7041 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7042 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7043 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7044 bytes += rx_ring->stats.bytes;
7045 packets += rx_ring->stats.packets;
7047 adapter->non_eop_descs = non_eop_descs;
7048 adapter->alloc_rx_page = alloc_rx_page;
7049 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7050 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7051 adapter->hw_csum_rx_error = hw_csum_rx_error;
7052 netdev->stats.rx_bytes = bytes;
7053 netdev->stats.rx_packets = packets;
7057 /* gather some stats to the adapter struct that are per queue */
7058 for (i = 0; i < adapter->num_tx_queues; i++) {
7059 struct ixgbe_ring *tx_ring = READ_ONCE(adapter->tx_ring[i]);
7063 restart_queue += tx_ring->tx_stats.restart_queue;
7064 tx_busy += tx_ring->tx_stats.tx_busy;
7065 bytes += tx_ring->stats.bytes;
7066 packets += tx_ring->stats.packets;
7068 for (i = 0; i < adapter->num_xdp_queues; i++) {
7069 struct ixgbe_ring *xdp_ring = READ_ONCE(adapter->xdp_ring[i]);
7073 restart_queue += xdp_ring->tx_stats.restart_queue;
7074 tx_busy += xdp_ring->tx_stats.tx_busy;
7075 bytes += xdp_ring->stats.bytes;
7076 packets += xdp_ring->stats.packets;
7078 adapter->restart_queue = restart_queue;
7079 adapter->tx_busy = tx_busy;
7080 netdev->stats.tx_bytes = bytes;
7081 netdev->stats.tx_packets = packets;
7083 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7085 /* 8 register reads */
7086 for (i = 0; i < 8; i++) {
7087 /* for packet buffers not used, the register should read 0 */
7088 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7090 hwstats->mpc[i] += mpc;
7091 total_mpc += hwstats->mpc[i];
7092 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7093 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7094 switch (hw->mac.type) {
7095 case ixgbe_mac_82598EB:
7096 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7097 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7098 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7099 hwstats->pxonrxc[i] +=
7100 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7102 case ixgbe_mac_82599EB:
7103 case ixgbe_mac_X540:
7104 case ixgbe_mac_X550:
7105 case ixgbe_mac_X550EM_x:
7106 case ixgbe_mac_x550em_a:
7107 hwstats->pxonrxc[i] +=
7108 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7115 /*16 register reads */
7116 for (i = 0; i < 16; i++) {
7117 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7118 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7119 if ((hw->mac.type == ixgbe_mac_82599EB) ||
7120 (hw->mac.type == ixgbe_mac_X540) ||
7121 (hw->mac.type == ixgbe_mac_X550) ||
7122 (hw->mac.type == ixgbe_mac_X550EM_x) ||
7123 (hw->mac.type == ixgbe_mac_x550em_a)) {
7124 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7125 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7126 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7127 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7131 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7132 /* work around hardware counting issue */
7133 hwstats->gprc -= missed_rx;
7135 ixgbe_update_xoff_received(adapter);
7137 /* 82598 hardware only has a 32 bit counter in the high register */
7138 switch (hw->mac.type) {
7139 case ixgbe_mac_82598EB:
7140 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7141 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7142 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7143 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7145 case ixgbe_mac_X540:
7146 case ixgbe_mac_X550:
7147 case ixgbe_mac_X550EM_x:
7148 case ixgbe_mac_x550em_a:
7149 /* OS2BMC stats are X540 and later */
7150 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7151 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7152 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7153 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7155 case ixgbe_mac_82599EB:
7156 for (i = 0; i < 16; i++)
7157 adapter->hw_rx_no_dma_resources +=
7158 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7159 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7160 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7161 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7162 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7163 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7164 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7165 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7166 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7167 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7169 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7170 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7171 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7172 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7173 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7174 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7175 /* Add up per cpu counters for total ddp aloc fail */
7176 if (adapter->fcoe.ddp_pool) {
7177 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7178 struct ixgbe_fcoe_ddp_pool *ddp_pool;
7180 u64 noddp = 0, noddp_ext_buff = 0;
7181 for_each_possible_cpu(cpu) {
7182 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7183 noddp += ddp_pool->noddp;
7184 noddp_ext_buff += ddp_pool->noddp_ext_buff;
7186 hwstats->fcoe_noddp = noddp;
7187 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7189 #endif /* IXGBE_FCOE */
7194 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7195 hwstats->bprc += bprc;
7196 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7197 if (hw->mac.type == ixgbe_mac_82598EB)
7198 hwstats->mprc -= bprc;
7199 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7200 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7201 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7202 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7203 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7204 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7205 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7206 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7207 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7208 hwstats->lxontxc += lxon;
7209 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7210 hwstats->lxofftxc += lxoff;
7211 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7212 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7214 * 82598 errata - tx of flow control packets is included in tx counters
7216 xon_off_tot = lxon + lxoff;
7217 hwstats->gptc -= xon_off_tot;
7218 hwstats->mptc -= xon_off_tot;
7219 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7220 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7221 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7222 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7223 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7224 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7225 hwstats->ptc64 -= xon_off_tot;
7226 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7227 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7228 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7229 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7230 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7231 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7233 /* Fill out the OS statistics structure */
7234 netdev->stats.multicast = hwstats->mprc;
7237 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7238 netdev->stats.rx_dropped = 0;
7239 netdev->stats.rx_length_errors = hwstats->rlec;
7240 netdev->stats.rx_crc_errors = hwstats->crcerrs;
7241 netdev->stats.rx_missed_errors = total_mpc;
7245 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7246 * @adapter: pointer to the device adapter structure
7248 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7250 struct ixgbe_hw *hw = &adapter->hw;
7253 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7256 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7258 /* if interface is down do nothing */
7259 if (test_bit(__IXGBE_DOWN, &adapter->state))
7262 /* do nothing if we are not using signature filters */
7263 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7266 adapter->fdir_overflow++;
7268 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7269 for (i = 0; i < adapter->num_tx_queues; i++)
7270 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7271 &(adapter->tx_ring[i]->state));
7272 for (i = 0; i < adapter->num_xdp_queues; i++)
7273 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7274 &adapter->xdp_ring[i]->state);
7275 /* re-enable flow director interrupts */
7276 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7278 e_err(probe, "failed to finish FDIR re-initialization, "
7279 "ignored adding FDIR ATR filters\n");
7284 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7285 * @adapter: pointer to the device adapter structure
7287 * This function serves two purposes. First it strobes the interrupt lines
7288 * in order to make certain interrupts are occurring. Secondly it sets the
7289 * bits needed to check for TX hangs. As a result we should immediately
7290 * determine if a hang has occurred.
7292 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7294 struct ixgbe_hw *hw = &adapter->hw;
7298 /* If we're down, removing or resetting, just bail */
7299 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7300 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7301 test_bit(__IXGBE_RESETTING, &adapter->state))
7304 /* Force detection of hung controller */
7305 if (netif_carrier_ok(adapter->netdev)) {
7306 for (i = 0; i < adapter->num_tx_queues; i++)
7307 set_check_for_tx_hang(adapter->tx_ring[i]);
7308 for (i = 0; i < adapter->num_xdp_queues; i++)
7309 set_check_for_tx_hang(adapter->xdp_ring[i]);
7312 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7314 * for legacy and MSI interrupts don't set any bits
7315 * that are enabled for EIAM, because this operation
7316 * would set *both* EIMS and EICS for any bit in EIAM
7318 IXGBE_WRITE_REG(hw, IXGBE_EICS,
7319 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7321 /* get one bit for every active tx/rx interrupt vector */
7322 for (i = 0; i < adapter->num_q_vectors; i++) {
7323 struct ixgbe_q_vector *qv = adapter->q_vector[i];
7324 if (qv->rx.ring || qv->tx.ring)
7329 /* Cause software interrupt to ensure rings are cleaned */
7330 ixgbe_irq_rearm_queues(adapter, eics);
7334 * ixgbe_watchdog_update_link - update the link status
7335 * @adapter: pointer to the device adapter structure
7337 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7339 struct ixgbe_hw *hw = &adapter->hw;
7340 u32 link_speed = adapter->link_speed;
7341 bool link_up = adapter->link_up;
7342 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7344 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7347 if (hw->mac.ops.check_link) {
7348 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7350 /* always assume link is up, if no check link function */
7351 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7355 if (adapter->ixgbe_ieee_pfc)
7356 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7358 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7359 hw->mac.ops.fc_enable(hw);
7360 ixgbe_set_rx_drop_en(adapter);
7364 time_after(jiffies, (adapter->link_check_timeout +
7365 IXGBE_TRY_LINK_TIMEOUT))) {
7366 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7367 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7368 IXGBE_WRITE_FLUSH(hw);
7371 adapter->link_up = link_up;
7372 adapter->link_speed = link_speed;
7375 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7377 #ifdef CONFIG_IXGBE_DCB
7378 struct net_device *netdev = adapter->netdev;
7379 struct dcb_app app = {
7380 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7385 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7386 up = dcb_ieee_getapp_mask(netdev, &app);
7388 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7393 * ixgbe_watchdog_link_is_up - update netif_carrier status and
7394 * print link up message
7395 * @adapter: pointer to the device adapter structure
7397 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7399 struct net_device *netdev = adapter->netdev;
7400 struct ixgbe_hw *hw = &adapter->hw;
7401 u32 link_speed = adapter->link_speed;
7402 const char *speed_str;
7403 bool flow_rx, flow_tx;
7405 /* only continue if link was previously down */
7406 if (netif_carrier_ok(netdev))
7409 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7411 switch (hw->mac.type) {
7412 case ixgbe_mac_82598EB: {
7413 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7414 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7415 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7416 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7419 case ixgbe_mac_X540:
7420 case ixgbe_mac_X550:
7421 case ixgbe_mac_X550EM_x:
7422 case ixgbe_mac_x550em_a:
7423 case ixgbe_mac_82599EB: {
7424 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7425 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7426 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7427 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7436 adapter->last_rx_ptp_check = jiffies;
7438 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7439 ixgbe_ptp_start_cyclecounter(adapter);
7441 switch (link_speed) {
7442 case IXGBE_LINK_SPEED_10GB_FULL:
7443 speed_str = "10 Gbps";
7445 case IXGBE_LINK_SPEED_5GB_FULL:
7446 speed_str = "5 Gbps";
7448 case IXGBE_LINK_SPEED_2_5GB_FULL:
7449 speed_str = "2.5 Gbps";
7451 case IXGBE_LINK_SPEED_1GB_FULL:
7452 speed_str = "1 Gbps";
7454 case IXGBE_LINK_SPEED_100_FULL:
7455 speed_str = "100 Mbps";
7457 case IXGBE_LINK_SPEED_10_FULL:
7458 speed_str = "10 Mbps";
7461 speed_str = "unknown speed";
7464 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7465 ((flow_rx && flow_tx) ? "RX/TX" :
7467 (flow_tx ? "TX" : "None"))));
7469 netif_carrier_on(netdev);
7470 ixgbe_check_vf_rate_limit(adapter);
7472 /* enable transmits */
7473 netif_tx_wake_all_queues(adapter->netdev);
7475 /* update the default user priority for VFs */
7476 ixgbe_update_default_up(adapter);
7478 /* ping all the active vfs to let them know link has changed */
7479 ixgbe_ping_all_vfs(adapter);
7483 * ixgbe_watchdog_link_is_down - update netif_carrier status and
7484 * print link down message
7485 * @adapter: pointer to the adapter structure
7487 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7489 struct net_device *netdev = adapter->netdev;
7490 struct ixgbe_hw *hw = &adapter->hw;
7492 adapter->link_up = false;
7493 adapter->link_speed = 0;
7495 /* only continue if link was up previously */
7496 if (!netif_carrier_ok(netdev))
7499 /* poll for SFP+ cable when link is down */
7500 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7501 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7503 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7504 ixgbe_ptp_start_cyclecounter(adapter);
7506 e_info(drv, "NIC Link is Down\n");
7507 netif_carrier_off(netdev);
7509 /* ping all the active vfs to let them know link has changed */
7510 ixgbe_ping_all_vfs(adapter);
7513 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7517 for (i = 0; i < adapter->num_tx_queues; i++) {
7518 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7520 if (tx_ring->next_to_use != tx_ring->next_to_clean)
7524 for (i = 0; i < adapter->num_xdp_queues; i++) {
7525 struct ixgbe_ring *ring = adapter->xdp_ring[i];
7527 if (ring->next_to_use != ring->next_to_clean)
7534 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7536 struct ixgbe_hw *hw = &adapter->hw;
7537 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7538 u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7542 if (!adapter->num_vfs)
7545 /* resetting the PF is only needed for MAC before X550 */
7546 if (hw->mac.type >= ixgbe_mac_X550)
7549 for (i = 0; i < adapter->num_vfs; i++) {
7550 for (j = 0; j < q_per_pool; j++) {
7553 h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7554 t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7565 * ixgbe_watchdog_flush_tx - flush queues on link down
7566 * @adapter: pointer to the device adapter structure
7568 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7570 if (!netif_carrier_ok(adapter->netdev)) {
7571 if (ixgbe_ring_tx_pending(adapter) ||
7572 ixgbe_vf_tx_pending(adapter)) {
7573 /* We've lost link, so the controller stops DMA,
7574 * but we've got queued Tx work that's never going
7575 * to get done, so reset controller to flush Tx.
