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_pcie_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 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 = rx_ring->rx_offset;
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 bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer,
1944 int rx_buffer_pgcnt)
1946 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1947 struct page *page = rx_buffer->page;
1949 /* avoid re-using remote and pfmemalloc pages */
1950 if (!dev_page_is_reusable(page))
1953 #if (PAGE_SIZE < 8192)
1954 /* if we are only owner of page we can reuse it */
1955 if (unlikely((rx_buffer_pgcnt - pagecnt_bias) > 1))
1958 /* The last offset is a bit aggressive in that we assume the
1959 * worst case of FCoE being enabled and using a 3K buffer.
1960 * However this should have minimal impact as the 1K extra is
1961 * still less than one buffer in size.
1963 #define IXGBE_LAST_OFFSET \
1964 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1965 if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1969 /* If we have drained the page fragment pool we need to update
1970 * the pagecnt_bias and page count so that we fully restock the
1971 * number of references the driver holds.
1973 if (unlikely(pagecnt_bias == 1)) {
1974 page_ref_add(page, USHRT_MAX - 1);
1975 rx_buffer->pagecnt_bias = USHRT_MAX;
1982 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1983 * @rx_ring: rx descriptor ring to transact packets on
1984 * @rx_buffer: buffer containing page to add
1985 * @skb: sk_buff to place the data into
1986 * @size: size of data in rx_buffer
1988 * This function will add the data contained in rx_buffer->page to the skb.
1989 * This is done either through a direct copy if the data in the buffer is
1990 * less than the skb header size, otherwise it will just attach the page as
1991 * a frag to the skb.
1993 * The function will then update the page offset if necessary and return
1994 * true if the buffer can be reused by the adapter.
1996 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1997 struct ixgbe_rx_buffer *rx_buffer,
1998 struct sk_buff *skb,
2001 #if (PAGE_SIZE < 8192)
2002 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2004 unsigned int truesize = rx_ring->rx_offset ?
2005 SKB_DATA_ALIGN(rx_ring->rx_offset + size) :
2006 SKB_DATA_ALIGN(size);
2008 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2009 rx_buffer->page_offset, size, truesize);
2010 #if (PAGE_SIZE < 8192)
2011 rx_buffer->page_offset ^= truesize;
2013 rx_buffer->page_offset += truesize;
2017 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2018 union ixgbe_adv_rx_desc *rx_desc,
2019 struct sk_buff **skb,
2020 const unsigned int size,
2021 int *rx_buffer_pgcnt)
2023 struct ixgbe_rx_buffer *rx_buffer;
2025 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2027 #if (PAGE_SIZE < 8192)
2028 page_count(rx_buffer->page);
2032 prefetchw(rx_buffer->page);
2033 *skb = rx_buffer->skb;
2035 /* Delay unmapping of the first packet. It carries the header
2036 * information, HW may still access the header after the writeback.
2037 * Only unmap it when EOP is reached
2039 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2044 ixgbe_dma_sync_frag(rx_ring, *skb);
2047 /* we are reusing so sync this buffer for CPU use */
2048 dma_sync_single_range_for_cpu(rx_ring->dev,
2050 rx_buffer->page_offset,
2054 rx_buffer->pagecnt_bias--;
2059 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2060 struct ixgbe_rx_buffer *rx_buffer,
2061 struct sk_buff *skb,
2062 int rx_buffer_pgcnt)
2064 if (ixgbe_can_reuse_rx_page(rx_buffer, rx_buffer_pgcnt)) {
2065 /* hand second half of page back to the ring */
2066 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2068 if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2069 /* the page has been released from the ring */
2070 IXGBE_CB(skb)->page_released = true;
2072 /* we are not reusing the buffer so unmap it */
2073 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2074 ixgbe_rx_pg_size(rx_ring),
2078 __page_frag_cache_drain(rx_buffer->page,
2079 rx_buffer->pagecnt_bias);
2082 /* clear contents of rx_buffer */
2083 rx_buffer->page = NULL;
2084 rx_buffer->skb = NULL;
2087 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2088 struct ixgbe_rx_buffer *rx_buffer,
2089 struct xdp_buff *xdp,
2090 union ixgbe_adv_rx_desc *rx_desc)
2092 unsigned int size = xdp->data_end - xdp->data;
2093 #if (PAGE_SIZE < 8192)
2094 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2096 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2097 xdp->data_hard_start);
2099 struct sk_buff *skb;
2101 /* prefetch first cache line of first page */
2102 net_prefetch(xdp->data);
2104 /* Note, we get here by enabling legacy-rx via:
2106 * ethtool --set-priv-flags <dev> legacy-rx on
2108 * In this mode, we currently get 0 extra XDP headroom as
2109 * opposed to having legacy-rx off, where we process XDP
2110 * packets going to stack via ixgbe_build_skb(). The latter
2111 * provides us currently with 192 bytes of headroom.
2113 * For ixgbe_construct_skb() mode it means that the
2114 * xdp->data_meta will always point to xdp->data, since
2115 * the helper cannot expand the head. Should this ever
2116 * change in future for legacy-rx mode on, then lets also
2117 * add xdp->data_meta handling here.
2120 /* allocate a skb to store the frags */
2121 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2125 if (size > IXGBE_RX_HDR_SIZE) {
2126 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2127 IXGBE_CB(skb)->dma = rx_buffer->dma;
2129 skb_add_rx_frag(skb, 0, rx_buffer->page,
2130 xdp->data - page_address(rx_buffer->page),
2132 #if (PAGE_SIZE < 8192)
2133 rx_buffer->page_offset ^= truesize;
2135 rx_buffer->page_offset += truesize;
2138 memcpy(__skb_put(skb, size),
2139 xdp->data, ALIGN(size, sizeof(long)));
2140 rx_buffer->pagecnt_bias++;
2146 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2147 struct ixgbe_rx_buffer *rx_buffer,
2148 struct xdp_buff *xdp,
2149 union ixgbe_adv_rx_desc *rx_desc)
2151 unsigned int metasize = xdp->data - xdp->data_meta;
2152 #if (PAGE_SIZE < 8192)
2153 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2155 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2156 SKB_DATA_ALIGN(xdp->data_end -
2157 xdp->data_hard_start);
2159 struct sk_buff *skb;
2161 /* Prefetch first cache line of first page. If xdp->data_meta
2162 * is unused, this points extactly as xdp->data, otherwise we
2163 * likely have a consumer accessing first few bytes of meta
2164 * data, and then actual data.
2166 net_prefetch(xdp->data_meta);
2168 /* build an skb to around the page buffer */
2169 skb = build_skb(xdp->data_hard_start, truesize);
2173 /* update pointers within the skb to store the data */
2174 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2175 __skb_put(skb, xdp->data_end - xdp->data);
2177 skb_metadata_set(skb, metasize);
2179 /* record DMA address if this is the start of a chain of buffers */
2180 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2181 IXGBE_CB(skb)->dma = rx_buffer->dma;
2183 /* update buffer offset */
2184 #if (PAGE_SIZE < 8192)
2185 rx_buffer->page_offset ^= truesize;
2187 rx_buffer->page_offset += truesize;
2193 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2194 struct ixgbe_ring *rx_ring,
2195 struct xdp_buff *xdp)
2197 int err, result = IXGBE_XDP_PASS;
2198 struct bpf_prog *xdp_prog;
2199 struct xdp_frame *xdpf;
2203 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2208 prefetchw(xdp->data_hard_start); /* xdp_frame write */
2210 act = bpf_prog_run_xdp(xdp_prog, xdp);
2215 xdpf = xdp_convert_buff_to_frame(xdp);
2216 if (unlikely(!xdpf)) {
2217 result = IXGBE_XDP_CONSUMED;
2220 result = ixgbe_xmit_xdp_ring(adapter, xdpf);
2223 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2225 result = IXGBE_XDP_REDIR;
2227 result = IXGBE_XDP_CONSUMED;
2230 bpf_warn_invalid_xdp_action(act);
2233 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2234 fallthrough; /* handle aborts by dropping packet */
2236 result = IXGBE_XDP_CONSUMED;
2241 return ERR_PTR(-result);
2244 static unsigned int ixgbe_rx_frame_truesize(struct ixgbe_ring *rx_ring,
2247 unsigned int truesize;
2249 #if (PAGE_SIZE < 8192)
2250 truesize = ixgbe_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
2252 truesize = rx_ring->rx_offset ?
2253 SKB_DATA_ALIGN(rx_ring->rx_offset + size) +
2254 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
2255 SKB_DATA_ALIGN(size);
2260 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2261 struct ixgbe_rx_buffer *rx_buffer,
2264 unsigned int truesize = ixgbe_rx_frame_truesize(rx_ring, size);
2265 #if (PAGE_SIZE < 8192)
2266 rx_buffer->page_offset ^= truesize;
2268 rx_buffer->page_offset += truesize;
2273 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2274 * @q_vector: structure containing interrupt and ring information
2275 * @rx_ring: rx descriptor ring to transact packets on
2276 * @budget: Total limit on number of packets to process
2278 * This function provides a "bounce buffer" approach to Rx interrupt
2279 * processing. The advantage to this is that on systems that have
2280 * expensive overhead for IOMMU access this provides a means of avoiding
2281 * it by maintaining the mapping of the page to the syste.
2283 * Returns amount of work completed
2285 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2286 struct ixgbe_ring *rx_ring,
2289 unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0;
2290 struct ixgbe_adapter *adapter = q_vector->adapter;
2293 unsigned int mss = 0;
2294 #endif /* IXGBE_FCOE */
2295 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2296 unsigned int offset = rx_ring->rx_offset;
2297 unsigned int xdp_xmit = 0;
2298 struct xdp_buff xdp;
2300 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
2301 #if (PAGE_SIZE < 8192)
2302 frame_sz = ixgbe_rx_frame_truesize(rx_ring, 0);
2304 xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
2306 while (likely(total_rx_packets < budget)) {
2307 union ixgbe_adv_rx_desc *rx_desc;
2308 struct ixgbe_rx_buffer *rx_buffer;
2309 struct sk_buff *skb;
2310 int rx_buffer_pgcnt;
2313 /* return some buffers to hardware, one at a time is too slow */
2314 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2315 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2319 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2320 size = le16_to_cpu(rx_desc->wb.upper.length);
2324 /* This memory barrier is needed to keep us from reading
2325 * any other fields out of the rx_desc until we know the
2326 * descriptor has been written back
2330 rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size, &rx_buffer_pgcnt);
2332 /* retrieve a buffer from the ring */
2334 unsigned char *hard_start;
2336 hard_start = page_address(rx_buffer->page) +
2337 rx_buffer->page_offset - offset;
2338 xdp_prepare_buff(&xdp, hard_start, offset, size, true);
2339 #if (PAGE_SIZE > 4096)
2340 /* At larger PAGE_SIZE, frame_sz depend on len size */
2341 xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, size);
2343 skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2347 unsigned int xdp_res = -PTR_ERR(skb);
2349 if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2350 xdp_xmit |= xdp_res;
2351 ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2353 rx_buffer->pagecnt_bias++;
2356 total_rx_bytes += size;
2358 ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2359 } else if (ring_uses_build_skb(rx_ring)) {
2360 skb = ixgbe_build_skb(rx_ring, rx_buffer,
2363 skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2367 /* exit if we failed to retrieve a buffer */
2369 rx_ring->rx_stats.alloc_rx_buff_failed++;
2370 rx_buffer->pagecnt_bias++;
2374 ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb, rx_buffer_pgcnt);
2377 /* place incomplete frames back on ring for completion */
2378 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2381 /* verify the packet layout is correct */
2382 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2385 /* probably a little skewed due to removing CRC */
2386 total_rx_bytes += skb->len;
2388 /* populate checksum, timestamp, VLAN, and protocol */
2389 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2392 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2393 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2394 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2395 /* include DDPed FCoE data */
2396 if (ddp_bytes > 0) {
2398 mss = rx_ring->netdev->mtu -
2399 sizeof(struct fcoe_hdr) -
2400 sizeof(struct fc_frame_header) -
2401 sizeof(struct fcoe_crc_eof);
2405 total_rx_bytes += ddp_bytes;
2406 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2410 dev_kfree_skb_any(skb);
2415 #endif /* IXGBE_FCOE */
2416 ixgbe_rx_skb(q_vector, skb);
2418 /* update budget accounting */
2422 if (xdp_xmit & IXGBE_XDP_REDIR)
2425 if (xdp_xmit & IXGBE_XDP_TX) {
2426 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2428 /* Force memory writes to complete before letting h/w
2429 * know there are new descriptors to fetch.
2432 writel(ring->next_to_use, ring->tail);
2435 u64_stats_update_begin(&rx_ring->syncp);
2436 rx_ring->stats.packets += total_rx_packets;
2437 rx_ring->stats.bytes += total_rx_bytes;
2438 u64_stats_update_end(&rx_ring->syncp);
2439 q_vector->rx.total_packets += total_rx_packets;
2440 q_vector->rx.total_bytes += total_rx_bytes;
2442 return total_rx_packets;
2446 * ixgbe_configure_msix - Configure MSI-X hardware
2447 * @adapter: board private structure
2449 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2452 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2454 struct ixgbe_q_vector *q_vector;
2458 /* Populate MSIX to EITR Select */
2459 if (adapter->num_vfs > 32) {
2460 u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2461 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2465 * Populate the IVAR table and set the ITR values to the
2466 * corresponding register.
2468 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2469 struct ixgbe_ring *ring;
2470 q_vector = adapter->q_vector[v_idx];
2472 ixgbe_for_each_ring(ring, q_vector->rx)
2473 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2475 ixgbe_for_each_ring(ring, q_vector->tx)
2476 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2478 ixgbe_write_eitr(q_vector);
2481 switch (adapter->hw.mac.type) {
2482 case ixgbe_mac_82598EB:
2483 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2486 case ixgbe_mac_82599EB:
2487 case ixgbe_mac_X540:
2488 case ixgbe_mac_X550:
2489 case ixgbe_mac_X550EM_x:
2490 case ixgbe_mac_x550em_a:
2491 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2496 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2498 /* set up to autoclear timer, and the vectors */
2499 mask = IXGBE_EIMS_ENABLE_MASK;
2500 mask &= ~(IXGBE_EIMS_OTHER |
2501 IXGBE_EIMS_MAILBOX |
2504 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2508 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2509 * @q_vector: structure containing interrupt and ring information
2510 * @ring_container: structure containing ring performance data
2512 * Stores a new ITR value based on packets and byte
2513 * counts during the last interrupt. The advantage of per interrupt
2514 * computation is faster updates and more accurate ITR for the current
2515 * traffic pattern. Constants in this function were computed
2516 * based on theoretical maximum wire speed and thresholds were set based
2517 * on testing data as well as attempting to minimize response time
2518 * while increasing bulk throughput.
2520 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2521 struct ixgbe_ring_container *ring_container)
2523 unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2524 IXGBE_ITR_ADAPTIVE_LATENCY;
2525 unsigned int avg_wire_size, packets, bytes;
2526 unsigned long next_update = jiffies;
2528 /* If we don't have any rings just leave ourselves set for maximum
2529 * possible latency so we take ourselves out of the equation.
2531 if (!ring_container->ring)
2534 /* If we didn't update within up to 1 - 2 jiffies we can assume
2535 * that either packets are coming in so slow there hasn't been
2536 * any work, or that there is so much work that NAPI is dealing
2537 * with interrupt moderation and we don't need to do anything.
2539 if (time_after(next_update, ring_container->next_update))
2542 packets = ring_container->total_packets;
2544 /* We have no packets to actually measure against. This means
2545 * either one of the other queues on this vector is active or
2546 * we are a Tx queue doing TSO with too high of an interrupt rate.
2548 * When this occurs just tick up our delay by the minimum value
2549 * and hope that this extra delay will prevent us from being called
2550 * without any work on our queue.
2553 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2554 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2555 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2556 itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2560 bytes = ring_container->total_bytes;
2562 /* If packets are less than 4 or bytes are less than 9000 assume
2563 * insufficient data to use bulk rate limiting approach. We are
2564 * likely latency driven.
2566 if (packets < 4 && bytes < 9000) {
2567 itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2568 goto adjust_by_size;
2571 /* Between 4 and 48 we can assume that our current interrupt delay
2572 * is only slightly too low. As such we should increase it by a small
2576 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2577 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2578 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2582 /* Between 48 and 96 is our "goldilocks" zone where we are working
2583 * out "just right". Just report that our current ITR is good for us.
2586 itr = q_vector->itr >> 2;
2590 /* If packet count is 96 or greater we are likely looking at a slight
2591 * overrun of the delay we want. Try halving our delay to see if that
2592 * will cut the number of packets in half per interrupt.
2594 if (packets < 256) {
2595 itr = q_vector->itr >> 3;
2596 if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2597 itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2601 /* The paths below assume we are dealing with a bulk ITR since number
2602 * of packets is 256 or greater. We are just going to have to compute
2603 * a value and try to bring the count under control, though for smaller
2604 * packet sizes there isn't much we can do as NAPI polling will likely
2605 * be kicking in sooner rather than later.
2607 itr = IXGBE_ITR_ADAPTIVE_BULK;
2610 /* If packet counts are 256 or greater we can assume we have a gross
2611 * overestimation of what the rate should be. Instead of trying to fine
2612 * tune it just use the formula below to try and dial in an exact value
2613 * give the current packet size of the frame.
2615 avg_wire_size = bytes / packets;
2617 /* The following is a crude approximation of:
2618 * wmem_default / (size + overhead) = desired_pkts_per_int
2619 * rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2620 * (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2622 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2623 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2626 * (170 * (size + 24)) / (size + 640) = ITR
2628 * We first do some math on the packet size and then finally bitshift
2629 * by 8 after rounding up. We also have to account for PCIe link speed
2630 * difference as ITR scales based on this.
2632 if (avg_wire_size <= 60) {
2633 /* Start at 50k ints/sec */
2634 avg_wire_size = 5120;
2635 } else if (avg_wire_size <= 316) {
2636 /* 50K ints/sec to 16K ints/sec */
2637 avg_wire_size *= 40;
2638 avg_wire_size += 2720;
2639 } else if (avg_wire_size <= 1084) {
2640 /* 16K ints/sec to 9.2K ints/sec */
2641 avg_wire_size *= 15;
2642 avg_wire_size += 11452;
2643 } else if (avg_wire_size < 1968) {
2644 /* 9.2K ints/sec to 8K ints/sec */
2646 avg_wire_size += 22420;
2648 /* plateau at a limit of 8K ints/sec */
2649 avg_wire_size = 32256;
2652 /* If we are in low latency mode half our delay which doubles the rate
2653 * to somewhere between 100K to 16K ints/sec
2655 if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2656 avg_wire_size >>= 1;
2658 /* Resultant value is 256 times larger than it needs to be. This
2659 * gives us room to adjust the value as needed to either increase
2660 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2662 * Use addition as we have already recorded the new latency flag
2663 * for the ITR value.
2665 switch (q_vector->adapter->link_speed) {
2666 case IXGBE_LINK_SPEED_10GB_FULL:
2667 case IXGBE_LINK_SPEED_100_FULL:
2669 itr += DIV_ROUND_UP(avg_wire_size,
2670 IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2671 IXGBE_ITR_ADAPTIVE_MIN_INC;
2673 case IXGBE_LINK_SPEED_2_5GB_FULL:
2674 case IXGBE_LINK_SPEED_1GB_FULL:
2675 case IXGBE_LINK_SPEED_10_FULL:
2676 if (avg_wire_size > 8064)
2677 avg_wire_size = 8064;
2678 itr += DIV_ROUND_UP(avg_wire_size,
2679 IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2680 IXGBE_ITR_ADAPTIVE_MIN_INC;
2685 /* write back value */
2686 ring_container->itr = itr;
2688 /* next update should occur within next jiffy */
2689 ring_container->next_update = next_update + 1;
2691 ring_container->total_bytes = 0;
2692 ring_container->total_packets = 0;
2696 * ixgbe_write_eitr - write EITR register in hardware specific way
2697 * @q_vector: structure containing interrupt and ring information
2699 * This function is made to be called by ethtool and by the driver
2700 * when it needs to update EITR registers at runtime. Hardware
2701 * specific quirks/differences are taken care of here.
2703 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2705 struct ixgbe_adapter *adapter = q_vector->adapter;
2706 struct ixgbe_hw *hw = &adapter->hw;
2707 int v_idx = q_vector->v_idx;
2708 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2710 switch (adapter->hw.mac.type) {
2711 case ixgbe_mac_82598EB:
2712 /* must write high and low 16 bits to reset counter */
2713 itr_reg |= (itr_reg << 16);
2715 case ixgbe_mac_82599EB:
2716 case ixgbe_mac_X540:
2717 case ixgbe_mac_X550:
2718 case ixgbe_mac_X550EM_x:
2719 case ixgbe_mac_x550em_a:
2721 * set the WDIS bit to not clear the timer bits and cause an
2722 * immediate assertion of the interrupt
2724 itr_reg |= IXGBE_EITR_CNT_WDIS;
2729 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2732 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2736 ixgbe_update_itr(q_vector, &q_vector->tx);
2737 ixgbe_update_itr(q_vector, &q_vector->rx);
2739 /* use the smallest value of new ITR delay calculations */
2740 new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2742 /* Clear latency flag if set, shift into correct position */
2743 new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2746 if (new_itr != q_vector->itr) {
2747 /* save the algorithm value here */
2748 q_vector->itr = new_itr;
2750 ixgbe_write_eitr(q_vector);
2755 * ixgbe_check_overtemp_subtask - check for over temperature
2756 * @adapter: pointer to adapter
2758 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2760 struct ixgbe_hw *hw = &adapter->hw;
2761 u32 eicr = adapter->interrupt_event;
2764 if (test_bit(__IXGBE_DOWN, &adapter->state))
2767 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2770 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2772 switch (hw->device_id) {
2773 case IXGBE_DEV_ID_82599_T3_LOM:
2775 * Since the warning interrupt is for both ports
2776 * we don't have to check if:
2777 * - This interrupt wasn't for our port.
2778 * - We may have missed the interrupt so always have to
2779 * check if we got a LSC
2781 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2782 !(eicr & IXGBE_EICR_LSC))
2785 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2787 bool link_up = false;
2789 hw->mac.ops.check_link(hw, &speed, &link_up, false);
2795 /* Check if this is not due to overtemp */
2796 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2800 case IXGBE_DEV_ID_X550EM_A_1G_T:
2801 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2802 rc = hw->phy.ops.check_overtemp(hw);
2803 if (rc != IXGBE_ERR_OVERTEMP)
2807 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2809 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2813 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2815 adapter->interrupt_event = 0;
2818 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2820 struct ixgbe_hw *hw = &adapter->hw;
2822 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2823 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2824 e_crit(probe, "Fan has stopped, replace the adapter\n");
2825 /* write to clear the interrupt */
2826 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2830 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2832 struct ixgbe_hw *hw = &adapter->hw;
2834 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2837 switch (adapter->hw.mac.type) {
2838 case ixgbe_mac_82599EB:
2840 * Need to check link state so complete overtemp check
2843 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2844 (eicr & IXGBE_EICR_LSC)) &&
2845 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2846 adapter->interrupt_event = eicr;
2847 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2848 ixgbe_service_event_schedule(adapter);
2852 case ixgbe_mac_x550em_a:
2853 if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2854 adapter->interrupt_event = eicr;
2855 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2856 ixgbe_service_event_schedule(adapter);
2857 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2858 IXGBE_EICR_GPI_SDP0_X550EM_a);
2859 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2860 IXGBE_EICR_GPI_SDP0_X550EM_a);
2863 case ixgbe_mac_X550:
2864 case ixgbe_mac_X540:
2865 if (!(eicr & IXGBE_EICR_TS))
2872 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2875 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2877 switch (hw->mac.type) {
2878 case ixgbe_mac_82598EB:
2879 if (hw->phy.type == ixgbe_phy_nl)
2882 case ixgbe_mac_82599EB:
2883 case ixgbe_mac_X550EM_x:
2884 case ixgbe_mac_x550em_a:
2885 switch (hw->mac.ops.get_media_type(hw)) {
2886 case ixgbe_media_type_fiber:
2887 case ixgbe_media_type_fiber_qsfp:
2897 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2899 struct ixgbe_hw *hw = &adapter->hw;
2900 u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2902 if (!ixgbe_is_sfp(hw))
2905 /* Later MAC's use different SDP */
2906 if (hw->mac.type >= ixgbe_mac_X540)
2907 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2909 if (eicr & eicr_mask) {
2910 /* Clear the interrupt */
2911 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2912 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2913 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2914 adapter->sfp_poll_time = 0;
2915 ixgbe_service_event_schedule(adapter);
2919 if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2920 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2921 /* Clear the interrupt */
2922 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2923 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2924 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2925 ixgbe_service_event_schedule(adapter);
2930 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2932 struct ixgbe_hw *hw = &adapter->hw;
2935 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2936 adapter->link_check_timeout = jiffies;
2937 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2938 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2939 IXGBE_WRITE_FLUSH(hw);
2940 ixgbe_service_event_schedule(adapter);
2944 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2948 struct ixgbe_hw *hw = &adapter->hw;
2950 switch (hw->mac.type) {
2951 case ixgbe_mac_82598EB:
2952 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2953 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2955 case ixgbe_mac_82599EB:
2956 case ixgbe_mac_X540:
2957 case ixgbe_mac_X550:
2958 case ixgbe_mac_X550EM_x:
2959 case ixgbe_mac_x550em_a:
2960 mask = (qmask & 0xFFFFFFFF);
2962 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2963 mask = (qmask >> 32);
2965 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2970 /* skip the flush */
2974 * ixgbe_irq_enable - Enable default interrupt generation settings
2975 * @adapter: board private structure
2976 * @queues: enable irqs for queues
2977 * @flush: flush register write
2979 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2982 struct ixgbe_hw *hw = &adapter->hw;
2983 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2985 /* don't reenable LSC while waiting for link */
2986 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2987 mask &= ~IXGBE_EIMS_LSC;
2989 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2990 switch (adapter->hw.mac.type) {
2991 case ixgbe_mac_82599EB:
2992 mask |= IXGBE_EIMS_GPI_SDP0(hw);
2994 case ixgbe_mac_X540:
2995 case ixgbe_mac_X550:
2996 case ixgbe_mac_X550EM_x:
2997 case ixgbe_mac_x550em_a:
2998 mask |= IXGBE_EIMS_TS;
3003 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3004 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3005 switch (adapter->hw.mac.type) {
3006 case ixgbe_mac_82599EB:
3007 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3008 mask |= IXGBE_EIMS_GPI_SDP2(hw);
3010 case ixgbe_mac_X540:
3011 case ixgbe_mac_X550:
3012 case ixgbe_mac_X550EM_x:
3013 case ixgbe_mac_x550em_a:
3014 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3015 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3016 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3017 mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3018 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3019 mask |= IXGBE_EICR_GPI_SDP0_X540;
3020 mask |= IXGBE_EIMS_ECC;
3021 mask |= IXGBE_EIMS_MAILBOX;
3027 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3028 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3029 mask |= IXGBE_EIMS_FLOW_DIR;
3031 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3033 ixgbe_irq_enable_queues(adapter, ~0);
3035 IXGBE_WRITE_FLUSH(&adapter->hw);
3038 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3040 struct ixgbe_adapter *adapter = data;
3041 struct ixgbe_hw *hw = &adapter->hw;
3045 * Workaround for Silicon errata. Use clear-by-write instead
3046 * of clear-by-read. Reading with EICS will return the
3047 * interrupt causes without clearing, which later be done
3048 * with the write to EICR.
