1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
4 #include <linux/types.h>
5 #include <linux/module.h>
7 #include <linux/netdevice.h>
8 #include <linux/vmalloc.h>
9 #include <linux/string.h>
11 #include <linux/interrupt.h>
13 #include <linux/tcp.h>
14 #include <linux/sctp.h>
15 #include <linux/pkt_sched.h>
16 #include <linux/ipv6.h>
17 #include <linux/slab.h>
18 #include <net/checksum.h>
19 #include <net/ip6_checksum.h>
20 #include <linux/etherdevice.h>
21 #include <linux/ethtool.h>
23 #include <linux/if_vlan.h>
24 #include <linux/if_macvlan.h>
25 #include <linux/if_bridge.h>
26 #include <linux/prefetch.h>
27 #include <linux/bpf.h>
28 #include <linux/bpf_trace.h>
29 #include <linux/atomic.h>
30 #include <linux/numa.h>
31 #include <scsi/fc/fc_fcoe.h>
32 #include <net/udp_tunnel.h>
33 #include <net/pkt_cls.h>
34 #include <net/tc_act/tc_gact.h>
35 #include <net/tc_act/tc_mirred.h>
36 #include <net/vxlan.h>
38 #include <net/xdp_sock.h>
42 #include "ixgbe_common.h"
43 #include "ixgbe_dcb_82599.h"
44 #include "ixgbe_phy.h"
45 #include "ixgbe_sriov.h"
46 #include "ixgbe_model.h"
47 #include "ixgbe_txrx_common.h"
49 char ixgbe_driver_name[] = "ixgbe";
50 static const char ixgbe_driver_string[] =
51 "Intel(R) 10 Gigabit PCI Express Network Driver";
53 char ixgbe_default_device_descr[] =
54 "Intel(R) 10 Gigabit Network Connection";
56 static char ixgbe_default_device_descr[] =
57 "Intel(R) 10 Gigabit Network Connection";
59 #define DRV_VERSION "5.1.0-k"
60 const char ixgbe_driver_version[] = DRV_VERSION;
61 static const char ixgbe_copyright[] =
62 "Copyright (c) 1999-2016 Intel Corporation.";
64 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";
66 static const struct ixgbe_info *ixgbe_info_tbl[] = {
67 [board_82598] = &ixgbe_82598_info,
68 [board_82599] = &ixgbe_82599_info,
69 [board_X540] = &ixgbe_X540_info,
70 [board_X550] = &ixgbe_X550_info,
71 [board_X550EM_x] = &ixgbe_X550EM_x_info,
72 [board_x550em_x_fw] = &ixgbe_x550em_x_fw_info,
73 [board_x550em_a] = &ixgbe_x550em_a_info,
74 [board_x550em_a_fw] = &ixgbe_x550em_a_fw_info,
77 /* ixgbe_pci_tbl - PCI Device ID Table
79 * Wildcard entries (PCI_ANY_ID) should come last
80 * Last entry must be all 0s
82 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
83 * Class, Class Mask, private data (not used) }
85 static const struct pci_device_id ixgbe_pci_tbl[] = {
86 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
88 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
117 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T1), board_X550},
118 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
119 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_XFI), board_X550EM_x},
120 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
121 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x},
122 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x},
123 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_1G_T), board_x550em_x_fw},
124 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR), board_x550em_a },
125 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR_L), board_x550em_a },
126 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP_N), board_x550em_a },
127 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII), board_x550em_a },
128 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII_L), board_x550em_a },
129 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_10G_T), board_x550em_a},
130 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP), board_x550em_a },
131 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T), board_x550em_a_fw },
132 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T_L), board_x550em_a_fw },
133 /* required last entry */
136 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
138 #ifdef CONFIG_IXGBE_DCA
139 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
141 static struct notifier_block dca_notifier = {
142 .notifier_call = ixgbe_notify_dca,
148 #ifdef CONFIG_PCI_IOV
149 static unsigned int max_vfs;
150 module_param(max_vfs, uint, 0);
151 MODULE_PARM_DESC(max_vfs,
152 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
153 #endif /* CONFIG_PCI_IOV */
155 static unsigned int allow_unsupported_sfp;
156 module_param(allow_unsupported_sfp, uint, 0);
157 MODULE_PARM_DESC(allow_unsupported_sfp,