7576 * (Do the reset outside of interrupt context).
7578 e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7579 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7584 #ifdef CONFIG_PCI_IOV
7585 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7587 struct ixgbe_hw *hw = &adapter->hw;
7588 struct pci_dev *pdev = adapter->pdev;
7592 if (!(netif_carrier_ok(adapter->netdev)))
7595 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7596 if (gpc) /* If incrementing then no need for the check below */
7598 /* Check to see if a bad DMA write target from an errant or
7599 * malicious VF has caused a PCIe error. If so then we can
7600 * issue a VFLR to the offending VF(s) and then resume without
7601 * requesting a full slot reset.
7607 /* check status reg for all VFs owned by this PF */
7608 for (vf = 0; vf < adapter->num_vfs; ++vf) {
7609 struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7614 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7615 if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7616 status_reg & PCI_STATUS_REC_MASTER_ABORT)
7621 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7625 /* Do not perform spoof check for 82598 or if not in IOV mode */
7626 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7627 adapter->num_vfs == 0)
7630 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7633 * ssvpc register is cleared on read, if zero then no
7634 * spoofed packets in the last interval.
7639 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7642 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7647 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7650 #endif /* CONFIG_PCI_IOV */
7654 * ixgbe_watchdog_subtask - check and bring link up
7655 * @adapter: pointer to the device adapter structure
7657 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7659 /* if interface is down, removing or resetting, do nothing */
7660 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7661 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7662 test_bit(__IXGBE_RESETTING, &adapter->state))
7665 ixgbe_watchdog_update_link(adapter);
7667 if (adapter->link_up)
7668 ixgbe_watchdog_link_is_up(adapter);
7670 ixgbe_watchdog_link_is_down(adapter);
7672 ixgbe_check_for_bad_vf(adapter);
7673 ixgbe_spoof_check(adapter);
7674 ixgbe_update_stats(adapter);
7676 ixgbe_watchdog_flush_tx(adapter);
7680 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7681 * @adapter: the ixgbe adapter structure
7683 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7685 struct ixgbe_hw *hw = &adapter->hw;
7688 /* not searching for SFP so there is nothing to do here */
7689 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7690 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7693 if (adapter->sfp_poll_time &&
7694 time_after(adapter->sfp_poll_time, jiffies))
7695 return; /* If not yet time to poll for SFP */
7697 /* someone else is in init, wait until next service event */
7698 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7701 adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7703 err = hw->phy.ops.identify_sfp(hw);
7704 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7707 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7708 /* If no cable is present, then we need to reset
7709 * the next time we find a good cable. */
7710 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7717 /* exit if reset not needed */
7718 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7721 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7724 * A module may be identified correctly, but the EEPROM may not have
7725 * support for that module. setup_sfp() will fail in that case, so
7726 * we should not allow that module to load.
7728 if (hw->mac.type == ixgbe_mac_82598EB)
7729 err = hw->phy.ops.reset(hw);
7731 err = hw->mac.ops.setup_sfp(hw);
7733 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7736 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7737 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7740 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7742 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7743 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7744 e_dev_err("failed to initialize because an unsupported "
7745 "SFP+ module type was detected.\n");
7746 e_dev_err("Reload the driver after installing a "
7747 "supported module.\n");
7748 unregister_netdev(adapter->netdev);
7753 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7754 * @adapter: the ixgbe adapter structure
7756 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7758 struct ixgbe_hw *hw = &adapter->hw;
7761 bool autoneg = false;
7763 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7766 /* someone else is in init, wait until next service event */
7767 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7770 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7772 hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7774 /* advertise highest capable link speed */
7775 if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7776 speed = IXGBE_LINK_SPEED_10GB_FULL;
7778 speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7779 IXGBE_LINK_SPEED_1GB_FULL);
7781 if (hw->mac.ops.setup_link)
7782 hw->mac.ops.setup_link(hw, speed, true);
7784 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7785 adapter->link_check_timeout = jiffies;
7786 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7790 * ixgbe_service_timer - Timer Call-back
7791 * @t: pointer to timer_list structure
7793 static void ixgbe_service_timer(struct timer_list *t)
7795 struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7796 unsigned long next_event_offset;
7798 /* poll faster when waiting for link */
7799 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7800 next_event_offset = HZ / 10;
7802 next_event_offset = HZ * 2;
7804 /* Reset the timer */
7805 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7807 ixgbe_service_event_schedule(adapter);
7810 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7812 struct ixgbe_hw *hw = &adapter->hw;
7815 if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7818 adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7820 if (!hw->phy.ops.handle_lasi)
7823 status = hw->phy.ops.handle_lasi(&adapter->hw);
7824 if (status != IXGBE_ERR_OVERTEMP)
7827 e_crit(drv, "%s\n", ixgbe_overheat_msg);
7830 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7832 if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7836 /* If we're already down, removing or resetting, just bail */
7837 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7838 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7839 test_bit(__IXGBE_RESETTING, &adapter->state)) {
7844 ixgbe_dump(adapter);
7845 netdev_err(adapter->netdev, "Reset adapter\n");
7846 adapter->tx_timeout_count++;
7848 ixgbe_reinit_locked(adapter);
7853 * ixgbe_check_fw_error - Check firmware for errors
7854 * @adapter: the adapter private structure
7856 * Check firmware errors in register FWSM
7858 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
7860 struct ixgbe_hw *hw = &adapter->hw;
7863 /* read fwsm.ext_err_ind register and log errors */
7864 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
7866 if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
7867 !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
7868 e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7871 if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
7872 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");
7880 * ixgbe_service_task - manages and runs subtasks
7881 * @work: pointer to work_struct containing our data
7883 static void ixgbe_service_task(struct work_struct *work)
7885 struct ixgbe_adapter *adapter = container_of(work,
7886 struct ixgbe_adapter,
7888 if (ixgbe_removed(adapter->hw.hw_addr)) {
7889 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7891 ixgbe_down(adapter);
7894 ixgbe_service_event_complete(adapter);
7897 if (ixgbe_check_fw_error(adapter)) {
7898 if (!test_bit(__IXGBE_DOWN, &adapter->state))
7899 unregister_netdev(adapter->netdev);
7900 ixgbe_service_event_complete(adapter);
7903 ixgbe_reset_subtask(adapter);
7904 ixgbe_phy_interrupt_subtask(adapter);
7905 ixgbe_sfp_detection_subtask(adapter);
7906 ixgbe_sfp_link_config_subtask(adapter);
7907 ixgbe_check_overtemp_subtask(adapter);
7908 ixgbe_watchdog_subtask(adapter);
7909 ixgbe_fdir_reinit_subtask(adapter);
7910 ixgbe_check_hang_subtask(adapter);
7912 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7913 ixgbe_ptp_overflow_check(adapter);
7914 if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
7915 ixgbe_ptp_rx_hang(adapter);
7916 ixgbe_ptp_tx_hang(adapter);
7919 ixgbe_service_event_complete(adapter);
7922 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7923 struct ixgbe_tx_buffer *first,
7925 struct ixgbe_ipsec_tx_data *itd)
7927 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7928 struct sk_buff *skb = first->skb;
7939 u32 paylen, l4_offset;
7940 u32 fceof_saidx = 0;
7943 if (skb->ip_summed != CHECKSUM_PARTIAL)
7946 if (!skb_is_gso(skb))
7949 err = skb_cow_head(skb, 0);
7953 if (eth_p_mpls(first->protocol))
7954 ip.hdr = skb_inner_network_header(skb);
7956 ip.hdr = skb_network_header(skb);
7957 l4.hdr = skb_checksum_start(skb);
7959 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7960 type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
7961 IXGBE_ADVTXD_TUCMD_L4T_UDP : IXGBE_ADVTXD_TUCMD_L4T_TCP;
7963 /* initialize outer IP header fields */
7964 if (ip.v4->version == 4) {
7965 unsigned char *csum_start = skb_checksum_start(skb);
7966 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
7967 int len = csum_start - trans_start;
7969 /* IP header will have to cancel out any data that
7970 * is not a part of the outer IP header, so set to
7971 * a reverse csum if needed, else init check to 0.
7973 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
7974 csum_fold(csum_partial(trans_start,
7976 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7979 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7980 IXGBE_TX_FLAGS_CSUM |
7981 IXGBE_TX_FLAGS_IPV4;
7983 ip.v6->payload_len = 0;
7984 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7985 IXGBE_TX_FLAGS_CSUM;
7988 /* determine offset of inner transport header */
7989 l4_offset = l4.hdr - skb->data;
7991 /* remove payload length from inner checksum */
7992 paylen = skb->len - l4_offset;
7994 if (type_tucmd & IXGBE_ADVTXD_TUCMD_L4T_TCP) {
7995 /* compute length of segmentation header */
7996 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
7997 csum_replace_by_diff(&l4.tcp->check,
7998 (__force __wsum)htonl(paylen));
8000 /* compute length of segmentation header */
8001 *hdr_len = sizeof(*l4.udp) + l4_offset;
8002 csum_replace_by_diff(&l4.udp->check,
8003 (__force __wsum)htonl(paylen));
8006 /* update gso size and bytecount with header size */
8007 first->gso_segs = skb_shinfo(skb)->gso_segs;
8008 first->bytecount += (first->gso_segs - 1) * *hdr_len;
8010 /* mss_l4len_id: use 0 as index for TSO */
8011 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8012 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8014 fceof_saidx |= itd->sa_idx;
8015 type_tucmd |= itd->flags | itd->trailer_len;
8017 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8018 vlan_macip_lens = l4.hdr - ip.hdr;
8019 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8020 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8022 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8028 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb)
8030 unsigned int offset = 0;
8032 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
8034 return offset == skb_checksum_start_offset(skb);
8037 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8038 struct ixgbe_tx_buffer *first,
8039 struct ixgbe_ipsec_tx_data *itd)
8041 struct sk_buff *skb = first->skb;
8042 u32 vlan_macip_lens = 0;
8043 u32 fceof_saidx = 0;
8046 if (skb->ip_summed != CHECKSUM_PARTIAL) {
8048 if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8049 IXGBE_TX_FLAGS_CC)))
8054 switch (skb->csum_offset) {
8055 case offsetof(struct tcphdr, check):
8056 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8058 case offsetof(struct udphdr, check):
8060 case offsetof(struct sctphdr, checksum):
8061 /* validate that this is actually an SCTP request */
8062 if (((first->protocol == htons(ETH_P_IP)) &&
8063 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
8064 ((first->protocol == htons(ETH_P_IPV6)) &&
8065 ixgbe_ipv6_csum_is_sctp(skb))) {
8066 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8071 skb_checksum_help(skb);
8075 /* update TX checksum flag */
8076 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8077 vlan_macip_lens = skb_checksum_start_offset(skb) -
8078 skb_network_offset(skb);
8080 /* vlan_macip_lens: MACLEN, VLAN tag */
8081 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8082 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8084 fceof_saidx |= itd->sa_idx;
8085 type_tucmd |= itd->flags | itd->trailer_len;
8087 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8090 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8091 ((_flag <= _result) ? \
8092 ((u32)(_input & _flag) * (_result / _flag)) : \
8093 ((u32)(_input & _flag) / (_flag / _result)))
8095 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8097 /* set type for advanced descriptor with frame checksum insertion */
8098 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8099 IXGBE_ADVTXD_DCMD_DEXT |
8100 IXGBE_ADVTXD_DCMD_IFCS;
8102 /* set HW vlan bit if vlan is present */
8103 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8104 IXGBE_ADVTXD_DCMD_VLE);
8106 /* set segmentation enable bits for TSO/FSO */
8107 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8108 IXGBE_ADVTXD_DCMD_TSE);
8110 /* set timestamp bit if present */
8111 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8112 IXGBE_ADVTXD_MAC_TSTAMP);
8114 /* insert frame checksum */
8115 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8120 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8121 u32 tx_flags, unsigned int paylen)
8123 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8125 /* enable L4 checksum for TSO and TX checksum offload */
8126 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8127 IXGBE_TX_FLAGS_CSUM,
8128 IXGBE_ADVTXD_POPTS_TXSM);
8130 /* enable IPv4 checksum for TSO */
8131 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8132 IXGBE_TX_FLAGS_IPV4,
8133 IXGBE_ADVTXD_POPTS_IXSM);
8136 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8137 IXGBE_TX_FLAGS_IPSEC,
8138 IXGBE_ADVTXD_POPTS_IPSEC);
8141 * Check Context must be set if Tx switch is enabled, which it
8142 * always is for case where virtual functions are running
8144 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8148 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8151 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8153 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8155 /* Herbert's original patch had:
8156 * smp_mb__after_netif_stop_queue();
8157 * but since that doesn't exist yet, just open code it.