3050 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3052 /* The lower 16bits of the EICR register are for the queue interrupts
3053 * which should be masked here in order to not accidentally clear them if
3054 * the bits are high when ixgbe_msix_other is called. There is a race
3055 * condition otherwise which results in possible performance loss
3056 * especially if the ixgbe_msix_other interrupt is triggering
3057 * consistently (as it would when PPS is turned on for the X540 device)
3061 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3063 if (eicr & IXGBE_EICR_LSC)
3064 ixgbe_check_lsc(adapter);
3066 if (eicr & IXGBE_EICR_MAILBOX)
3067 ixgbe_msg_task(adapter);
3069 switch (hw->mac.type) {
3070 case ixgbe_mac_82599EB:
3071 case ixgbe_mac_X540:
3072 case ixgbe_mac_X550:
3073 case ixgbe_mac_X550EM_x:
3074 case ixgbe_mac_x550em_a:
3075 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3076 (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3077 adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3078 ixgbe_service_event_schedule(adapter);
3079 IXGBE_WRITE_REG(hw, IXGBE_EICR,
3080 IXGBE_EICR_GPI_SDP0_X540);
3082 if (eicr & IXGBE_EICR_ECC) {
3083 e_info(link, "Received ECC Err, initiating reset\n");
3084 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3085 ixgbe_service_event_schedule(adapter);
3086 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3088 /* Handle Flow Director Full threshold interrupt */
3089 if (eicr & IXGBE_EICR_FLOW_DIR) {
3090 int reinit_count = 0;
3092 for (i = 0; i < adapter->num_tx_queues; i++) {
3093 struct ixgbe_ring *ring = adapter->tx_ring[i];
3094 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3099 /* no more flow director interrupts until after init */
3100 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3101 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3102 ixgbe_service_event_schedule(adapter);
3105 ixgbe_check_sfp_event(adapter, eicr);
3106 ixgbe_check_overtemp_event(adapter, eicr);
3112 ixgbe_check_fan_failure(adapter, eicr);
3114 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3115 ixgbe_ptp_check_pps_event(adapter);
3117 /* re-enable the original interrupt state, no lsc, no queues */
3118 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3119 ixgbe_irq_enable(adapter, false, false);
3124 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3126 struct ixgbe_q_vector *q_vector = data;
3128 /* EIAM disabled interrupts (on this vector) for us */
3130 if (q_vector->rx.ring || q_vector->tx.ring)
3131 napi_schedule_irqoff(&q_vector->napi);
3137 * ixgbe_poll - NAPI Rx polling callback
3138 * @napi: structure for representing this polling device
3139 * @budget: how many packets driver is allowed to clean
3141 * This function is used for legacy and MSI, NAPI mode
3143 int ixgbe_poll(struct napi_struct *napi, int budget)
3145 struct ixgbe_q_vector *q_vector =
3146 container_of(napi, struct ixgbe_q_vector, napi);
3147 struct ixgbe_adapter *adapter = q_vector->adapter;
3148 struct ixgbe_ring *ring;
3149 int per_ring_budget, work_done = 0;
3150 bool clean_complete = true;
3152 #ifdef CONFIG_IXGBE_DCA
3153 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3154 ixgbe_update_dca(q_vector);
3157 ixgbe_for_each_ring(ring, q_vector->tx) {
3158 bool wd = ring->xsk_pool ?
3159 ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3160 ixgbe_clean_tx_irq(q_vector, ring, budget);
3163 clean_complete = false;
3166 /* Exit if we are called by netpoll */
3170 /* attempt to distribute budget to each queue fairly, but don't allow
3171 * the budget to go below 1 because we'll exit polling */
3172 if (q_vector->rx.count > 1)
3173 per_ring_budget = max(budget/q_vector->rx.count, 1);
3175 per_ring_budget = budget;
3177 ixgbe_for_each_ring(ring, q_vector->rx) {
3178 int cleaned = ring->xsk_pool ?
3179 ixgbe_clean_rx_irq_zc(q_vector, ring,
3181 ixgbe_clean_rx_irq(q_vector, ring,
3184 work_done += cleaned;
3185 if (cleaned >= per_ring_budget)
3186 clean_complete = false;
3189 /* If all work not completed, return budget and keep polling */
3190 if (!clean_complete)
3193 /* all work done, exit the polling mode */
3194 if (likely(napi_complete_done(napi, work_done))) {
3195 if (adapter->rx_itr_setting & 1)
3196 ixgbe_set_itr(q_vector);
3197 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3198 ixgbe_irq_enable_queues(adapter,
3199 BIT_ULL(q_vector->v_idx));
3202 return min(work_done, budget - 1);
3206 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3207 * @adapter: board private structure
3209 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3210 * interrupts from the kernel.
3212 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3214 struct net_device *netdev = adapter->netdev;
3215 unsigned int ri = 0, ti = 0;
3218 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3219 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3220 struct msix_entry *entry = &adapter->msix_entries[vector];
3222 if (q_vector->tx.ring && q_vector->rx.ring) {
3223 snprintf(q_vector->name, sizeof(q_vector->name),
3224 "%s-TxRx-%u", netdev->name, ri++);
3226 } else if (q_vector->rx.ring) {
3227 snprintf(q_vector->name, sizeof(q_vector->name),
3228 "%s-rx-%u", netdev->name, ri++);
3229 } else if (q_vector->tx.ring) {
3230 snprintf(q_vector->name, sizeof(q_vector->name),
3231 "%s-tx-%u", netdev->name, ti++);
3233 /* skip this unused q_vector */
3236 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3237 q_vector->name, q_vector);
3239 e_err(probe, "request_irq failed for MSIX interrupt "
3240 "Error: %d\n", err);
3241 goto free_queue_irqs;
3243 /* If Flow Director is enabled, set interrupt affinity */
3244 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3245 /* assign the mask for this irq */
3246 irq_set_affinity_hint(entry->vector,
3247 &q_vector->affinity_mask);
3251 err = request_irq(adapter->msix_entries[vector].vector,
3252 ixgbe_msix_other, 0, netdev->name, adapter);
3254 e_err(probe, "request_irq for msix_other failed: %d\n", err);
3255 goto free_queue_irqs;
3263 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3265 free_irq(adapter->msix_entries[vector].vector,
3266 adapter->q_vector[vector]);
3268 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3269 pci_disable_msix(adapter->pdev);
3270 kfree(adapter->msix_entries);
3271 adapter->msix_entries = NULL;
3276 * ixgbe_intr - legacy mode Interrupt Handler
3277 * @irq: interrupt number
3278 * @data: pointer to a network interface device structure
3280 static irqreturn_t ixgbe_intr(int irq, void *data)
3282 struct ixgbe_adapter *adapter = data;
3283 struct ixgbe_hw *hw = &adapter->hw;
3284 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3288 * Workaround for silicon errata #26 on 82598. Mask the interrupt
3289 * before the read of EICR.
3291 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3293 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3294 * therefore no explicit interrupt disable is necessary */
3295 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3298 * shared interrupt alert!
3299 * make sure interrupts are enabled because the read will
3300 * have disabled interrupts due to EIAM
3301 * finish the workaround of silicon errata on 82598. Unmask
3302 * the interrupt that we masked before the EICR read.
3304 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3305 ixgbe_irq_enable(adapter, true, true);
3306 return IRQ_NONE; /* Not our interrupt */
3309 if (eicr & IXGBE_EICR_LSC)
3310 ixgbe_check_lsc(adapter);
3312 switch (hw->mac.type) {
3313 case ixgbe_mac_82599EB:
3314 ixgbe_check_sfp_event(adapter, eicr);
3316 case ixgbe_mac_X540:
3317 case ixgbe_mac_X550:
3318 case ixgbe_mac_X550EM_x:
3319 case ixgbe_mac_x550em_a:
3320 if (eicr & IXGBE_EICR_ECC) {
3321 e_info(link, "Received ECC Err, initiating reset\n");
3322 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3323 ixgbe_service_event_schedule(adapter);
3324 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3326 ixgbe_check_overtemp_event(adapter, eicr);
3332 ixgbe_check_fan_failure(adapter, eicr);
3333 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3334 ixgbe_ptp_check_pps_event(adapter);
3336 /* would disable interrupts here but EIAM disabled it */
3337 napi_schedule_irqoff(&q_vector->napi);
3340 * re-enable link(maybe) and non-queue interrupts, no flush.
3341 * ixgbe_poll will re-enable the queue interrupts
3343 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3344 ixgbe_irq_enable(adapter, false, false);
3350 * ixgbe_request_irq - initialize interrupts
3351 * @adapter: board private structure
3353 * Attempts to configure interrupts using the best available
3354 * capabilities of the hardware and kernel.
3356 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3358 struct net_device *netdev = adapter->netdev;
3361 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3362 err = ixgbe_request_msix_irqs(adapter);
3363 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3364 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3365 netdev->name, adapter);
3367 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3368 netdev->name, adapter);
3371 e_err(probe, "request_irq failed, Error %d\n", err);
3376 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3380 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3381 free_irq(adapter->pdev->irq, adapter);
3385 if (!adapter->msix_entries)
3388 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3389 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3390 struct msix_entry *entry = &adapter->msix_entries[vector];
3392 /* free only the irqs that were actually requested */
3393 if (!q_vector->rx.ring && !q_vector->tx.ring)
3396 /* clear the affinity_mask in the IRQ descriptor */
3397 irq_set_affinity_hint(entry->vector, NULL);
3399 free_irq(entry->vector, q_vector);
3402 free_irq(adapter->msix_entries[vector].vector, adapter);
3406 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3407 * @adapter: board private structure
3409 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3411 switch (adapter->hw.mac.type) {
3412 case ixgbe_mac_82598EB:
3413 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3415 case ixgbe_mac_82599EB:
3416 case ixgbe_mac_X540:
3417 case ixgbe_mac_X550:
3418 case ixgbe_mac_X550EM_x:
3419 case ixgbe_mac_x550em_a:
3420 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3421 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3422 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3427 IXGBE_WRITE_FLUSH(&adapter->hw);
3428 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3431 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3432 synchronize_irq(adapter->msix_entries[vector].vector);
3434 synchronize_irq(adapter->msix_entries[vector++].vector);
3436 synchronize_irq(adapter->pdev->irq);
3441 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3442 * @adapter: board private structure
3445 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3447 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3449 ixgbe_write_eitr(q_vector);
3451 ixgbe_set_ivar(adapter, 0, 0, 0);
3452 ixgbe_set_ivar(adapter, 1, 0, 0);
3454 e_info(hw, "Legacy interrupt IVAR setup done\n");
3458 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3459 * @adapter: board private structure
3460 * @ring: structure containing ring specific data
3462 * Configure the Tx descriptor ring after a reset.
3464 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3465 struct ixgbe_ring *ring)
3467 struct ixgbe_hw *hw = &adapter->hw;
3468 u64 tdba = ring->dma;
3470 u32 txdctl = IXGBE_TXDCTL_ENABLE;
3471 u8 reg_idx = ring->reg_idx;
3473 ring->xsk_pool = NULL;
3474 if (ring_is_xdp(ring))
3475 ring->xsk_pool = ixgbe_xsk_pool(adapter, ring);
3477 /* disable queue to avoid issues while updating state */
3478 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3479 IXGBE_WRITE_FLUSH(hw);
3481 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3482 (tdba & DMA_BIT_MASK(32)));
3483 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3484 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3485 ring->count * sizeof(union ixgbe_adv_tx_desc));
3486 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3487 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3488 ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3491 * set WTHRESH to encourage burst writeback, it should not be set
3492 * higher than 1 when:
3493 * - ITR is 0 as it could cause false TX hangs
3494 * - ITR is set to > 100k int/sec and BQL is enabled
3496 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3497 * to or less than the number of on chip descriptors, which is
3500 if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3501 txdctl |= 1u << 16; /* WTHRESH = 1 */
3503 txdctl |= 8u << 16; /* WTHRESH = 8 */
3506 * Setting PTHRESH to 32 both improves performance
3507 * and avoids a TX hang with DFP enabled
3509 txdctl |= (1u << 8) | /* HTHRESH = 1 */
3510 32; /* PTHRESH = 32 */
3512 /* reinitialize flowdirector state */
3513 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3514 ring->atr_sample_rate = adapter->atr_sample_rate;
3515 ring->atr_count = 0;
3516 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3518 ring->atr_sample_rate = 0;
3521 /* initialize XPS */
3522 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3523 struct ixgbe_q_vector *q_vector = ring->q_vector;
3526 netif_set_xps_queue(ring->netdev,
3527 &q_vector->affinity_mask,
3531 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3533 /* reinitialize tx_buffer_info */
3534 memset(ring->tx_buffer_info, 0,
3535 sizeof(struct ixgbe_tx_buffer) * ring->count);
3538 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3540 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3541 if (hw->mac.type == ixgbe_mac_82598EB &&
3542 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3545 /* poll to verify queue is enabled */
3547 usleep_range(1000, 2000);
3548 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3549 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3551 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3554 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3556 struct ixgbe_hw *hw = &adapter->hw;
3558 u8 tcs = adapter->hw_tcs;
3560 if (hw->mac.type == ixgbe_mac_82598EB)
3563 /* disable the arbiter while setting MTQC */
3564 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3565 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3566 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3568 /* set transmit pool layout */
3569 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3570 mtqc = IXGBE_MTQC_VT_ENA;
3572 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3574 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3575 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3576 IXGBE_82599_VMDQ_4Q_MASK)
3577 mtqc |= IXGBE_MTQC_32VF;
3579 mtqc |= IXGBE_MTQC_64VF;
3582 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3583 } else if (tcs > 1) {
3584 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3586 u8 max_txq = adapter->num_tx_queues +
3587 adapter->num_xdp_queues;
3589 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3591 mtqc = IXGBE_MTQC_64Q_1PB;
3595 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3597 /* Enable Security TX Buffer IFG for multiple pb */
3599 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3600 sectx |= IXGBE_SECTX_DCB;
3601 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3604 /* re-enable the arbiter */
3605 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3606 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3610 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3611 * @adapter: board private structure
3613 * Configure the Tx unit of the MAC after a reset.
3615 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3617 struct ixgbe_hw *hw = &adapter->hw;
3621 ixgbe_setup_mtqc(adapter);
3623 if (hw->mac.type != ixgbe_mac_82598EB) {
3624 /* DMATXCTL.EN must be before Tx queues are enabled */
3625 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3626 dmatxctl |= IXGBE_DMATXCTL_TE;
3627 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3630 /* Setup the HW Tx Head and Tail descriptor pointers */
3631 for (i = 0; i < adapter->num_tx_queues; i++)
3632 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3633 for (i = 0; i < adapter->num_xdp_queues; i++)
3634 ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3637 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3638 struct ixgbe_ring *ring)
3640 struct ixgbe_hw *hw = &adapter->hw;
3641 u8 reg_idx = ring->reg_idx;
3642 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3644 srrctl |= IXGBE_SRRCTL_DROP_EN;
3646 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3649 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3650 struct ixgbe_ring *ring)
3652 struct ixgbe_hw *hw = &adapter->hw;
3653 u8 reg_idx = ring->reg_idx;
3654 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3656 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3658 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3661 #ifdef CONFIG_IXGBE_DCB
3662 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3664 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3668 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3670 if (adapter->ixgbe_ieee_pfc)
3671 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3674 * We should set the drop enable bit if:
3677 * Number of Rx queues > 1 and flow control is disabled
3679 * This allows us to avoid head of line blocking for security
3680 * and performance reasons.
3682 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3683 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3684 for (i = 0; i < adapter->num_rx_queues; i++)
3685 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3687 for (i = 0; i < adapter->num_rx_queues; i++)
3688 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3692 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3694 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3695 struct ixgbe_ring *rx_ring)
3697 struct ixgbe_hw *hw = &adapter->hw;
3699 u8 reg_idx = rx_ring->reg_idx;
3701 if (hw->mac.type == ixgbe_mac_82598EB) {
3702 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3705 * if VMDq is not active we must program one srrctl register
3706 * per RSS queue since we have enabled RDRXCTL.MVMEN
3711 /* configure header buffer length, needed for RSC */
3712 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3714 /* configure the packet buffer length */
3715 if (rx_ring->xsk_pool) {
3716 u32 xsk_buf_len = xsk_pool_get_rx_frame_size(rx_ring->xsk_pool);
3718 /* If the MAC support setting RXDCTL.RLPML, the
3719 * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3720 * RXDCTL.RLPML is set to the actual UMEM buffer
3721 * size. If not, then we are stuck with a 1k buffer
3722 * size resolution. In this case frames larger than
3723 * the UMEM buffer size viewed in a 1k resolution will
3726 if (hw->mac.type != ixgbe_mac_82599EB)
3727 srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3729 srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3730 } else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3731 srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3733 srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3736 /* configure descriptor type */
3737 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3739 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3743 * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3744 * @adapter: device handle
3746 * - 82598/82599/X540: 128
3747 * - X550(non-SRIOV mode): 512
3748 * - X550(SRIOV mode): 64
3750 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3752 if (adapter->hw.mac.type < ixgbe_mac_X550)
3754 else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3761 * ixgbe_store_key - Write the RSS key to HW
3762 * @adapter: device handle
3764 * Write the RSS key stored in adapter.rss_key to HW.
3766 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3768 struct ixgbe_hw *hw = &adapter->hw;
3771 for (i = 0; i < 10; i++)
3772 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3776 * ixgbe_init_rss_key - Initialize adapter RSS key
3777 * @adapter: device handle
3779 * Allocates and initializes the RSS key if it is not allocated.
3781 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3785 if (!adapter->rss_key) {
3786 rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3787 if (unlikely(!rss_key))
3790 netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3791 adapter->rss_key = rss_key;
3798 * ixgbe_store_reta - Write the RETA table to HW
3799 * @adapter: device handle
3801 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3803 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3805 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3806 struct ixgbe_hw *hw = &adapter->hw;
3809 u8 *indir_tbl = adapter->rss_indir_tbl;
3811 /* Fill out the redirection table as follows:
3812 * - 82598: 8 bit wide entries containing pair of 4 bit RSS
3814 * - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3815 * - X550: 8 bit wide entries containing 6 bit RSS index
3817 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3818 indices_multi = 0x11;
3820 indices_multi = 0x1;
3822 /* Write redirection table to HW */
3823 for (i = 0; i < reta_entries; i++) {
3824 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3827 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3829 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3837 * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3838 * @adapter: device handle
3840 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3842 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3844 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3845 struct ixgbe_hw *hw = &adapter->hw;
3848 /* Write redirection table to HW */
3849 for (i = 0; i < reta_entries; i++) {
3850 u16 pool = adapter->num_rx_pools;
3852 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3858 IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3864 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3867 u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3868 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3870 /* Program table for at least 4 queues w/ SR-IOV so that VFs can
3871 * make full use of any rings they may have. We will use the
3872 * PSRTYPE register to control how many rings we use within the PF.
3874 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3877 /* Fill out hash function seeds */
3878 ixgbe_store_key(adapter);
3880 /* Fill out redirection table */
3881 memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3883 for (i = 0, j = 0; i < reta_entries; i++, j++) {
3887 adapter->rss_indir_tbl[i] = j;
3890 ixgbe_store_reta(adapter);
3893 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3895 struct ixgbe_hw *hw = &adapter->hw;
3896 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3899 /* Fill out hash function seeds */
3900 for (i = 0; i < 10; i++) {
3901 u16 pool = adapter->num_rx_pools;
3905 IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3906 *(adapter->rss_key + i));
3909 /* Fill out the redirection table */
3910 for (i = 0, j = 0; i < 64; i++, j++) {
3914 adapter->rss_indir_tbl[i] = j;
3917 ixgbe_store_vfreta(adapter);
3920 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3922 struct ixgbe_hw *hw = &adapter->hw;
3923 u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3926 /* Disable indicating checksum in descriptor, enables RSS hash */
3927 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3928 rxcsum |= IXGBE_RXCSUM_PCSD;
3929 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3931 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3932 if (adapter->ring_feature[RING_F_RSS].mask)
3933 mrqc = IXGBE_MRQC_RSSEN;
3935 u8 tcs = adapter->hw_tcs;
3937 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3939 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3941 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3942 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3943 IXGBE_82599_VMDQ_4Q_MASK)
3944 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3946 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3948 /* Enable L3/L4 for Tx Switched packets only for X550,
3949 * older devices do not support this feature
3951 if (hw->mac.type >= ixgbe_mac_X550)
3952 mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3955 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3957 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3959 mrqc = IXGBE_MRQC_RSSEN;
3963 /* Perform hash on these packet types */
3964 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3965 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3966 IXGBE_MRQC_RSS_FIELD_IPV6 |
3967 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3969 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3970 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3971 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3972 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3974 if ((hw->mac.type >= ixgbe_mac_X550) &&
3975 (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3976 u16 pool = adapter->num_rx_pools;
3978 /* Enable VF RSS mode */
3979 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3980 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3982 /* Setup RSS through the VF registers */
3983 ixgbe_setup_vfreta(adapter);
3984 vfmrqc = IXGBE_MRQC_RSSEN;
3985 vfmrqc |= rss_field;
3989 IXGBE_PFVFMRQC(VMDQ_P(pool)),
3992 ixgbe_setup_reta(adapter);
3994 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3999 * ixgbe_configure_rscctl - enable RSC for the indicated ring
4000 * @adapter: address of board private structure
4001 * @ring: structure containing ring specific data
4003 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4004 struct ixgbe_ring *ring)
4006 struct ixgbe_hw *hw = &adapter->hw;
4008 u8 reg_idx = ring->reg_idx;
4010 if (!ring_is_rsc_enabled(ring))
4013 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4014 rscctrl |= IXGBE_RSCCTL_RSCEN;
4016 * we must limit the number of descriptors so that the
4017 * total size of max desc * buf_len is not greater
4020 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4021 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4024 #define IXGBE_MAX_RX_DESC_POLL 10
4025 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4026 struct ixgbe_ring *ring)
4028 struct ixgbe_hw *hw = &adapter->hw;
4029 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4031 u8 reg_idx = ring->reg_idx;
4033 if (ixgbe_removed(hw->hw_addr))
4035 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4036 if (hw->mac.type == ixgbe_mac_82598EB &&
4037 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4041 usleep_range(1000, 2000);
4042 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4043 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4046 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4047 "the polling period\n", reg_idx);
4051 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4052 struct ixgbe_ring *ring)
4054 struct ixgbe_hw *hw = &adapter->hw;
4055 union ixgbe_adv_rx_desc *rx_desc;
4056 u64 rdba = ring->dma;
4058 u8 reg_idx = ring->reg_idx;
4060 xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4061 ring->xsk_pool = ixgbe_xsk_pool(adapter, ring);
4062 if (ring->xsk_pool) {
4063 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4064 MEM_TYPE_XSK_BUFF_POOL,
4066 xsk_pool_set_rxq_info(ring->xsk_pool, &ring->xdp_rxq);
4068 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4069 MEM_TYPE_PAGE_SHARED, NULL));
4072 /* disable queue to avoid use of these values while updating state */
4073 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4074 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4076 /* write value back with RXDCTL.ENABLE bit cleared */
4077 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4078 IXGBE_WRITE_FLUSH(hw);
4080 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4081 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4082 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4083 ring->count * sizeof(union ixgbe_adv_rx_desc));
4084 /* Force flushing of IXGBE_RDLEN to prevent MDD */
4085 IXGBE_WRITE_FLUSH(hw);
4087 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4088 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4089 ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4091 ixgbe_configure_srrctl(adapter, ring);
4092 ixgbe_configure_rscctl(adapter, ring);
4094 if (hw->mac.type == ixgbe_mac_82598EB) {
4096 * enable cache line friendly hardware writes:
4097 * PTHRESH=32 descriptors (half the internal cache),
4098 * this also removes ugly rx_no_buffer_count increment
4099 * HTHRESH=4 descriptors (to minimize latency on fetch)
4100 * WTHRESH=8 burst writeback up to two cache lines
4102 rxdctl &= ~0x3FFFFF;
4104 #if (PAGE_SIZE < 8192)
4105 /* RXDCTL.RLPML does not work on 82599 */
4106 } else if (hw->mac.type != ixgbe_mac_82599EB) {
4107 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4108 IXGBE_RXDCTL_RLPML_EN);
4110 /* Limit the maximum frame size so we don't overrun the skb.
4111 * This can happen in SRIOV mode when the MTU of the VF is
4112 * higher than the MTU of the PF.
4114 if (ring_uses_build_skb(ring) &&
4115 !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4116 rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4117 IXGBE_RXDCTL_RLPML_EN;
4121 ring->rx_offset = ixgbe_rx_offset(ring);
4123 if (ring->xsk_pool && hw->mac.type != ixgbe_mac_82599EB) {
4124 u32 xsk_buf_len = xsk_pool_get_rx_frame_size(ring->xsk_pool);
4126 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4127 IXGBE_RXDCTL_RLPML_EN);
4128 rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4130 ring->rx_buf_len = xsk_buf_len;
4133 /* initialize rx_buffer_info */
4134 memset(ring->rx_buffer_info, 0,
4135 sizeof(struct ixgbe_rx_buffer) * ring->count);
4137 /* initialize Rx descriptor 0 */
4138 rx_desc = IXGBE_RX_DESC(ring, 0);
4139 rx_desc->wb.upper.length = 0;
4141 /* enable receive descriptor ring */
4142 rxdctl |= IXGBE_RXDCTL_ENABLE;
4143 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4145 ixgbe_rx_desc_queue_enable(adapter, ring);
4147 ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4149 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4152 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4154 struct ixgbe_hw *hw = &adapter->hw;
4155 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4156 u16 pool = adapter->num_rx_pools;
4158 /* PSRTYPE must be initialized in non 82598 adapters */
4159 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4160 IXGBE_PSRTYPE_UDPHDR |
4161 IXGBE_PSRTYPE_IPV4HDR |
4162 IXGBE_PSRTYPE_L2HDR |
4163 IXGBE_PSRTYPE_IPV6HDR;
4165 if (hw->mac.type == ixgbe_mac_82598EB)
4169 psrtype |= 2u << 29;
4171 psrtype |= 1u << 29;
4174 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4177 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4179 struct ixgbe_hw *hw = &adapter->hw;
4180 u16 pool = adapter->num_rx_pools;
4181 u32 reg_offset, vf_shift, vmolr;
4182 u32 gcr_ext, vmdctl;
4185 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4188 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4189 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4190 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4191 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4192 vmdctl |= IXGBE_VT_CTL_REPLEN;
4193 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4195 /* accept untagged packets until a vlan tag is
4196 * specifically set for the VMDQ queue/pool
4198 vmolr = IXGBE_VMOLR_AUPE;
4200 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4202 vf_shift = VMDQ_P(0) % 32;
4203 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4205 /* Enable only the PF's pool for Tx/Rx */
4206 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4207 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4208 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4209 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4210 if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4211 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4213 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4214 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4216 /* clear VLAN promisc flag so VFTA will be updated if necessary */
4217 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4220 * Set up VF register offsets for selected VT Mode,
4221 * i.e. 32 or 64 VFs for SR-IOV
4223 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4224 case IXGBE_82599_VMDQ_8Q_MASK:
4225 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4227 case IXGBE_82599_VMDQ_4Q_MASK:
4228 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4231 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4235 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4237 for (i = 0; i < adapter->num_vfs; i++) {
4238 /* configure spoof checking */
4239 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4240 adapter->vfinfo[i].spoofchk_enabled);
4242 /* Enable/Disable RSS query feature */
4243 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4244 adapter->vfinfo[i].rss_query_enabled);
4248 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4250 struct ixgbe_hw *hw = &adapter->hw;
4251 struct net_device *netdev = adapter->netdev;
4252 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4253 struct ixgbe_ring *rx_ring;
4258 /* adjust max frame to be able to do baby jumbo for FCoE */
4259 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4260 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4261 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4263 #endif /* IXGBE_FCOE */
4265 /* adjust max frame to be at least the size of a standard frame */
4266 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4267 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4269 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4270 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4271 mhadd &= ~IXGBE_MHADD_MFS_MASK;
4272 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4274 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4277 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4278 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4279 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4280 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4283 * Setup the HW Rx Head and Tail Descriptor Pointers and
4284 * the Base and Length of the Rx Descriptor Ring
4286 for (i = 0; i < adapter->num_rx_queues; i++) {
4287 rx_ring = adapter->rx_ring[i];
4289 clear_ring_rsc_enabled(rx_ring);
4290 clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4291 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4293 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4294 set_ring_rsc_enabled(rx_ring);
4296 if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4297 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4299 if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4302 set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4304 #if (PAGE_SIZE < 8192)
4305 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4306 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4308 if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4309 (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4310 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4315 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4317 struct ixgbe_hw *hw = &adapter->hw;
4318 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4320 switch (hw->mac.type) {
4321 case ixgbe_mac_82598EB:
4323 * For VMDq support of different descriptor types or
4324 * buffer sizes through the use of multiple SRRCTL
4325 * registers, RDRXCTL.MVMEN must be set to 1
4327 * also, the manual doesn't mention it clearly but DCA hints
4328 * will only use queue 0's tags unless this bit is set. Side
4329 * effects of setting this bit are only that SRRCTL must be
4330 * fully programmed [0..15]
4332 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4334 case ixgbe_mac_X550:
4335 case ixgbe_mac_X550EM_x:
4336 case ixgbe_mac_x550em_a:
4337 if (adapter->num_vfs)
4338 rdrxctl |= IXGBE_RDRXCTL_PSP;
4340 case ixgbe_mac_82599EB:
4341 case ixgbe_mac_X540:
4342 /* Disable RSC for ACK packets */
4343 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4344 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4345 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4346 /* hardware requires some bits to be set by default */
4347 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4348 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4351 /* We should do nothing since we don't know this hardware */
4355 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4359 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4360 * @adapter: board private structure
4362 * Configure the Rx unit of the MAC after a reset.