158 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
160 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
161 static int debug = -1;
162 module_param(debug, int, 0);
163 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
165 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
166 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
167 MODULE_LICENSE("GPL v2");
168 MODULE_VERSION(DRV_VERSION);
170 static struct workqueue_struct *ixgbe_wq;
172 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
173 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *);
175 static const struct net_device_ops ixgbe_netdev_ops;
177 static bool netif_is_ixgbe(struct net_device *dev)
179 return dev && (dev->netdev_ops == &ixgbe_netdev_ops);
182 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
185 struct pci_dev *parent_dev;
186 struct pci_bus *parent_bus;
188 parent_bus = adapter->pdev->bus->parent;
192 parent_dev = parent_bus->self;
196 if (!pci_is_pcie(parent_dev))
199 pcie_capability_read_word(parent_dev, reg, value);
200 if (*value == IXGBE_FAILED_READ_CFG_WORD &&
201 ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
206 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
208 struct ixgbe_hw *hw = &adapter->hw;
212 hw->bus.type = ixgbe_bus_type_pci_express;
214 /* Get the negotiated link width and speed from PCI config space of the
215 * parent, as this device is behind a switch
217 err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
219 /* assume caller will handle error case */
223 hw->bus.width = ixgbe_convert_bus_width(link_status);
224 hw->bus.speed = ixgbe_convert_bus_speed(link_status);
230 * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
231 * @hw: hw specific details
233 * This function is used by probe to determine whether a device's PCI-Express
234 * bandwidth details should be gathered from the parent bus instead of from the
235 * device. Used to ensure that various locations all have the correct device ID
238 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
240 switch (hw->device_id) {
241 case IXGBE_DEV_ID_82599_SFP_SF_QP:
242 case IXGBE_DEV_ID_82599_QSFP_SF_QP:
249 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
252 struct ixgbe_hw *hw = &adapter->hw;
253 struct pci_dev *pdev;
255 /* Some devices are not connected over PCIe and thus do not negotiate
256 * speed. These devices do not have valid bus info, and thus any report
257 * we generate may not be correct.
259 if (hw->bus.type == ixgbe_bus_type_internal)
262 /* determine whether to use the parent device */
263 if (ixgbe_pcie_from_parent(&adapter->hw))
264 pdev = adapter->pdev->bus->parent->self;
266 pdev = adapter->pdev;
268 pcie_print_link_status(pdev);
271 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
273 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
274 !test_bit(__IXGBE_REMOVING, &adapter->state) &&
275 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
276 queue_work(ixgbe_wq, &adapter->service_task);
279 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
281 struct ixgbe_adapter *adapter = hw->back;
286 e_dev_err("Adapter removed\n");
287 if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
288 ixgbe_service_event_schedule(adapter);
291 static u32 ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
293 u8 __iomem *reg_addr;
297 reg_addr = READ_ONCE(hw->hw_addr);
298 if (ixgbe_removed(reg_addr))
299 return IXGBE_FAILED_READ_REG;
301 /* Register read of 0xFFFFFFF can indicate the adapter has been removed,
302 * so perform several status register reads to determine if the adapter
305 for (i = 0; i < IXGBE_FAILED_READ_RETRIES; i++) {
306 value = readl(reg_addr + IXGBE_STATUS);
307 if (value != IXGBE_FAILED_READ_REG)
312 if (value == IXGBE_FAILED_READ_REG)
313 ixgbe_remove_adapter(hw);
315 value = readl(reg_addr + reg);
320 * ixgbe_read_reg - Read from device register
321 * @hw: hw specific details
322 * @reg: offset of register to read
324 * Returns : value read or IXGBE_FAILED_READ_REG if removed
326 * This function is used to read device registers. It checks for device
327 * removal by confirming any read that returns all ones by checking the
328 * status register value for all ones. This function avoids reading from
329 * the hardware if a removal was previously detected in which case it
330 * returns IXGBE_FAILED_READ_REG (all ones).