8161 /* We need to check again in a case another CPU has just
8162 * made room available.
8164 if (likely(ixgbe_desc_unused(tx_ring) < size))
8167 /* A reprieve! - use start_queue because it doesn't call schedule */
8168 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8169 ++tx_ring->tx_stats.restart_queue;
8173 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8175 if (likely(ixgbe_desc_unused(tx_ring) >= size))
8178 return __ixgbe_maybe_stop_tx(tx_ring, size);
8181 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8182 struct ixgbe_tx_buffer *first,
8185 struct sk_buff *skb = first->skb;
8186 struct ixgbe_tx_buffer *tx_buffer;
8187 union ixgbe_adv_tx_desc *tx_desc;
8190 unsigned int data_len, size;
8191 u32 tx_flags = first->tx_flags;
8192 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8193 u16 i = tx_ring->next_to_use;
8195 tx_desc = IXGBE_TX_DESC(tx_ring, i);
8197 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8199 size = skb_headlen(skb);
8200 data_len = skb->data_len;
8203 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8204 if (data_len < sizeof(struct fcoe_crc_eof)) {
8205 size -= sizeof(struct fcoe_crc_eof) - data_len;
8208 data_len -= sizeof(struct fcoe_crc_eof);
8213 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8217 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8218 if (dma_mapping_error(tx_ring->dev, dma))
8221 /* record length, and DMA address */
8222 dma_unmap_len_set(tx_buffer, len, size);
8223 dma_unmap_addr_set(tx_buffer, dma, dma);
8225 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8227 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8228 tx_desc->read.cmd_type_len =
8229 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8233 if (i == tx_ring->count) {
8234 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8237 tx_desc->read.olinfo_status = 0;
8239 dma += IXGBE_MAX_DATA_PER_TXD;
8240 size -= IXGBE_MAX_DATA_PER_TXD;
8242 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8245 if (likely(!data_len))
8248 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8252 if (i == tx_ring->count) {
8253 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8256 tx_desc->read.olinfo_status = 0;
8259 size = min_t(unsigned int, data_len, skb_frag_size(frag));
8261 size = skb_frag_size(frag);
8265 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8268 tx_buffer = &tx_ring->tx_buffer_info[i];
8271 /* write last descriptor with RS and EOP bits */
8272 cmd_type |= size | IXGBE_TXD_CMD;
8273 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8275 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8277 /* set the timestamp */
8278 first->time_stamp = jiffies;
8280 skb_tx_timestamp(skb);
8283 * Force memory writes to complete before letting h/w know there
8284 * are new descriptors to fetch. (Only applicable for weak-ordered
8285 * memory model archs, such as IA-64).
8287 * We also need this memory barrier to make certain all of the
8288 * status bits have been updated before next_to_watch is written.
8292 /* set next_to_watch value indicating a packet is present */
8293 first->next_to_watch = tx_desc;
8296 if (i == tx_ring->count)
8299 tx_ring->next_to_use = i;
8301 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8303 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
8304 writel(i, tx_ring->tail);
8309 dev_err(tx_ring->dev, "TX DMA map failed\n");
8311 /* clear dma mappings for failed tx_buffer_info map */
8313 tx_buffer = &tx_ring->tx_buffer_info[i];
8314 if (dma_unmap_len(tx_buffer, len))
8315 dma_unmap_page(tx_ring->dev,
8316 dma_unmap_addr(tx_buffer, dma),
8317 dma_unmap_len(tx_buffer, len),
8319 dma_unmap_len_set(tx_buffer, len, 0);
8320 if (tx_buffer == first)
8323 i += tx_ring->count;
8327 dev_kfree_skb_any(first->skb);
8330 tx_ring->next_to_use = i;
8335 static void ixgbe_atr(struct ixgbe_ring *ring,
8336 struct ixgbe_tx_buffer *first)
8338 struct ixgbe_q_vector *q_vector = ring->q_vector;
8339 union ixgbe_atr_hash_dword input = { .dword = 0 };
8340 union ixgbe_atr_hash_dword common = { .dword = 0 };
8342 unsigned char *network;
8344 struct ipv6hdr *ipv6;
8348 struct sk_buff *skb;
8352 /* if ring doesn't have a interrupt vector, cannot perform ATR */
8356 /* do nothing if sampling is disabled */
8357 if (!ring->atr_sample_rate)
8362 /* currently only IPv4/IPv6 with TCP is supported */
8363 if ((first->protocol != htons(ETH_P_IP)) &&
8364 (first->protocol != htons(ETH_P_IPV6)))
8367 /* snag network header to get L4 type and address */
8369 hdr.network = skb_network_header(skb);
8370 if (unlikely(hdr.network <= skb->data))
8372 if (skb->encapsulation &&
8373 first->protocol == htons(ETH_P_IP) &&
8374 hdr.ipv4->protocol == IPPROTO_UDP) {
8375 struct ixgbe_adapter *adapter = q_vector->adapter;
8377 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8381 /* verify the port is recognized as VXLAN */
8382 if (adapter->vxlan_port &&
8383 udp_hdr(skb)->dest == adapter->vxlan_port)
8384 hdr.network = skb_inner_network_header(skb);
8386 if (adapter->geneve_port &&
8387 udp_hdr(skb)->dest == adapter->geneve_port)
8388 hdr.network = skb_inner_network_header(skb);
8391 /* Make sure we have at least [minimum IPv4 header + TCP]
8392 * or [IPv6 header] bytes
8394 if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8397 /* Currently only IPv4/IPv6 with TCP is supported */
8398 switch (hdr.ipv4->version) {
8400 /* access ihl as u8 to avoid unaligned access on ia64 */
8401 hlen = (hdr.network[0] & 0x0F) << 2;
8402 l4_proto = hdr.ipv4->protocol;
8405 hlen = hdr.network - skb->data;
8406 l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8407 hlen -= hdr.network - skb->data;
8413 if (l4_proto != IPPROTO_TCP)
8416 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8417 hlen + sizeof(struct tcphdr)))
8420 th = (struct tcphdr *)(hdr.network + hlen);
8422 /* skip this packet since the socket is closing */
8426 /* sample on all syn packets or once every atr sample count */
8427 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8430 /* reset sample count */
8431 ring->atr_count = 0;
8433 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8436 * src and dst are inverted, think how the receiver sees them
8438 * The input is broken into two sections, a non-compressed section
8439 * containing vm_pool, vlan_id, and flow_type. The rest of the data
8440 * is XORed together and stored in the compressed dword.
8442 input.formatted.vlan_id = vlan_id;
8445 * since src port and flex bytes occupy the same word XOR them together
8446 * and write the value to source port portion of compressed dword
8448 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8449 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8451 common.port.src ^= th->dest ^ first->protocol;
8452 common.port.dst ^= th->source;
8454 switch (hdr.ipv4->version) {
8456 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8457 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8460 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8461 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8462 hdr.ipv6->saddr.s6_addr32[1] ^
8463 hdr.ipv6->saddr.s6_addr32[2] ^
8464 hdr.ipv6->saddr.s6_addr32[3] ^
8465 hdr.ipv6->daddr.s6_addr32[0] ^
8466 hdr.ipv6->daddr.s6_addr32[1] ^
8467 hdr.ipv6->daddr.s6_addr32[2] ^
8468 hdr.ipv6->daddr.s6_addr32[3];
8474 if (hdr.network != skb_network_header(skb))
8475 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8477 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
8478 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8479 input, common, ring->queue_index);
8483 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8484 struct net_device *sb_dev)
8486 struct ixgbe_adapter *adapter;
8487 struct ixgbe_ring_feature *f;
8491 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8492 struct net_device *vdev = sb_dev;
8494 txq = vdev->tc_to_txq[tc].offset;
8495 txq += reciprocal_scale(skb_get_hash(skb),
8496 vdev->tc_to_txq[tc].count);
8502 * only execute the code below if protocol is FCoE
8503 * or FIP and we have FCoE enabled on the adapter
8505 switch (vlan_get_protocol(skb)) {
8506 case htons(ETH_P_FCOE):
8507 case htons(ETH_P_FIP):
8508 adapter = netdev_priv(dev);
8510 if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8514 return netdev_pick_tx(dev, skb, sb_dev);
8517 f = &adapter->ring_feature[RING_F_FCOE];
8519 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8522 while (txq >= f->indices)
8525 return txq + f->offset;
8529 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8530 struct xdp_frame *xdpf)
8532 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8533 struct ixgbe_tx_buffer *tx_buffer;
8534 union ixgbe_adv_tx_desc *tx_desc;
8541 if (unlikely(!ixgbe_desc_unused(ring)))
8542 return IXGBE_XDP_CONSUMED;
8544 dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
8545 if (dma_mapping_error(ring->dev, dma))
8546 return IXGBE_XDP_CONSUMED;
8548 /* record the location of the first descriptor for this packet */
8549 tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8550 tx_buffer->bytecount = len;
8551 tx_buffer->gso_segs = 1;
8552 tx_buffer->protocol = 0;
8554 i = ring->next_to_use;
8555 tx_desc = IXGBE_TX_DESC(ring, i);
8557 dma_unmap_len_set(tx_buffer, len, len);
8558 dma_unmap_addr_set(tx_buffer, dma, dma);
8559 tx_buffer->xdpf = xdpf;
8561 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8563 /* put descriptor type bits */
8564 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8565 IXGBE_ADVTXD_DCMD_DEXT |
8566 IXGBE_ADVTXD_DCMD_IFCS;
8567 cmd_type |= len | IXGBE_TXD_CMD;
8568 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8569 tx_desc->read.olinfo_status =
8570 cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8572 /* Avoid any potential race with xdp_xmit and cleanup */
8575 /* set next_to_watch value indicating a packet is present */
8577 if (i == ring->count)
8580 tx_buffer->next_to_watch = tx_desc;
8581 ring->next_to_use = i;
8583 return IXGBE_XDP_TX;
8586 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8587 struct ixgbe_adapter *adapter,
8588 struct ixgbe_ring *tx_ring)
8590 struct ixgbe_tx_buffer *first;
8594 u16 count = TXD_USE_COUNT(skb_headlen(skb));
8595 struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8596 __be16 protocol = skb->protocol;
8600 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8601 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8602 * + 2 desc gap to keep tail from touching head,
8603 * + 1 desc for context descriptor,
8604 * otherwise try next time
8606 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8607 count += TXD_USE_COUNT(skb_frag_size(
8608 &skb_shinfo(skb)->frags[f]));
8610 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8611 tx_ring->tx_stats.tx_busy++;
8612 return NETDEV_TX_BUSY;
8615 /* record the location of the first descriptor for this packet */
8616 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8618 first->bytecount = skb->len;
8619 first->gso_segs = 1;
8621 /* if we have a HW VLAN tag being added default to the HW one */
8622 if (skb_vlan_tag_present(skb)) {
8623 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8624 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8625 /* else if it is a SW VLAN check the next protocol and store the tag */
8626 } else if (protocol == htons(ETH_P_8021Q)) {
8627 struct vlan_hdr *vhdr, _vhdr;
8628 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8632 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8633 IXGBE_TX_FLAGS_VLAN_SHIFT;
8634 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8636 protocol = vlan_get_protocol(skb);
8638 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8639 adapter->ptp_clock) {
8640 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
8641 !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8643 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8644 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8646 /* schedule check for Tx timestamp */
8647 adapter->ptp_tx_skb = skb_get(skb);
8648 adapter->ptp_tx_start = jiffies;
8649 schedule_work(&adapter->ptp_tx_work);
8651 adapter->tx_hwtstamp_skipped++;
8655 #ifdef CONFIG_PCI_IOV
8657 * Use the l2switch_enable flag - would be false if the DMA
8658 * Tx switch had been disabled.
8660 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8661 tx_flags |= IXGBE_TX_FLAGS_CC;
8664 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8665 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8666 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8667 (skb->priority != TC_PRIO_CONTROL))) {
8668 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8669 tx_flags |= (skb->priority & 0x7) <<
8670 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8671 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8672 struct vlan_ethhdr *vhdr;
8674 if (skb_cow_head(skb, 0))
8676 vhdr = (struct vlan_ethhdr *)skb->data;
8677 vhdr->h_vlan_TCI = htons(tx_flags >>
8678 IXGBE_TX_FLAGS_VLAN_SHIFT);
8680 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8684 /* record initial flags and protocol */
8685 first->tx_flags = tx_flags;
8686 first->protocol = protocol;
8689 /* setup tx offload for FCoE */
8690 if ((protocol == htons(ETH_P_FCOE)) &&
8691 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8692 tso = ixgbe_fso(tx_ring, first, &hdr_len);
8699 #endif /* IXGBE_FCOE */
8701 #ifdef CONFIG_IXGBE_IPSEC
8702 if (xfrm_offload(skb) &&
8703 !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8706 tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8710 ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8712 /* add the ATR filter if ATR is on */
8713 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8714 ixgbe_atr(tx_ring, first);
8718 #endif /* IXGBE_FCOE */
8719 if (ixgbe_tx_map(tx_ring, first, hdr_len))
8720 goto cleanup_tx_timestamp;
8722 return NETDEV_TX_OK;
8725 dev_kfree_skb_any(first->skb);
8727 cleanup_tx_timestamp:
8728 if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8729 dev_kfree_skb_any(adapter->ptp_tx_skb);
8730 adapter->ptp_tx_skb = NULL;
8731 cancel_work_sync(&adapter->ptp_tx_work);
8732 clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8735 return NETDEV_TX_OK;
8738 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8739 struct net_device *netdev,
8740 struct ixgbe_ring *ring)
8742 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8743 struct ixgbe_ring *tx_ring;
8746 * The minimum packet size for olinfo paylen is 17 so pad the skb
8747 * in order to meet this minimum size requirement.