4364 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4366 struct ixgbe_hw *hw = &adapter->hw;
4370 /* disable receives while setting up the descriptors */
4371 hw->mac.ops.disable_rx(hw);
4373 ixgbe_setup_psrtype(adapter);
4374 ixgbe_setup_rdrxctl(adapter);
4377 rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4378 rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4379 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4380 rfctl |= IXGBE_RFCTL_RSC_DIS;
4382 /* disable NFS filtering */
4383 rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4384 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4386 /* Program registers for the distribution of queues */
4387 ixgbe_setup_mrqc(adapter);
4389 /* set_rx_buffer_len must be called before ring initialization */
4390 ixgbe_set_rx_buffer_len(adapter);
4393 * Setup the HW Rx Head and Tail Descriptor Pointers and
4394 * the Base and Length of the Rx Descriptor Ring
4396 for (i = 0; i < adapter->num_rx_queues; i++)
4397 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4399 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4400 /* disable drop enable for 82598 parts */
4401 if (hw->mac.type == ixgbe_mac_82598EB)
4402 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4404 /* enable all receives */
4405 rxctrl |= IXGBE_RXCTRL_RXEN;
4406 hw->mac.ops.enable_rx_dma(hw, rxctrl);
4409 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4410 __be16 proto, u16 vid)
4412 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4413 struct ixgbe_hw *hw = &adapter->hw;
4415 /* add VID to filter table */
4416 if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4417 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4419 set_bit(vid, adapter->active_vlans);
4424 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4429 /* short cut the special case */
4433 /* Search for the vlan id in the VLVF entries */
4434 for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4435 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4436 if ((vlvf & VLAN_VID_MASK) == vlan)
4443 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4445 struct ixgbe_hw *hw = &adapter->hw;
4449 idx = ixgbe_find_vlvf_entry(hw, vid);
4453 /* See if any other pools are set for this VLAN filter
4454 * entry other than the PF.
4456 word = idx * 2 + (VMDQ_P(0) / 32);
4457 bits = ~BIT(VMDQ_P(0) % 32);
4458 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4460 /* Disable the filter so this falls into the default pool. */
4461 if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4462 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4463 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4464 IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4468 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4469 __be16 proto, u16 vid)
4471 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4472 struct ixgbe_hw *hw = &adapter->hw;
4474 /* remove VID from filter table */
4475 if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4476 hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4478 clear_bit(vid, adapter->active_vlans);
4484 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4485 * @adapter: driver data
4487 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4489 struct ixgbe_hw *hw = &adapter->hw;
4493 switch (hw->mac.type) {
4494 case ixgbe_mac_82598EB:
4495 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4496 vlnctrl &= ~IXGBE_VLNCTRL_VME;
4497 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4499 case ixgbe_mac_82599EB:
4500 case ixgbe_mac_X540:
4501 case ixgbe_mac_X550:
4502 case ixgbe_mac_X550EM_x:
4503 case ixgbe_mac_x550em_a:
4504 for (i = 0; i < adapter->num_rx_queues; i++) {
4505 struct ixgbe_ring *ring = adapter->rx_ring[i];
4507 if (!netif_is_ixgbe(ring->netdev))
4511 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4512 vlnctrl &= ~IXGBE_RXDCTL_VME;
4513 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4522 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4523 * @adapter: driver data
4525 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4527 struct ixgbe_hw *hw = &adapter->hw;
4531 switch (hw->mac.type) {
4532 case ixgbe_mac_82598EB:
4533 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4534 vlnctrl |= IXGBE_VLNCTRL_VME;
4535 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4537 case ixgbe_mac_82599EB:
4538 case ixgbe_mac_X540:
4539 case ixgbe_mac_X550:
4540 case ixgbe_mac_X550EM_x:
4541 case ixgbe_mac_x550em_a:
4542 for (i = 0; i < adapter->num_rx_queues; i++) {
4543 struct ixgbe_ring *ring = adapter->rx_ring[i];
4545 if (!netif_is_ixgbe(ring->netdev))
4549 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4550 vlnctrl |= IXGBE_RXDCTL_VME;
4551 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4559 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4561 struct ixgbe_hw *hw = &adapter->hw;
4564 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4566 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4567 /* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4568 vlnctrl |= IXGBE_VLNCTRL_VFE;
4569 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4571 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4572 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4576 /* Nothing to do for 82598 */
4577 if (hw->mac.type == ixgbe_mac_82598EB)
4580 /* We are already in VLAN promisc, nothing to do */
4581 if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4584 /* Set flag so we don't redo unnecessary work */
4585 adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4587 /* Add PF to all active pools */
4588 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4589 u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4590 u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4592 vlvfb |= BIT(VMDQ_P(0) % 32);
4593 IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4596 /* Set all bits in the VLAN filter table array */
4597 for (i = hw->mac.vft_size; i--;)
4598 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4601 #define VFTA_BLOCK_SIZE 8
4602 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4604 struct ixgbe_hw *hw = &adapter->hw;
4605 u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4606 u32 vid_start = vfta_offset * 32;
4607 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4608 u32 i, vid, word, bits;
4610 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4611 u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4613 /* pull VLAN ID from VLVF */
4614 vid = vlvf & VLAN_VID_MASK;
4616 /* only concern outselves with a certain range */
4617 if (vid < vid_start || vid >= vid_end)
4621 /* record VLAN ID in VFTA */
4622 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4624 /* if PF is part of this then continue */
4625 if (test_bit(vid, adapter->active_vlans))
4629 /* remove PF from the pool */
4630 word = i * 2 + VMDQ_P(0) / 32;
4631 bits = ~BIT(VMDQ_P(0) % 32);
4632 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4633 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4636 /* extract values from active_vlans and write back to VFTA */
4637 for (i = VFTA_BLOCK_SIZE; i--;) {
4638 vid = (vfta_offset + i) * 32;
4639 word = vid / BITS_PER_LONG;
4640 bits = vid % BITS_PER_LONG;
4642 vfta[i] |= adapter->active_vlans[word] >> bits;
4644 IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4648 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4650 struct ixgbe_hw *hw = &adapter->hw;
4653 /* Set VLAN filtering to enabled */
4654 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4655 vlnctrl |= IXGBE_VLNCTRL_VFE;
4656 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4658 if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4659 hw->mac.type == ixgbe_mac_82598EB)
4662 /* We are not in VLAN promisc, nothing to do */
4663 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4666 /* Set flag so we don't redo unnecessary work */
4667 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4669 for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4670 ixgbe_scrub_vfta(adapter, i);
4673 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4677 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4679 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4680 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4684 * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4685 * @netdev: network interface device structure
4687 * Writes multicast address list to the MTA hash table.
4688 * Returns: -ENOMEM on failure
4689 * 0 on no addresses written
4690 * X on writing X addresses to MTA
4692 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4694 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4695 struct ixgbe_hw *hw = &adapter->hw;
4697 if (!netif_running(netdev))
4700 if (hw->mac.ops.update_mc_addr_list)
4701 hw->mac.ops.update_mc_addr_list(hw, netdev);
4705 #ifdef CONFIG_PCI_IOV
4706 ixgbe_restore_vf_multicasts(adapter);
4709 return netdev_mc_count(netdev);
4712 #ifdef CONFIG_PCI_IOV
4713 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4715 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4716 struct ixgbe_hw *hw = &adapter->hw;
4719 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4720 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4722 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4723 hw->mac.ops.set_rar(hw, i,
4728 hw->mac.ops.clear_rar(hw, i);
4733 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4735 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4736 struct ixgbe_hw *hw = &adapter->hw;
4739 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4740 if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4743 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4745 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4746 hw->mac.ops.set_rar(hw, i,
4751 hw->mac.ops.clear_rar(hw, i);
4755 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4757 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4758 struct ixgbe_hw *hw = &adapter->hw;
4761 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4762 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4763 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4766 ixgbe_sync_mac_table(adapter);
4769 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4771 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4772 struct ixgbe_hw *hw = &adapter->hw;
4775 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4776 /* do not count default RAR as available */
4777 if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4780 /* only count unused and addresses that belong to us */
4781 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4782 if (mac_table->pool != pool)
4792 /* this function destroys the first RAR entry */
4793 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4795 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4796 struct ixgbe_hw *hw = &adapter->hw;
4798 memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4799 mac_table->pool = VMDQ_P(0);
4801 mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4803 hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4807 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4808 const u8 *addr, u16 pool)
4810 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4811 struct ixgbe_hw *hw = &adapter->hw;
4814 if (is_zero_ether_addr(addr))
4817 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4818 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4821 ether_addr_copy(mac_table->addr, addr);
4822 mac_table->pool = pool;
4824 mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4825 IXGBE_MAC_STATE_IN_USE;
4827 ixgbe_sync_mac_table(adapter);
4835 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4836 const u8 *addr, u16 pool)
4838 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4839 struct ixgbe_hw *hw = &adapter->hw;
4842 if (is_zero_ether_addr(addr))
4845 /* search table for addr, if found clear IN_USE flag and sync */
4846 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4847 /* we can only delete an entry if it is in use */
4848 if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4850 /* we only care about entries that belong to the given pool */
4851 if (mac_table->pool != pool)
4853 /* we only care about a specific MAC address */
4854 if (!ether_addr_equal(addr, mac_table->addr))
4857 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4858 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4860 ixgbe_sync_mac_table(adapter);
4868 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4870 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4873 ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4875 return min_t(int, ret, 0);
4878 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4880 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4882 ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4888 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4889 * @netdev: network interface device structure
4891 * The set_rx_method entry point is called whenever the unicast/multicast
4892 * address list or the network interface flags are updated. This routine is
4893 * responsible for configuring the hardware for proper unicast, multicast and
4896 void ixgbe_set_rx_mode(struct net_device *netdev)
4898 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4899 struct ixgbe_hw *hw = &adapter->hw;
4900 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4901 netdev_features_t features = netdev->features;
4904 /* Check for Promiscuous and All Multicast modes */
4905 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4907 /* set all bits that we expect to always be set */
4908 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4909 fctrl |= IXGBE_FCTRL_BAM;
4910 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4911 fctrl |= IXGBE_FCTRL_PMCF;
4913 /* clear the bits we are changing the status of */
4914 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4915 if (netdev->flags & IFF_PROMISC) {
4916 hw->addr_ctrl.user_set_promisc = true;
4917 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4918 vmolr |= IXGBE_VMOLR_MPE;
4919 features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4921 if (netdev->flags & IFF_ALLMULTI) {
4922 fctrl |= IXGBE_FCTRL_MPE;
4923 vmolr |= IXGBE_VMOLR_MPE;
4925 hw->addr_ctrl.user_set_promisc = false;
4929 * Write addresses to available RAR registers, if there is not
4930 * sufficient space to store all the addresses then enable
4931 * unicast promiscuous mode
4933 if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4934 fctrl |= IXGBE_FCTRL_UPE;
4935 vmolr |= IXGBE_VMOLR_ROPE;
4938 /* Write addresses to the MTA, if the attempt fails
4939 * then we should just turn on promiscuous mode so
4940 * that we can at least receive multicast traffic
4942 count = ixgbe_write_mc_addr_list(netdev);
4944 fctrl |= IXGBE_FCTRL_MPE;
4945 vmolr |= IXGBE_VMOLR_MPE;
4947 vmolr |= IXGBE_VMOLR_ROMPE;
4950 if (hw->mac.type != ixgbe_mac_82598EB) {
4951 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4952 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4954 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4957 /* This is useful for sniffing bad packets. */
4958 if (features & NETIF_F_RXALL) {
4959 /* UPE and MPE will be handled by normal PROMISC logic
4960 * in e1000e_set_rx_mode */
4961 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4962 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4963 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4965 fctrl &= ~(IXGBE_FCTRL_DPF);
4966 /* NOTE: VLAN filtering is disabled by setting PROMISC */
4969 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4971 if (features & NETIF_F_HW_VLAN_CTAG_RX)
4972 ixgbe_vlan_strip_enable(adapter);
4974 ixgbe_vlan_strip_disable(adapter);
4976 if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4977 ixgbe_vlan_promisc_disable(adapter);
4979 ixgbe_vlan_promisc_enable(adapter);
4982 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4986 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4987 napi_enable(&adapter->q_vector[q_idx]->napi);
4990 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4994 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4995 napi_disable(&adapter->q_vector[q_idx]->napi);
4998 static int ixgbe_udp_tunnel_sync(struct net_device *dev, unsigned int table)
5000 struct ixgbe_adapter *adapter = netdev_priv(dev);
5001 struct ixgbe_hw *hw = &adapter->hw;
5002 struct udp_tunnel_info ti;
5004 udp_tunnel_nic_get_port(dev, table, 0, &ti);
5005 if (ti.type == UDP_TUNNEL_TYPE_VXLAN)
5006 adapter->vxlan_port = ti.port;
5008 adapter->geneve_port = ti.port;
5010 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL,
5011 ntohs(adapter->vxlan_port) |
5012 ntohs(adapter->geneve_port) <<
5013 IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT);
5017 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550 = {
5018 .sync_table = ixgbe_udp_tunnel_sync,
5019 .flags = UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5021 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
5025 static const struct udp_tunnel_nic_info ixgbe_udp_tunnels_x550em_a = {
5026 .sync_table = ixgbe_udp_tunnel_sync,
5027 .flags = UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
5029 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
5030 { .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_GENEVE, },
5034 #ifdef CONFIG_IXGBE_DCB
5036 * ixgbe_configure_dcb - Configure DCB hardware
5037 * @adapter: ixgbe adapter struct
5039 * This is called by the driver on open to configure the DCB hardware.
5040 * This is also called by the gennetlink interface when reconfiguring
5043 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5045 struct ixgbe_hw *hw = &adapter->hw;
5046 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5048 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5049 if (hw->mac.type == ixgbe_mac_82598EB)
5050 netif_set_gso_max_size(adapter->netdev, 65536);
5054 if (hw->mac.type == ixgbe_mac_82598EB)
5055 netif_set_gso_max_size(adapter->netdev, 32768);
5058 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5059 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5062 /* reconfigure the hardware */
5063 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5064 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5066 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5068 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5069 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5070 ixgbe_dcb_hw_ets(&adapter->hw,
5071 adapter->ixgbe_ieee_ets,
5073 ixgbe_dcb_hw_pfc_config(&adapter->hw,
5074 adapter->ixgbe_ieee_pfc->pfc_en,
5075 adapter->ixgbe_ieee_ets->prio_tc);
5078 /* Enable RSS Hash per TC */
5079 if (hw->mac.type != ixgbe_mac_82598EB) {
5081 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5088 /* write msb to all 8 TCs in one write */
5089 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5094 /* Additional bittime to account for IXGBE framing */
5095 #define IXGBE_ETH_FRAMING 20
5098 * ixgbe_hpbthresh - calculate high water mark for flow control
5100 * @adapter: board private structure to calculate for
5101 * @pb: packet buffer to calculate
5103 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5105 struct ixgbe_hw *hw = &adapter->hw;
5106 struct net_device *dev = adapter->netdev;
5107 int link, tc, kb, marker;
5110 /* Calculate max LAN frame size */
5111 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5114 /* FCoE traffic class uses FCOE jumbo frames */
5115 if ((dev->features & NETIF_F_FCOE_MTU) &&
5116 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5117 (pb == ixgbe_fcoe_get_tc(adapter)))
5118 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5121 /* Calculate delay value for device */
5122 switch (hw->mac.type) {
5123 case ixgbe_mac_X540:
5124 case ixgbe_mac_X550:
5125 case ixgbe_mac_X550EM_x:
5126 case ixgbe_mac_x550em_a:
5127 dv_id = IXGBE_DV_X540(link, tc);
5130 dv_id = IXGBE_DV(link, tc);
5134 /* Loopback switch introduces additional latency */
5135 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5136 dv_id += IXGBE_B2BT(tc);
5138 /* Delay value is calculated in bit times convert to KB */
5139 kb = IXGBE_BT2KB(dv_id);
5140 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5142 marker = rx_pba - kb;
5144 /* It is possible that the packet buffer is not large enough
5145 * to provide required headroom. In this case throw an error
5146 * to user and a do the best we can.
5149 e_warn(drv, "Packet Buffer(%i) can not provide enough"
5150 "headroom to support flow control."
5151 "Decrease MTU or number of traffic classes\n", pb);
5159 * ixgbe_lpbthresh - calculate low water mark for for flow control
5161 * @adapter: board private structure to calculate for
5162 * @pb: packet buffer to calculate
5164 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5166 struct ixgbe_hw *hw = &adapter->hw;
5167 struct net_device *dev = adapter->netdev;
5171 /* Calculate max LAN frame size */
5172 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5175 /* FCoE traffic class uses FCOE jumbo frames */
5176 if ((dev->features & NETIF_F_FCOE_MTU) &&
5177 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5178 (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5179 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5182 /* Calculate delay value for device */
5183 switch (hw->mac.type) {
5184 case ixgbe_mac_X540:
5185 case ixgbe_mac_X550:
5186 case ixgbe_mac_X550EM_x:
5187 case ixgbe_mac_x550em_a:
5188 dv_id = IXGBE_LOW_DV_X540(tc);
5191 dv_id = IXGBE_LOW_DV(tc);
5195 /* Delay value is calculated in bit times convert to KB */
5196 return IXGBE_BT2KB(dv_id);
5200 * ixgbe_pbthresh_setup - calculate and setup high low water marks
5202 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5204 struct ixgbe_hw *hw = &adapter->hw;
5205 int num_tc = adapter->hw_tcs;
5211 for (i = 0; i < num_tc; i++) {
5212 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5213 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5215 /* Low water marks must not be larger than high water marks */
5216 if (hw->fc.low_water[i] > hw->fc.high_water[i])
5217 hw->fc.low_water[i] = 0;
5220 for (; i < MAX_TRAFFIC_CLASS; i++)
5221 hw->fc.high_water[i] = 0;
5224 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5226 struct ixgbe_hw *hw = &adapter->hw;
5228 u8 tc = adapter->hw_tcs;
5230 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5231 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5232 hdrm = 32 << adapter->fdir_pballoc;
5236 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5237 ixgbe_pbthresh_setup(adapter);
5240 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5242 struct ixgbe_hw *hw = &adapter->hw;
5243 struct hlist_node *node2;
5244 struct ixgbe_fdir_filter *filter;
5247 spin_lock(&adapter->fdir_perfect_lock);
5249 if (!hlist_empty(&adapter->fdir_filter_list))
5250 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5252 hlist_for_each_entry_safe(filter, node2,
5253 &adapter->fdir_filter_list, fdir_node) {
5254 if (filter->action == IXGBE_FDIR_DROP_QUEUE) {
5255 queue = IXGBE_FDIR_DROP_QUEUE;
5257 u32 ring = ethtool_get_flow_spec_ring(filter->action);
5258 u8 vf = ethtool_get_flow_spec_ring_vf(filter->action);
5260 if (!vf && (ring >= adapter->num_rx_queues)) {
5261 e_err(drv, "FDIR restore failed without VF, ring: %u\n",
5265 ((vf > adapter->num_vfs) ||
5266 ring >= adapter->num_rx_queues_per_pool)) {
5267 e_err(drv, "FDIR restore failed with VF, vf: %hhu, ring: %u\n",
5272 /* Map the ring onto the absolute queue index */
5274 queue = adapter->rx_ring[ring]->reg_idx;
5277 adapter->num_rx_queues_per_pool) + ring;
5280 ixgbe_fdir_write_perfect_filter_82599(hw,
5281 &filter->filter, filter->sw_idx, queue);
5284 spin_unlock(&adapter->fdir_perfect_lock);
5288 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5289 * @rx_ring: ring to free buffers from
5291 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5293 u16 i = rx_ring->next_to_clean;
5294 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5296 if (rx_ring->xsk_pool) {
5297 ixgbe_xsk_clean_rx_ring(rx_ring);
5301 /* Free all the Rx ring sk_buffs */
5302 while (i != rx_ring->next_to_alloc) {
5303 if (rx_buffer->skb) {
5304 struct sk_buff *skb = rx_buffer->skb;
5305 if (IXGBE_CB(skb)->page_released)
5306 dma_unmap_page_attrs(rx_ring->dev,
5308 ixgbe_rx_pg_size(rx_ring),
5314 /* Invalidate cache lines that may have been written to by
5315 * device so that we avoid corrupting memory.
5317 dma_sync_single_range_for_cpu(rx_ring->dev,
5319 rx_buffer->page_offset,
5320 ixgbe_rx_bufsz(rx_ring),
5323 /* free resources associated with mapping */
5324 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5325 ixgbe_rx_pg_size(rx_ring),
5328 __page_frag_cache_drain(rx_buffer->page,
5329 rx_buffer->pagecnt_bias);
5333 if (i == rx_ring->count) {
5335 rx_buffer = rx_ring->rx_buffer_info;
5340 rx_ring->next_to_alloc = 0;
5341 rx_ring->next_to_clean = 0;
5342 rx_ring->next_to_use = 0;
5345 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5346 struct ixgbe_fwd_adapter *accel)
5348 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5349 int num_tc = netdev_get_num_tc(adapter->netdev);
5350 struct net_device *vdev = accel->netdev;
5353 baseq = accel->pool * adapter->num_rx_queues_per_pool;
5354 netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5355 accel->pool, adapter->num_rx_pools,
5356 baseq, baseq + adapter->num_rx_queues_per_pool);
5358 accel->rx_base_queue = baseq;
5359 accel->tx_base_queue = baseq;
5361 /* record configuration for macvlan interface in vdev */
5362 for (i = 0; i < num_tc; i++)
5363 netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5364 i, rss_i, baseq + (rss_i * i));
5366 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5367 adapter->rx_ring[baseq + i]->netdev = vdev;
5369 /* Guarantee all rings are updated before we update the
5370 * MAC address filter.
5374 /* ixgbe_add_mac_filter will return an index if it succeeds, so we
5375 * need to only treat it as an error value if it is negative.
5377 err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5378 VMDQ_P(accel->pool));
5382 /* if we cannot add the MAC rule then disable the offload */
5383 macvlan_release_l2fw_offload(vdev);
5385 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5386 adapter->rx_ring[baseq + i]->netdev = NULL;
5388 netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5390 /* unbind the queues and drop the subordinate channel config */
5391 netdev_unbind_sb_channel(adapter->netdev, vdev);
5392 netdev_set_sb_channel(vdev, 0);
5394 clear_bit(accel->pool, adapter->fwd_bitmask);
5400 static int ixgbe_macvlan_up(struct net_device *vdev,
5401 struct netdev_nested_priv *priv)
5403 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)priv->data;
5404 struct ixgbe_fwd_adapter *accel;
5406 if (!netif_is_macvlan(vdev))
5409 accel = macvlan_accel_priv(vdev);
5413 ixgbe_fwd_ring_up(adapter, accel);
5418 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5420 struct netdev_nested_priv priv = {
5421 .data = (void *)adapter,
5424 netdev_walk_all_upper_dev_rcu(adapter->netdev,
5425 ixgbe_macvlan_up, &priv);
5428 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5430 struct ixgbe_hw *hw = &adapter->hw;
5432 ixgbe_configure_pb(adapter);
5433 #ifdef CONFIG_IXGBE_DCB
5434 ixgbe_configure_dcb(adapter);
5437 * We must restore virtualization before VLANs or else
5438 * the VLVF registers will not be populated
5440 ixgbe_configure_virtualization(adapter);
5442 ixgbe_set_rx_mode(adapter->netdev);
5443 ixgbe_restore_vlan(adapter);
5444 ixgbe_ipsec_restore(adapter);
5446 switch (hw->mac.type) {
5447 case ixgbe_mac_82599EB:
5448 case ixgbe_mac_X540:
5449 hw->mac.ops.disable_rx_buff(hw);
5455 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5456 ixgbe_init_fdir_signature_82599(&adapter->hw,
5457 adapter->fdir_pballoc);
5458 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5459 ixgbe_init_fdir_perfect_82599(&adapter->hw,
5460 adapter->fdir_pballoc);
5461 ixgbe_fdir_filter_restore(adapter);
5464 switch (hw->mac.type) {
5465 case ixgbe_mac_82599EB:
5466 case ixgbe_mac_X540:
5467 hw->mac.ops.enable_rx_buff(hw);
5473 #ifdef CONFIG_IXGBE_DCA
5475 if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5476 ixgbe_setup_dca(adapter);
5477 #endif /* CONFIG_IXGBE_DCA */
5480 /* configure FCoE L2 filters, redirection table, and Rx control */
5481 ixgbe_configure_fcoe(adapter);
5483 #endif /* IXGBE_FCOE */
5484 ixgbe_configure_tx(adapter);
5485 ixgbe_configure_rx(adapter);
5486 ixgbe_configure_dfwd(adapter);
5490 * ixgbe_sfp_link_config - set up SFP+ link
5491 * @adapter: pointer to private adapter struct
5493 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5496 * We are assuming the worst case scenario here, and that
5497 * is that an SFP was inserted/removed after the reset
5498 * but before SFP detection was enabled. As such the best
5499 * solution is to just start searching as soon as we start
5501 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5502 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5504 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5505 adapter->sfp_poll_time = 0;
5509 * ixgbe_non_sfp_link_config - set up non-SFP+ link
5510 * @hw: pointer to private hardware struct
5512 * Returns 0 on success, negative on failure
5514 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5517 bool autoneg, link_up = false;
5518 int ret = IXGBE_ERR_LINK_SETUP;
5520 if (hw->mac.ops.check_link)
5521 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5526 speed = hw->phy.autoneg_advertised;
5527 if (!speed && hw->mac.ops.get_link_capabilities) {
5528 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5530 speed &= ~(IXGBE_LINK_SPEED_5GB_FULL |
5531 IXGBE_LINK_SPEED_2_5GB_FULL);
5537 if (hw->mac.ops.setup_link)
5538 ret = hw->mac.ops.setup_link(hw, speed, link_up);
5543 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5545 struct ixgbe_hw *hw = &adapter->hw;
5548 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5549 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5551 gpie |= IXGBE_GPIE_EIAME;
5553 * use EIAM to auto-mask when MSI-X interrupt is asserted
5554 * this saves a register write for every interrupt
5556 switch (hw->mac.type) {
5557 case ixgbe_mac_82598EB:
5558 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5560 case ixgbe_mac_82599EB:
5561 case ixgbe_mac_X540:
5562 case ixgbe_mac_X550:
5563 case ixgbe_mac_X550EM_x:
5564 case ixgbe_mac_x550em_a:
5566 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5567 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5571 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
5572 * specifically only auto mask tx and rx interrupts */
5573 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5576 /* XXX: to interrupt immediately for EICS writes, enable this */
5577 /* gpie |= IXGBE_GPIE_EIMEN; */
5579 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5580 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5582 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5583 case IXGBE_82599_VMDQ_8Q_MASK:
5584 gpie |= IXGBE_GPIE_VTMODE_16;
5586 case IXGBE_82599_VMDQ_4Q_MASK:
5587 gpie |= IXGBE_GPIE_VTMODE_32;
5590 gpie |= IXGBE_GPIE_VTMODE_64;
5595 /* Enable Thermal over heat sensor interrupt */
5596 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5597 switch (adapter->hw.mac.type) {
5598 case ixgbe_mac_82599EB:
5599 gpie |= IXGBE_SDP0_GPIEN_8259X;
5606 /* Enable fan failure interrupt */
5607 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5608 gpie |= IXGBE_SDP1_GPIEN(hw);
5610 switch (hw->mac.type) {
5611 case ixgbe_mac_82599EB:
5612 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5614 case ixgbe_mac_X550EM_x:
5615 case ixgbe_mac_x550em_a:
5616 gpie |= IXGBE_SDP0_GPIEN_X540;
5622 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5625 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5627 struct ixgbe_hw *hw = &adapter->hw;
5631 ixgbe_get_hw_control(adapter);
5632 ixgbe_setup_gpie(adapter);
5634 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5635 ixgbe_configure_msix(adapter);
5637 ixgbe_configure_msi_and_legacy(adapter);
5639 /* enable the optics for 82599 SFP+ fiber */
5640 if (hw->mac.ops.enable_tx_laser)
5641 hw->mac.ops.enable_tx_laser(hw);
5643 if (hw->phy.ops.set_phy_power)
5644 hw->phy.ops.set_phy_power(hw, true);
5646 smp_mb__before_atomic();
5647 clear_bit(__IXGBE_DOWN, &adapter->state);
5648 ixgbe_napi_enable_all(adapter);
5650 if (ixgbe_is_sfp(hw)) {
5651 ixgbe_sfp_link_config(adapter);
5653 err = ixgbe_non_sfp_link_config(hw);
5655 e_err(probe, "link_config FAILED %d\n", err);
5658 /* clear any pending interrupts, may auto mask */
5659 IXGBE_READ_REG(hw, IXGBE_EICR);
5660 ixgbe_irq_enable(adapter, true, true);
5663 * If this adapter has a fan, check to see if we had a failure
5664 * before we enabled the interrupt.