332 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
334 u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
337 if (ixgbe_removed(reg_addr))
338 return IXGBE_FAILED_READ_REG;
339 if (unlikely(hw->phy.nw_mng_if_sel &
340 IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE)) {
341 struct ixgbe_adapter *adapter;
344 for (i = 0; i < 200; ++i) {
345 value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY);
347 goto writes_completed;
348 if (value == IXGBE_FAILED_READ_REG) {
349 ixgbe_remove_adapter(hw);
350 return IXGBE_FAILED_READ_REG;
356 e_warn(hw, "register writes incomplete %08x\n", value);
360 value = readl(reg_addr + reg);
361 if (unlikely(value == IXGBE_FAILED_READ_REG))
362 value = ixgbe_check_remove(hw, reg);
366 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
370 pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
371 if (value == IXGBE_FAILED_READ_CFG_WORD) {
372 ixgbe_remove_adapter(hw);
378 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
380 struct ixgbe_adapter *adapter = hw->back;
383 if (ixgbe_removed(hw->hw_addr))
384 return IXGBE_FAILED_READ_CFG_WORD;
385 pci_read_config_word(adapter->pdev, reg, &value);
386 if (value == IXGBE_FAILED_READ_CFG_WORD &&
387 ixgbe_check_cfg_remove(hw, adapter->pdev))
388 return IXGBE_FAILED_READ_CFG_WORD;
392 #ifdef CONFIG_PCI_IOV
393 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
395 struct ixgbe_adapter *adapter = hw->back;
398 if (ixgbe_removed(hw->hw_addr))
399 return IXGBE_FAILED_READ_CFG_DWORD;
400 pci_read_config_dword(adapter->pdev, reg, &value);
401 if (value == IXGBE_FAILED_READ_CFG_DWORD &&
402 ixgbe_check_cfg_remove(hw, adapter->pdev))
403 return IXGBE_FAILED_READ_CFG_DWORD;
406 #endif /* CONFIG_PCI_IOV */
408 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
410 struct ixgbe_adapter *adapter = hw->back;
412 if (ixgbe_removed(hw->hw_addr))
414 pci_write_config_word(adapter->pdev, reg, value);
417 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
419 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
421 /* flush memory to make sure state is correct before next watchdog */
422 smp_mb__before_atomic();
423 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
426 struct ixgbe_reg_info {
431 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
433 /* General Registers */
434 {IXGBE_CTRL, "CTRL"},
435 {IXGBE_STATUS, "STATUS"},
436 {IXGBE_CTRL_EXT, "CTRL_EXT"},
438 /* Interrupt Registers */
439 {IXGBE_EICR, "EICR"},
442 {IXGBE_SRRCTL(0), "SRRCTL"},
443 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
444 {IXGBE_RDLEN(0), "RDLEN"},
445 {IXGBE_RDH(0), "RDH"},
446 {IXGBE_RDT(0), "RDT"},
447 {IXGBE_RXDCTL(0), "RXDCTL"},
448 {IXGBE_RDBAL(0), "RDBAL"},
449 {IXGBE_RDBAH(0), "RDBAH"},
452 {IXGBE_TDBAL(0), "TDBAL"},
453 {IXGBE_TDBAH(0), "TDBAH"},
454 {IXGBE_TDLEN(0), "TDLEN"},
455 {IXGBE_TDH(0), "TDH"},
456 {IXGBE_TDT(0), "TDT"},
457 {IXGBE_TXDCTL(0), "TXDCTL"},
459 /* List Terminator */
465 * ixgbe_regdump - register printout routine
467 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
473 switch (reginfo->ofs) {
474 case IXGBE_SRRCTL(0):
475 for (i = 0; i < 64; i++)
476 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
478 case IXGBE_DCA_RXCTRL(0):
479 for (i = 0; i < 64; i++)
480 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
483 for (i = 0; i < 64; i++)
484 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
487 for (i = 0; i < 64; i++)
488 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
491 for (i = 0; i < 64; i++)
492 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
494 case IXGBE_RXDCTL(0):
495 for (i = 0; i < 64; i++)
496 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
499 for (i = 0; i < 64; i++)
500 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
503 for (i = 0; i < 64; i++)
504 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
507 for (i = 0; i < 64; i++)
508 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
511 for (i = 0; i < 64; i++)
512 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
515 for (i = 0; i < 64; i++)
516 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
519 for (i = 0; i < 64; i++)
520 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
523 for (i = 0; i < 64; i++)
524 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
526 case IXGBE_TXDCTL(0):
527 for (i = 0; i < 64; i++)
528 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
531 pr_info("%-15s %08x\n",
532 reginfo->name, IXGBE_READ_REG(hw, reginfo->ofs));
542 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i, i + 7);
543 for (j = 0; j < 8; j++)
544 p += sprintf(p, " %08x", regs[i++]);
545 pr_err("%-15s%s\n", rname, buf);
550 static void ixgbe_print_buffer(struct ixgbe_ring *ring, int n)