8749 if (skb_put_padto(skb, 17))
8750 return NETDEV_TX_OK;
8752 tx_ring = ring ? ring : adapter->tx_ring[skb_get_queue_mapping(skb)];
8753 if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8754 return NETDEV_TX_BUSY;
8756 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8759 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8760 struct net_device *netdev)
8762 return __ixgbe_xmit_frame(skb, netdev, NULL);
8766 * ixgbe_set_mac - Change the Ethernet Address of the NIC
8767 * @netdev: network interface device structure
8768 * @p: pointer to an address structure
8770 * Returns 0 on success, negative on failure
8772 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8774 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8775 struct ixgbe_hw *hw = &adapter->hw;
8776 struct sockaddr *addr = p;
8778 if (!is_valid_ether_addr(addr->sa_data))
8779 return -EADDRNOTAVAIL;
8781 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8782 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8784 ixgbe_mac_set_default_filter(adapter);
8790 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8792 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8793 struct ixgbe_hw *hw = &adapter->hw;
8797 if (adapter->mii_bus) {
8800 if (devad != MDIO_DEVAD_NONE)
8801 regnum |= (devad << 16) | MII_ADDR_C45;
8803 return mdiobus_read(adapter->mii_bus, prtad, regnum);
8806 if (prtad != hw->phy.mdio.prtad)
8808 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8814 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8815 u16 addr, u16 value)
8817 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8818 struct ixgbe_hw *hw = &adapter->hw;
8820 if (adapter->mii_bus) {
8823 if (devad != MDIO_DEVAD_NONE)
8824 regnum |= (devad << 16) | MII_ADDR_C45;
8826 return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8829 if (prtad != hw->phy.mdio.prtad)
8831 return hw->phy.ops.write_reg(hw, addr, devad, value);
8834 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8836 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8840 return ixgbe_ptp_set_ts_config(adapter, req);
8842 return ixgbe_ptp_get_ts_config(adapter, req);
8844 if (!adapter->hw.phy.ops.read_reg)
8848 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8853 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8855 * @dev: network interface device structure
8857 * Returns non-zero on failure
8859 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8862 struct ixgbe_adapter *adapter = netdev_priv(dev);
8863 struct ixgbe_hw *hw = &adapter->hw;
8865 if (is_valid_ether_addr(hw->mac.san_addr)) {
8867 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8870 /* update SAN MAC vmdq pool selection */
8871 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8877 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8879 * @dev: network interface device structure
8881 * Returns non-zero on failure
8883 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8886 struct ixgbe_adapter *adapter = netdev_priv(dev);
8887 struct ixgbe_mac_info *mac = &adapter->hw.mac;
8889 if (is_valid_ether_addr(mac->san_addr)) {
8891 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8897 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8898 struct ixgbe_ring *ring)
8905 start = u64_stats_fetch_begin_irq(&ring->syncp);
8906 packets = ring->stats.packets;
8907 bytes = ring->stats.bytes;
8908 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8909 stats->tx_packets += packets;
8910 stats->tx_bytes += bytes;
8914 static void ixgbe_get_stats64(struct net_device *netdev,
8915 struct rtnl_link_stats64 *stats)
8917 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8921 for (i = 0; i < adapter->num_rx_queues; i++) {
8922 struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8928 start = u64_stats_fetch_begin_irq(&ring->syncp);
8929 packets = ring->stats.packets;
8930 bytes = ring->stats.bytes;
8931 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8932 stats->rx_packets += packets;
8933 stats->rx_bytes += bytes;
8937 for (i = 0; i < adapter->num_tx_queues; i++) {
8938 struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8940 ixgbe_get_ring_stats64(stats, ring);
8942 for (i = 0; i < adapter->num_xdp_queues; i++) {
8943 struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8945 ixgbe_get_ring_stats64(stats, ring);
8949 /* following stats updated by ixgbe_watchdog_task() */
8950 stats->multicast = netdev->stats.multicast;
8951 stats->rx_errors = netdev->stats.rx_errors;
8952 stats->rx_length_errors = netdev->stats.rx_length_errors;
8953 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
8954 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
8957 #ifdef CONFIG_IXGBE_DCB
8959 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8960 * @adapter: pointer to ixgbe_adapter
8961 * @tc: number of traffic classes currently enabled
8963 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8964 * 802.1Q priority maps to a packet buffer that exists.
8966 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8968 struct ixgbe_hw *hw = &adapter->hw;
8972 /* 82598 have a static priority to TC mapping that can not
8973 * be changed so no validation is needed.
8975 if (hw->mac.type == ixgbe_mac_82598EB)
8978 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
8981 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
8982 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
8984 /* If up2tc is out of bounds default to zero */
8986 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
8990 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
8996 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
8997 * @adapter: Pointer to adapter struct
8999 * Populate the netdev user priority to tc map
9001 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
9003 struct net_device *dev = adapter->netdev;
9004 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9005 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9008 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9011 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9012 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9014 tc = ets->prio_tc[prio];
9016 netdev_set_prio_tc_map(dev, prio, tc);
9020 #endif /* CONFIG_IXGBE_DCB */
9021 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev, void *data)
9023 struct ixgbe_adapter *adapter = data;
9024 struct ixgbe_fwd_adapter *accel;
9027 /* we only care about macvlans... */
9028 if (!netif_is_macvlan(vdev))
9031 /* that have hardware offload enabled... */
9032 accel = macvlan_accel_priv(vdev);
9036 /* If we can relocate to a different bit do so */
9037 pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9038 if (pool < adapter->num_rx_pools) {
9039 set_bit(pool, adapter->fwd_bitmask);
9044 /* if we cannot find a free pool then disable the offload */
9045 netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9046 macvlan_release_l2fw_offload(vdev);
9048 /* unbind the queues and drop the subordinate channel config */
9049 netdev_unbind_sb_channel(adapter->netdev, vdev);
9050 netdev_set_sb_channel(vdev, 0);
9057 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9059 struct ixgbe_adapter *adapter = netdev_priv(dev);
9061 /* flush any stale bits out of the fwd bitmask */
9062 bitmap_clear(adapter->fwd_bitmask, 1, 63);
9064 /* walk through upper devices reassigning pools */
9065 netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9070 * ixgbe_setup_tc - configure net_device for multiple traffic classes
9072 * @dev: net device to configure
9073 * @tc: number of traffic classes to enable
9075 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9077 struct ixgbe_adapter *adapter = netdev_priv(dev);
9078 struct ixgbe_hw *hw = &adapter->hw;
9080 /* Hardware supports up to 8 traffic classes */
9081 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9084 if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9087 /* Hardware has to reinitialize queues and interrupts to
9088 * match packet buffer alignment. Unfortunately, the
9089 * hardware is not flexible enough to do this dynamically.
9091 if (netif_running(dev))
9094 ixgbe_reset(adapter);
9096 ixgbe_clear_interrupt_scheme(adapter);
9098 #ifdef CONFIG_IXGBE_DCB
9100 if (adapter->xdp_prog) {
9101 e_warn(probe, "DCB is not supported with XDP\n");
9103 ixgbe_init_interrupt_scheme(adapter);
9104 if (netif_running(dev))
9109 netdev_set_num_tc(dev, tc);
9110 ixgbe_set_prio_tc_map(adapter);
9112 adapter->hw_tcs = tc;
9113 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9115 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9116 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9117 adapter->hw.fc.requested_mode = ixgbe_fc_none;
9120 netdev_reset_tc(dev);
9122 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9123 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9125 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9126 adapter->hw_tcs = tc;
9128 adapter->temp_dcb_cfg.pfc_mode_enable = false;
9129 adapter->dcb_cfg.pfc_mode_enable = false;
9132 ixgbe_validate_rtr(adapter, tc);
9134 #endif /* CONFIG_IXGBE_DCB */
9135 ixgbe_init_interrupt_scheme(adapter);
9137 ixgbe_defrag_macvlan_pools(dev);
9139 if (netif_running(dev))
9140 return ixgbe_open(dev);
9145 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9146 struct tc_cls_u32_offload *cls)
9148 u32 hdl = cls->knode.handle;
9149 u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9150 u32 loc = cls->knode.handle & 0xfffff;
9152 struct ixgbe_jump_table *jump = NULL;
9154 if (loc > IXGBE_MAX_HW_ENTRIES)
9157 if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9160 /* Clear this filter in the link data it is associated with */
9161 if (uhtid != 0x800) {
9162 jump = adapter->jump_tables[uhtid];
9165 if (!test_bit(loc - 1, jump->child_loc_map))
9167 clear_bit(loc - 1, jump->child_loc_map);
9170 /* Check if the filter being deleted is a link */
9171 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9172 jump = adapter->jump_tables[i];
9173 if (jump && jump->link_hdl == hdl) {
9174 /* Delete filters in the hardware in the child hash
9175 * table associated with this link
9177 for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9178 if (!test_bit(j, jump->child_loc_map))
9180 spin_lock(&adapter->fdir_perfect_lock);
9181 err = ixgbe_update_ethtool_fdir_entry(adapter,
9184 spin_unlock(&adapter->fdir_perfect_lock);
9185 clear_bit(j, jump->child_loc_map);
9187 /* Remove resources for this link */
9191 adapter->jump_tables[i] = NULL;
9196 spin_lock(&adapter->fdir_perfect_lock);
9197 err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9198 spin_unlock(&adapter->fdir_perfect_lock);
9202 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9203 struct tc_cls_u32_offload *cls)
9205 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9207 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9210 /* This ixgbe devices do not support hash tables at the moment
9211 * so abort when given hash tables.
9213 if (cls->hnode.divisor > 0)
9216 set_bit(uhtid - 1, &adapter->tables);
9220 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9221 struct tc_cls_u32_offload *cls)
9223 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9225 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9228 clear_bit(uhtid - 1, &adapter->tables);
9232 #ifdef CONFIG_NET_CLS_ACT
9233 struct upper_walk_data {
9234 struct ixgbe_adapter *adapter;
9240 static int get_macvlan_queue(struct net_device *upper, void *_data)
9242 if (netif_is_macvlan(upper)) {
9243 struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9244 struct upper_walk_data *data = _data;
9245 struct ixgbe_adapter *adapter = data->adapter;
9246 int ifindex = data->ifindex;
9248 if (vadapter && upper->ifindex == ifindex) {
9249 data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9250 data->action = data->queue;
9258 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9259 u8 *queue, u64 *action)
9261 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9262 unsigned int num_vfs = adapter->num_vfs, vf;
9263 struct upper_walk_data data;
9264 struct net_device *upper;
9266 /* redirect to a SRIOV VF */
9267 for (vf = 0; vf < num_vfs; ++vf) {
9268 upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9269 if (upper->ifindex == ifindex) {
9270 *queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9272 *action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9277 /* redirect to a offloaded macvlan netdev */
9278 data.adapter = adapter;
9279 data.ifindex = ifindex;
9282 if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9283 get_macvlan_queue, &data)) {
9284 *action = data.action;
9285 *queue = data.queue;
9293 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9294 struct tcf_exts *exts, u64 *action, u8 *queue)
9296 const struct tc_action *a;
9299 if (!tcf_exts_has_actions(exts))
9302 tcf_exts_for_each_action(i, a, exts) {
9304 if (is_tcf_gact_shot(a)) {
9305 *action = IXGBE_FDIR_DROP_QUEUE;
9306 *queue = IXGBE_FDIR_DROP_QUEUE;
9310 /* Redirect to a VF or a offloaded macvlan */
9311 if (is_tcf_mirred_egress_redirect(a)) {
9312 struct net_device *dev = tcf_mirred_dev(a);
9316 return handle_redirect_action(adapter, dev->ifindex,
9326 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9327 struct tcf_exts *exts, u64 *action, u8 *queue)
9331 #endif /* CONFIG_NET_CLS_ACT */
9333 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9334 union ixgbe_atr_input *mask,
9335 struct tc_cls_u32_offload *cls,
9336 struct ixgbe_mat_field *field_ptr,
9337 struct ixgbe_nexthdr *nexthdr)
9341 bool found_entry = false, found_jump_field = false;
9343 for (i = 0; i < cls->knode.sel->nkeys; i++) {
9344 off = cls->knode.sel->keys[i].off;
9345 val = cls->knode.sel->keys[i].val;
9346 m = cls->knode.sel->keys[i].mask;
9348 for (j = 0; field_ptr[j].val; j++) {
9349 if (field_ptr[j].off == off) {
9350 field_ptr[j].val(input, mask, (__force u32)val,
9352 input->filter.formatted.flow_type |=
9359 if (nexthdr->off == cls->knode.sel->keys[i].off &&
9361 (__force u32)cls->knode.sel->keys[i].val &&
9363 (__force u32)cls->knode.sel->keys[i].mask)
9364 found_jump_field = true;
9370 if (nexthdr && !found_jump_field)
9376 mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9377 IXGBE_ATR_L4TYPE_MASK;
9379 if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9380 mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9385 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9386 struct tc_cls_u32_offload *cls)
9388 __be16 protocol = cls->common.protocol;
9389 u32 loc = cls->knode.handle & 0xfffff;
9390 struct ixgbe_hw *hw = &adapter->hw;
9391 struct ixgbe_mat_field *field_ptr;
9392 struct ixgbe_fdir_filter *input = NULL;
9393 union ixgbe_atr_input *mask = NULL;
9394 struct ixgbe_jump_table *jump = NULL;
9395 int i, err = -EINVAL;
9397 u32 uhtid, link_uhtid;
9399 uhtid = TC_U32_USERHTID(cls->knode.handle);
9400 link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9402 /* At the moment cls_u32 jumps to network layer and skips past
9403 * L2 headers. The canonical method to match L2 frames is to use
9404 * negative values. However this is error prone at best but really
9405 * just broken because there is no way to "know" what sort of hdr
9406 * is in front of the network layer. Fix cls_u32 to support L2
9407 * headers when needed.