5666 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5667 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5668 if (esdp & IXGBE_ESDP_SDP1)
5669 e_crit(drv, "Fan has stopped, replace the adapter\n");
5672 /* bring the link up in the watchdog, this could race with our first
5673 * link up interrupt but shouldn't be a problem */
5674 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5675 adapter->link_check_timeout = jiffies;
5676 mod_timer(&adapter->service_timer, jiffies);
5678 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5679 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5680 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5681 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5684 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5686 /* put off any impending NetWatchDogTimeout */
5687 netif_trans_update(adapter->netdev);
5689 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5690 usleep_range(1000, 2000);
5691 if (adapter->hw.phy.type == ixgbe_phy_fw)
5692 ixgbe_watchdog_link_is_down(adapter);
5693 ixgbe_down(adapter);
5695 * If SR-IOV enabled then wait a bit before bringing the adapter
5696 * back up to give the VFs time to respond to the reset. The
5697 * two second wait is based upon the watchdog timer cycle in
5700 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5703 clear_bit(__IXGBE_RESETTING, &adapter->state);
5706 void ixgbe_up(struct ixgbe_adapter *adapter)
5708 /* hardware has been reset, we need to reload some things */
5709 ixgbe_configure(adapter);
5711 ixgbe_up_complete(adapter);
5714 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5718 pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5720 switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5721 case IXGBE_PCIDEVCTRL2_17_34s:
5722 case IXGBE_PCIDEVCTRL2_4_8s:
5723 /* For now we cap the upper limit on delay to 2 seconds
5724 * as we end up going up to 34 seconds of delay in worst
5725 * case timeout value.
5727 case IXGBE_PCIDEVCTRL2_1_2s:
5728 return 2000000ul; /* 2.0 s */
5729 case IXGBE_PCIDEVCTRL2_260_520ms:
5730 return 520000ul; /* 520 ms */
5731 case IXGBE_PCIDEVCTRL2_65_130ms:
5732 return 130000ul; /* 130 ms */
5733 case IXGBE_PCIDEVCTRL2_16_32ms:
5734 return 32000ul; /* 32 ms */
5735 case IXGBE_PCIDEVCTRL2_1_2ms:
5736 return 2000ul; /* 2 ms */
5737 case IXGBE_PCIDEVCTRL2_50_100us:
5738 return 100ul; /* 100 us */
5739 case IXGBE_PCIDEVCTRL2_16_32ms_def:
5740 return 32000ul; /* 32 ms */
5745 /* We shouldn't need to hit this path, but just in case default as
5746 * though completion timeout is not supported and support 32ms.
5751 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5753 unsigned long wait_delay, delay_interval;
5754 struct ixgbe_hw *hw = &adapter->hw;
5758 /* disable receives */
5759 hw->mac.ops.disable_rx(hw);
5761 if (ixgbe_removed(hw->hw_addr))
5764 /* disable all enabled Rx queues */
5765 for (i = 0; i < adapter->num_rx_queues; i++) {
5766 struct ixgbe_ring *ring = adapter->rx_ring[i];
5767 u8 reg_idx = ring->reg_idx;
5769 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5770 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5771 rxdctl |= IXGBE_RXDCTL_SWFLSH;
5773 /* write value back with RXDCTL.ENABLE bit cleared */
5774 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5777 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5778 if (hw->mac.type == ixgbe_mac_82598EB &&
5779 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5782 /* Determine our minimum delay interval. We will increase this value
5783 * with each subsequent test. This way if the device returns quickly
5784 * we should spend as little time as possible waiting, however as
5785 * the time increases we will wait for larger periods of time.
5787 * The trick here is that we increase the interval using the
5788 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5789 * of that wait is that it totals up to 100x whatever interval we
5790 * choose. Since our minimum wait is 100us we can just divide the
5791 * total timeout by 100 to get our minimum delay interval.
5793 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5795 wait_loop = IXGBE_MAX_RX_DESC_POLL;
5796 wait_delay = delay_interval;
5798 while (wait_loop--) {
5799 usleep_range(wait_delay, wait_delay + 10);
5800 wait_delay += delay_interval * 2;
5803 /* OR together the reading of all the active RXDCTL registers,
5804 * and then test the result. We need the disable to complete
5805 * before we start freeing the memory and invalidating the
5808 for (i = 0; i < adapter->num_rx_queues; i++) {
5809 struct ixgbe_ring *ring = adapter->rx_ring[i];
5810 u8 reg_idx = ring->reg_idx;
5812 rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5815 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5820 "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5823 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5825 unsigned long wait_delay, delay_interval;
5826 struct ixgbe_hw *hw = &adapter->hw;
5830 if (ixgbe_removed(hw->hw_addr))
5833 /* disable all enabled Tx queues */
5834 for (i = 0; i < adapter->num_tx_queues; i++) {
5835 struct ixgbe_ring *ring = adapter->tx_ring[i];
5836 u8 reg_idx = ring->reg_idx;
5838 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5841 /* disable all enabled XDP Tx queues */
5842 for (i = 0; i < adapter->num_xdp_queues; i++) {
5843 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5844 u8 reg_idx = ring->reg_idx;
5846 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5849 /* If the link is not up there shouldn't be much in the way of
5850 * pending transactions. Those that are left will be flushed out
5851 * when the reset logic goes through the flush sequence to clean out
5852 * the pending Tx transactions.
5854 if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5855 goto dma_engine_disable;
5857 /* Determine our minimum delay interval. We will increase this value
5858 * with each subsequent test. This way if the device returns quickly
5859 * we should spend as little time as possible waiting, however as
5860 * the time increases we will wait for larger periods of time.
5862 * The trick here is that we increase the interval using the
5863 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5864 * of that wait is that it totals up to 100x whatever interval we
5865 * choose. Since our minimum wait is 100us we can just divide the
5866 * total timeout by 100 to get our minimum delay interval.
5868 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5870 wait_loop = IXGBE_MAX_RX_DESC_POLL;
5871 wait_delay = delay_interval;
5873 while (wait_loop--) {
5874 usleep_range(wait_delay, wait_delay + 10);
5875 wait_delay += delay_interval * 2;
5878 /* OR together the reading of all the active TXDCTL registers,
5879 * and then test the result. We need the disable to complete
5880 * before we start freeing the memory and invalidating the
5883 for (i = 0; i < adapter->num_tx_queues; i++) {
5884 struct ixgbe_ring *ring = adapter->tx_ring[i];
5885 u8 reg_idx = ring->reg_idx;
5887 txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5889 for (i = 0; i < adapter->num_xdp_queues; i++) {
5890 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5891 u8 reg_idx = ring->reg_idx;
5893 txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5896 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5897 goto dma_engine_disable;
5901 "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5904 /* Disable the Tx DMA engine on 82599 and later MAC */
5905 switch (hw->mac.type) {
5906 case ixgbe_mac_82599EB:
5907 case ixgbe_mac_X540:
5908 case ixgbe_mac_X550:
5909 case ixgbe_mac_X550EM_x:
5910 case ixgbe_mac_x550em_a:
5911 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5912 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5913 ~IXGBE_DMATXCTL_TE));
5920 void ixgbe_reset(struct ixgbe_adapter *adapter)
5922 struct ixgbe_hw *hw = &adapter->hw;
5923 struct net_device *netdev = adapter->netdev;
5926 if (ixgbe_removed(hw->hw_addr))
5928 /* lock SFP init bit to prevent race conditions with the watchdog */
5929 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5930 usleep_range(1000, 2000);
5932 /* clear all SFP and link config related flags while holding SFP_INIT */
5933 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5934 IXGBE_FLAG2_SFP_NEEDS_RESET);
5935 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5937 err = hw->mac.ops.init_hw(hw);
5940 case IXGBE_ERR_SFP_NOT_PRESENT:
5941 case IXGBE_ERR_SFP_NOT_SUPPORTED:
5943 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5944 e_dev_err("master disable timed out\n");
5946 case IXGBE_ERR_EEPROM_VERSION:
5947 /* We are running on a pre-production device, log a warning */
5948 e_dev_warn("This device is a pre-production adapter/LOM. "
5949 "Please be aware there may be issues associated with "
5950 "your hardware. If you are experiencing problems "
5951 "please contact your Intel or hardware "
5952 "representative who provided you with this "
5956 e_dev_err("Hardware Error: %d\n", err);
5959 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5961 /* flush entries out of MAC table */
5962 ixgbe_flush_sw_mac_table(adapter);
5963 __dev_uc_unsync(netdev, NULL);
5965 /* do not flush user set addresses */
5966 ixgbe_mac_set_default_filter(adapter);
5968 /* update SAN MAC vmdq pool selection */
5969 if (hw->mac.san_mac_rar_index)
5970 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5972 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5973 ixgbe_ptp_reset(adapter);
5975 if (hw->phy.ops.set_phy_power) {
5976 if (!netif_running(adapter->netdev) && !adapter->wol)
5977 hw->phy.ops.set_phy_power(hw, false);
5979 hw->phy.ops.set_phy_power(hw, true);
5984 * ixgbe_clean_tx_ring - Free Tx Buffers
5985 * @tx_ring: ring to be cleaned
5987 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5989 u16 i = tx_ring->next_to_clean;
5990 struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
5992 if (tx_ring->xsk_pool) {
5993 ixgbe_xsk_clean_tx_ring(tx_ring);
5997 while (i != tx_ring->next_to_use) {
5998 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
6000 /* Free all the Tx ring sk_buffs */
6001 if (ring_is_xdp(tx_ring))
6002 xdp_return_frame(tx_buffer->xdpf);
6004 dev_kfree_skb_any(tx_buffer->skb);
6006 /* unmap skb header data */
6007 dma_unmap_single(tx_ring->dev,
6008 dma_unmap_addr(tx_buffer, dma),
6009 dma_unmap_len(tx_buffer, len),
6012 /* check for eop_desc to determine the end of the packet */
6013 eop_desc = tx_buffer->next_to_watch;
6014 tx_desc = IXGBE_TX_DESC(tx_ring, i);
6016 /* unmap remaining buffers */
6017 while (tx_desc != eop_desc) {
6021 if (unlikely(i == tx_ring->count)) {
6023 tx_buffer = tx_ring->tx_buffer_info;
6024 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6027 /* unmap any remaining paged data */
6028 if (dma_unmap_len(tx_buffer, len))
6029 dma_unmap_page(tx_ring->dev,
6030 dma_unmap_addr(tx_buffer, dma),
6031 dma_unmap_len(tx_buffer, len),
6035 /* move us one more past the eop_desc for start of next pkt */
6038 if (unlikely(i == tx_ring->count)) {
6040 tx_buffer = tx_ring->tx_buffer_info;
6044 /* reset BQL for queue */
6045 if (!ring_is_xdp(tx_ring))
6046 netdev_tx_reset_queue(txring_txq(tx_ring));
6049 /* reset next_to_use and next_to_clean */
6050 tx_ring->next_to_use = 0;
6051 tx_ring->next_to_clean = 0;
6055 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6056 * @adapter: board private structure
6058 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6062 for (i = 0; i < adapter->num_rx_queues; i++)
6063 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6067 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6068 * @adapter: board private structure
6070 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6074 for (i = 0; i < adapter->num_tx_queues; i++)
6075 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6076 for (i = 0; i < adapter->num_xdp_queues; i++)
6077 ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6080 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6082 struct hlist_node *node2;
6083 struct ixgbe_fdir_filter *filter;
6085 spin_lock(&adapter->fdir_perfect_lock);
6087 hlist_for_each_entry_safe(filter, node2,
6088 &adapter->fdir_filter_list, fdir_node) {
6089 hlist_del(&filter->fdir_node);
6092 adapter->fdir_filter_count = 0;
6094 spin_unlock(&adapter->fdir_perfect_lock);
6097 void ixgbe_down(struct ixgbe_adapter *adapter)
6099 struct net_device *netdev = adapter->netdev;
6100 struct ixgbe_hw *hw = &adapter->hw;
6103 /* signal that we are down to the interrupt handler */
6104 if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6105 return; /* do nothing if already down */
6107 /* Shut off incoming Tx traffic */
6108 netif_tx_stop_all_queues(netdev);
6110 /* call carrier off first to avoid false dev_watchdog timeouts */
6111 netif_carrier_off(netdev);
6112 netif_tx_disable(netdev);
6115 ixgbe_disable_rx(adapter);
6117 /* synchronize_rcu() needed for pending XDP buffers to drain */
6118 if (adapter->xdp_ring[0])
6121 ixgbe_irq_disable(adapter);
6123 ixgbe_napi_disable_all(adapter);
6125 clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6126 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6127 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6129 del_timer_sync(&adapter->service_timer);
6131 if (adapter->num_vfs) {
6132 /* Clear EITR Select mapping */
6133 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6135 /* Mark all the VFs as inactive */
6136 for (i = 0 ; i < adapter->num_vfs; i++)
6137 adapter->vfinfo[i].clear_to_send = false;
6139 /* ping all the active vfs to let them know we are going down */
6140 ixgbe_ping_all_vfs(adapter);
6142 /* Disable all VFTE/VFRE TX/RX */
6143 ixgbe_disable_tx_rx(adapter);
6146 /* disable transmits in the hardware now that interrupts are off */
6147 ixgbe_disable_tx(adapter);
6149 if (!pci_channel_offline(adapter->pdev))
6150 ixgbe_reset(adapter);
6152 /* power down the optics for 82599 SFP+ fiber */
6153 if (hw->mac.ops.disable_tx_laser)
6154 hw->mac.ops.disable_tx_laser(hw);
6156 ixgbe_clean_all_tx_rings(adapter);
6157 ixgbe_clean_all_rx_rings(adapter);
6161 * ixgbe_set_eee_capable - helper function to determine EEE support on X550
6162 * @adapter: board private structure
6164 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6166 struct ixgbe_hw *hw = &adapter->hw;
6168 switch (hw->device_id) {
6169 case IXGBE_DEV_ID_X550EM_A_1G_T:
6170 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6171 if (!hw->phy.eee_speeds_supported)
6173 adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6174 if (!hw->phy.eee_speeds_advertised)
6176 adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6179 adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6180 adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6186 * ixgbe_tx_timeout - Respond to a Tx Hang
6187 * @netdev: network interface device structure
6188 * @txqueue: queue number that timed out
6190 static void ixgbe_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
6192 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6194 /* Do the reset outside of interrupt context */
6195 ixgbe_tx_timeout_reset(adapter);
6198 #ifdef CONFIG_IXGBE_DCB
6199 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6201 struct ixgbe_hw *hw = &adapter->hw;
6202 struct tc_configuration *tc;
6205 switch (hw->mac.type) {
6206 case ixgbe_mac_82598EB:
6207 case ixgbe_mac_82599EB:
6208 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6209 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6211 case ixgbe_mac_X540:
6212 case ixgbe_mac_X550:
6213 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6214 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6216 case ixgbe_mac_X550EM_x:
6217 case ixgbe_mac_x550em_a:
6219 adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6220 adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6224 /* Configure DCB traffic classes */
6225 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6226 tc = &adapter->dcb_cfg.tc_config[j];
6227 tc->path[DCB_TX_CONFIG].bwg_id = 0;
6228 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6229 tc->path[DCB_RX_CONFIG].bwg_id = 0;
6230 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6231 tc->dcb_pfc = pfc_disabled;
6234 /* Initialize default user to priority mapping, UPx->TC0 */
6235 tc = &adapter->dcb_cfg.tc_config[0];
6236 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6237 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6239 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6240 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6241 adapter->dcb_cfg.pfc_mode_enable = false;
6242 adapter->dcb_set_bitmap = 0x00;
6243 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6244 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6245 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6246 sizeof(adapter->temp_dcb_cfg));
6251 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6252 * @adapter: board private structure to initialize
6253 * @ii: pointer to ixgbe_info for device
6255 * ixgbe_sw_init initializes the Adapter private data structure.
6256 * Fields are initialized based on PCI device information and
6257 * OS network device settings (MTU size).
6259 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6260 const struct ixgbe_info *ii)
6262 struct ixgbe_hw *hw = &adapter->hw;
6263 struct pci_dev *pdev = adapter->pdev;
6264 unsigned int rss, fdir;
6268 /* PCI config space info */
6270 hw->vendor_id = pdev->vendor;
6271 hw->device_id = pdev->device;
6272 hw->revision_id = pdev->revision;
6273 hw->subsystem_vendor_id = pdev->subsystem_vendor;
6274 hw->subsystem_device_id = pdev->subsystem_device;
6276 /* get_invariants needs the device IDs */
6277 ii->get_invariants(hw);
6279 /* Set common capability flags and settings */
6280 rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6281 adapter->ring_feature[RING_F_RSS].limit = rss;
6282 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6283 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6284 adapter->atr_sample_rate = 20;
6285 fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6286 adapter->ring_feature[RING_F_FDIR].limit = fdir;
6287 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6288 adapter->ring_feature[RING_F_VMDQ].limit = 1;
6289 #ifdef CONFIG_IXGBE_DCA
6290 adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6292 #ifdef CONFIG_IXGBE_DCB
6293 adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6294 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6297 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6298 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6299 #ifdef CONFIG_IXGBE_DCB
6300 /* Default traffic class to use for FCoE */
6301 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6302 #endif /* CONFIG_IXGBE_DCB */
6303 #endif /* IXGBE_FCOE */
6305 /* initialize static ixgbe jump table entries */
6306 adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6308 if (!adapter->jump_tables[0])
6310 adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6312 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6313 adapter->jump_tables[i] = NULL;
6315 adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6316 sizeof(struct ixgbe_mac_addr),
6318 if (!adapter->mac_table)
6321 if (ixgbe_init_rss_key(adapter))
6324 adapter->af_xdp_zc_qps = bitmap_zalloc(MAX_XDP_QUEUES, GFP_KERNEL);
6325 if (!adapter->af_xdp_zc_qps)
6328 /* Set MAC specific capability flags and exceptions */
6329 switch (hw->mac.type) {
6330 case ixgbe_mac_82598EB:
6331 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6333 if (hw->device_id == IXGBE_DEV_ID_82598AT)
6334 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6336 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6337 adapter->ring_feature[RING_F_FDIR].limit = 0;
6338 adapter->atr_sample_rate = 0;
6339 adapter->fdir_pballoc = 0;
6341 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6342 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6343 #ifdef CONFIG_IXGBE_DCB
6344 adapter->fcoe.up = 0;
6345 #endif /* IXGBE_DCB */
6346 #endif /* IXGBE_FCOE */
6348 case ixgbe_mac_82599EB:
6349 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6350 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6352 case ixgbe_mac_X540:
6353 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6354 if (fwsm & IXGBE_FWSM_TS_ENABLED)
6355 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6357 case ixgbe_mac_x550em_a:
6358 switch (hw->device_id) {
6359 case IXGBE_DEV_ID_X550EM_A_1G_T:
6360 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6361 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6367 case ixgbe_mac_X550EM_x:
6368 #ifdef CONFIG_IXGBE_DCB
6369 adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6372 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6373 #ifdef CONFIG_IXGBE_DCB
6374 adapter->fcoe.up = 0;
6375 #endif /* IXGBE_DCB */
6376 #endif /* IXGBE_FCOE */
6378 case ixgbe_mac_X550:
6379 if (hw->mac.type == ixgbe_mac_X550)
6380 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6381 #ifdef CONFIG_IXGBE_DCA
6382 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6390 /* FCoE support exists, always init the FCoE lock */
6391 spin_lock_init(&adapter->fcoe.lock);
6394 /* n-tuple support exists, always init our spinlock */
6395 spin_lock_init(&adapter->fdir_perfect_lock);
6397 #ifdef CONFIG_IXGBE_DCB
6398 ixgbe_init_dcb(adapter);
6400 ixgbe_init_ipsec_offload(adapter);
6402 /* default flow control settings */
6403 hw->fc.requested_mode = ixgbe_fc_full;
6404 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
6405 ixgbe_pbthresh_setup(adapter);
6406 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6407 hw->fc.send_xon = true;
6408 hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6410 #ifdef CONFIG_PCI_IOV
6412 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6414 /* assign number of SR-IOV VFs */
6415 if (hw->mac.type != ixgbe_mac_82598EB) {
6416 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6418 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6421 #endif /* CONFIG_PCI_IOV */
6423 /* enable itr by default in dynamic mode */
6424 adapter->rx_itr_setting = 1;
6425 adapter->tx_itr_setting = 1;
6427 /* set default ring sizes */
6428 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6429 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6431 /* set default work limits */
6432 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6434 /* initialize eeprom parameters */
6435 if (ixgbe_init_eeprom_params_generic(hw)) {
6436 e_dev_err("EEPROM initialization failed\n");
6440 /* PF holds first pool slot */
6441 set_bit(0, adapter->fwd_bitmask);
6442 set_bit(__IXGBE_DOWN, &adapter->state);
6448 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6449 * @tx_ring: tx descriptor ring (for a specific queue) to setup
6451 * Return 0 on success, negative on failure
6453 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6455 struct device *dev = tx_ring->dev;
6456 int orig_node = dev_to_node(dev);
6457 int ring_node = NUMA_NO_NODE;
6460 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6462 if (tx_ring->q_vector)
6463 ring_node = tx_ring->q_vector->numa_node;
6465 tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6466 if (!tx_ring->tx_buffer_info)
6467 tx_ring->tx_buffer_info = vmalloc(size);
6468 if (!tx_ring->tx_buffer_info)
6471 /* round up to nearest 4K */
6472 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6473 tx_ring->size = ALIGN(tx_ring->size, 4096);
6475 set_dev_node(dev, ring_node);
6476 tx_ring->desc = dma_alloc_coherent(dev,
6480 set_dev_node(dev, orig_node);
6482 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6483 &tx_ring->dma, GFP_KERNEL);
6487 tx_ring->next_to_use = 0;
6488 tx_ring->next_to_clean = 0;
6492 vfree(tx_ring->tx_buffer_info);
6493 tx_ring->tx_buffer_info = NULL;
6494 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6499 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6500 * @adapter: board private structure
6502 * If this function returns with an error, then it's possible one or
6503 * more of the rings is populated (while the rest are not). It is the
6504 * callers duty to clean those orphaned rings.
6506 * Return 0 on success, negative on failure
6508 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6510 int i, j = 0, err = 0;
6512 for (i = 0; i < adapter->num_tx_queues; i++) {
6513 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6517 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6520 for (j = 0; j < adapter->num_xdp_queues; j++) {
6521 err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6525 e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6531 /* rewind the index freeing the rings as we go */
6533 ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6535 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6540 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6541 * @adapter: pointer to ixgbe_adapter
6542 * @rx_ring: rx descriptor ring (for a specific queue) to setup
6544 * Returns 0 on success, negative on failure
6546 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6547 struct ixgbe_ring *rx_ring)
6549 struct device *dev = rx_ring->dev;
6550 int orig_node = dev_to_node(dev);
6551 int ring_node = NUMA_NO_NODE;
6554 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6556 if (rx_ring->q_vector)
6557 ring_node = rx_ring->q_vector->numa_node;
6559 rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6560 if (!rx_ring->rx_buffer_info)
6561 rx_ring->rx_buffer_info = vmalloc(size);
6562 if (!rx_ring->rx_buffer_info)
6565 /* Round up to nearest 4K */
6566 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6567 rx_ring->size = ALIGN(rx_ring->size, 4096);
6569 set_dev_node(dev, ring_node);
6570 rx_ring->desc = dma_alloc_coherent(dev,
6574 set_dev_node(dev, orig_node);
6576 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6577 &rx_ring->dma, GFP_KERNEL);
6581 rx_ring->next_to_clean = 0;
6582 rx_ring->next_to_use = 0;
6584 /* XDP RX-queue info */
6585 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6586 rx_ring->queue_index, rx_ring->q_vector->napi.napi_id) < 0)
6589 rx_ring->xdp_prog = adapter->xdp_prog;
6593 vfree(rx_ring->rx_buffer_info);
6594 rx_ring->rx_buffer_info = NULL;
6595 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6600 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6601 * @adapter: board private structure
6603 * If this function returns with an error, then it's possible one or
6604 * more of the rings is populated (while the rest are not). It is the
6605 * callers duty to clean those orphaned rings.