552 struct ixgbe_tx_buffer *tx_buffer;
554 tx_buffer = &ring->tx_buffer_info[ring->next_to_clean];
555 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
556 n, ring->next_to_use, ring->next_to_clean,
557 (u64)dma_unmap_addr(tx_buffer, dma),
558 dma_unmap_len(tx_buffer, len),
559 tx_buffer->next_to_watch,
560 (u64)tx_buffer->time_stamp);
564 * ixgbe_dump - Print registers, tx-rings and rx-rings
566 static void ixgbe_dump(struct ixgbe_adapter *adapter)
568 struct net_device *netdev = adapter->netdev;
569 struct ixgbe_hw *hw = &adapter->hw;
570 struct ixgbe_reg_info *reginfo;
572 struct ixgbe_ring *ring;
573 struct ixgbe_tx_buffer *tx_buffer;
574 union ixgbe_adv_tx_desc *tx_desc;
575 struct my_u0 { u64 a; u64 b; } *u0;
576 struct ixgbe_ring *rx_ring;
577 union ixgbe_adv_rx_desc *rx_desc;
578 struct ixgbe_rx_buffer *rx_buffer_info;
581 if (!netif_msg_hw(adapter))
584 /* Print netdevice Info */
586 dev_info(&adapter->pdev->dev, "Net device Info\n");
587 pr_info("Device Name state "
589 pr_info("%-15s %016lX %016lX\n",
592 dev_trans_start(netdev));
595 /* Print Registers */
596 dev_info(&adapter->pdev->dev, "Register Dump\n");
597 pr_info(" Register Name Value\n");
598 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
599 reginfo->name; reginfo++) {
600 ixgbe_regdump(hw, reginfo);
603 /* Print TX Ring Summary */
604 if (!netdev || !netif_running(netdev))
607 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
608 pr_info(" %s %s %s %s\n",
609 "Queue [NTU] [NTC] [bi(ntc)->dma ]",
610 "leng", "ntw", "timestamp");
611 for (n = 0; n < adapter->num_tx_queues; n++) {
612 ring = adapter->tx_ring[n];
613 ixgbe_print_buffer(ring, n);
616 for (n = 0; n < adapter->num_xdp_queues; n++) {
617 ring = adapter->xdp_ring[n];
618 ixgbe_print_buffer(ring, n);
622 if (!netif_msg_tx_done(adapter))
623 goto rx_ring_summary;
625 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
627 /* Transmit Descriptor Formats
629 * 82598 Advanced Transmit Descriptor
630 * +--------------------------------------------------------------+
631 * 0 | Buffer Address [63:0] |
632 * +--------------------------------------------------------------+
633 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
634 * +--------------------------------------------------------------+
635 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
637 * 82598 Advanced Transmit Descriptor (Write-Back Format)
638 * +--------------------------------------------------------------+
640 * +--------------------------------------------------------------+
641 * 8 | RSV | STA | NXTSEQ |
642 * +--------------------------------------------------------------+
645 * 82599+ Advanced Transmit Descriptor
646 * +--------------------------------------------------------------+
647 * 0 | Buffer Address [63:0] |
648 * +--------------------------------------------------------------+
649 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN |
650 * +--------------------------------------------------------------+
651 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0
653 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
654 * +--------------------------------------------------------------+
656 * +--------------------------------------------------------------+
657 * 8 | RSV | STA | RSV |
658 * +--------------------------------------------------------------+
662 for (n = 0; n < adapter->num_tx_queues; n++) {
663 ring = adapter->tx_ring[n];
664 pr_info("------------------------------------\n");
665 pr_info("TX QUEUE INDEX = %d\n", ring->queue_index);
666 pr_info("------------------------------------\n");
667 pr_info("%s%s %s %s %s %s\n",
668 "T [desc] [address 63:0 ] ",
669 "[PlPOIdStDDt Ln] [bi->dma ] ",
670 "leng", "ntw", "timestamp", "bi->skb");
672 for (i = 0; ring->desc && (i < ring->count); i++) {
673 tx_desc = IXGBE_TX_DESC(ring, i);
674 tx_buffer = &ring->tx_buffer_info[i];
675 u0 = (struct my_u0 *)tx_desc;
676 if (dma_unmap_len(tx_buffer, len) > 0) {
677 const char *ring_desc;
679 if (i == ring->next_to_use &&
680 i == ring->next_to_clean)
681 ring_desc = " NTC/U";
682 else if (i == ring->next_to_use)
684 else if (i == ring->next_to_clean)
688 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p%s",
690 le64_to_cpu((__force __le64)u0->a),
691 le64_to_cpu((__force __le64)u0->b),
692 (u64)dma_unmap_addr(tx_buffer, dma),
693 dma_unmap_len(tx_buffer, len),
694 tx_buffer->next_to_watch,
695 (u64)tx_buffer->time_stamp,
699 if (netif_msg_pktdata(adapter) &&
701 print_hex_dump(KERN_INFO, "",
702 DUMP_PREFIX_ADDRESS, 16, 1,
703 tx_buffer->skb->data,
704 dma_unmap_len(tx_buffer, len),
710 /* Print RX Rings Summary */
712 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