9409 if (protocol != htons(ETH_P_IP))
9412 if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9413 e_err(drv, "Location out of range\n");
9417 /* cls u32 is a graph starting at root node 0x800. The driver tracks
9418 * links and also the fields used to advance the parser across each
9419 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9420 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9421 * To add support for new nodes update ixgbe_model.h parse structures
9422 * this function _should_ be generic try not to hardcode values here.
9424 if (uhtid == 0x800) {
9425 field_ptr = (adapter->jump_tables[0])->mat;
9427 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9429 if (!adapter->jump_tables[uhtid])
9431 field_ptr = (adapter->jump_tables[uhtid])->mat;
9437 /* At this point we know the field_ptr is valid and need to either
9438 * build cls_u32 link or attach filter. Because adding a link to
9439 * a handle that does not exist is invalid and the same for adding
9440 * rules to handles that don't exist.
9444 struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9446 if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9449 if (!test_bit(link_uhtid - 1, &adapter->tables))
9452 /* Multiple filters as links to the same hash table are not
9453 * supported. To add a new filter with the same next header
9454 * but different match/jump conditions, create a new hash table
9457 if (adapter->jump_tables[link_uhtid] &&
9458 (adapter->jump_tables[link_uhtid])->link_hdl) {
9459 e_err(drv, "Link filter exists for link: %x\n",
9464 for (i = 0; nexthdr[i].jump; i++) {
9465 if (nexthdr[i].o != cls->knode.sel->offoff ||
9466 nexthdr[i].s != cls->knode.sel->offshift ||
9468 (__force u32)cls->knode.sel->offmask)
9471 jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9474 input = kzalloc(sizeof(*input), GFP_KERNEL);
9479 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9484 jump->input = input;
9486 jump->link_hdl = cls->knode.handle;
9488 err = ixgbe_clsu32_build_input(input, mask, cls,
9489 field_ptr, &nexthdr[i]);
9491 jump->mat = nexthdr[i].jump;
9492 adapter->jump_tables[link_uhtid] = jump;
9503 input = kzalloc(sizeof(*input), GFP_KERNEL);
9506 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9512 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9513 if ((adapter->jump_tables[uhtid])->input)
9514 memcpy(input, (adapter->jump_tables[uhtid])->input,
9516 if ((adapter->jump_tables[uhtid])->mask)
9517 memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9520 /* Lookup in all child hash tables if this location is already
9521 * filled with a filter
9523 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9524 struct ixgbe_jump_table *link = adapter->jump_tables[i];
9526 if (link && (test_bit(loc - 1, link->child_loc_map))) {
9527 e_err(drv, "Filter exists in location: %x\n",
9534 err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9538 err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9543 input->sw_idx = loc;
9545 spin_lock(&adapter->fdir_perfect_lock);
9547 if (hlist_empty(&adapter->fdir_filter_list)) {
9548 memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9549 err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9551 goto err_out_w_lock;
9552 } else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9554 goto err_out_w_lock;
9557 ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9558 err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9559 input->sw_idx, queue);
9561 ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9562 spin_unlock(&adapter->fdir_perfect_lock);
9564 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9565 set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9570 spin_unlock(&adapter->fdir_perfect_lock);
9580 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9581 struct tc_cls_u32_offload *cls_u32)
9583 switch (cls_u32->command) {
9584 case TC_CLSU32_NEW_KNODE:
9585 case TC_CLSU32_REPLACE_KNODE:
9586 return ixgbe_configure_clsu32(adapter, cls_u32);
9587 case TC_CLSU32_DELETE_KNODE:
9588 return ixgbe_delete_clsu32(adapter, cls_u32);
9589 case TC_CLSU32_NEW_HNODE:
9590 case TC_CLSU32_REPLACE_HNODE:
9591 return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9592 case TC_CLSU32_DELETE_HNODE:
9593 return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9599 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9602 struct ixgbe_adapter *adapter = cb_priv;
9604 if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9608 case TC_SETUP_CLSU32:
9609 return ixgbe_setup_tc_cls_u32(adapter, type_data);
9615 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9616 struct tc_mqprio_qopt *mqprio)
9618 mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9619 return ixgbe_setup_tc(dev, mqprio->num_tc);
9622 static LIST_HEAD(ixgbe_block_cb_list);
9624 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9627 struct ixgbe_adapter *adapter = netdev_priv(dev);
9630 case TC_SETUP_BLOCK:
9631 return flow_block_cb_setup_simple(type_data,
9632 &ixgbe_block_cb_list,
9633 ixgbe_setup_tc_block_cb,
9634 adapter, adapter, true);
9635 case TC_SETUP_QDISC_MQPRIO:
9636 return ixgbe_setup_tc_mqprio(dev, type_data);
9642 #ifdef CONFIG_PCI_IOV
9643 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9645 struct net_device *netdev = adapter->netdev;
9648 ixgbe_setup_tc(netdev, adapter->hw_tcs);
9653 void ixgbe_do_reset(struct net_device *netdev)
9655 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9657 if (netif_running(netdev))
9658 ixgbe_reinit_locked(adapter);
9660 ixgbe_reset(adapter);
9663 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9664 netdev_features_t features)
9666 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9668 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9669 if (!(features & NETIF_F_RXCSUM))
9670 features &= ~NETIF_F_LRO;
9672 /* Turn off LRO if not RSC capable */
9673 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9674 features &= ~NETIF_F_LRO;
9676 if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9677 e_dev_err("LRO is not supported with XDP\n");
9678 features &= ~NETIF_F_LRO;
9684 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9686 int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9689 /* go back to full RSS if we're not running SR-IOV */
9690 if (!adapter->ring_feature[RING_F_VMDQ].offset)
9691 adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9692 IXGBE_FLAG_SRIOV_ENABLED);
9694 adapter->ring_feature[RING_F_RSS].limit = rss;
9695 adapter->ring_feature[RING_F_VMDQ].limit = 1;
9697 ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9700 static int ixgbe_set_features(struct net_device *netdev,
9701 netdev_features_t features)
9703 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9704 netdev_features_t changed = netdev->features ^ features;
9705 bool need_reset = false;
9707 /* Make sure RSC matches LRO, reset if change */
9708 if (!(features & NETIF_F_LRO)) {
9709 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9711 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9712 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9713 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9714 if (adapter->rx_itr_setting == 1 ||
9715 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9716 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9718 } else if ((changed ^ features) & NETIF_F_LRO) {
9719 e_info(probe, "rx-usecs set too low, "
9725 * Check if Flow Director n-tuple support or hw_tc support was
9726 * enabled or disabled. If the state changed, we need to reset.
9728 if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9729 /* turn off ATR, enable perfect filters and reset */
9730 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9733 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9734 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9736 /* turn off perfect filters, enable ATR and reset */
9737 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9740 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9742 /* We cannot enable ATR if SR-IOV is enabled */
9743 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9744 /* We cannot enable ATR if we have 2 or more tcs */
9745 (adapter->hw_tcs > 1) ||
9746 /* We cannot enable ATR if RSS is disabled */
9747 (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9748 /* A sample rate of 0 indicates ATR disabled */
9749 (!adapter->atr_sample_rate))
9750 ; /* do nothing not supported */
9751 else /* otherwise supported and set the flag */
9752 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9755 if (changed & NETIF_F_RXALL)
9758 netdev->features = features;
9760 if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9761 ixgbe_reset_l2fw_offload(adapter);
9762 else if (need_reset)
9763 ixgbe_do_reset(netdev);
9764 else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9765 NETIF_F_HW_VLAN_CTAG_FILTER))
9766 ixgbe_set_rx_mode(netdev);
9771 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9772 struct net_device *dev,
9773 const unsigned char *addr, u16 vid,
9775 struct netlink_ext_ack *extack)
9777 /* guarantee we can provide a unique filter for the unicast address */
9778 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9779 struct ixgbe_adapter *adapter = netdev_priv(dev);
9780 u16 pool = VMDQ_P(0);
9782 if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9786 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9790 * ixgbe_configure_bridge_mode - set various bridge modes
9791 * @adapter: the private structure
9792 * @mode: requested bridge mode
9794 * Configure some settings require for various bridge modes.
9796 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9799 struct ixgbe_hw *hw = &adapter->hw;
9800 unsigned int p, num_pools;
9804 case BRIDGE_MODE_VEPA:
9805 /* disable Tx loopback, rely on switch hairpin mode */
9806 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9808 /* must enable Rx switching replication to allow multicast
9809 * packet reception on all VFs, and to enable source address
9812 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9813 vmdctl |= IXGBE_VT_CTL_REPLEN;
9814 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9816 /* enable Rx source address pruning. Note, this requires
9817 * replication to be enabled or else it does nothing.
9819 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9820 for (p = 0; p < num_pools; p++) {
9821 if (hw->mac.ops.set_source_address_pruning)
9822 hw->mac.ops.set_source_address_pruning(hw,
9827 case BRIDGE_MODE_VEB:
9828 /* enable Tx loopback for internal VF/PF communication */
9829 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9830 IXGBE_PFDTXGSWC_VT_LBEN);
9832 /* disable Rx switching replication unless we have SR-IOV
9835 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9836 if (!adapter->num_vfs)
9837 vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9838 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9840 /* disable Rx source address pruning, since we don't expect to
9841 * be receiving external loopback of our transmitted frames.
9843 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9844 for (p = 0; p < num_pools; p++) {
9845 if (hw->mac.ops.set_source_address_pruning)
9846 hw->mac.ops.set_source_address_pruning(hw,
9855 adapter->bridge_mode = mode;
9857 e_info(drv, "enabling bridge mode: %s\n",
9858 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
9863 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
9864 struct nlmsghdr *nlh, u16 flags,
9865 struct netlink_ext_ack *extack)
9867 struct ixgbe_adapter *adapter = netdev_priv(dev);
9868 struct nlattr *attr, *br_spec;
9871 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9874 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
9878 nla_for_each_nested(attr, br_spec, rem) {
9882 if (nla_type(attr) != IFLA_BRIDGE_MODE)
9885 if (nla_len(attr) < sizeof(mode))
9888 mode = nla_get_u16(attr);
9889 status = ixgbe_configure_bridge_mode(adapter, mode);
9899 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
9900 struct net_device *dev,
9901 u32 filter_mask, int nlflags)
9903 struct ixgbe_adapter *adapter = netdev_priv(dev);
9905 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9908 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
9909 adapter->bridge_mode, 0, 0, nlflags,
9913 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
9915 struct ixgbe_adapter *adapter = netdev_priv(pdev);
9916 struct ixgbe_fwd_adapter *accel;
9917 int tcs = adapter->hw_tcs ? : 1;
9920 if (adapter->xdp_prog) {
9921 e_warn(probe, "L2FW offload is not supported with XDP\n");
9922 return ERR_PTR(-EINVAL);
9925 /* The hardware supported by ixgbe only filters on the destination MAC
9926 * address. In order to avoid issues we only support offloading modes
9927 * where the hardware can actually provide the functionality.
9929 if (!macvlan_supports_dest_filter(vdev))
9930 return ERR_PTR(-EMEDIUMTYPE);
9932 /* We need to lock down the macvlan to be a single queue device so that
9933 * we can reuse the tc_to_txq field in the macvlan netdev to represent
9934 * the queue mapping to our netdev.