6607 * Return 0 on success, negative on failure
6609 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6613 for (i = 0; i < adapter->num_rx_queues; i++) {
6614 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6618 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6623 err = ixgbe_setup_fcoe_ddp_resources(adapter);
6628 /* rewind the index freeing the rings as we go */
6630 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6635 * ixgbe_free_tx_resources - Free Tx Resources per Queue
6636 * @tx_ring: Tx descriptor ring for a specific queue
6638 * Free all transmit software resources
6640 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6642 ixgbe_clean_tx_ring(tx_ring);
6644 vfree(tx_ring->tx_buffer_info);
6645 tx_ring->tx_buffer_info = NULL;
6647 /* if not set, then don't free */
6651 dma_free_coherent(tx_ring->dev, tx_ring->size,
6652 tx_ring->desc, tx_ring->dma);
6654 tx_ring->desc = NULL;
6658 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6659 * @adapter: board private structure
6661 * Free all transmit software resources
6663 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6667 for (i = 0; i < adapter->num_tx_queues; i++)
6668 if (adapter->tx_ring[i]->desc)
6669 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6670 for (i = 0; i < adapter->num_xdp_queues; i++)
6671 if (adapter->xdp_ring[i]->desc)
6672 ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6676 * ixgbe_free_rx_resources - Free Rx Resources
6677 * @rx_ring: ring to clean the resources from
6679 * Free all receive software resources
6681 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6683 ixgbe_clean_rx_ring(rx_ring);
6685 rx_ring->xdp_prog = NULL;
6686 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6687 vfree(rx_ring->rx_buffer_info);
6688 rx_ring->rx_buffer_info = NULL;
6690 /* if not set, then don't free */
6694 dma_free_coherent(rx_ring->dev, rx_ring->size,
6695 rx_ring->desc, rx_ring->dma);
6697 rx_ring->desc = NULL;
6701 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6702 * @adapter: board private structure
6704 * Free all receive software resources
6706 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6711 ixgbe_free_fcoe_ddp_resources(adapter);
6714 for (i = 0; i < adapter->num_rx_queues; i++)
6715 if (adapter->rx_ring[i]->desc)
6716 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6720 * ixgbe_change_mtu - Change the Maximum Transfer Unit
6721 * @netdev: network interface device structure
6722 * @new_mtu: new value for maximum frame size
6724 * Returns 0 on success, negative on failure
6726 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6728 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6730 if (adapter->xdp_prog) {
6731 int new_frame_size = new_mtu + ETH_HLEN + ETH_FCS_LEN +
6735 for (i = 0; i < adapter->num_rx_queues; i++) {
6736 struct ixgbe_ring *ring = adapter->rx_ring[i];
6738 if (new_frame_size > ixgbe_rx_bufsz(ring)) {
6739 e_warn(probe, "Requested MTU size is not supported with XDP\n");
6746 * For 82599EB we cannot allow legacy VFs to enable their receive
6747 * paths when MTU greater than 1500 is configured. So display a
6748 * warning that legacy VFs will be disabled.
6750 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6751 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6752 (new_mtu > ETH_DATA_LEN))
6753 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6755 netdev_dbg(netdev, "changing MTU from %d to %d\n",
6756 netdev->mtu, new_mtu);
6758 /* must set new MTU before calling down or up */
6759 netdev->mtu = new_mtu;
6761 if (netif_running(netdev))
6762 ixgbe_reinit_locked(adapter);
6768 * ixgbe_open - Called when a network interface is made active
6769 * @netdev: network interface device structure
6771 * Returns 0 on success, negative value on failure
6773 * The open entry point is called when a network interface is made
6774 * active by the system (IFF_UP). At this point all resources needed
6775 * for transmit and receive operations are allocated, the interrupt
6776 * handler is registered with the OS, the watchdog timer is started,
6777 * and the stack is notified that the interface is ready.
6779 int ixgbe_open(struct net_device *netdev)
6781 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6782 struct ixgbe_hw *hw = &adapter->hw;
6785 /* disallow open during test */
6786 if (test_bit(__IXGBE_TESTING, &adapter->state))
6789 netif_carrier_off(netdev);
6791 /* allocate transmit descriptors */
6792 err = ixgbe_setup_all_tx_resources(adapter);
6796 /* allocate receive descriptors */
6797 err = ixgbe_setup_all_rx_resources(adapter);
6801 ixgbe_configure(adapter);
6803 err = ixgbe_request_irq(adapter);
6807 /* Notify the stack of the actual queue counts. */
6808 queues = adapter->num_tx_queues;
6809 err = netif_set_real_num_tx_queues(netdev, queues);
6811 goto err_set_queues;
6813 queues = adapter->num_rx_queues;
6814 err = netif_set_real_num_rx_queues(netdev, queues);
6816 goto err_set_queues;
6818 ixgbe_ptp_init(adapter);
6820 ixgbe_up_complete(adapter);
6822 udp_tunnel_nic_reset_ntf(netdev);
6827 ixgbe_free_irq(adapter);
6829 ixgbe_free_all_rx_resources(adapter);
6830 if (hw->phy.ops.set_phy_power && !adapter->wol)
6831 hw->phy.ops.set_phy_power(&adapter->hw, false);
6833 ixgbe_free_all_tx_resources(adapter);
6835 ixgbe_reset(adapter);
6840 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6842 ixgbe_ptp_suspend(adapter);
6844 if (adapter->hw.phy.ops.enter_lplu) {
6845 adapter->hw.phy.reset_disable = true;
6846 ixgbe_down(adapter);
6847 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6848 adapter->hw.phy.reset_disable = false;
6850 ixgbe_down(adapter);
6853 ixgbe_free_irq(adapter);
6855 ixgbe_free_all_tx_resources(adapter);
6856 ixgbe_free_all_rx_resources(adapter);
6860 * ixgbe_close - Disables a network interface
6861 * @netdev: network interface device structure
6863 * Returns 0, this is not allowed to fail
6865 * The close entry point is called when an interface is de-activated
6866 * by the OS. The hardware is still under the drivers control, but
6867 * needs to be disabled. A global MAC reset is issued to stop the
6868 * hardware, and all transmit and receive resources are freed.
6870 int ixgbe_close(struct net_device *netdev)
6872 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6874 ixgbe_ptp_stop(adapter);
6876 if (netif_device_present(netdev))
6877 ixgbe_close_suspend(adapter);
6879 ixgbe_fdir_filter_exit(adapter);
6881 ixgbe_release_hw_control(adapter);
6886 static int __maybe_unused ixgbe_resume(struct device *dev_d)
6888 struct pci_dev *pdev = to_pci_dev(dev_d);
6889 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6890 struct net_device *netdev = adapter->netdev;
6893 adapter->hw.hw_addr = adapter->io_addr;
6895 smp_mb__before_atomic();
6896 clear_bit(__IXGBE_DISABLED, &adapter->state);
6897 pci_set_master(pdev);
6899 device_wakeup_disable(dev_d);
6901 ixgbe_reset(adapter);
6903 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6906 err = ixgbe_init_interrupt_scheme(adapter);
6907 if (!err && netif_running(netdev))
6908 err = ixgbe_open(netdev);
6912 netif_device_attach(netdev);
6918 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6920 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6921 struct net_device *netdev = adapter->netdev;
6922 struct ixgbe_hw *hw = &adapter->hw;
6924 u32 wufc = adapter->wol;
6927 netif_device_detach(netdev);
6929 if (netif_running(netdev))
6930 ixgbe_close_suspend(adapter);
6932 ixgbe_clear_interrupt_scheme(adapter);
6935 if (hw->mac.ops.stop_link_on_d3)
6936 hw->mac.ops.stop_link_on_d3(hw);
6941 ixgbe_set_rx_mode(netdev);
6943 /* enable the optics for 82599 SFP+ fiber as we can WoL */
6944 if (hw->mac.ops.enable_tx_laser)
6945 hw->mac.ops.enable_tx_laser(hw);
6947 /* enable the reception of multicast packets */
6948 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6949 fctrl |= IXGBE_FCTRL_MPE;
6950 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6952 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6953 ctrl |= IXGBE_CTRL_GIO_DIS;
6954 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6956 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6958 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6959 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6962 switch (hw->mac.type) {
6963 case ixgbe_mac_82598EB:
6964 pci_wake_from_d3(pdev, false);
6966 case ixgbe_mac_82599EB:
6967 case ixgbe_mac_X540:
6968 case ixgbe_mac_X550:
6969 case ixgbe_mac_X550EM_x:
6970 case ixgbe_mac_x550em_a:
6971 pci_wake_from_d3(pdev, !!wufc);
6977 *enable_wake = !!wufc;
6978 if (hw->phy.ops.set_phy_power && !*enable_wake)
6979 hw->phy.ops.set_phy_power(hw, false);
6981 ixgbe_release_hw_control(adapter);
6983 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
6984 pci_disable_device(pdev);
6989 static int __maybe_unused ixgbe_suspend(struct device *dev_d)
6991 struct pci_dev *pdev = to_pci_dev(dev_d);
6995 retval = __ixgbe_shutdown(pdev, &wake);
6997 device_set_wakeup_enable(dev_d, wake);
7002 static void ixgbe_shutdown(struct pci_dev *pdev)
7006 __ixgbe_shutdown(pdev, &wake);
7008 if (system_state == SYSTEM_POWER_OFF) {
7009 pci_wake_from_d3(pdev, wake);
7010 pci_set_power_state(pdev, PCI_D3hot);
7015 * ixgbe_update_stats - Update the board statistics counters.
7016 * @adapter: board private structure
7018 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7020 struct net_device *netdev = adapter->netdev;
7021 struct ixgbe_hw *hw = &adapter->hw;
7022 struct ixgbe_hw_stats *hwstats = &adapter->stats;
7024 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7025 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7026 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7027 u64 alloc_rx_page = 0;
7028 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7030 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7031 test_bit(__IXGBE_RESETTING, &adapter->state))
7034 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7037 for (i = 0; i < adapter->num_rx_queues; i++) {
7038 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7039 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7041 adapter->rsc_total_count = rsc_count;
7042 adapter->rsc_total_flush = rsc_flush;
7045 for (i = 0; i < adapter->num_rx_queues; i++) {
7046 struct ixgbe_ring *rx_ring = READ_ONCE(adapter->rx_ring[i]);
7050 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7051 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7052 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7053 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7054 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7055 bytes += rx_ring->stats.bytes;
7056 packets += rx_ring->stats.packets;
7058 adapter->non_eop_descs = non_eop_descs;
7059 adapter->alloc_rx_page = alloc_rx_page;
7060 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7061 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7062 adapter->hw_csum_rx_error = hw_csum_rx_error;
7063 netdev->stats.rx_bytes = bytes;
7064 netdev->stats.rx_packets = packets;
7068 /* gather some stats to the adapter struct that are per queue */
7069 for (i = 0; i < adapter->num_tx_queues; i++) {
7070 struct ixgbe_ring *tx_ring = READ_ONCE(adapter->tx_ring[i]);
7074 restart_queue += tx_ring->tx_stats.restart_queue;
7075 tx_busy += tx_ring->tx_stats.tx_busy;
7076 bytes += tx_ring->stats.bytes;
7077 packets += tx_ring->stats.packets;
7079 for (i = 0; i < adapter->num_xdp_queues; i++) {
7080 struct ixgbe_ring *xdp_ring = READ_ONCE(adapter->xdp_ring[i]);
7084 restart_queue += xdp_ring->tx_stats.restart_queue;
7085 tx_busy += xdp_ring->tx_stats.tx_busy;
7086 bytes += xdp_ring->stats.bytes;
7087 packets += xdp_ring->stats.packets;
7089 adapter->restart_queue = restart_queue;
7090 adapter->tx_busy = tx_busy;
7091 netdev->stats.tx_bytes = bytes;
7092 netdev->stats.tx_packets = packets;
7094 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7096 /* 8 register reads */
7097 for (i = 0; i < 8; i++) {
7098 /* for packet buffers not used, the register should read 0 */
7099 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7101 hwstats->mpc[i] += mpc;
7102 total_mpc += hwstats->mpc[i];
7103 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7104 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7105 switch (hw->mac.type) {
7106 case ixgbe_mac_82598EB:
7107 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7108 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7109 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7110 hwstats->pxonrxc[i] +=
7111 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7113 case ixgbe_mac_82599EB:
7114 case ixgbe_mac_X540:
7115 case ixgbe_mac_X550:
7116 case ixgbe_mac_X550EM_x:
7117 case ixgbe_mac_x550em_a:
7118 hwstats->pxonrxc[i] +=
7119 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7126 /*16 register reads */
7127 for (i = 0; i < 16; i++) {
7128 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7129 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7130 if ((hw->mac.type == ixgbe_mac_82599EB) ||
7131 (hw->mac.type == ixgbe_mac_X540) ||
7132 (hw->mac.type == ixgbe_mac_X550) ||
7133 (hw->mac.type == ixgbe_mac_X550EM_x) ||
7134 (hw->mac.type == ixgbe_mac_x550em_a)) {
7135 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7136 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7137 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7138 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7142 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7143 /* work around hardware counting issue */
7144 hwstats->gprc -= missed_rx;
7146 ixgbe_update_xoff_received(adapter);
7148 /* 82598 hardware only has a 32 bit counter in the high register */
7149 switch (hw->mac.type) {
7150 case ixgbe_mac_82598EB:
7151 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7152 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7153 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7154 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7156 case ixgbe_mac_X540:
7157 case ixgbe_mac_X550:
7158 case ixgbe_mac_X550EM_x:
7159 case ixgbe_mac_x550em_a:
7160 /* OS2BMC stats are X540 and later */
7161 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7162 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7163 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7164 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7166 case ixgbe_mac_82599EB:
7167 for (i = 0; i < 16; i++)
7168 adapter->hw_rx_no_dma_resources +=
7169 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7170 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7171 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7172 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7173 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7174 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7175 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7176 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7177 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7178 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7180 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7181 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7182 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7183 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7184 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7185 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7186 /* Add up per cpu counters for total ddp aloc fail */
7187 if (adapter->fcoe.ddp_pool) {
7188 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7189 struct ixgbe_fcoe_ddp_pool *ddp_pool;
7191 u64 noddp = 0, noddp_ext_buff = 0;
7192 for_each_possible_cpu(cpu) {
7193 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7194 noddp += ddp_pool->noddp;
7195 noddp_ext_buff += ddp_pool->noddp_ext_buff;
7197 hwstats->fcoe_noddp = noddp;
7198 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7200 #endif /* IXGBE_FCOE */
7205 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7206 hwstats->bprc += bprc;
7207 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7208 if (hw->mac.type == ixgbe_mac_82598EB)
7209 hwstats->mprc -= bprc;
7210 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7211 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7212 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7213 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7214 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7215 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7216 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7217 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7218 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7219 hwstats->lxontxc += lxon;
7220 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7221 hwstats->lxofftxc += lxoff;
7222 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7223 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7225 * 82598 errata - tx of flow control packets is included in tx counters
7227 xon_off_tot = lxon + lxoff;
7228 hwstats->gptc -= xon_off_tot;
7229 hwstats->mptc -= xon_off_tot;
7230 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7231 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7232 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7233 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7234 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7235 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7236 hwstats->ptc64 -= xon_off_tot;
7237 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7238 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7239 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7240 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7241 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7242 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7244 /* Fill out the OS statistics structure */
7245 netdev->stats.multicast = hwstats->mprc;
7248 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7249 netdev->stats.rx_dropped = 0;
7250 netdev->stats.rx_length_errors = hwstats->rlec;
7251 netdev->stats.rx_crc_errors = hwstats->crcerrs;
7252 netdev->stats.rx_missed_errors = total_mpc;
7256 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7257 * @adapter: pointer to the device adapter structure
7259 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7261 struct ixgbe_hw *hw = &adapter->hw;
7264 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7267 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7269 /* if interface is down do nothing */
7270 if (test_bit(__IXGBE_DOWN, &adapter->state))
7273 /* do nothing if we are not using signature filters */
7274 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7277 adapter->fdir_overflow++;
7279 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7280 for (i = 0; i < adapter->num_tx_queues; i++)
7281 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7282 &(adapter->tx_ring[i]->state));
7283 for (i = 0; i < adapter->num_xdp_queues; i++)
7284 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7285 &adapter->xdp_ring[i]->state);
7286 /* re-enable flow director interrupts */
7287 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7289 e_err(probe, "failed to finish FDIR re-initialization, "
7290 "ignored adding FDIR ATR filters\n");
7295 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7296 * @adapter: pointer to the device adapter structure
7298 * This function serves two purposes. First it strobes the interrupt lines
7299 * in order to make certain interrupts are occurring. Secondly it sets the
7300 * bits needed to check for TX hangs. As a result we should immediately
7301 * determine if a hang has occurred.
7303 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7305 struct ixgbe_hw *hw = &adapter->hw;
7309 /* If we're down, removing or resetting, just bail */
7310 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7311 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7312 test_bit(__IXGBE_RESETTING, &adapter->state))
7315 /* Force detection of hung controller */
7316 if (netif_carrier_ok(adapter->netdev)) {
7317 for (i = 0; i < adapter->num_tx_queues; i++)
7318 set_check_for_tx_hang(adapter->tx_ring[i]);
7319 for (i = 0; i < adapter->num_xdp_queues; i++)
7320 set_check_for_tx_hang(adapter->xdp_ring[i]);
7323 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7325 * for legacy and MSI interrupts don't set any bits
7326 * that are enabled for EIAM, because this operation
7327 * would set *both* EIMS and EICS for any bit in EIAM
7329 IXGBE_WRITE_REG(hw, IXGBE_EICS,
7330 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7332 /* get one bit for every active tx/rx interrupt vector */
7333 for (i = 0; i < adapter->num_q_vectors; i++) {
7334 struct ixgbe_q_vector *qv = adapter->q_vector[i];
7335 if (qv->rx.ring || qv->tx.ring)
7340 /* Cause software interrupt to ensure rings are cleaned */
7341 ixgbe_irq_rearm_queues(adapter, eics);
7345 * ixgbe_watchdog_update_link - update the link status
7346 * @adapter: pointer to the device adapter structure
7348 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7350 struct ixgbe_hw *hw = &adapter->hw;
7351 u32 link_speed = adapter->link_speed;
7352 bool link_up = adapter->link_up;
7353 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7355 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7358 if (hw->mac.ops.check_link) {
7359 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7361 /* always assume link is up, if no check link function */
7362 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7366 if (adapter->ixgbe_ieee_pfc)
7367 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7369 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7370 hw->mac.ops.fc_enable(hw);
7371 ixgbe_set_rx_drop_en(adapter);
7375 time_after(jiffies, (adapter->link_check_timeout +
7376 IXGBE_TRY_LINK_TIMEOUT))) {
7377 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7378 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7379 IXGBE_WRITE_FLUSH(hw);
7382 adapter->link_up = link_up;
7383 adapter->link_speed = link_speed;
7386 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7388 #ifdef CONFIG_IXGBE_DCB
7389 struct net_device *netdev = adapter->netdev;
7390 struct dcb_app app = {
7391 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7396 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7397 up = dcb_ieee_getapp_mask(netdev, &app);
7399 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7404 * ixgbe_watchdog_link_is_up - update netif_carrier status and
7405 * print link up message
7406 * @adapter: pointer to the device adapter structure
7408 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7410 struct net_device *netdev = adapter->netdev;
7411 struct ixgbe_hw *hw = &adapter->hw;
7412 u32 link_speed = adapter->link_speed;
7413 const char *speed_str;
7414 bool flow_rx, flow_tx;
7416 /* only continue if link was previously down */
7417 if (netif_carrier_ok(netdev))
7420 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7422 switch (hw->mac.type) {
7423 case ixgbe_mac_82598EB: {
7424 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7425 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7426 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7427 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7430 case ixgbe_mac_X540:
7431 case ixgbe_mac_X550:
7432 case ixgbe_mac_X550EM_x:
7433 case ixgbe_mac_x550em_a:
7434 case ixgbe_mac_82599EB: {
7435 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7436 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7437 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7438 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7447 adapter->last_rx_ptp_check = jiffies;
7449 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7450 ixgbe_ptp_start_cyclecounter(adapter);
7452 switch (link_speed) {
7453 case IXGBE_LINK_SPEED_10GB_FULL:
7454 speed_str = "10 Gbps";
7456 case IXGBE_LINK_SPEED_5GB_FULL:
7457 speed_str = "5 Gbps";
7459 case IXGBE_LINK_SPEED_2_5GB_FULL:
7460 speed_str = "2.5 Gbps";
7462 case IXGBE_LINK_SPEED_1GB_FULL:
7463 speed_str = "1 Gbps";
7465 case IXGBE_LINK_SPEED_100_FULL:
7466 speed_str = "100 Mbps";
7468 case IXGBE_LINK_SPEED_10_FULL:
7469 speed_str = "10 Mbps";
7472 speed_str = "unknown speed";
7475 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7476 ((flow_rx && flow_tx) ? "RX/TX" :
7478 (flow_tx ? "TX" : "None"))));
7480 netif_carrier_on(netdev);
7481 ixgbe_check_vf_rate_limit(adapter);
7483 /* enable transmits */
7484 netif_tx_wake_all_queues(adapter->netdev);
7486 /* update the default user priority for VFs */
7487 ixgbe_update_default_up(adapter);
7489 /* ping all the active vfs to let them know link has changed */
7490 ixgbe_ping_all_vfs(adapter);
7494 * ixgbe_watchdog_link_is_down - update netif_carrier status and
7495 * print link down message
7496 * @adapter: pointer to the adapter structure
7498 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7500 struct net_device *netdev = adapter->netdev;
7501 struct ixgbe_hw *hw = &adapter->hw;
7503 adapter->link_up = false;
7504 adapter->link_speed = 0;
7506 /* only continue if link was up previously */
7507 if (!netif_carrier_ok(netdev))
7510 /* poll for SFP+ cable when link is down */
7511 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7512 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7514 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7515 ixgbe_ptp_start_cyclecounter(adapter);
7517 e_info(drv, "NIC Link is Down\n");
7518 netif_carrier_off(netdev);
7520 /* ping all the active vfs to let them know link has changed */
7521 ixgbe_ping_all_vfs(adapter);
7524 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7528 for (i = 0; i < adapter->num_tx_queues; i++) {
7529 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7531 if (tx_ring->next_to_use != tx_ring->next_to_clean)
7535 for (i = 0; i < adapter->num_xdp_queues; i++) {
7536 struct ixgbe_ring *ring = adapter->xdp_ring[i];
7538 if (ring->next_to_use != ring->next_to_clean)
7545 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7547 struct ixgbe_hw *hw = &adapter->hw;
7548 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7549 u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7553 if (!adapter->num_vfs)
7556 /* resetting the PF is only needed for MAC before X550 */
7557 if (hw->mac.type >= ixgbe_mac_X550)
7560 for (i = 0; i < adapter->num_vfs; i++) {
7561 for (j = 0; j < q_per_pool; j++) {
7564 h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7565 t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7576 * ixgbe_watchdog_flush_tx - flush queues on link down
7577 * @adapter: pointer to the device adapter structure
7579 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7581 if (!netif_carrier_ok(adapter->netdev)) {
7582 if (ixgbe_ring_tx_pending(adapter) ||
7583 ixgbe_vf_tx_pending(adapter)) {
7584 /* We've lost link, so the controller stops DMA,
7585 * but we've got queued Tx work that's never going
7586 * to get done, so reset controller to flush Tx.
7587 * (Do the reset outside of interrupt context).
7589 e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7590 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7595 #ifdef CONFIG_PCI_IOV
7596 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7598 struct ixgbe_hw *hw = &adapter->hw;
7599 struct pci_dev *pdev = adapter->pdev;
7603 if (!(netif_carrier_ok(adapter->netdev)))
7606 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7607 if (gpc) /* If incrementing then no need for the check below */
7609 /* Check to see if a bad DMA write target from an errant or
7610 * malicious VF has caused a PCIe error. If so then we can
7611 * issue a VFLR to the offending VF(s) and then resume without
7612 * requesting a full slot reset.
7618 /* check status reg for all VFs owned by this PF */
7619 for (vf = 0; vf < adapter->num_vfs; ++vf) {
7620 struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7625 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7626 if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7627 status_reg & PCI_STATUS_REC_MASTER_ABORT)
7632 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7636 /* Do not perform spoof check for 82598 or if not in IOV mode */
7637 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7638 adapter->num_vfs == 0)
7641 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7644 * ssvpc register is cleared on read, if zero then no
7645 * spoofed packets in the last interval.
7650 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7653 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7658 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7661 #endif /* CONFIG_PCI_IOV */
7665 * ixgbe_watchdog_subtask - check and bring link up
7666 * @adapter: pointer to the device adapter structure
7668 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7670 /* if interface is down, removing or resetting, do nothing */
7671 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7672 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7673 test_bit(__IXGBE_RESETTING, &adapter->state))
7676 ixgbe_watchdog_update_link(adapter);
7678 if (adapter->link_up)
7679 ixgbe_watchdog_link_is_up(adapter);
7681 ixgbe_watchdog_link_is_down(adapter);
7683 ixgbe_check_for_bad_vf(adapter);
7684 ixgbe_spoof_check(adapter);
7685 ixgbe_update_stats(adapter);
7687 ixgbe_watchdog_flush_tx(adapter);
7691 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7692 * @adapter: the ixgbe adapter structure
7694 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7696 struct ixgbe_hw *hw = &adapter->hw;
7699 /* not searching for SFP so there is nothing to do here */
7700 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7701 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7704 if (adapter->sfp_poll_time &&
7705 time_after(adapter->sfp_poll_time, jiffies))
7706 return; /* If not yet time to poll for SFP */
7708 /* someone else is in init, wait until next service event */
7709 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7712 adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7714 err = hw->phy.ops.identify_sfp(hw);
7715 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7718 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7719 /* If no cable is present, then we need to reset
7720 * the next time we find a good cable. */
7721 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7728 /* exit if reset not needed */
7729 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7732 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7735 * A module may be identified correctly, but the EEPROM may not have
7736 * support for that module. setup_sfp() will fail in that case, so
7737 * we should not allow that module to load.
7739 if (hw->mac.type == ixgbe_mac_82598EB)
7740 err = hw->phy.ops.reset(hw);
7742 err = hw->mac.ops.setup_sfp(hw);
7744 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7747 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7748 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7751 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7753 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7754 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7755 e_dev_err("failed to initialize because an unsupported "
7756 "SFP+ module type was detected.\n");
7757 e_dev_err("Reload the driver after installing a "
7758 "supported module.\n");
7759 unregister_netdev(adapter->netdev);
7764 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7765 * @adapter: the ixgbe adapter structure
7767 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7769 struct ixgbe_hw *hw = &adapter->hw;
7772 bool autoneg = false;
7774 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7777 /* someone else is in init, wait until next service event */
7778 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7781 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7783 hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7785 /* advertise highest capable link speed */
7786 if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7787 speed = IXGBE_LINK_SPEED_10GB_FULL;
7789 speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7790 IXGBE_LINK_SPEED_1GB_FULL);
7792 if (hw->mac.ops.setup_link)
7793 hw->mac.ops.setup_link(hw, speed, true);
7795 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7796 adapter->link_check_timeout = jiffies;
7797 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7801 * ixgbe_service_timer - Timer Call-back
7802 * @t: pointer to timer_list structure
7804 static void ixgbe_service_timer(struct timer_list *t)
7806 struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7807 unsigned long next_event_offset;
7809 /* poll faster when waiting for link */
7810 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7811 next_event_offset = HZ / 10;
7813 next_event_offset = HZ * 2;
7815 /* Reset the timer */
7816 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7818 ixgbe_service_event_schedule(adapter);
7821 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7823 struct ixgbe_hw *hw = &adapter->hw;
7826 if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7829 adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7831 if (!hw->phy.ops.handle_lasi)
7834 status = hw->phy.ops.handle_lasi(&adapter->hw);
7835 if (status != IXGBE_ERR_OVERTEMP)
7838 e_crit(drv, "%s\n", ixgbe_overheat_msg);
7841 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7843 if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7847 /* If we're already down, removing or resetting, just bail */
7848 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7849 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7850 test_bit(__IXGBE_RESETTING, &adapter->state)) {
7855 ixgbe_dump(adapter);
7856 netdev_err(adapter->netdev, "Reset adapter\n");
7857 adapter->tx_timeout_count++;
7859 ixgbe_reinit_locked(adapter);
7864 * ixgbe_check_fw_error - Check firmware for errors
7865 * @adapter: the adapter private structure
7867 * Check firmware errors in register FWSM
7869 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
7871 struct ixgbe_hw *hw = &adapter->hw;
7874 /* read fwsm.ext_err_ind register and log errors */
7875 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
7877 if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
7878 !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
7879 e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7882 if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
7883 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");
7891 * ixgbe_service_task - manages and runs subtasks
7892 * @work: pointer to work_struct containing our data
7894 static void ixgbe_service_task(struct work_struct *work)
7896 struct ixgbe_adapter *adapter = container_of(work,
7897 struct ixgbe_adapter,
7899 if (ixgbe_removed(adapter->hw.hw_addr)) {
7900 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7902 ixgbe_down(adapter);
7905 ixgbe_service_event_complete(adapter);
7908 if (ixgbe_check_fw_error(adapter)) {
7909 if (!test_bit(__IXGBE_DOWN, &adapter->state))
7910 unregister_netdev(adapter->netdev);
7911 ixgbe_service_event_complete(adapter);
7914 ixgbe_reset_subtask(adapter);
7915 ixgbe_phy_interrupt_subtask(adapter);
7916 ixgbe_sfp_detection_subtask(adapter);
7917 ixgbe_sfp_link_config_subtask(adapter);
7918 ixgbe_check_overtemp_subtask(adapter);
7919 ixgbe_watchdog_subtask(adapter);
7920 ixgbe_fdir_reinit_subtask(adapter);
7921 ixgbe_check_hang_subtask(adapter);
7923 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7924 ixgbe_ptp_overflow_check(adapter);
7925 if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
7926 ixgbe_ptp_rx_hang(adapter);
7927 ixgbe_ptp_tx_hang(adapter);
7930 ixgbe_service_event_complete(adapter);
7933 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7934 struct ixgbe_tx_buffer *first,
7936 struct ixgbe_ipsec_tx_data *itd)
7938 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7939 struct sk_buff *skb = first->skb;
7950 u32 paylen, l4_offset;
7951 u32 fceof_saidx = 0;
7954 if (skb->ip_summed != CHECKSUM_PARTIAL)
7957 if (!skb_is_gso(skb))
7960 err = skb_cow_head(skb, 0);
7964 if (eth_p_mpls(first->protocol))
7965 ip.hdr = skb_inner_network_header(skb);
7967 ip.hdr = skb_network_header(skb);
7968 l4.hdr = skb_checksum_start(skb);
7970 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7971 type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
7972 IXGBE_ADVTXD_TUCMD_L4T_UDP : IXGBE_ADVTXD_TUCMD_L4T_TCP;
7974 /* initialize outer IP header fields */
7975 if (ip.v4->version == 4) {
7976 unsigned char *csum_start = skb_checksum_start(skb);
7977 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
7978 int len = csum_start - trans_start;
7980 /* IP header will have to cancel out any data that
7981 * is not a part of the outer IP header, so set to
7982 * a reverse csum if needed, else init check to 0.
7984 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
7985 csum_fold(csum_partial(trans_start,
7987 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7990 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7991 IXGBE_TX_FLAGS_CSUM |
7992 IXGBE_TX_FLAGS_IPV4;
7994 ip.v6->payload_len = 0;
7995 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7996 IXGBE_TX_FLAGS_CSUM;
7999 /* determine offset of inner transport header */
8000 l4_offset = l4.hdr - skb->data;
8002 /* remove payload length from inner checksum */
8003 paylen = skb->len - l4_offset;
8005 if (type_tucmd & IXGBE_ADVTXD_TUCMD_L4T_TCP) {
8006 /* compute length of segmentation header */
8007 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
8008 csum_replace_by_diff(&l4.tcp->check,
8009 (__force __wsum)htonl(paylen));
8011 /* compute length of segmentation header */
8012 *hdr_len = sizeof(*l4.udp) + l4_offset;
8013 csum_replace_by_diff(&l4.udp->check,
8014 (__force __wsum)htonl(paylen));
8017 /* update gso size and bytecount with header size */
8018 first->gso_segs = skb_shinfo(skb)->gso_segs;
8019 first->bytecount += (first->gso_segs - 1) * *hdr_len;
8021 /* mss_l4len_id: use 0 as index for TSO */
8022 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8023 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8025 fceof_saidx |= itd->sa_idx;
8026 type_tucmd |= itd->flags | itd->trailer_len;
8028 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8029 vlan_macip_lens = l4.hdr - ip.hdr;
8030 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8031 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8033 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8039 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8040 struct ixgbe_tx_buffer *first,
8041 struct ixgbe_ipsec_tx_data *itd)
8043 struct sk_buff *skb = first->skb;
8044 u32 vlan_macip_lens = 0;
8045 u32 fceof_saidx = 0;
8048 if (skb->ip_summed != CHECKSUM_PARTIAL) {
8050 if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8051 IXGBE_TX_FLAGS_CC)))
8056 switch (skb->csum_offset) {
8057 case offsetof(struct tcphdr, check):
8058 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8060 case offsetof(struct udphdr, check):
8062 case offsetof(struct sctphdr, checksum):
8063 /* validate that this is actually an SCTP request */
8064 if (skb_csum_is_sctp(skb)) {
8065 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8070 skb_checksum_help(skb);
8074 /* update TX checksum flag */
8075 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8076 vlan_macip_lens = skb_checksum_start_offset(skb) -
8077 skb_network_offset(skb);
8079 /* vlan_macip_lens: MACLEN, VLAN tag */
8080 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8081 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8083 fceof_saidx |= itd->sa_idx;
8084 type_tucmd |= itd->flags | itd->trailer_len;
8086 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8089 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8090 ((_flag <= _result) ? \
8091 ((u32)(_input & _flag) * (_result / _flag)) : \
8092 ((u32)(_input & _flag) / (_flag / _result)))
8094 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8096 /* set type for advanced descriptor with frame checksum insertion */
8097 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8098 IXGBE_ADVTXD_DCMD_DEXT |
8099 IXGBE_ADVTXD_DCMD_IFCS;
8101 /* set HW vlan bit if vlan is present */
8102 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8103 IXGBE_ADVTXD_DCMD_VLE);
8105 /* set segmentation enable bits for TSO/FSO */
8106 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8107 IXGBE_ADVTXD_DCMD_TSE);
8109 /* set timestamp bit if present */
8110 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8111 IXGBE_ADVTXD_MAC_TSTAMP);
8113 /* insert frame checksum */
8114 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8119 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8120 u32 tx_flags, unsigned int paylen)
8122 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8124 /* enable L4 checksum for TSO and TX checksum offload */
8125 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8126 IXGBE_TX_FLAGS_CSUM,
8127 IXGBE_ADVTXD_POPTS_TXSM);
8129 /* enable IPv4 checksum for TSO */
8130 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8131 IXGBE_TX_FLAGS_IPV4,
8132 IXGBE_ADVTXD_POPTS_IXSM);
8135 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8136 IXGBE_TX_FLAGS_IPSEC,
8137 IXGBE_ADVTXD_POPTS_IPSEC);
8140 * Check Context must be set if Tx switch is enabled, which it
8141 * always is for case where virtual functions are running
8143 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8147 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8150 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8152 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8154 /* Herbert's original patch had:
8155 * smp_mb__after_netif_stop_queue();
8156 * but since that doesn't exist yet, just open code it.
8160 /* We need to check again in a case another CPU has just
8161 * made room available.
8163 if (likely(ixgbe_desc_unused(tx_ring) < size))
8166 /* A reprieve! - use start_queue because it doesn't call schedule */
8167 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8168 ++tx_ring->tx_stats.restart_queue;
8172 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8174 if (likely(ixgbe_desc_unused(tx_ring) >= size))
8177 return __ixgbe_maybe_stop_tx(tx_ring, size);
8180 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8181 struct ixgbe_tx_buffer *first,
8184 struct sk_buff *skb = first->skb;
8185 struct ixgbe_tx_buffer *tx_buffer;
8186 union ixgbe_adv_tx_desc *tx_desc;
8189 unsigned int data_len, size;
8190 u32 tx_flags = first->tx_flags;
8191 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8192 u16 i = tx_ring->next_to_use;
8194 tx_desc = IXGBE_TX_DESC(tx_ring, i);
8196 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8198 size = skb_headlen(skb);
8199 data_len = skb->data_len;
8202 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8203 if (data_len < sizeof(struct fcoe_crc_eof)) {
8204 size -= sizeof(struct fcoe_crc_eof) - data_len;
8207 data_len -= sizeof(struct fcoe_crc_eof);
8212 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8216 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8217 if (dma_mapping_error(tx_ring->dev, dma))
8220 /* record length, and DMA address */
8221 dma_unmap_len_set(tx_buffer, len, size);
8222 dma_unmap_addr_set(tx_buffer, dma, dma);
8224 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8226 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8227 tx_desc->read.cmd_type_len =
8228 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8232 if (i == tx_ring->count) {
8233 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8236 tx_desc->read.olinfo_status = 0;
8238 dma += IXGBE_MAX_DATA_PER_TXD;
8239 size -= IXGBE_MAX_DATA_PER_TXD;
8241 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8244 if (likely(!data_len))
8247 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8251 if (i == tx_ring->count) {
8252 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8255 tx_desc->read.olinfo_status = 0;
8258 size = min_t(unsigned int, data_len, skb_frag_size(frag));
8260 size = skb_frag_size(frag);
8264 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8267 tx_buffer = &tx_ring->tx_buffer_info[i];
8270 /* write last descriptor with RS and EOP bits */
8271 cmd_type |= size | IXGBE_TXD_CMD;
8272 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8274 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8276 /* set the timestamp */
8277 first->time_stamp = jiffies;
8279 skb_tx_timestamp(skb);
8282 * Force memory writes to complete before letting h/w know there
8283 * are new descriptors to fetch. (Only applicable for weak-ordered
8284 * memory model archs, such as IA-64).
8286 * We also need this memory barrier to make certain all of the
8287 * status bits have been updated before next_to_watch is written.
8291 /* set next_to_watch value indicating a packet is present */
8292 first->next_to_watch = tx_desc;
8295 if (i == tx_ring->count)
8298 tx_ring->next_to_use = i;
8300 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8302 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
8303 writel(i, tx_ring->tail);
8308 dev_err(tx_ring->dev, "TX DMA map failed\n");
8310 /* clear dma mappings for failed tx_buffer_info map */
8312 tx_buffer = &tx_ring->tx_buffer_info[i];
8313 if (dma_unmap_len(tx_buffer, len))
8314 dma_unmap_page(tx_ring->dev,
8315 dma_unmap_addr(tx_buffer, dma),
8316 dma_unmap_len(tx_buffer, len),
8318 dma_unmap_len_set(tx_buffer, len, 0);
8319 if (tx_buffer == first)
8322 i += tx_ring->count;
8326 dev_kfree_skb_any(first->skb);
8329 tx_ring->next_to_use = i;
8334 static void ixgbe_atr(struct ixgbe_ring *ring,
8335 struct ixgbe_tx_buffer *first)
8337 struct ixgbe_q_vector *q_vector = ring->q_vector;
8338 union ixgbe_atr_hash_dword input = { .dword = 0 };
8339 union ixgbe_atr_hash_dword common = { .dword = 0 };
8341 unsigned char *network;
8343 struct ipv6hdr *ipv6;
8347 struct sk_buff *skb;
8351 /* if ring doesn't have a interrupt vector, cannot perform ATR */
8355 /* do nothing if sampling is disabled */
8356 if (!ring->atr_sample_rate)
8361 /* currently only IPv4/IPv6 with TCP is supported */
8362 if ((first->protocol != htons(ETH_P_IP)) &&
8363 (first->protocol != htons(ETH_P_IPV6)))
8366 /* snag network header to get L4 type and address */
8368 hdr.network = skb_network_header(skb);
8369 if (unlikely(hdr.network <= skb->data))
8371 if (skb->encapsulation &&
8372 first->protocol == htons(ETH_P_IP) &&
8373 hdr.ipv4->protocol == IPPROTO_UDP) {
8374 struct ixgbe_adapter *adapter = q_vector->adapter;
8376 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8380 /* verify the port is recognized as VXLAN */
8381 if (adapter->vxlan_port &&
8382 udp_hdr(skb)->dest == adapter->vxlan_port)
8383 hdr.network = skb_inner_network_header(skb);
8385 if (adapter->geneve_port &&
8386 udp_hdr(skb)->dest == adapter->geneve_port)
8387 hdr.network = skb_inner_network_header(skb);
8390 /* Make sure we have at least [minimum IPv4 header + TCP]
8391 * or [IPv6 header] bytes
8393 if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8396 /* Currently only IPv4/IPv6 with TCP is supported */
8397 switch (hdr.ipv4->version) {
8399 /* access ihl as u8 to avoid unaligned access on ia64 */
8400 hlen = (hdr.network[0] & 0x0F) << 2;
8401 l4_proto = hdr.ipv4->protocol;
8404 hlen = hdr.network - skb->data;
8405 l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8406 hlen -= hdr.network - skb->data;
8412 if (l4_proto != IPPROTO_TCP)
8415 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8416 hlen + sizeof(struct tcphdr)))
8419 th = (struct tcphdr *)(hdr.network + hlen);
8421 /* skip this packet since the socket is closing */
8425 /* sample on all syn packets or once every atr sample count */
8426 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8429 /* reset sample count */
8430 ring->atr_count = 0;
8432 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8435 * src and dst are inverted, think how the receiver sees them
8437 * The input is broken into two sections, a non-compressed section
8438 * containing vm_pool, vlan_id, and flow_type. The rest of the data
8439 * is XORed together and stored in the compressed dword.
8441 input.formatted.vlan_id = vlan_id;
8444 * since src port and flex bytes occupy the same word XOR them together
8445 * and write the value to source port portion of compressed dword
8447 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8448 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8450 common.port.src ^= th->dest ^ first->protocol;
8451 common.port.dst ^= th->source;
8453 switch (hdr.ipv4->version) {
8455 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8456 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8459 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8460 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8461 hdr.ipv6->saddr.s6_addr32[1] ^
8462 hdr.ipv6->saddr.s6_addr32[2] ^
8463 hdr.ipv6->saddr.s6_addr32[3] ^
8464 hdr.ipv6->daddr.s6_addr32[0] ^
8465 hdr.ipv6->daddr.s6_addr32[1] ^
8466 hdr.ipv6->daddr.s6_addr32[2] ^
8467 hdr.ipv6->daddr.s6_addr32[3];
8473 if (hdr.network != skb_network_header(skb))
8474 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8476 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
8477 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8478 input, common, ring->queue_index);
8482 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8483 struct net_device *sb_dev)
8485 struct ixgbe_adapter *adapter;
8486 struct ixgbe_ring_feature *f;
8490 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8491 struct net_device *vdev = sb_dev;
8493 txq = vdev->tc_to_txq[tc].offset;
8494 txq += reciprocal_scale(skb_get_hash(skb),
8495 vdev->tc_to_txq[tc].count);
8501 * only execute the code below if protocol is FCoE
8502 * or FIP and we have FCoE enabled on the adapter
8504 switch (vlan_get_protocol(skb)) {
8505 case htons(ETH_P_FCOE):
8506 case htons(ETH_P_FIP):
8507 adapter = netdev_priv(dev);
8509 if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8513 return netdev_pick_tx(dev, skb, sb_dev);
8516 f = &adapter->ring_feature[RING_F_FCOE];
8518 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8521 while (txq >= f->indices)
8524 return txq + f->offset;
8528 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8529 struct xdp_frame *xdpf)
8531 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8532 struct ixgbe_tx_buffer *tx_buffer;
8533 union ixgbe_adv_tx_desc *tx_desc;
8540 if (unlikely(!ixgbe_desc_unused(ring)))
8541 return IXGBE_XDP_CONSUMED;
8543 dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
8544 if (dma_mapping_error(ring->dev, dma))
8545 return IXGBE_XDP_CONSUMED;
8547 /* record the location of the first descriptor for this packet */
8548 tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8549 tx_buffer->bytecount = len;
8550 tx_buffer->gso_segs = 1;
8551 tx_buffer->protocol = 0;
8553 i = ring->next_to_use;
8554 tx_desc = IXGBE_TX_DESC(ring, i);
8556 dma_unmap_len_set(tx_buffer, len, len);
8557 dma_unmap_addr_set(tx_buffer, dma, dma);
8558 tx_buffer->xdpf = xdpf;
8560 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8562 /* put descriptor type bits */
8563 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8564 IXGBE_ADVTXD_DCMD_DEXT |
8565 IXGBE_ADVTXD_DCMD_IFCS;
8566 cmd_type |= len | IXGBE_TXD_CMD;
8567 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8568 tx_desc->read.olinfo_status =
8569 cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8571 /* Avoid any potential race with xdp_xmit and cleanup */
8574 /* set next_to_watch value indicating a packet is present */
8576 if (i == ring->count)
8579 tx_buffer->next_to_watch = tx_desc;
8580 ring->next_to_use = i;
8582 return IXGBE_XDP_TX;
8585 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8586 struct ixgbe_adapter *adapter,
8587 struct ixgbe_ring *tx_ring)
8589 struct ixgbe_tx_buffer *first;
8593 u16 count = TXD_USE_COUNT(skb_headlen(skb));
8594 struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8595 __be16 protocol = skb->protocol;
8599 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8600 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8601 * + 2 desc gap to keep tail from touching head,
8602 * + 1 desc for context descriptor,
8603 * otherwise try next time
8605 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8606 count += TXD_USE_COUNT(skb_frag_size(
8607 &skb_shinfo(skb)->frags[f]));
8609 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8610 tx_ring->tx_stats.tx_busy++;
8611 return NETDEV_TX_BUSY;
8614 /* record the location of the first descriptor for this packet */
8615 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8617 first->bytecount = skb->len;
8618 first->gso_segs = 1;
8620 /* if we have a HW VLAN tag being added default to the HW one */
8621 if (skb_vlan_tag_present(skb)) {
8622 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8623 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8624 /* else if it is a SW VLAN check the next protocol and store the tag */
8625 } else if (protocol == htons(ETH_P_8021Q)) {
8626 struct vlan_hdr *vhdr, _vhdr;
8627 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8631 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8632 IXGBE_TX_FLAGS_VLAN_SHIFT;
8633 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8635 protocol = vlan_get_protocol(skb);
8637 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8638 adapter->ptp_clock) {
8639 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
8640 !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8642 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8643 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8645 /* schedule check for Tx timestamp */
8646 adapter->ptp_tx_skb = skb_get(skb);
8647 adapter->ptp_tx_start = jiffies;
8648 schedule_work(&adapter->ptp_tx_work);
8650 adapter->tx_hwtstamp_skipped++;
8654 #ifdef CONFIG_PCI_IOV
8656 * Use the l2switch_enable flag - would be false if the DMA
8657 * Tx switch had been disabled.
8659 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8660 tx_flags |= IXGBE_TX_FLAGS_CC;
8663 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8664 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8665 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8666 (skb->priority != TC_PRIO_CONTROL))) {
8667 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8668 tx_flags |= (skb->priority & 0x7) <<
8669 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8670 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8671 struct vlan_ethhdr *vhdr;
8673 if (skb_cow_head(skb, 0))
8675 vhdr = (struct vlan_ethhdr *)skb->data;
8676 vhdr->h_vlan_TCI = htons(tx_flags >>
8677 IXGBE_TX_FLAGS_VLAN_SHIFT);
8679 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8683 /* record initial flags and protocol */
8684 first->tx_flags = tx_flags;
8685 first->protocol = protocol;
8688 /* setup tx offload for FCoE */
8689 if ((protocol == htons(ETH_P_FCOE)) &&
8690 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8691 tso = ixgbe_fso(tx_ring, first, &hdr_len);
8698 #endif /* IXGBE_FCOE */
8700 #ifdef CONFIG_IXGBE_IPSEC
8701 if (xfrm_offload(skb) &&
8702 !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8705 tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8709 ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8711 /* add the ATR filter if ATR is on */
8712 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8713 ixgbe_atr(tx_ring, first);
8717 #endif /* IXGBE_FCOE */
8718 if (ixgbe_tx_map(tx_ring, first, hdr_len))
8719 goto cleanup_tx_timestamp;
8721 return NETDEV_TX_OK;
8724 dev_kfree_skb_any(first->skb);
8726 cleanup_tx_timestamp:
8727 if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8728 dev_kfree_skb_any(adapter->ptp_tx_skb);
8729 adapter->ptp_tx_skb = NULL;
8730 cancel_work_sync(&adapter->ptp_tx_work);
8731 clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8734 return NETDEV_TX_OK;
8737 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8738 struct net_device *netdev,
8739 struct ixgbe_ring *ring)
8741 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8742 struct ixgbe_ring *tx_ring;
8745 * The minimum packet size for olinfo paylen is 17 so pad the skb
8746 * in order to meet this minimum size requirement.
8748 if (skb_put_padto(skb, 17))
8749 return NETDEV_TX_OK;
8751 tx_ring = ring ? ring : adapter->tx_ring[skb_get_queue_mapping(skb)];
8752 if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8753 return NETDEV_TX_BUSY;
8755 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8758 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8759 struct net_device *netdev)
8761 return __ixgbe_xmit_frame(skb, netdev, NULL);
8765 * ixgbe_set_mac - Change the Ethernet Address of the NIC
8766 * @netdev: network interface device structure
8767 * @p: pointer to an address structure
8769 * Returns 0 on success, negative on failure
8771 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8773 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8774 struct ixgbe_hw *hw = &adapter->hw;
8775 struct sockaddr *addr = p;
8777 if (!is_valid_ether_addr(addr->sa_data))
8778 return -EADDRNOTAVAIL;
8780 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8781 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8783 ixgbe_mac_set_default_filter(adapter);
8789 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8791 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8792 struct ixgbe_hw *hw = &adapter->hw;
8796 if (adapter->mii_bus) {
8799 if (devad != MDIO_DEVAD_NONE)
8800 regnum |= (devad << 16) | MII_ADDR_C45;
8802 return mdiobus_read(adapter->mii_bus, prtad, regnum);
8805 if (prtad != hw->phy.mdio.prtad)
8807 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8813 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8814 u16 addr, u16 value)
8816 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8817 struct ixgbe_hw *hw = &adapter->hw;
8819 if (adapter->mii_bus) {
8822 if (devad != MDIO_DEVAD_NONE)
8823 regnum |= (devad << 16) | MII_ADDR_C45;
8825 return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8828 if (prtad != hw->phy.mdio.prtad)
8830 return hw->phy.ops.write_reg(hw, addr, devad, value);
8833 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8835 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8839 return ixgbe_ptp_set_ts_config(adapter, req);
8841 return ixgbe_ptp_get_ts_config(adapter, req);
8843 if (!adapter->hw.phy.ops.read_reg)
8847 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8852 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8854 * @dev: network interface device structure
8856 * Returns non-zero on failure
8858 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8861 struct ixgbe_adapter *adapter = netdev_priv(dev);
8862 struct ixgbe_hw *hw = &adapter->hw;
8864 if (is_valid_ether_addr(hw->mac.san_addr)) {
8866 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8869 /* update SAN MAC vmdq pool selection */
8870 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8876 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8878 * @dev: network interface device structure
8880 * Returns non-zero on failure
8882 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8885 struct ixgbe_adapter *adapter = netdev_priv(dev);
8886 struct ixgbe_mac_info *mac = &adapter->hw.mac;
8888 if (is_valid_ether_addr(mac->san_addr)) {
8890 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8896 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8897 struct ixgbe_ring *ring)
8904 start = u64_stats_fetch_begin_irq(&ring->syncp);
8905 packets = ring->stats.packets;
8906 bytes = ring->stats.bytes;
8907 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8908 stats->tx_packets += packets;
8909 stats->tx_bytes += bytes;
8913 static void ixgbe_get_stats64(struct net_device *netdev,
8914 struct rtnl_link_stats64 *stats)
8916 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8920 for (i = 0; i < adapter->num_rx_queues; i++) {
8921 struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8927 start = u64_stats_fetch_begin_irq(&ring->syncp);
8928 packets = ring->stats.packets;
8929 bytes = ring->stats.bytes;
8930 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8931 stats->rx_packets += packets;
8932 stats->rx_bytes += bytes;
8936 for (i = 0; i < adapter->num_tx_queues; i++) {
8937 struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8939 ixgbe_get_ring_stats64(stats, ring);
8941 for (i = 0; i < adapter->num_xdp_queues; i++) {
8942 struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8944 ixgbe_get_ring_stats64(stats, ring);
8948 /* following stats updated by ixgbe_watchdog_task() */
8949 stats->multicast = netdev->stats.multicast;
8950 stats->rx_errors = netdev->stats.rx_errors;
8951 stats->rx_length_errors = netdev->stats.rx_length_errors;
8952 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
8953 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
8956 #ifdef CONFIG_IXGBE_DCB
8958 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8959 * @adapter: pointer to ixgbe_adapter
8960 * @tc: number of traffic classes currently enabled
8962 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8963 * 802.1Q priority maps to a packet buffer that exists.
8965 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8967 struct ixgbe_hw *hw = &adapter->hw;
8971 /* 82598 have a static priority to TC mapping that can not
8972 * be changed so no validation is needed.
8974 if (hw->mac.type == ixgbe_mac_82598EB)
8977 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
8980 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
8981 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
8983 /* If up2tc is out of bounds default to zero */
8985 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
8989 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
8995 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
8996 * @adapter: Pointer to adapter struct
8998 * Populate the netdev user priority to tc map
9000 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
9002 struct net_device *dev = adapter->netdev;
9003 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9004 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9007 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9010 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9011 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9013 tc = ets->prio_tc[prio];
9015 netdev_set_prio_tc_map(dev, prio, tc);
9019 #endif /* CONFIG_IXGBE_DCB */
9020 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev,
9021 struct netdev_nested_priv *priv)
9023 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)priv->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);
9060 struct netdev_nested_priv priv = {
9061 .data = (void *)adapter,
9064 /* flush any stale bits out of the fwd bitmask */
9065 bitmap_clear(adapter->fwd_bitmask, 1, 63);
9067 /* walk through upper devices reassigning pools */
9068 netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9073 * ixgbe_setup_tc - configure net_device for multiple traffic classes
9075 * @dev: net device to configure
9076 * @tc: number of traffic classes to enable
9078 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9080 struct ixgbe_adapter *adapter = netdev_priv(dev);
9081 struct ixgbe_hw *hw = &adapter->hw;
9083 /* Hardware supports up to 8 traffic classes */
9084 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9087 if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9090 /* Hardware has to reinitialize queues and interrupts to
9091 * match packet buffer alignment. Unfortunately, the
9092 * hardware is not flexible enough to do this dynamically.