9936 if (netif_is_multiqueue(vdev))
9937 return ERR_PTR(-ERANGE);
9939 pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9940 if (pool == adapter->num_rx_pools) {
9941 u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
9944 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
9945 adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
9946 adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
9947 return ERR_PTR(-EBUSY);
9949 /* Hardware has a limited number of available pools. Each VF,
9950 * and the PF require a pool. Check to ensure we don't
9951 * attempt to use more then the available number of pools.
9953 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
9954 return ERR_PTR(-EBUSY);
9956 /* Enable VMDq flag so device will be set in VM mode */
9957 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
9958 IXGBE_FLAG_SRIOV_ENABLED;
9960 /* Try to reserve as many queues per pool as possible,
9961 * we start with the configurations that support 4 queues
9962 * per pools, followed by 2, and then by just 1 per pool.
9964 if (used_pools < 32 && adapter->num_rx_pools < 16)
9965 reserved_pools = min_t(u16,
9967 16 - adapter->num_rx_pools);
9968 else if (adapter->num_rx_pools < 32)
9969 reserved_pools = min_t(u16,
9971 32 - adapter->num_rx_pools);
9973 reserved_pools = 64 - used_pools;
9976 if (!reserved_pools)
9977 return ERR_PTR(-EBUSY);
9979 adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
9981 /* Force reinit of ring allocation with VMDQ enabled */
9982 err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
9984 return ERR_PTR(err);
9986 if (pool >= adapter->num_rx_pools)
9987 return ERR_PTR(-ENOMEM);
9990 accel = kzalloc(sizeof(*accel), GFP_KERNEL);
9992 return ERR_PTR(-ENOMEM);
9994 set_bit(pool, adapter->fwd_bitmask);
9995 netdev_set_sb_channel(vdev, pool);
9997 accel->netdev = vdev;
9999 if (!netif_running(pdev))
10002 err = ixgbe_fwd_ring_up(adapter, accel);
10004 return ERR_PTR(err);
10009 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10011 struct ixgbe_fwd_adapter *accel = priv;
10012 struct ixgbe_adapter *adapter = netdev_priv(pdev);
10013 unsigned int rxbase = accel->rx_base_queue;
10016 /* delete unicast filter associated with offloaded interface */
10017 ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10018 VMDQ_P(accel->pool));
10020 /* Allow remaining Rx packets to get flushed out of the
10021 * Rx FIFO before we drop the netdev for the ring.
10023 usleep_range(10000, 20000);
10025 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10026 struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10027 struct ixgbe_q_vector *qv = ring->q_vector;
10029 /* Make sure we aren't processing any packets and clear
10030 * netdev to shut down the ring.
10032 if (netif_running(adapter->netdev))
10033 napi_synchronize(&qv->napi);
10034 ring->netdev = NULL;
10037 /* unbind the queues and drop the subordinate channel config */
10038 netdev_unbind_sb_channel(pdev, accel->netdev);
10039 netdev_set_sb_channel(accel->netdev, 0);
10041 clear_bit(accel->pool, adapter->fwd_bitmask);
10045 #define IXGBE_MAX_MAC_HDR_LEN 127
10046 #define IXGBE_MAX_NETWORK_HDR_LEN 511
10048 static netdev_features_t
10049 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10050 netdev_features_t features)
10052 unsigned int network_hdr_len, mac_hdr_len;
10054 /* Make certain the headers can be described by a context descriptor */
10055 mac_hdr_len = skb_network_header(skb) - skb->data;
10056 if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10057 return features & ~(NETIF_F_HW_CSUM |
10059 NETIF_F_GSO_UDP_L4 |
10060 NETIF_F_HW_VLAN_CTAG_TX |
10064 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10065 if (unlikely(network_hdr_len > IXGBE_MAX_NETWORK_HDR_LEN))
10066 return features & ~(NETIF_F_HW_CSUM |
10068 NETIF_F_GSO_UDP_L4 |
10072 /* We can only support IPV4 TSO in tunnels if we can mangle the
10073 * inner IP ID field, so strip TSO if MANGLEID is not supported.
10074 * IPsec offoad sets skb->encapsulation but still can handle
10075 * the TSO, so it's the exception.
10077 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10078 #ifdef CONFIG_IXGBE_IPSEC
10079 if (!secpath_exists(skb))
10081 features &= ~NETIF_F_TSO;
10087 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10089 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10090 struct ixgbe_adapter *adapter = netdev_priv(dev);
10091 struct bpf_prog *old_prog;
10094 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10097 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10100 /* verify ixgbe ring attributes are sufficient for XDP */
10101 for (i = 0; i < adapter->num_rx_queues; i++) {
10102 struct ixgbe_ring *ring = adapter->rx_ring[i];
10104 if (ring_is_rsc_enabled(ring))
10107 if (frame_size > ixgbe_rx_bufsz(ring))
10111 if (nr_cpu_ids > MAX_XDP_QUEUES)
10114 old_prog = xchg(&adapter->xdp_prog, prog);
10115 need_reset = (!!prog != !!old_prog);
10117 /* If transitioning XDP modes reconfigure rings */
10122 /* Wait until ndo_xsk_wakeup completes. */
10124 err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10127 rcu_assign_pointer(adapter->xdp_prog, old_prog);
10131 for (i = 0; i < adapter->num_rx_queues; i++)
10132 (void)xchg(&adapter->rx_ring[i]->xdp_prog,
10133 adapter->xdp_prog);
10137 bpf_prog_put(old_prog);
10139 /* Kick start the NAPI context if there is an AF_XDP socket open
10140 * on that queue id. This so that receiving will start.
10142 if (need_reset && prog)
10143 for (i = 0; i < adapter->num_rx_queues; i++)
10144 if (adapter->xdp_ring[i]->xsk_pool)
10145 (void)ixgbe_xsk_wakeup(adapter->netdev, i,
10151 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10153 struct ixgbe_adapter *adapter = netdev_priv(dev);
10155 switch (xdp->command) {
10156 case XDP_SETUP_PROG:
10157 return ixgbe_xdp_setup(dev, xdp->prog);
10158 case XDP_SETUP_XSK_POOL:
10159 return ixgbe_xsk_pool_setup(adapter, xdp->xsk.pool,
10160 xdp->xsk.queue_id);
10167 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10169 /* Force memory writes to complete before letting h/w know there
10170 * are new descriptors to fetch.
10173 writel(ring->next_to_use, ring->tail);
10176 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10177 struct xdp_frame **frames, u32 flags)
10179 struct ixgbe_adapter *adapter = netdev_priv(dev);
10180 struct ixgbe_ring *ring;
10184 if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10187 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10190 /* During program transitions its possible adapter->xdp_prog is assigned
10191 * but ring has not been configured yet. In this case simply abort xmit.
10193 ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10194 if (unlikely(!ring))
10197 if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10200 for (i = 0; i < n; i++) {
10201 struct xdp_frame *xdpf = frames[i];
10204 err = ixgbe_xmit_xdp_ring(adapter, xdpf);
10205 if (err != IXGBE_XDP_TX) {
10206 xdp_return_frame_rx_napi(xdpf);
10211 if (unlikely(flags & XDP_XMIT_FLUSH))
10212 ixgbe_xdp_ring_update_tail(ring);
10217 static const struct net_device_ops ixgbe_netdev_ops = {
10218 .ndo_open = ixgbe_open,
10219 .ndo_stop = ixgbe_close,
10220 .ndo_start_xmit = ixgbe_xmit_frame,
10221 .ndo_set_rx_mode = ixgbe_set_rx_mode,
10222 .ndo_validate_addr = eth_validate_addr,
10223 .ndo_set_mac_address = ixgbe_set_mac,
10224 .ndo_change_mtu = ixgbe_change_mtu,
10225 .ndo_tx_timeout = ixgbe_tx_timeout,
10226 .ndo_set_tx_maxrate = ixgbe_tx_maxrate,
10227 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
10228 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
10229 .ndo_do_ioctl = ixgbe_ioctl,
10230 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
10231 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
10232 .ndo_set_vf_rate = ixgbe_ndo_set_vf_bw,
10233 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
10234 .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10235 .ndo_set_vf_trust = ixgbe_ndo_set_vf_trust,
10236 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
10237 .ndo_get_stats64 = ixgbe_get_stats64,
10238 .ndo_setup_tc = __ixgbe_setup_tc,
10240 .ndo_select_queue = ixgbe_select_queue,
10241 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10242 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10243 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10244 .ndo_fcoe_enable = ixgbe_fcoe_enable,
10245 .ndo_fcoe_disable = ixgbe_fcoe_disable,
10246 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10247 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10248 #endif /* IXGBE_FCOE */
10249 .ndo_set_features = ixgbe_set_features,
10250 .ndo_fix_features = ixgbe_fix_features,
10251 .ndo_fdb_add = ixgbe_ndo_fdb_add,
10252 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
10253 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
10254 .ndo_dfwd_add_station = ixgbe_fwd_add,
10255 .ndo_dfwd_del_station = ixgbe_fwd_del,
10256 .ndo_udp_tunnel_add = udp_tunnel_nic_add_port,
10257 .ndo_udp_tunnel_del = udp_tunnel_nic_del_port,
10258 .ndo_features_check = ixgbe_features_check,
10259 .ndo_bpf = ixgbe_xdp,
10260 .ndo_xdp_xmit = ixgbe_xdp_xmit,
10261 .ndo_xsk_wakeup = ixgbe_xsk_wakeup,
10264 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10265 struct ixgbe_ring *tx_ring)
10267 unsigned long wait_delay, delay_interval;
10268 struct ixgbe_hw *hw = &adapter->hw;
10269 u8 reg_idx = tx_ring->reg_idx;
10273 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10275 /* delay mechanism from ixgbe_disable_tx */
10276 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10278 wait_loop = IXGBE_MAX_RX_DESC_POLL;
10279 wait_delay = delay_interval;
10281 while (wait_loop--) {
10282 usleep_range(wait_delay, wait_delay + 10);
10283 wait_delay += delay_interval * 2;
10284 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10286 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10290 e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10293 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10294 struct ixgbe_ring *tx_ring)
10296 set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10297 ixgbe_disable_txr_hw(adapter, tx_ring);
10300 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10301 struct ixgbe_ring *rx_ring)
10303 unsigned long wait_delay, delay_interval;
10304 struct ixgbe_hw *hw = &adapter->hw;
10305 u8 reg_idx = rx_ring->reg_idx;
10309 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10310 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10311 rxdctl |= IXGBE_RXDCTL_SWFLSH;
10313 /* write value back with RXDCTL.ENABLE bit cleared */
10314 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10316 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10317 if (hw->mac.type == ixgbe_mac_82598EB &&
10318 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10321 /* delay mechanism from ixgbe_disable_rx */
10322 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10324 wait_loop = IXGBE_MAX_RX_DESC_POLL;
10325 wait_delay = delay_interval;
10327 while (wait_loop--) {
10328 usleep_range(wait_delay, wait_delay + 10);
10329 wait_delay += delay_interval * 2;
10330 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10332 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10336 e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10339 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10341 memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10342 memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10345 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10347 memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10348 memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10352 * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10353 * @adapter: adapter structure
10354 * @ring: ring index
10356 * This function disables a certain Rx/Tx/XDP Tx ring. The function
10357 * assumes that the netdev is running.
10359 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10361 struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10363 rx_ring = adapter->rx_ring[ring];
10364 tx_ring = adapter->tx_ring[ring];
10365 xdp_ring = adapter->xdp_ring[ring];
10367 ixgbe_disable_txr(adapter, tx_ring);
10369 ixgbe_disable_txr(adapter, xdp_ring);
10370 ixgbe_disable_rxr_hw(adapter, rx_ring);
10375 /* Rx/Tx/XDP Tx share the same napi context. */
10376 napi_disable(&rx_ring->q_vector->napi);
10378 ixgbe_clean_tx_ring(tx_ring);
10380 ixgbe_clean_tx_ring(xdp_ring);
10381 ixgbe_clean_rx_ring(rx_ring);
10383 ixgbe_reset_txr_stats(tx_ring);
10385 ixgbe_reset_txr_stats(xdp_ring);
10386 ixgbe_reset_rxr_stats(rx_ring);
10390 * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10391 * @adapter: adapter structure
10392 * @ring: ring index
10394 * This function enables a certain Rx/Tx/XDP Tx ring. The function
10395 * assumes that the netdev is running.
10397 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10399 struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10401 rx_ring = adapter->rx_ring[ring];
10402 tx_ring = adapter->tx_ring[ring];
10403 xdp_ring = adapter->xdp_ring[ring];
10405 /* Rx/Tx/XDP Tx share the same napi context. */
10406 napi_enable(&rx_ring->q_vector->napi);
10408 ixgbe_configure_tx_ring(adapter, tx_ring);
10410 ixgbe_configure_tx_ring(adapter, xdp_ring);
10411 ixgbe_configure_rx_ring(adapter, rx_ring);
10413 clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10415 clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10419 * ixgbe_enumerate_functions - Get the number of ports this device has
10420 * @adapter: adapter structure
10422 * This function enumerates the phsyical functions co-located on a single slot,
10423 * in order to determine how many ports a device has. This is most useful in
10424 * determining the required GT/s of PCIe bandwidth necessary for optimal
10427 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10429 struct pci_dev *entry, *pdev = adapter->pdev;
10432 /* Some cards can not use the generic count PCIe functions method,
10433 * because they are behind a parent switch, so we hardcode these with
10434 * the correct number of functions.