9094 if (netif_running(dev))
9097 ixgbe_reset(adapter);
9099 ixgbe_clear_interrupt_scheme(adapter);
9101 #ifdef CONFIG_IXGBE_DCB
9103 if (adapter->xdp_prog) {
9104 e_warn(probe, "DCB is not supported with XDP\n");
9106 ixgbe_init_interrupt_scheme(adapter);
9107 if (netif_running(dev))
9112 netdev_set_num_tc(dev, tc);
9113 ixgbe_set_prio_tc_map(adapter);
9115 adapter->hw_tcs = tc;
9116 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9118 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9119 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9120 adapter->hw.fc.requested_mode = ixgbe_fc_none;
9123 netdev_reset_tc(dev);
9125 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9126 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9128 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9129 adapter->hw_tcs = tc;
9131 adapter->temp_dcb_cfg.pfc_mode_enable = false;
9132 adapter->dcb_cfg.pfc_mode_enable = false;
9135 ixgbe_validate_rtr(adapter, tc);
9137 #endif /* CONFIG_IXGBE_DCB */
9138 ixgbe_init_interrupt_scheme(adapter);
9140 ixgbe_defrag_macvlan_pools(dev);
9142 if (netif_running(dev))
9143 return ixgbe_open(dev);
9148 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9149 struct tc_cls_u32_offload *cls)
9151 u32 hdl = cls->knode.handle;
9152 u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9153 u32 loc = cls->knode.handle & 0xfffff;
9155 struct ixgbe_jump_table *jump = NULL;
9157 if (loc > IXGBE_MAX_HW_ENTRIES)
9160 if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9163 /* Clear this filter in the link data it is associated with */
9164 if (uhtid != 0x800) {
9165 jump = adapter->jump_tables[uhtid];
9168 if (!test_bit(loc - 1, jump->child_loc_map))
9170 clear_bit(loc - 1, jump->child_loc_map);
9173 /* Check if the filter being deleted is a link */
9174 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9175 jump = adapter->jump_tables[i];
9176 if (jump && jump->link_hdl == hdl) {
9177 /* Delete filters in the hardware in the child hash
9178 * table associated with this link
9180 for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9181 if (!test_bit(j, jump->child_loc_map))
9183 spin_lock(&adapter->fdir_perfect_lock);
9184 err = ixgbe_update_ethtool_fdir_entry(adapter,
9187 spin_unlock(&adapter->fdir_perfect_lock);
9188 clear_bit(j, jump->child_loc_map);
9190 /* Remove resources for this link */
9194 adapter->jump_tables[i] = NULL;
9199 spin_lock(&adapter->fdir_perfect_lock);
9200 err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9201 spin_unlock(&adapter->fdir_perfect_lock);
9205 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9206 struct tc_cls_u32_offload *cls)
9208 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9210 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9213 /* This ixgbe devices do not support hash tables at the moment
9214 * so abort when given hash tables.
9216 if (cls->hnode.divisor > 0)
9219 set_bit(uhtid - 1, &adapter->tables);
9223 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9224 struct tc_cls_u32_offload *cls)
9226 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9228 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9231 clear_bit(uhtid - 1, &adapter->tables);
9235 #ifdef CONFIG_NET_CLS_ACT
9236 struct upper_walk_data {
9237 struct ixgbe_adapter *adapter;
9243 static int get_macvlan_queue(struct net_device *upper,
9244 struct netdev_nested_priv *priv)
9246 if (netif_is_macvlan(upper)) {
9247 struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9248 struct ixgbe_adapter *adapter;
9249 struct upper_walk_data *data;
9252 data = (struct upper_walk_data *)priv->data;
9253 ifindex = data->ifindex;
9254 adapter = data->adapter;
9255 if (vadapter && upper->ifindex == ifindex) {
9256 data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9257 data->action = data->queue;
9265 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9266 u8 *queue, u64 *action)
9268 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9269 unsigned int num_vfs = adapter->num_vfs, vf;
9270 struct netdev_nested_priv priv;
9271 struct upper_walk_data data;
9272 struct net_device *upper;
9274 /* redirect to a SRIOV VF */
9275 for (vf = 0; vf < num_vfs; ++vf) {
9276 upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9277 if (upper->ifindex == ifindex) {
9278 *queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9280 *action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9285 /* redirect to a offloaded macvlan netdev */
9286 data.adapter = adapter;
9287 data.ifindex = ifindex;
9290 priv.data = (void *)&data;
9291 if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9292 get_macvlan_queue, &priv)) {
9293 *action = data.action;
9294 *queue = data.queue;
9302 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9303 struct tcf_exts *exts, u64 *action, u8 *queue)
9305 const struct tc_action *a;
9308 if (!tcf_exts_has_actions(exts))
9311 tcf_exts_for_each_action(i, a, exts) {
9313 if (is_tcf_gact_shot(a)) {
9314 *action = IXGBE_FDIR_DROP_QUEUE;
9315 *queue = IXGBE_FDIR_DROP_QUEUE;
9319 /* Redirect to a VF or a offloaded macvlan */
9320 if (is_tcf_mirred_egress_redirect(a)) {
9321 struct net_device *dev = tcf_mirred_dev(a);
9325 return handle_redirect_action(adapter, dev->ifindex,
9335 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9336 struct tcf_exts *exts, u64 *action, u8 *queue)
9340 #endif /* CONFIG_NET_CLS_ACT */
9342 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9343 union ixgbe_atr_input *mask,
9344 struct tc_cls_u32_offload *cls,
9345 struct ixgbe_mat_field *field_ptr,
9346 struct ixgbe_nexthdr *nexthdr)
9350 bool found_entry = false, found_jump_field = false;
9352 for (i = 0; i < cls->knode.sel->nkeys; i++) {
9353 off = cls->knode.sel->keys[i].off;
9354 val = cls->knode.sel->keys[i].val;
9355 m = cls->knode.sel->keys[i].mask;
9357 for (j = 0; field_ptr[j].val; j++) {
9358 if (field_ptr[j].off == off) {
9359 field_ptr[j].val(input, mask, (__force u32)val,
9361 input->filter.formatted.flow_type |=
9368 if (nexthdr->off == cls->knode.sel->keys[i].off &&
9370 (__force u32)cls->knode.sel->keys[i].val &&
9372 (__force u32)cls->knode.sel->keys[i].mask)
9373 found_jump_field = true;
9379 if (nexthdr && !found_jump_field)
9385 mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9386 IXGBE_ATR_L4TYPE_MASK;
9388 if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9389 mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9394 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9395 struct tc_cls_u32_offload *cls)
9397 __be16 protocol = cls->common.protocol;
9398 u32 loc = cls->knode.handle & 0xfffff;
9399 struct ixgbe_hw *hw = &adapter->hw;
9400 struct ixgbe_mat_field *field_ptr;
9401 struct ixgbe_fdir_filter *input = NULL;
9402 union ixgbe_atr_input *mask = NULL;
9403 struct ixgbe_jump_table *jump = NULL;
9404 int i, err = -EINVAL;
9406 u32 uhtid, link_uhtid;
9408 uhtid = TC_U32_USERHTID(cls->knode.handle);
9409 link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9411 /* At the moment cls_u32 jumps to network layer and skips past
9412 * L2 headers. The canonical method to match L2 frames is to use
9413 * negative values. However this is error prone at best but really
9414 * just broken because there is no way to "know" what sort of hdr
9415 * is in front of the network layer. Fix cls_u32 to support L2
9416 * headers when needed.
9418 if (protocol != htons(ETH_P_IP))
9421 if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9422 e_err(drv, "Location out of range\n");
9426 /* cls u32 is a graph starting at root node 0x800. The driver tracks
9427 * links and also the fields used to advance the parser across each
9428 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9429 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9430 * To add support for new nodes update ixgbe_model.h parse structures
9431 * this function _should_ be generic try not to hardcode values here.
9433 if (uhtid == 0x800) {
9434 field_ptr = (adapter->jump_tables[0])->mat;
9436 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9438 if (!adapter->jump_tables[uhtid])
9440 field_ptr = (adapter->jump_tables[uhtid])->mat;
9446 /* At this point we know the field_ptr is valid and need to either
9447 * build cls_u32 link or attach filter. Because adding a link to
9448 * a handle that does not exist is invalid and the same for adding
9449 * rules to handles that don't exist.
9453 struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9455 if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9458 if (!test_bit(link_uhtid - 1, &adapter->tables))
9461 /* Multiple filters as links to the same hash table are not
9462 * supported. To add a new filter with the same next header
9463 * but different match/jump conditions, create a new hash table
9466 if (adapter->jump_tables[link_uhtid] &&
9467 (adapter->jump_tables[link_uhtid])->link_hdl) {
9468 e_err(drv, "Link filter exists for link: %x\n",
9473 for (i = 0; nexthdr[i].jump; i++) {
9474 if (nexthdr[i].o != cls->knode.sel->offoff ||
9475 nexthdr[i].s != cls->knode.sel->offshift ||
9477 (__force u32)cls->knode.sel->offmask)
9480 jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9483 input = kzalloc(sizeof(*input), GFP_KERNEL);
9488 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9493 jump->input = input;
9495 jump->link_hdl = cls->knode.handle;
9497 err = ixgbe_clsu32_build_input(input, mask, cls,
9498 field_ptr, &nexthdr[i]);
9500 jump->mat = nexthdr[i].jump;
9501 adapter->jump_tables[link_uhtid] = jump;
9512 input = kzalloc(sizeof(*input), GFP_KERNEL);
9515 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9521 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9522 if ((adapter->jump_tables[uhtid])->input)
9523 memcpy(input, (adapter->jump_tables[uhtid])->input,
9525 if ((adapter->jump_tables[uhtid])->mask)
9526 memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9529 /* Lookup in all child hash tables if this location is already
9530 * filled with a filter
9532 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9533 struct ixgbe_jump_table *link = adapter->jump_tables[i];
9535 if (link && (test_bit(loc - 1, link->child_loc_map))) {
9536 e_err(drv, "Filter exists in location: %x\n",
9543 err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9547 err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9552 input->sw_idx = loc;
9554 spin_lock(&adapter->fdir_perfect_lock);
9556 if (hlist_empty(&adapter->fdir_filter_list)) {
9557 memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9558 err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9560 goto err_out_w_lock;
9561 } else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9563 goto err_out_w_lock;
9566 ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9567 err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9568 input->sw_idx, queue);
9570 goto err_out_w_lock;
9572 ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9573 spin_unlock(&adapter->fdir_perfect_lock);
9575 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9576 set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9581 spin_unlock(&adapter->fdir_perfect_lock);
9591 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9592 struct tc_cls_u32_offload *cls_u32)
9594 switch (cls_u32->command) {
9595 case TC_CLSU32_NEW_KNODE:
9596 case TC_CLSU32_REPLACE_KNODE:
9597 return ixgbe_configure_clsu32(adapter, cls_u32);
9598 case TC_CLSU32_DELETE_KNODE:
9599 return ixgbe_delete_clsu32(adapter, cls_u32);
9600 case TC_CLSU32_NEW_HNODE:
9601 case TC_CLSU32_REPLACE_HNODE:
9602 return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9603 case TC_CLSU32_DELETE_HNODE:
9604 return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9610 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9613 struct ixgbe_adapter *adapter = cb_priv;
9615 if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9619 case TC_SETUP_CLSU32:
9620 return ixgbe_setup_tc_cls_u32(adapter, type_data);
9626 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9627 struct tc_mqprio_qopt *mqprio)
9629 mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9630 return ixgbe_setup_tc(dev, mqprio->num_tc);
9633 static LIST_HEAD(ixgbe_block_cb_list);
9635 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9638 struct ixgbe_adapter *adapter = netdev_priv(dev);
9641 case TC_SETUP_BLOCK:
9642 return flow_block_cb_setup_simple(type_data,
9643 &ixgbe_block_cb_list,
9644 ixgbe_setup_tc_block_cb,
9645 adapter, adapter, true);
9646 case TC_SETUP_QDISC_MQPRIO:
9647 return ixgbe_setup_tc_mqprio(dev, type_data);
9653 #ifdef CONFIG_PCI_IOV
9654 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9656 struct net_device *netdev = adapter->netdev;
9659 ixgbe_setup_tc(netdev, adapter->hw_tcs);
9664 void ixgbe_do_reset(struct net_device *netdev)
9666 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9668 if (netif_running(netdev))
9669 ixgbe_reinit_locked(adapter);
9671 ixgbe_reset(adapter);
9674 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9675 netdev_features_t features)
9677 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9679 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9680 if (!(features & NETIF_F_RXCSUM))
9681 features &= ~NETIF_F_LRO;
9683 /* Turn off LRO if not RSC capable */
9684 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9685 features &= ~NETIF_F_LRO;
9687 if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9688 e_dev_err("LRO is not supported with XDP\n");
9689 features &= ~NETIF_F_LRO;
9695 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9697 int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9700 /* go back to full RSS if we're not running SR-IOV */
9701 if (!adapter->ring_feature[RING_F_VMDQ].offset)
9702 adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9703 IXGBE_FLAG_SRIOV_ENABLED);
9705 adapter->ring_feature[RING_F_RSS].limit = rss;
9706 adapter->ring_feature[RING_F_VMDQ].limit = 1;
9708 ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9711 static int ixgbe_set_features(struct net_device *netdev,
9712 netdev_features_t features)
9714 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9715 netdev_features_t changed = netdev->features ^ features;
9716 bool need_reset = false;
9718 /* Make sure RSC matches LRO, reset if change */
9719 if (!(features & NETIF_F_LRO)) {
9720 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9722 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9723 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9724 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9725 if (adapter->rx_itr_setting == 1 ||
9726 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9727 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9729 } else if ((changed ^ features) & NETIF_F_LRO) {
9730 e_info(probe, "rx-usecs set too low, "
9736 * Check if Flow Director n-tuple support or hw_tc support was
9737 * enabled or disabled. If the state changed, we need to reset.
9739 if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9740 /* turn off ATR, enable perfect filters and reset */
9741 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9744 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9745 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9747 /* turn off perfect filters, enable ATR and reset */
9748 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9751 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9753 /* We cannot enable ATR if SR-IOV is enabled */
9754 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9755 /* We cannot enable ATR if we have 2 or more tcs */
9756 (adapter->hw_tcs > 1) ||
9757 /* We cannot enable ATR if RSS is disabled */
9758 (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9759 /* A sample rate of 0 indicates ATR disabled */
9760 (!adapter->atr_sample_rate))
9761 ; /* do nothing not supported */
9762 else /* otherwise supported and set the flag */
9763 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9766 if (changed & NETIF_F_RXALL)
9769 netdev->features = features;
9771 if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9772 ixgbe_reset_l2fw_offload(adapter);
9773 else if (need_reset)
9774 ixgbe_do_reset(netdev);
9775 else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9776 NETIF_F_HW_VLAN_CTAG_FILTER))
9777 ixgbe_set_rx_mode(netdev);
9782 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9783 struct net_device *dev,
9784 const unsigned char *addr, u16 vid,
9786 struct netlink_ext_ack *extack)
9788 /* guarantee we can provide a unique filter for the unicast address */
9789 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9790 struct ixgbe_adapter *adapter = netdev_priv(dev);
9791 u16 pool = VMDQ_P(0);
9793 if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9797 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9801 * ixgbe_configure_bridge_mode - set various bridge modes
9802 * @adapter: the private structure
9803 * @mode: requested bridge mode
9805 * Configure some settings require for various bridge modes.
9807 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9810 struct ixgbe_hw *hw = &adapter->hw;
9811 unsigned int p, num_pools;
9815 case BRIDGE_MODE_VEPA:
9816 /* disable Tx loopback, rely on switch hairpin mode */
9817 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9819 /* must enable Rx switching replication to allow multicast
9820 * packet reception on all VFs, and to enable source address
9823 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9824 vmdctl |= IXGBE_VT_CTL_REPLEN;
9825 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9827 /* enable Rx source address pruning. Note, this requires
9828 * replication to be enabled or else it does nothing.
9830 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9831 for (p = 0; p < num_pools; p++) {
9832 if (hw->mac.ops.set_source_address_pruning)
9833 hw->mac.ops.set_source_address_pruning(hw,
9838 case BRIDGE_MODE_VEB:
9839 /* enable Tx loopback for internal VF/PF communication */
9840 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9841 IXGBE_PFDTXGSWC_VT_LBEN);
9843 /* disable Rx switching replication unless we have SR-IOV
9846 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9847 if (!adapter->num_vfs)
9848 vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9849 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9851 /* disable Rx source address pruning, since we don't expect to
9852 * be receiving external loopback of our transmitted frames.
9854 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9855 for (p = 0; p < num_pools; p++) {
9856 if (hw->mac.ops.set_source_address_pruning)
9857 hw->mac.ops.set_source_address_pruning(hw,
9866 adapter->bridge_mode = mode;
9868 e_info(drv, "enabling bridge mode: %s\n",
9869 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
9874 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
9875 struct nlmsghdr *nlh, u16 flags,
9876 struct netlink_ext_ack *extack)
9878 struct ixgbe_adapter *adapter = netdev_priv(dev);
9879 struct nlattr *attr, *br_spec;
9882 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9885 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
9889 nla_for_each_nested(attr, br_spec, rem) {
9893 if (nla_type(attr) != IFLA_BRIDGE_MODE)
9896 if (nla_len(attr) < sizeof(mode))
9899 mode = nla_get_u16(attr);
9900 status = ixgbe_configure_bridge_mode(adapter, mode);
9910 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
9911 struct net_device *dev,
9912 u32 filter_mask, int nlflags)
9914 struct ixgbe_adapter *adapter = netdev_priv(dev);
9916 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9919 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
9920 adapter->bridge_mode, 0, 0, nlflags,
9924 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
9926 struct ixgbe_adapter *adapter = netdev_priv(pdev);
9927 struct ixgbe_fwd_adapter *accel;
9928 int tcs = adapter->hw_tcs ? : 1;
9931 if (adapter->xdp_prog) {
9932 e_warn(probe, "L2FW offload is not supported with XDP\n");
9933 return ERR_PTR(-EINVAL);
9936 /* The hardware supported by ixgbe only filters on the destination MAC
9937 * address. In order to avoid issues we only support offloading modes
9938 * where the hardware can actually provide the functionality.
9940 if (!macvlan_supports_dest_filter(vdev))
9941 return ERR_PTR(-EMEDIUMTYPE);
9943 /* We need to lock down the macvlan to be a single queue device so that
9944 * we can reuse the tc_to_txq field in the macvlan netdev to represent
9945 * the queue mapping to our netdev.
9947 if (netif_is_multiqueue(vdev))
9948 return ERR_PTR(-ERANGE);
9950 pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9951 if (pool == adapter->num_rx_pools) {
9952 u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
9955 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
9956 adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
9957 adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
9958 return ERR_PTR(-EBUSY);
9960 /* Hardware has a limited number of available pools. Each VF,
9961 * and the PF require a pool. Check to ensure we don't
9962 * attempt to use more then the available number of pools.
9964 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
9965 return ERR_PTR(-EBUSY);
9967 /* Enable VMDq flag so device will be set in VM mode */
9968 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
9969 IXGBE_FLAG_SRIOV_ENABLED;
9971 /* Try to reserve as many queues per pool as possible,
9972 * we start with the configurations that support 4 queues
9973 * per pools, followed by 2, and then by just 1 per pool.
9975 if (used_pools < 32 && adapter->num_rx_pools < 16)
9976 reserved_pools = min_t(u16,
9978 16 - adapter->num_rx_pools);
9979 else if (adapter->num_rx_pools < 32)
9980 reserved_pools = min_t(u16,
9982 32 - adapter->num_rx_pools);
9984 reserved_pools = 64 - used_pools;
9987 if (!reserved_pools)
9988 return ERR_PTR(-EBUSY);
9990 adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
9992 /* Force reinit of ring allocation with VMDQ enabled */
9993 err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
9995 return ERR_PTR(err);
9997 if (pool >= adapter->num_rx_pools)
9998 return ERR_PTR(-ENOMEM);
10001 accel = kzalloc(sizeof(*accel), GFP_KERNEL);
10003 return ERR_PTR(-ENOMEM);
10005 set_bit(pool, adapter->fwd_bitmask);
10006 netdev_set_sb_channel(vdev, pool);
10007 accel->pool = pool;
10008 accel->netdev = vdev;
10010 if (!netif_running(pdev))
10013 err = ixgbe_fwd_ring_up(adapter, accel);
10015 return ERR_PTR(err);
10020 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10022 struct ixgbe_fwd_adapter *accel = priv;
10023 struct ixgbe_adapter *adapter = netdev_priv(pdev);
10024 unsigned int rxbase = accel->rx_base_queue;
10027 /* delete unicast filter associated with offloaded interface */
10028 ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10029 VMDQ_P(accel->pool));
10031 /* Allow remaining Rx packets to get flushed out of the
10032 * Rx FIFO before we drop the netdev for the ring.
10034 usleep_range(10000, 20000);
10036 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10037 struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10038 struct ixgbe_q_vector *qv = ring->q_vector;
10040 /* Make sure we aren't processing any packets and clear
10041 * netdev to shut down the ring.
10043 if (netif_running(adapter->netdev))
10044 napi_synchronize(&qv->napi);
10045 ring->netdev = NULL;
10048 /* unbind the queues and drop the subordinate channel config */
10049 netdev_unbind_sb_channel(pdev, accel->netdev);
10050 netdev_set_sb_channel(accel->netdev, 0);
10052 clear_bit(accel->pool, adapter->fwd_bitmask);
10056 #define IXGBE_MAX_MAC_HDR_LEN 127
10057 #define IXGBE_MAX_NETWORK_HDR_LEN 511
10059 static netdev_features_t
10060 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10061 netdev_features_t features)
10063 unsigned int network_hdr_len, mac_hdr_len;
10065 /* Make certain the headers can be described by a context descriptor */
10066 mac_hdr_len = skb_network_header(skb) - skb->data;
10067 if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10068 return features & ~(NETIF_F_HW_CSUM |
10070 NETIF_F_GSO_UDP_L4 |
10071 NETIF_F_HW_VLAN_CTAG_TX |
10075 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10076 if (unlikely(network_hdr_len > IXGBE_MAX_NETWORK_HDR_LEN))
10077 return features & ~(NETIF_F_HW_CSUM |
10079 NETIF_F_GSO_UDP_L4 |
10083 /* We can only support IPV4 TSO in tunnels if we can mangle the
10084 * inner IP ID field, so strip TSO if MANGLEID is not supported.
10085 * IPsec offoad sets skb->encapsulation but still can handle
10086 * the TSO, so it's the exception.
10088 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10089 #ifdef CONFIG_IXGBE_IPSEC
10090 if (!secpath_exists(skb))
10092 features &= ~NETIF_F_TSO;
10098 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10100 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10101 struct ixgbe_adapter *adapter = netdev_priv(dev);
10102 struct bpf_prog *old_prog;
10105 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10108 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10111 /* verify ixgbe ring attributes are sufficient for XDP */
10112 for (i = 0; i < adapter->num_rx_queues; i++) {
10113 struct ixgbe_ring *ring = adapter->rx_ring[i];
10115 if (ring_is_rsc_enabled(ring))
10118 if (frame_size > ixgbe_rx_bufsz(ring))
10122 if (nr_cpu_ids > MAX_XDP_QUEUES)
10125 old_prog = xchg(&adapter->xdp_prog, prog);
10126 need_reset = (!!prog != !!old_prog);
10128 /* If transitioning XDP modes reconfigure rings */
10133 /* Wait until ndo_xsk_wakeup completes. */
10135 err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10138 rcu_assign_pointer(adapter->xdp_prog, old_prog);
10142 for (i = 0; i < adapter->num_rx_queues; i++)
10143 (void)xchg(&adapter->rx_ring[i]->xdp_prog,
10144 adapter->xdp_prog);
10148 bpf_prog_put(old_prog);
10150 /* Kick start the NAPI context if there is an AF_XDP socket open
10151 * on that queue id. This so that receiving will start.
10153 if (need_reset && prog)
10154 for (i = 0; i < adapter->num_rx_queues; i++)
10155 if (adapter->xdp_ring[i]->xsk_pool)
10156 (void)ixgbe_xsk_wakeup(adapter->netdev, i,
10162 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10164 struct ixgbe_adapter *adapter = netdev_priv(dev);
10166 switch (xdp->command) {
10167 case XDP_SETUP_PROG:
10168 return ixgbe_xdp_setup(dev, xdp->prog);
10169 case XDP_SETUP_XSK_POOL:
10170 return ixgbe_xsk_pool_setup(adapter, xdp->xsk.pool,
10171 xdp->xsk.queue_id);
10178 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10180 /* Force memory writes to complete before letting h/w know there
10181 * are new descriptors to fetch.
10184 writel(ring->next_to_use, ring->tail);
10187 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10188 struct xdp_frame **frames, u32 flags)
10190 struct ixgbe_adapter *adapter = netdev_priv(dev);
10191 struct ixgbe_ring *ring;
10195 if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10198 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10201 /* During program transitions its possible adapter->xdp_prog is assigned
10202 * but ring has not been configured yet. In this case simply abort xmit.
10204 ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10205 if (unlikely(!ring))
10208 if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10211 for (i = 0; i < n; i++) {
10212 struct xdp_frame *xdpf = frames[i];
10215 err = ixgbe_xmit_xdp_ring(adapter, xdpf);
10216 if (err != IXGBE_XDP_TX)
10221 if (unlikely(flags & XDP_XMIT_FLUSH))
10222 ixgbe_xdp_ring_update_tail(ring);
10227 static const struct net_device_ops ixgbe_netdev_ops = {
10228 .ndo_open = ixgbe_open,
10229 .ndo_stop = ixgbe_close,
10230 .ndo_start_xmit = ixgbe_xmit_frame,
10231 .ndo_set_rx_mode = ixgbe_set_rx_mode,
10232 .ndo_validate_addr = eth_validate_addr,
10233 .ndo_set_mac_address = ixgbe_set_mac,
10234 .ndo_change_mtu = ixgbe_change_mtu,
10235 .ndo_tx_timeout = ixgbe_tx_timeout,
10236 .ndo_set_tx_maxrate = ixgbe_tx_maxrate,
10237 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
10238 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
10239 .ndo_do_ioctl = ixgbe_ioctl,
10240 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
10241 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
10242 .ndo_set_vf_rate = ixgbe_ndo_set_vf_bw,
10243 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
10244 .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10245 .ndo_set_vf_trust = ixgbe_ndo_set_vf_trust,
10246 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
10247 .ndo_get_stats64 = ixgbe_get_stats64,
10248 .ndo_setup_tc = __ixgbe_setup_tc,
10250 .ndo_select_queue = ixgbe_select_queue,
10251 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10252 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10253 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10254 .ndo_fcoe_enable = ixgbe_fcoe_enable,
10255 .ndo_fcoe_disable = ixgbe_fcoe_disable,
10256 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10257 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10258 #endif /* IXGBE_FCOE */
10259 .ndo_set_features = ixgbe_set_features,
10260 .ndo_fix_features = ixgbe_fix_features,
10261 .ndo_fdb_add = ixgbe_ndo_fdb_add,
10262 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
10263 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
10264 .ndo_dfwd_add_station = ixgbe_fwd_add,
10265 .ndo_dfwd_del_station = ixgbe_fwd_del,
10266 .ndo_features_check = ixgbe_features_check,
10267 .ndo_bpf = ixgbe_xdp,
10268 .ndo_xdp_xmit = ixgbe_xdp_xmit,
10269 .ndo_xsk_wakeup = ixgbe_xsk_wakeup,
10272 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10273 struct ixgbe_ring *tx_ring)
10275 unsigned long wait_delay, delay_interval;
10276 struct ixgbe_hw *hw = &adapter->hw;
10277 u8 reg_idx = tx_ring->reg_idx;
10281 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10283 /* delay mechanism from ixgbe_disable_tx */
10284 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10286 wait_loop = IXGBE_MAX_RX_DESC_POLL;
10287 wait_delay = delay_interval;
10289 while (wait_loop--) {
10290 usleep_range(wait_delay, wait_delay + 10);
10291 wait_delay += delay_interval * 2;
10292 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10294 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10298 e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10301 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10302 struct ixgbe_ring *tx_ring)
10304 set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10305 ixgbe_disable_txr_hw(adapter, tx_ring);
10308 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10309 struct ixgbe_ring *rx_ring)
10311 unsigned long wait_delay, delay_interval;
10312 struct ixgbe_hw *hw = &adapter->hw;
10313 u8 reg_idx = rx_ring->reg_idx;
10317 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10318 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10319 rxdctl |= IXGBE_RXDCTL_SWFLSH;
10321 /* write value back with RXDCTL.ENABLE bit cleared */
10322 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10324 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10325 if (hw->mac.type == ixgbe_mac_82598EB &&
10326 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10329 /* delay mechanism from ixgbe_disable_rx */
10330 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10332 wait_loop = IXGBE_MAX_RX_DESC_POLL;
10333 wait_delay = delay_interval;
10335 while (wait_loop--) {
10336 usleep_range(wait_delay, wait_delay + 10);
10337 wait_delay += delay_interval * 2;
10338 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10340 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10344 e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10347 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10349 memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10350 memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10353 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10355 memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10356 memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10360 * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10361 * @adapter: adapter structure
10362 * @ring: ring index
10364 * This function disables a certain Rx/Tx/XDP Tx ring. The function
10365 * assumes that the netdev is running.