10436 if (ixgbe_pcie_from_parent(&adapter->hw))
10439 list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10440 /* don't count virtual functions */
10441 if (entry->is_virtfn)
10444 /* When the devices on the bus don't all match our device ID,
10445 * we can't reliably determine the correct number of
10446 * functions. This can occur if a function has been direct
10447 * attached to a virtual machine using VT-d, for example. In
10448 * this case, simply return -1 to indicate this.
10450 if ((entry->vendor != pdev->vendor) ||
10451 (entry->device != pdev->device))
10461 * ixgbe_wol_supported - Check whether device supports WoL
10462 * @adapter: the adapter private structure
10463 * @device_id: the device ID
10464 * @subdevice_id: the subsystem device ID
10466 * This function is used by probe and ethtool to determine
10467 * which devices have WoL support
10470 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10473 struct ixgbe_hw *hw = &adapter->hw;
10474 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10476 /* WOL not supported on 82598 */
10477 if (hw->mac.type == ixgbe_mac_82598EB)
10480 /* check eeprom to see if WOL is enabled for X540 and newer */
10481 if (hw->mac.type >= ixgbe_mac_X540) {
10482 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10483 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10484 (hw->bus.func == 0)))
10488 /* WOL is determined based on device IDs for 82599 MACs */
10489 switch (device_id) {
10490 case IXGBE_DEV_ID_82599_SFP:
10491 /* Only these subdevices could supports WOL */
10492 switch (subdevice_id) {
10493 case IXGBE_SUBDEV_ID_82599_560FLR:
10494 case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10495 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10496 case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10497 /* only support first port */
10498 if (hw->bus.func != 0)
10501 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10502 case IXGBE_SUBDEV_ID_82599_SFP:
10503 case IXGBE_SUBDEV_ID_82599_RNDC:
10504 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10505 case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10506 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10507 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10511 case IXGBE_DEV_ID_82599EN_SFP:
10512 /* Only these subdevices support WOL */
10513 switch (subdevice_id) {
10514 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10518 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10519 /* All except this subdevice support WOL */
10520 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10523 case IXGBE_DEV_ID_82599_KX4:
10533 * ixgbe_set_fw_version - Set FW version
10534 * @adapter: the adapter private structure
10536 * This function is used by probe and ethtool to determine the FW version to
10537 * format to display. The FW version is taken from the EEPROM/NVM.
10539 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10541 struct ixgbe_hw *hw = &adapter->hw;
10542 struct ixgbe_nvm_version nvm_ver;
10544 ixgbe_get_oem_prod_version(hw, &nvm_ver);
10545 if (nvm_ver.oem_valid) {
10546 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10547 "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10548 nvm_ver.oem_release);
10552 ixgbe_get_etk_id(hw, &nvm_ver);
10553 ixgbe_get_orom_version(hw, &nvm_ver);
10555 if (nvm_ver.or_valid) {
10556 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10557 "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10558 nvm_ver.or_build, nvm_ver.or_patch);
10562 /* Set ETrack ID format */
10563 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10564 "0x%08x", nvm_ver.etk_id);
10568 * ixgbe_probe - Device Initialization Routine
10569 * @pdev: PCI device information struct
10570 * @ent: entry in ixgbe_pci_tbl
10572 * Returns 0 on success, negative on failure
10574 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10575 * The OS initialization, configuring of the adapter private structure,
10576 * and a hardware reset occur.
10578 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10580 struct net_device *netdev;
10581 struct ixgbe_adapter *adapter = NULL;
10582 struct ixgbe_hw *hw;
10583 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10584 int i, err, pci_using_dac, expected_gts;
10585 unsigned int indices = MAX_TX_QUEUES;
10586 u8 part_str[IXGBE_PBANUM_LENGTH];
10587 bool disable_dev = false;
10593 /* Catch broken hardware that put the wrong VF device ID in
10594 * the PCIe SR-IOV capability.
10596 if (pdev->is_virtfn) {
10597 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10598 pci_name(pdev), pdev->vendor, pdev->device);
10602 err = pci_enable_device_mem(pdev);
10606 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10609 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10611 dev_err(&pdev->dev,
10612 "No usable DMA configuration, aborting\n");
10618 err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10620 dev_err(&pdev->dev,
10621 "pci_request_selected_regions failed 0x%x\n", err);
10625 pci_enable_pcie_error_reporting(pdev);
10627 pci_set_master(pdev);
10628 pci_save_state(pdev);
10630 if (ii->mac == ixgbe_mac_82598EB) {
10631 #ifdef CONFIG_IXGBE_DCB
10632 /* 8 TC w/ 4 queues per TC */
10633 indices = 4 * MAX_TRAFFIC_CLASS;
10635 indices = IXGBE_MAX_RSS_INDICES;
10639 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10642 goto err_alloc_etherdev;
10645 SET_NETDEV_DEV(netdev, &pdev->dev);
10647 adapter = netdev_priv(netdev);
10649 adapter->netdev = netdev;
10650 adapter->pdev = pdev;
10652 hw->back = adapter;
10653 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10655 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10656 pci_resource_len(pdev, 0));
10657 adapter->io_addr = hw->hw_addr;
10658 if (!hw->hw_addr) {
10663 netdev->netdev_ops = &ixgbe_netdev_ops;
10664 ixgbe_set_ethtool_ops(netdev);
10665 netdev->watchdog_timeo = 5 * HZ;
10666 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10669 hw->mac.ops = *ii->mac_ops;
10670 hw->mac.type = ii->mac;
10671 hw->mvals = ii->mvals;
10673 hw->link.ops = *ii->link_ops;
10676 hw->eeprom.ops = *ii->eeprom_ops;
10677 eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10678 if (ixgbe_removed(hw->hw_addr)) {
10682 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10683 if (!(eec & BIT(8)))
10684 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10687 hw->phy.ops = *ii->phy_ops;
10688 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10689 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
10690 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10691 hw->phy.mdio.mmds = 0;
10692 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10693 hw->phy.mdio.dev = netdev;
10694 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10695 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10697 /* setup the private structure */
10698 err = ixgbe_sw_init(adapter, ii);
10702 switch (adapter->hw.mac.type) {
10703 case ixgbe_mac_X550:
10704 case ixgbe_mac_X550EM_x:
10705 netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550;
10707 case ixgbe_mac_x550em_a:
10708 netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550em_a;
10714 /* Make sure the SWFW semaphore is in a valid state */
10715 if (hw->mac.ops.init_swfw_sync)
10716 hw->mac.ops.init_swfw_sync(hw);
10718 /* Make it possible the adapter to be woken up via WOL */
10719 switch (adapter->hw.mac.type) {
10720 case ixgbe_mac_82599EB:
10721 case ixgbe_mac_X540:
10722 case ixgbe_mac_X550:
10723 case ixgbe_mac_X550EM_x:
10724 case ixgbe_mac_x550em_a:
10725 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10732 * If there is a fan on this device and it has failed log the
10735 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10736 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10737 if (esdp & IXGBE_ESDP_SDP1)
10738 e_crit(probe, "Fan has stopped, replace the adapter\n");
10741 if (allow_unsupported_sfp)
10742 hw->allow_unsupported_sfp = allow_unsupported_sfp;
10744 /* reset_hw fills in the perm_addr as well */
10745 hw->phy.reset_if_overtemp = true;
10746 err = hw->mac.ops.reset_hw(hw);
10747 hw->phy.reset_if_overtemp = false;
10748 ixgbe_set_eee_capable(adapter);
10749 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10751 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10752 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10753 e_dev_err("Reload the driver after installing a supported module.\n");
10756 e_dev_err("HW Init failed: %d\n", err);
10760 #ifdef CONFIG_PCI_IOV
10761 /* SR-IOV not supported on the 82598 */
10762 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10765 ixgbe_init_mbx_params_pf(hw);
10766 hw->mbx.ops = ii->mbx_ops;
10767 pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10768 ixgbe_enable_sriov(adapter, max_vfs);
10772 netdev->features = NETIF_F_SG |
10779 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10780 NETIF_F_GSO_GRE_CSUM | \
10781 NETIF_F_GSO_IPXIP4 | \
10782 NETIF_F_GSO_IPXIP6 | \
10783 NETIF_F_GSO_UDP_TUNNEL | \
10784 NETIF_F_GSO_UDP_TUNNEL_CSUM)
10786 netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10787 netdev->features |= NETIF_F_GSO_PARTIAL |
10788 IXGBE_GSO_PARTIAL_FEATURES;
10790 if (hw->mac.type >= ixgbe_mac_82599EB)
10791 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
10793 #ifdef CONFIG_IXGBE_IPSEC
10794 #define IXGBE_ESP_FEATURES (NETIF_F_HW_ESP | \
10795 NETIF_F_HW_ESP_TX_CSUM | \
10798 if (adapter->ipsec)
10799 netdev->features |= IXGBE_ESP_FEATURES;
10801 /* copy netdev features into list of user selectable features */
10802 netdev->hw_features |= netdev->features |
10803 NETIF_F_HW_VLAN_CTAG_FILTER |
10804 NETIF_F_HW_VLAN_CTAG_RX |
10805 NETIF_F_HW_VLAN_CTAG_TX |
10807 NETIF_F_HW_L2FW_DOFFLOAD;
10809 if (hw->mac.type >= ixgbe_mac_82599EB)
10810 netdev->hw_features |= NETIF_F_NTUPLE |
10814 netdev->features |= NETIF_F_HIGHDMA;
10816 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10817 netdev->hw_enc_features |= netdev->vlan_features;
10818 netdev->mpls_features |= NETIF_F_SG |
10822 netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10824 /* set this bit last since it cannot be part of vlan_features */
10825 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10826 NETIF_F_HW_VLAN_CTAG_RX |
10827 NETIF_F_HW_VLAN_CTAG_TX;
10829 netdev->priv_flags |= IFF_UNICAST_FLT;
10830 netdev->priv_flags |= IFF_SUPP_NOFCS;
10832 /* MTU range: 68 - 9710 */
10833 netdev->min_mtu = ETH_MIN_MTU;
10834 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
10836 #ifdef CONFIG_IXGBE_DCB
10837 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
10838 netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
10842 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
10843 unsigned int fcoe_l;
10845 if (hw->mac.ops.get_device_caps) {
10846 hw->mac.ops.get_device_caps(hw, &device_caps);
10847 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
10848 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
10852 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
10853 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
10855 netdev->features |= NETIF_F_FSO |
10858 netdev->vlan_features |= NETIF_F_FSO |
10862 #endif /* IXGBE_FCOE */
10863 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
10864 netdev->hw_features |= NETIF_F_LRO;
10865 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
10866 netdev->features |= NETIF_F_LRO;
10868 if (ixgbe_check_fw_error(adapter)) {
10873 /* make sure the EEPROM is good */
10874 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
10875 e_dev_err("The EEPROM Checksum Is Not Valid\n");
10880 eth_platform_get_mac_address(&adapter->pdev->dev,
10881 adapter->hw.mac.perm_addr);
10883 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
10885 if (!is_valid_ether_addr(netdev->dev_addr)) {
10886 e_dev_err("invalid MAC address\n");
10891 /* Set hw->mac.addr to permanent MAC address */
10892 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
10893 ixgbe_mac_set_default_filter(adapter);
10895 timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
10897 if (ixgbe_removed(hw->hw_addr)) {
10901 INIT_WORK(&adapter->service_task, ixgbe_service_task);
10902 set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
10903 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
10905 err = ixgbe_init_interrupt_scheme(adapter);
10909 for (i = 0; i < adapter->num_rx_queues; i++)
10910 u64_stats_init(&adapter->rx_ring[i]->syncp);
10911 for (i = 0; i < adapter->num_tx_queues; i++)
10912 u64_stats_init(&adapter->tx_ring[i]->syncp);
10913 for (i = 0; i < adapter->num_xdp_queues; i++)
10914 u64_stats_init(&adapter->xdp_ring[i]->syncp);
10916 /* WOL not supported for all devices */
10918 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
10919 hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
10920 pdev->subsystem_device);
10921 if (hw->wol_enabled)
10922 adapter->wol = IXGBE_WUFC_MAG;
10924 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
10926 /* save off EEPROM version number */
10927 ixgbe_set_fw_version(adapter);
10929 /* pick up the PCI bus settings for reporting later */
10930 if (ixgbe_pcie_from_parent(hw))
10931 ixgbe_get_parent_bus_info(adapter);
10933 hw->mac.ops.get_bus_info(hw);
10935 /* calculate the expected PCIe bandwidth required for optimal
10936 * performance. Note that some older parts will never have enough
10937 * bandwidth due to being older generation PCIe parts. We clamp these
10938 * parts to ensure no warning is displayed if it can't be fixed.
10940 switch (hw->mac.type) {
10941 case ixgbe_mac_82598EB:
10942 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
10945 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
10949 /* don't check link if we failed to enumerate functions */
10950 if (expected_gts > 0)
10951 ixgbe_check_minimum_link(adapter, expected_gts);
10953 err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
10955 strlcpy(part_str, "Unknown", sizeof(part_str));
10956 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
10957 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
10958 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
10961 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
10962 hw->mac.type, hw->phy.type, part_str);
10964 e_dev_info("%pM\n", netdev->dev_addr);
10966 /* reset the hardware with the new settings */
10967 err = hw->mac.ops.start_hw(hw);
10968 if (err == IXGBE_ERR_EEPROM_VERSION) {
10969 /* We are running on a pre-production device, log a warning */
10970 e_dev_warn("This device is a pre-production adapter/LOM. "
10971 "Please be aware there may be issues associated "
10972 "with your hardware. If you are experiencing "
10973 "problems please contact your Intel or hardware "
10974 "representative who provided you with this "
10977 strcpy(netdev->name, "eth%d");
10978 pci_set_drvdata(pdev, adapter);
10979 err = register_netdev(netdev);
10984 /* power down the optics for 82599 SFP+ fiber */
10985 if (hw->mac.ops.disable_tx_laser)
10986 hw->mac.ops.disable_tx_laser(hw);
10988 /* carrier off reporting is important to ethtool even BEFORE open */
10989 netif_carrier_off(netdev);
10991 #ifdef CONFIG_IXGBE_DCA
10992 if (dca_add_requester(&pdev->dev) == 0) {
10993 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
10994 ixgbe_setup_dca(adapter);
10997 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
10998 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
10999 for (i = 0; i < adapter->num_vfs; i++)
11000 ixgbe_vf_configuration(pdev, (i | 0x10000000));
11003 /* firmware requires driver version to be 0xFFFFFFFF
11004 * since os does not support feature
11006 if (hw->mac.ops.set_fw_drv_ver)
11007 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11008 sizeof(UTS_RELEASE) - 1,
11011 /* add san mac addr to netdev */
11012 ixgbe_add_sanmac_netdev(netdev);
11014 e_dev_info("%s\n", ixgbe_default_device_descr);
11016 #ifdef CONFIG_IXGBE_HWMON
11017 if (ixgbe_sysfs_init(adapter))
11018 e_err(probe, "failed to allocate sysfs resources\n");
11019 #endif /* CONFIG_IXGBE_HWMON */
11021 ixgbe_dbg_adapter_init(adapter);
11023 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11024 if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11025 hw->mac.ops.setup_link(hw,
11026 IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11029 err = ixgbe_mii_bus_init(hw);
11036 unregister_netdev(netdev);
11038 ixgbe_release_hw_control(adapter);
11039 ixgbe_clear_interrupt_scheme(adapter);
11041 ixgbe_disable_sriov(adapter);
11042 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11043 iounmap(adapter->io_addr);
11044 kfree(adapter->jump_tables[0]);
11045 kfree(adapter->mac_table);
11046 kfree(adapter->rss_key);
11047 bitmap_free(adapter->af_xdp_zc_qps);
11049 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11050 free_netdev(netdev);
11051 err_alloc_etherdev:
11052 pci_release_mem_regions(pdev);
11055 if (!adapter || disable_dev)
11056 pci_disable_device(pdev);
11061 * ixgbe_remove - Device Removal Routine
11062 * @pdev: PCI device information struct
11064 * ixgbe_remove is called by the PCI subsystem to alert the driver
11065 * that it should release a PCI device. The could be caused by a
11066 * Hot-Plug event, or because the driver is going to be removed from
11069 static void ixgbe_remove(struct pci_dev *pdev)
11071 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11072 struct net_device *netdev;
11076 /* if !adapter then we already cleaned up in probe */
11080 netdev = adapter->netdev;
11081 ixgbe_dbg_adapter_exit(adapter);
11083 set_bit(__IXGBE_REMOVING, &adapter->state);
11084 cancel_work_sync(&adapter->service_task);
11086 if (adapter->mii_bus)
11087 mdiobus_unregister(adapter->mii_bus);
11089 #ifdef CONFIG_IXGBE_DCA
11090 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11091 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11092 dca_remove_requester(&pdev->dev);
11093 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11094 IXGBE_DCA_CTRL_DCA_DISABLE);
11098 #ifdef CONFIG_IXGBE_HWMON
11099 ixgbe_sysfs_exit(adapter);
11100 #endif /* CONFIG_IXGBE_HWMON */
11102 /* remove the added san mac */
11103 ixgbe_del_sanmac_netdev(netdev);
11105 #ifdef CONFIG_PCI_IOV
11106 ixgbe_disable_sriov(adapter);
11108 if (netdev->reg_state == NETREG_REGISTERED)
11109 unregister_netdev(netdev);
11111 ixgbe_stop_ipsec_offload(adapter);
11112 ixgbe_clear_interrupt_scheme(adapter);
11114 ixgbe_release_hw_control(adapter);
11117 kfree(adapter->ixgbe_ieee_pfc);
11118 kfree(adapter->ixgbe_ieee_ets);
11121 iounmap(adapter->io_addr);
11122 pci_release_mem_regions(pdev);
11124 e_dev_info("complete\n");
11126 for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11127 if (adapter->jump_tables[i]) {
11128 kfree(adapter->jump_tables[i]->input);
11129 kfree(adapter->jump_tables[i]->mask);
11131 kfree(adapter->jump_tables[i]);
11134 kfree(adapter->mac_table);
11135 kfree(adapter->rss_key);
11136 bitmap_free(adapter->af_xdp_zc_qps);
11137 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11138 free_netdev(netdev);
11140 pci_disable_pcie_error_reporting(pdev);
11143 pci_disable_device(pdev);
11147 * ixgbe_io_error_detected - called when PCI error is detected
11148 * @pdev: Pointer to PCI device
11149 * @state: The current pci connection state
11151 * This function is called after a PCI bus error affecting
11152 * this device has been detected.
11154 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11155 pci_channel_state_t state)
11157 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11158 struct net_device *netdev = adapter->netdev;
11160 #ifdef CONFIG_PCI_IOV
11161 struct ixgbe_hw *hw = &adapter->hw;
11162 struct pci_dev *bdev, *vfdev;
11163 u32 dw0, dw1, dw2, dw3;
11165 u16 req_id, pf_func;
11167 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11168 adapter->num_vfs == 0)
11169 goto skip_bad_vf_detection;
11171 bdev = pdev->bus->self;
11172 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11173 bdev = bdev->bus->self;
11176 goto skip_bad_vf_detection;
11178 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11180 goto skip_bad_vf_detection;
11182 dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11183 dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11184 dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11185 dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11186 if (ixgbe_removed(hw->hw_addr))
11187 goto skip_bad_vf_detection;
11189 req_id = dw1 >> 16;
11190 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11191 if (!(req_id & 0x0080))
11192 goto skip_bad_vf_detection;
11194 pf_func = req_id & 0x01;
11195 if ((pf_func & 1) == (pdev->devfn & 1)) {
11196 unsigned int device_id;
11198 vf = (req_id & 0x7F) >> 1;
11199 e_dev_err("VF %d has caused a PCIe error\n", vf);
11200 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11201 "%8.8x\tdw3: %8.8x\n",
11202 dw0, dw1, dw2, dw3);
11203 switch (adapter->hw.mac.type) {
11204 case ixgbe_mac_82599EB:
11205 device_id = IXGBE_82599_VF_DEVICE_ID;
11207 case ixgbe_mac_X540:
11208 device_id = IXGBE_X540_VF_DEVICE_ID;
11210 case ixgbe_mac_X550:
11211 device_id = IXGBE_DEV_ID_X550_VF;
11213 case ixgbe_mac_X550EM_x:
11214 device_id = IXGBE_DEV_ID_X550EM_X_VF;
11216 case ixgbe_mac_x550em_a:
11217 device_id = IXGBE_DEV_ID_X550EM_A_VF;
11224 /* Find the pci device of the offending VF */
11225 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11227 if (vfdev->devfn == (req_id & 0xFF))
11229 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11233 * There's a slim chance the VF could have been hot plugged,
11234 * so if it is no longer present we don't need to issue the
11235 * VFLR. Just clean up the AER in that case.
11239 /* Free device reference count */
11240 pci_dev_put(vfdev);
11245 * Even though the error may have occurred on the other port
11246 * we still need to increment the vf error reference count for
11247 * both ports because the I/O resume function will be called
11248 * for both of them.
11250 adapter->vferr_refcount++;
11252 return PCI_ERS_RESULT_RECOVERED;
11254 skip_bad_vf_detection:
11255 #endif /* CONFIG_PCI_IOV */
11256 if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11257 return PCI_ERS_RESULT_DISCONNECT;
11259 if (!netif_device_present(netdev))
11260 return PCI_ERS_RESULT_DISCONNECT;
11263 netif_device_detach(netdev);
11265 if (netif_running(netdev))
11266 ixgbe_close_suspend(adapter);
11268 if (state == pci_channel_io_perm_failure) {
11270 return PCI_ERS_RESULT_DISCONNECT;
11273 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11274 pci_disable_device(pdev);
11277 /* Request a slot reset. */
11278 return PCI_ERS_RESULT_NEED_RESET;
11282 * ixgbe_io_slot_reset - called after the pci bus has been reset.
11283 * @pdev: Pointer to PCI device
11285 * Restart the card from scratch, as if from a cold-boot.
11287 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11289 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11290 pci_ers_result_t result;
11292 if (pci_enable_device_mem(pdev)) {
11293 e_err(probe, "Cannot re-enable PCI device after reset.\n");
11294 result = PCI_ERS_RESULT_DISCONNECT;
11296 smp_mb__before_atomic();
11297 clear_bit(__IXGBE_DISABLED, &adapter->state);
11298 adapter->hw.hw_addr = adapter->io_addr;
11299 pci_set_master(pdev);
11300 pci_restore_state(pdev);
11301 pci_save_state(pdev);
11303 pci_wake_from_d3(pdev, false);
11305 ixgbe_reset(adapter);
11306 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11307 result = PCI_ERS_RESULT_RECOVERED;
11314 * ixgbe_io_resume - called when traffic can start flowing again.
11315 * @pdev: Pointer to PCI device
11317 * This callback is called when the error recovery driver tells us that
11318 * its OK to resume normal operation.
11320 static void ixgbe_io_resume(struct pci_dev *pdev)
11322 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11323 struct net_device *netdev = adapter->netdev;
11325 #ifdef CONFIG_PCI_IOV
11326 if (adapter->vferr_refcount) {
11327 e_info(drv, "Resuming after VF err\n");
11328 adapter->vferr_refcount--;
11334 if (netif_running(netdev))
11335 ixgbe_open(netdev);
11337 netif_device_attach(netdev);
11341 static const struct pci_error_handlers ixgbe_err_handler = {
11342 .error_detected = ixgbe_io_error_detected,
11343 .slot_reset = ixgbe_io_slot_reset,
11344 .resume = ixgbe_io_resume,
11347 static SIMPLE_DEV_PM_OPS(ixgbe_pm_ops, ixgbe_suspend, ixgbe_resume);
11349 static struct pci_driver ixgbe_driver = {
11350 .name = ixgbe_driver_name,
11351 .id_table = ixgbe_pci_tbl,
11352 .probe = ixgbe_probe,
11353 .remove = ixgbe_remove,
11354 .driver.pm = &ixgbe_pm_ops,
11355 .shutdown = ixgbe_shutdown,
11356 .sriov_configure = ixgbe_pci_sriov_configure,
11357 .err_handler = &ixgbe_err_handler
11361 * ixgbe_init_module - Driver Registration Routine
11363 * ixgbe_init_module is the first routine called when the driver is
11364 * loaded. All it does is register with the PCI subsystem.
11366 static int __init ixgbe_init_module(void)
11369 pr_info("%s\n", ixgbe_driver_string);
11370 pr_info("%s\n", ixgbe_copyright);
11372 ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11374 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11380 ret = pci_register_driver(&ixgbe_driver);
11382 destroy_workqueue(ixgbe_wq);
11387 #ifdef CONFIG_IXGBE_DCA
11388 dca_register_notify(&dca_notifier);
11394 module_init(ixgbe_init_module);
11397 * ixgbe_exit_module - Driver Exit Cleanup Routine
11399 * ixgbe_exit_module is called just before the driver is removed
11402 static void __exit ixgbe_exit_module(void)
11404 #ifdef CONFIG_IXGBE_DCA
11405 dca_unregister_notify(&dca_notifier);
11407 pci_unregister_driver(&ixgbe_driver);
11411 destroy_workqueue(ixgbe_wq);
11416 #ifdef CONFIG_IXGBE_DCA
11417 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11422 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11423 __ixgbe_notify_dca);
11425 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11428 #endif /* CONFIG_IXGBE_DCA */
11430 module_exit(ixgbe_exit_module);