10367 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10369 struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10371 rx_ring = adapter->rx_ring[ring];
10372 tx_ring = adapter->tx_ring[ring];
10373 xdp_ring = adapter->xdp_ring[ring];
10375 ixgbe_disable_txr(adapter, tx_ring);
10377 ixgbe_disable_txr(adapter, xdp_ring);
10378 ixgbe_disable_rxr_hw(adapter, rx_ring);
10383 /* Rx/Tx/XDP Tx share the same napi context. */
10384 napi_disable(&rx_ring->q_vector->napi);
10386 ixgbe_clean_tx_ring(tx_ring);
10388 ixgbe_clean_tx_ring(xdp_ring);
10389 ixgbe_clean_rx_ring(rx_ring);
10391 ixgbe_reset_txr_stats(tx_ring);
10393 ixgbe_reset_txr_stats(xdp_ring);
10394 ixgbe_reset_rxr_stats(rx_ring);
10398 * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10399 * @adapter: adapter structure
10400 * @ring: ring index
10402 * This function enables a certain Rx/Tx/XDP Tx ring. The function
10403 * assumes that the netdev is running.
10405 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10407 struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10409 rx_ring = adapter->rx_ring[ring];
10410 tx_ring = adapter->tx_ring[ring];
10411 xdp_ring = adapter->xdp_ring[ring];
10413 /* Rx/Tx/XDP Tx share the same napi context. */
10414 napi_enable(&rx_ring->q_vector->napi);
10416 ixgbe_configure_tx_ring(adapter, tx_ring);
10418 ixgbe_configure_tx_ring(adapter, xdp_ring);
10419 ixgbe_configure_rx_ring(adapter, rx_ring);
10421 clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10423 clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10427 * ixgbe_enumerate_functions - Get the number of ports this device has
10428 * @adapter: adapter structure
10430 * This function enumerates the phsyical functions co-located on a single slot,
10431 * in order to determine how many ports a device has. This is most useful in
10432 * determining the required GT/s of PCIe bandwidth necessary for optimal
10435 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10437 struct pci_dev *entry, *pdev = adapter->pdev;
10440 /* Some cards can not use the generic count PCIe functions method,
10441 * because they are behind a parent switch, so we hardcode these with
10442 * the correct number of functions.
10444 if (ixgbe_pcie_from_parent(&adapter->hw))
10447 list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10448 /* don't count virtual functions */
10449 if (entry->is_virtfn)
10452 /* When the devices on the bus don't all match our device ID,
10453 * we can't reliably determine the correct number of
10454 * functions. This can occur if a function has been direct
10455 * attached to a virtual machine using VT-d, for example. In
10456 * this case, simply return -1 to indicate this.
10458 if ((entry->vendor != pdev->vendor) ||
10459 (entry->device != pdev->device))
10469 * ixgbe_wol_supported - Check whether device supports WoL
10470 * @adapter: the adapter private structure
10471 * @device_id: the device ID
10472 * @subdevice_id: the subsystem device ID
10474 * This function is used by probe and ethtool to determine
10475 * which devices have WoL support
10478 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10481 struct ixgbe_hw *hw = &adapter->hw;
10482 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10484 /* WOL not supported on 82598 */
10485 if (hw->mac.type == ixgbe_mac_82598EB)
10488 /* check eeprom to see if WOL is enabled for X540 and newer */
10489 if (hw->mac.type >= ixgbe_mac_X540) {
10490 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10491 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10492 (hw->bus.func == 0)))
10496 /* WOL is determined based on device IDs for 82599 MACs */
10497 switch (device_id) {
10498 case IXGBE_DEV_ID_82599_SFP:
10499 /* Only these subdevices could supports WOL */
10500 switch (subdevice_id) {
10501 case IXGBE_SUBDEV_ID_82599_560FLR:
10502 case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10503 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10504 case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10505 /* only support first port */
10506 if (hw->bus.func != 0)
10509 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10510 case IXGBE_SUBDEV_ID_82599_SFP:
10511 case IXGBE_SUBDEV_ID_82599_RNDC:
10512 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10513 case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10514 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10515 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10519 case IXGBE_DEV_ID_82599EN_SFP:
10520 /* Only these subdevices support WOL */
10521 switch (subdevice_id) {
10522 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10526 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10527 /* All except this subdevice support WOL */
10528 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10531 case IXGBE_DEV_ID_82599_KX4:
10541 * ixgbe_set_fw_version - Set FW version
10542 * @adapter: the adapter private structure
10544 * This function is used by probe and ethtool to determine the FW version to
10545 * format to display. The FW version is taken from the EEPROM/NVM.
10547 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10549 struct ixgbe_hw *hw = &adapter->hw;
10550 struct ixgbe_nvm_version nvm_ver;
10552 ixgbe_get_oem_prod_version(hw, &nvm_ver);
10553 if (nvm_ver.oem_valid) {
10554 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10555 "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10556 nvm_ver.oem_release);
10560 ixgbe_get_etk_id(hw, &nvm_ver);
10561 ixgbe_get_orom_version(hw, &nvm_ver);
10563 if (nvm_ver.or_valid) {
10564 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10565 "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10566 nvm_ver.or_build, nvm_ver.or_patch);
10570 /* Set ETrack ID format */
10571 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10572 "0x%08x", nvm_ver.etk_id);
10576 * ixgbe_probe - Device Initialization Routine
10577 * @pdev: PCI device information struct
10578 * @ent: entry in ixgbe_pci_tbl
10580 * Returns 0 on success, negative on failure
10582 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10583 * The OS initialization, configuring of the adapter private structure,
10584 * and a hardware reset occur.
10586 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10588 struct net_device *netdev;
10589 struct ixgbe_adapter *adapter = NULL;
10590 struct ixgbe_hw *hw;
10591 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10592 int i, err, pci_using_dac, expected_gts;
10593 unsigned int indices = MAX_TX_QUEUES;
10594 u8 part_str[IXGBE_PBANUM_LENGTH];
10595 bool disable_dev = false;
10601 /* Catch broken hardware that put the wrong VF device ID in
10602 * the PCIe SR-IOV capability.
10604 if (pdev->is_virtfn) {
10605 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10606 pci_name(pdev), pdev->vendor, pdev->device);
10610 err = pci_enable_device_mem(pdev);
10614 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10617 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10619 dev_err(&pdev->dev,
10620 "No usable DMA configuration, aborting\n");
10626 err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10628 dev_err(&pdev->dev,
10629 "pci_request_selected_regions failed 0x%x\n", err);
10633 pci_enable_pcie_error_reporting(pdev);
10635 pci_set_master(pdev);
10636 pci_save_state(pdev);
10638 if (ii->mac == ixgbe_mac_82598EB) {
10639 #ifdef CONFIG_IXGBE_DCB
10640 /* 8 TC w/ 4 queues per TC */
10641 indices = 4 * MAX_TRAFFIC_CLASS;
10643 indices = IXGBE_MAX_RSS_INDICES;
10647 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10650 goto err_alloc_etherdev;
10653 SET_NETDEV_DEV(netdev, &pdev->dev);
10655 adapter = netdev_priv(netdev);
10657 adapter->netdev = netdev;
10658 adapter->pdev = pdev;
10660 hw->back = adapter;
10661 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10663 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10664 pci_resource_len(pdev, 0));
10665 adapter->io_addr = hw->hw_addr;
10666 if (!hw->hw_addr) {
10671 netdev->netdev_ops = &ixgbe_netdev_ops;
10672 ixgbe_set_ethtool_ops(netdev);
10673 netdev->watchdog_timeo = 5 * HZ;
10674 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10677 hw->mac.ops = *ii->mac_ops;
10678 hw->mac.type = ii->mac;
10679 hw->mvals = ii->mvals;
10681 hw->link.ops = *ii->link_ops;
10684 hw->eeprom.ops = *ii->eeprom_ops;
10685 eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10686 if (ixgbe_removed(hw->hw_addr)) {
10690 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10691 if (!(eec & BIT(8)))
10692 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10695 hw->phy.ops = *ii->phy_ops;
10696 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10697 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
10698 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10699 hw->phy.mdio.mmds = 0;
10700 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10701 hw->phy.mdio.dev = netdev;
10702 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10703 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10705 /* setup the private structure */
10706 err = ixgbe_sw_init(adapter, ii);
10710 switch (adapter->hw.mac.type) {
10711 case ixgbe_mac_X550:
10712 case ixgbe_mac_X550EM_x:
10713 netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550;
10715 case ixgbe_mac_x550em_a:
10716 netdev->udp_tunnel_nic_info = &ixgbe_udp_tunnels_x550em_a;
10722 /* Make sure the SWFW semaphore is in a valid state */
10723 if (hw->mac.ops.init_swfw_sync)
10724 hw->mac.ops.init_swfw_sync(hw);
10726 /* Make it possible the adapter to be woken up via WOL */
10727 switch (adapter->hw.mac.type) {
10728 case ixgbe_mac_82599EB:
10729 case ixgbe_mac_X540:
10730 case ixgbe_mac_X550:
10731 case ixgbe_mac_X550EM_x:
10732 case ixgbe_mac_x550em_a:
10733 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10740 * If there is a fan on this device and it has failed log the
10743 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10744 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10745 if (esdp & IXGBE_ESDP_SDP1)
10746 e_crit(probe, "Fan has stopped, replace the adapter\n");
10749 if (allow_unsupported_sfp)
10750 hw->allow_unsupported_sfp = allow_unsupported_sfp;
10752 /* reset_hw fills in the perm_addr as well */
10753 hw->phy.reset_if_overtemp = true;
10754 err = hw->mac.ops.reset_hw(hw);
10755 hw->phy.reset_if_overtemp = false;
10756 ixgbe_set_eee_capable(adapter);
10757 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10759 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10760 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10761 e_dev_err("Reload the driver after installing a supported module.\n");
10764 e_dev_err("HW Init failed: %d\n", err);
10768 #ifdef CONFIG_PCI_IOV
10769 /* SR-IOV not supported on the 82598 */
10770 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10773 ixgbe_init_mbx_params_pf(hw);
10774 hw->mbx.ops = ii->mbx_ops;
10775 pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10776 ixgbe_enable_sriov(adapter, max_vfs);
10780 netdev->features = NETIF_F_SG |
10787 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10788 NETIF_F_GSO_GRE_CSUM | \
10789 NETIF_F_GSO_IPXIP4 | \
10790 NETIF_F_GSO_IPXIP6 | \
10791 NETIF_F_GSO_UDP_TUNNEL | \
10792 NETIF_F_GSO_UDP_TUNNEL_CSUM)
10794 netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10795 netdev->features |= NETIF_F_GSO_PARTIAL |
10796 IXGBE_GSO_PARTIAL_FEATURES;
10798 if (hw->mac.type >= ixgbe_mac_82599EB)
10799 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
10801 #ifdef CONFIG_IXGBE_IPSEC
10802 #define IXGBE_ESP_FEATURES (NETIF_F_HW_ESP | \
10803 NETIF_F_HW_ESP_TX_CSUM | \
10806 if (adapter->ipsec)
10807 netdev->features |= IXGBE_ESP_FEATURES;
10809 /* copy netdev features into list of user selectable features */
10810 netdev->hw_features |= netdev->features |
10811 NETIF_F_HW_VLAN_CTAG_FILTER |
10812 NETIF_F_HW_VLAN_CTAG_RX |
10813 NETIF_F_HW_VLAN_CTAG_TX |
10815 NETIF_F_HW_L2FW_DOFFLOAD;
10817 if (hw->mac.type >= ixgbe_mac_82599EB)
10818 netdev->hw_features |= NETIF_F_NTUPLE |
10822 netdev->features |= NETIF_F_HIGHDMA;
10824 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10825 netdev->hw_enc_features |= netdev->vlan_features;
10826 netdev->mpls_features |= NETIF_F_SG |
10830 netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10832 /* set this bit last since it cannot be part of vlan_features */
10833 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10834 NETIF_F_HW_VLAN_CTAG_RX |
10835 NETIF_F_HW_VLAN_CTAG_TX;
10837 netdev->priv_flags |= IFF_UNICAST_FLT;
10838 netdev->priv_flags |= IFF_SUPP_NOFCS;
10840 /* MTU range: 68 - 9710 */
10841 netdev->min_mtu = ETH_MIN_MTU;
10842 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
10844 #ifdef CONFIG_IXGBE_DCB
10845 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
10846 netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
10850 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
10851 unsigned int fcoe_l;
10853 if (hw->mac.ops.get_device_caps) {
10854 hw->mac.ops.get_device_caps(hw, &device_caps);
10855 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
10856 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
10860 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
10861 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
10863 netdev->features |= NETIF_F_FSO |
10866 netdev->vlan_features |= NETIF_F_FSO |
10870 #endif /* IXGBE_FCOE */
10871 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
10872 netdev->hw_features |= NETIF_F_LRO;
10873 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
10874 netdev->features |= NETIF_F_LRO;
10876 if (ixgbe_check_fw_error(adapter)) {
10881 /* make sure the EEPROM is good */
10882 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
10883 e_dev_err("The EEPROM Checksum Is Not Valid\n");
10888 eth_platform_get_mac_address(&adapter->pdev->dev,
10889 adapter->hw.mac.perm_addr);
10891 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
10893 if (!is_valid_ether_addr(netdev->dev_addr)) {
10894 e_dev_err("invalid MAC address\n");
10899 /* Set hw->mac.addr to permanent MAC address */
10900 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
10901 ixgbe_mac_set_default_filter(adapter);
10903 timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
10905 if (ixgbe_removed(hw->hw_addr)) {
10909 INIT_WORK(&adapter->service_task, ixgbe_service_task);
10910 set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
10911 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
10913 err = ixgbe_init_interrupt_scheme(adapter);
10917 for (i = 0; i < adapter->num_rx_queues; i++)
10918 u64_stats_init(&adapter->rx_ring[i]->syncp);
10919 for (i = 0; i < adapter->num_tx_queues; i++)
10920 u64_stats_init(&adapter->tx_ring[i]->syncp);
10921 for (i = 0; i < adapter->num_xdp_queues; i++)
10922 u64_stats_init(&adapter->xdp_ring[i]->syncp);
10924 /* WOL not supported for all devices */
10926 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
10927 hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
10928 pdev->subsystem_device);
10929 if (hw->wol_enabled)
10930 adapter->wol = IXGBE_WUFC_MAG;
10932 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
10934 /* save off EEPROM version number */
10935 ixgbe_set_fw_version(adapter);
10937 /* pick up the PCI bus settings for reporting later */
10938 if (ixgbe_pcie_from_parent(hw))
10939 ixgbe_get_parent_bus_info(adapter);
10941 hw->mac.ops.get_bus_info(hw);
10943 /* calculate the expected PCIe bandwidth required for optimal
10944 * performance. Note that some older parts will never have enough
10945 * bandwidth due to being older generation PCIe parts. We clamp these
10946 * parts to ensure no warning is displayed if it can't be fixed.
10948 switch (hw->mac.type) {
10949 case ixgbe_mac_82598EB:
10950 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
10953 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
10957 /* don't check link if we failed to enumerate functions */
10958 if (expected_gts > 0)
10959 ixgbe_check_minimum_link(adapter, expected_gts);
10961 err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
10963 strlcpy(part_str, "Unknown", sizeof(part_str));
10964 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
10965 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
10966 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
10969 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
10970 hw->mac.type, hw->phy.type, part_str);
10972 e_dev_info("%pM\n", netdev->dev_addr);
10974 /* reset the hardware with the new settings */
10975 err = hw->mac.ops.start_hw(hw);
10976 if (err == IXGBE_ERR_EEPROM_VERSION) {
10977 /* We are running on a pre-production device, log a warning */
10978 e_dev_warn("This device is a pre-production adapter/LOM. "
10979 "Please be aware there may be issues associated "
10980 "with your hardware. If you are experiencing "
10981 "problems please contact your Intel or hardware "
10982 "representative who provided you with this "
10985 strcpy(netdev->name, "eth%d");
10986 pci_set_drvdata(pdev, adapter);
10987 err = register_netdev(netdev);
10992 /* power down the optics for 82599 SFP+ fiber */
10993 if (hw->mac.ops.disable_tx_laser)
10994 hw->mac.ops.disable_tx_laser(hw);
10996 /* carrier off reporting is important to ethtool even BEFORE open */
10997 netif_carrier_off(netdev);
10999 #ifdef CONFIG_IXGBE_DCA
11000 if (dca_add_requester(&pdev->dev) == 0) {
11001 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11002 ixgbe_setup_dca(adapter);
11005 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11006 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11007 for (i = 0; i < adapter->num_vfs; i++)
11008 ixgbe_vf_configuration(pdev, (i | 0x10000000));
11011 /* firmware requires driver version to be 0xFFFFFFFF
11012 * since os does not support feature
11014 if (hw->mac.ops.set_fw_drv_ver)
11015 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11016 sizeof(UTS_RELEASE) - 1,
11019 /* add san mac addr to netdev */
11020 ixgbe_add_sanmac_netdev(netdev);
11022 e_dev_info("%s\n", ixgbe_default_device_descr);
11024 #ifdef CONFIG_IXGBE_HWMON
11025 if (ixgbe_sysfs_init(adapter))
11026 e_err(probe, "failed to allocate sysfs resources\n");
11027 #endif /* CONFIG_IXGBE_HWMON */
11029 ixgbe_dbg_adapter_init(adapter);
11031 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11032 if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11033 hw->mac.ops.setup_link(hw,
11034 IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11037 err = ixgbe_mii_bus_init(hw);
11044 unregister_netdev(netdev);
11046 ixgbe_release_hw_control(adapter);
11047 ixgbe_clear_interrupt_scheme(adapter);
11049 ixgbe_disable_sriov(adapter);
11050 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11051 iounmap(adapter->io_addr);
11052 kfree(adapter->jump_tables[0]);
11053 kfree(adapter->mac_table);
11054 kfree(adapter->rss_key);
11055 bitmap_free(adapter->af_xdp_zc_qps);
11057 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11058 free_netdev(netdev);
11059 err_alloc_etherdev:
11060 pci_release_mem_regions(pdev);
11063 if (!adapter || disable_dev)
11064 pci_disable_device(pdev);
11069 * ixgbe_remove - Device Removal Routine
11070 * @pdev: PCI device information struct
11072 * ixgbe_remove is called by the PCI subsystem to alert the driver
11073 * that it should release a PCI device. The could be caused by a
11074 * Hot-Plug event, or because the driver is going to be removed from
11077 static void ixgbe_remove(struct pci_dev *pdev)
11079 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11080 struct net_device *netdev;
11084 /* if !adapter then we already cleaned up in probe */
11088 netdev = adapter->netdev;
11089 ixgbe_dbg_adapter_exit(adapter);
11091 set_bit(__IXGBE_REMOVING, &adapter->state);
11092 cancel_work_sync(&adapter->service_task);
11094 if (adapter->mii_bus)
11095 mdiobus_unregister(adapter->mii_bus);
11097 #ifdef CONFIG_IXGBE_DCA
11098 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11099 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11100 dca_remove_requester(&pdev->dev);
11101 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11102 IXGBE_DCA_CTRL_DCA_DISABLE);
11106 #ifdef CONFIG_IXGBE_HWMON
11107 ixgbe_sysfs_exit(adapter);
11108 #endif /* CONFIG_IXGBE_HWMON */
11110 /* remove the added san mac */
11111 ixgbe_del_sanmac_netdev(netdev);
11113 #ifdef CONFIG_PCI_IOV
11114 ixgbe_disable_sriov(adapter);
11116 if (netdev->reg_state == NETREG_REGISTERED)
11117 unregister_netdev(netdev);
11119 ixgbe_stop_ipsec_offload(adapter);
11120 ixgbe_clear_interrupt_scheme(adapter);
11122 ixgbe_release_hw_control(adapter);
11125 kfree(adapter->ixgbe_ieee_pfc);
11126 kfree(adapter->ixgbe_ieee_ets);
11129 iounmap(adapter->io_addr);
11130 pci_release_mem_regions(pdev);
11132 e_dev_info("complete\n");
11134 for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11135 if (adapter->jump_tables[i]) {
11136 kfree(adapter->jump_tables[i]->input);
11137 kfree(adapter->jump_tables[i]->mask);
11139 kfree(adapter->jump_tables[i]);
11142 kfree(adapter->mac_table);
11143 kfree(adapter->rss_key);
11144 bitmap_free(adapter->af_xdp_zc_qps);
11145 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11146 free_netdev(netdev);
11148 pci_disable_pcie_error_reporting(pdev);
11151 pci_disable_device(pdev);
11155 * ixgbe_io_error_detected - called when PCI error is detected
11156 * @pdev: Pointer to PCI device
11157 * @state: The current pci connection state
11159 * This function is called after a PCI bus error affecting
11160 * this device has been detected.
11162 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11163 pci_channel_state_t state)
11165 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11166 struct net_device *netdev = adapter->netdev;
11168 #ifdef CONFIG_PCI_IOV
11169 struct ixgbe_hw *hw = &adapter->hw;
11170 struct pci_dev *bdev, *vfdev;
11171 u32 dw0, dw1, dw2, dw3;
11173 u16 req_id, pf_func;
11175 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11176 adapter->num_vfs == 0)
11177 goto skip_bad_vf_detection;
11179 bdev = pdev->bus->self;
11180 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11181 bdev = bdev->bus->self;
11184 goto skip_bad_vf_detection;
11186 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11188 goto skip_bad_vf_detection;
11190 dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11191 dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11192 dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11193 dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11194 if (ixgbe_removed(hw->hw_addr))
11195 goto skip_bad_vf_detection;
11197 req_id = dw1 >> 16;
11198 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11199 if (!(req_id & 0x0080))
11200 goto skip_bad_vf_detection;
11202 pf_func = req_id & 0x01;
11203 if ((pf_func & 1) == (pdev->devfn & 1)) {
11204 unsigned int device_id;
11206 vf = (req_id & 0x7F) >> 1;
11207 e_dev_err("VF %d has caused a PCIe error\n", vf);
11208 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11209 "%8.8x\tdw3: %8.8x\n",
11210 dw0, dw1, dw2, dw3);
11211 switch (adapter->hw.mac.type) {
11212 case ixgbe_mac_82599EB:
11213 device_id = IXGBE_82599_VF_DEVICE_ID;
11215 case ixgbe_mac_X540:
11216 device_id = IXGBE_X540_VF_DEVICE_ID;
11218 case ixgbe_mac_X550:
11219 device_id = IXGBE_DEV_ID_X550_VF;
11221 case ixgbe_mac_X550EM_x:
11222 device_id = IXGBE_DEV_ID_X550EM_X_VF;
11224 case ixgbe_mac_x550em_a:
11225 device_id = IXGBE_DEV_ID_X550EM_A_VF;
11232 /* Find the pci device of the offending VF */
11233 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11235 if (vfdev->devfn == (req_id & 0xFF))
11237 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11241 * There's a slim chance the VF could have been hot plugged,
11242 * so if it is no longer present we don't need to issue the
11243 * VFLR. Just clean up the AER in that case.
11247 /* Free device reference count */
11248 pci_dev_put(vfdev);
11253 * Even though the error may have occurred on the other port
11254 * we still need to increment the vf error reference count for
11255 * both ports because the I/O resume function will be called
11256 * for both of them.
11258 adapter->vferr_refcount++;
11260 return PCI_ERS_RESULT_RECOVERED;
11262 skip_bad_vf_detection:
11263 #endif /* CONFIG_PCI_IOV */
11264 if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11265 return PCI_ERS_RESULT_DISCONNECT;
11267 if (!netif_device_present(netdev))
11268 return PCI_ERS_RESULT_DISCONNECT;
11271 netif_device_detach(netdev);
11273 if (netif_running(netdev))
11274 ixgbe_close_suspend(adapter);
11276 if (state == pci_channel_io_perm_failure) {
11278 return PCI_ERS_RESULT_DISCONNECT;
11281 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11282 pci_disable_device(pdev);
11285 /* Request a slot reset. */
11286 return PCI_ERS_RESULT_NEED_RESET;
11290 * ixgbe_io_slot_reset - called after the pci bus has been reset.
11291 * @pdev: Pointer to PCI device
11293 * Restart the card from scratch, as if from a cold-boot.
11295 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11297 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11298 pci_ers_result_t result;
11300 if (pci_enable_device_mem(pdev)) {
11301 e_err(probe, "Cannot re-enable PCI device after reset.\n");
11302 result = PCI_ERS_RESULT_DISCONNECT;
11304 smp_mb__before_atomic();
11305 clear_bit(__IXGBE_DISABLED, &adapter->state);
11306 adapter->hw.hw_addr = adapter->io_addr;
11307 pci_set_master(pdev);
11308 pci_restore_state(pdev);
11309 pci_save_state(pdev);
11311 pci_wake_from_d3(pdev, false);
11313 ixgbe_reset(adapter);
11314 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11315 result = PCI_ERS_RESULT_RECOVERED;
11322 * ixgbe_io_resume - called when traffic can start flowing again.
11323 * @pdev: Pointer to PCI device
11325 * This callback is called when the error recovery driver tells us that
11326 * its OK to resume normal operation.
11328 static void ixgbe_io_resume(struct pci_dev *pdev)
11330 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11331 struct net_device *netdev = adapter->netdev;
11333 #ifdef CONFIG_PCI_IOV
11334 if (adapter->vferr_refcount) {
11335 e_info(drv, "Resuming after VF err\n");
11336 adapter->vferr_refcount--;
11342 if (netif_running(netdev))
11343 ixgbe_open(netdev);
11345 netif_device_attach(netdev);
11349 static const struct pci_error_handlers ixgbe_err_handler = {
11350 .error_detected = ixgbe_io_error_detected,
11351 .slot_reset = ixgbe_io_slot_reset,
11352 .resume = ixgbe_io_resume,
11355 static SIMPLE_DEV_PM_OPS(ixgbe_pm_ops, ixgbe_suspend, ixgbe_resume);
11357 static struct pci_driver ixgbe_driver = {
11358 .name = ixgbe_driver_name,
11359 .id_table = ixgbe_pci_tbl,
11360 .probe = ixgbe_probe,
11361 .remove = ixgbe_remove,
11362 .driver.pm = &ixgbe_pm_ops,
11363 .shutdown = ixgbe_shutdown,
11364 .sriov_configure = ixgbe_pci_sriov_configure,
11365 .err_handler = &ixgbe_err_handler
11369 * ixgbe_init_module - Driver Registration Routine
11371 * ixgbe_init_module is the first routine called when the driver is
11372 * loaded. All it does is register with the PCI subsystem.
11374 static int __init ixgbe_init_module(void)
11377 pr_info("%s\n", ixgbe_driver_string);
11378 pr_info("%s\n", ixgbe_copyright);
11380 ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11382 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11388 ret = pci_register_driver(&ixgbe_driver);
11390 destroy_workqueue(ixgbe_wq);
11395 #ifdef CONFIG_IXGBE_DCA
11396 dca_register_notify(&dca_notifier);
11402 module_init(ixgbe_init_module);
11405 * ixgbe_exit_module - Driver Exit Cleanup Routine
11407 * ixgbe_exit_module is called just before the driver is removed
11410 static void __exit ixgbe_exit_module(void)
11412 #ifdef CONFIG_IXGBE_DCA
11413 dca_unregister_notify(&dca_notifier);
11415 pci_unregister_driver(&ixgbe_driver);
11419 destroy_workqueue(ixgbe_wq);
11424 #ifdef CONFIG_IXGBE_DCA
11425 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11430 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11431 __ixgbe_notify_dca);
11433 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11436 #endif /* CONFIG_IXGBE_DCA */
11438 module_exit(ixgbe_exit_module);