713 pr_info("Queue [NTU] [NTC]\n");
714 for (n = 0; n < adapter->num_rx_queues; n++) {
715 rx_ring = adapter->rx_ring[n];
716 pr_info("%5d %5X %5X\n",
717 n, rx_ring->next_to_use, rx_ring->next_to_clean);
721 if (!netif_msg_rx_status(adapter))
724 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
726 /* Receive Descriptor Formats
728 * 82598 Advanced Receive Descriptor (Read) Format
730 * +-----------------------------------------------------+
731 * 0 | Packet Buffer Address [63:1] |A0/NSE|
732 * +----------------------------------------------+------+
733 * 8 | Header Buffer Address [63:1] | DD |
734 * +-----------------------------------------------------+
737 * 82598 Advanced Receive Descriptor (Write-Back) Format
739 * 63 48 47 32 31 30 21 20 16 15 4 3 0
740 * +------------------------------------------------------+
741 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS |
742 * | Packet | IP | | | | Type | Type |
743 * | Checksum | Ident | | | | | |
744 * +------------------------------------------------------+
745 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
746 * +------------------------------------------------------+
747 * 63 48 47 32 31 20 19 0
749 * 82599+ Advanced Receive Descriptor (Read) Format
751 * +-----------------------------------------------------+
752 * 0 | Packet Buffer Address [63:1] |A0/NSE|
753 * +----------------------------------------------+------+
754 * 8 | Header Buffer Address [63:1] | DD |
755 * +-----------------------------------------------------+
758 * 82599+ Advanced Receive Descriptor (Write-Back) Format
760 * 63 48 47 32 31 30 21 20 17 16 4 3 0
761 * +------------------------------------------------------+
762 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS |
763 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type |
764 * |/ Flow Dir Flt ID | | | | | |
765 * +------------------------------------------------------+
766 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
767 * +------------------------------------------------------+
768 * 63 48 47 32 31 20 19 0
771 for (n = 0; n < adapter->num_rx_queues; n++) {
772 rx_ring = adapter->rx_ring[n];
773 pr_info("------------------------------------\n");
774 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
775 pr_info("------------------------------------\n");
777 "R [desc] [ PktBuf A0] ",
778 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ",
779 "<-- Adv Rx Read format");
781 "RWB[desc] [PcsmIpSHl PtRs] ",
782 "[vl er S cks ln] ---------------- [bi->skb ] ",
783 "<-- Adv Rx Write-Back format");
785 for (i = 0; i < rx_ring->count; i++) {
786 const char *ring_desc;
788 if (i == rx_ring->next_to_use)
790 else if (i == rx_ring->next_to_clean)
795 rx_buffer_info = &rx_ring->rx_buffer_info[i];
796 rx_desc = IXGBE_RX_DESC(rx_ring, i);
797 u0 = (struct my_u0 *)rx_desc;
798 if (rx_desc->wb.upper.length) {
799 /* Descriptor Done */
800 pr_info("RWB[0x%03X] %016llX %016llX ---------------- %p%s\n",
802 le64_to_cpu((__force __le64)u0->a),
803 le64_to_cpu((__force __le64)u0->b),
807 pr_info("R [0x%03X] %016llX %016llX %016llX %p%s\n",
809 le64_to_cpu((__force __le64)u0->a),
810 le64_to_cpu((__force __le64)u0->b),
811 (u64)rx_buffer_info->dma,
815 if (netif_msg_pktdata(adapter) &&
816 rx_buffer_info->dma) {
817 print_hex_dump(KERN_INFO, "",
818 DUMP_PREFIX_ADDRESS, 16, 1,
819 page_address(rx_buffer_info->page) +
820 rx_buffer_info->page_offset,
821 ixgbe_rx_bufsz(rx_ring), true);
828 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
832 /* Let firmware take over control of h/w */
833 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
834 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
835 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
838 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
842 /* Let firmware know the driver has taken over */
843 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
844 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
845 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
849 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
850 * @adapter: pointer to adapter struct
851 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
852 * @queue: queue to map the corresponding interrupt to
853 * @msix_vector: the vector to map to the corresponding queue
856 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
857 u8 queue, u8 msix_vector)
860 struct ixgbe_hw *hw = &adapter->hw;
861 switch (hw->mac.type) {
862 case ixgbe_mac_82598EB:
863 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
866 index = (((direction * 64) + queue) >> 2) & 0x1F;
867 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
868 ivar &= ~(0xFF << (8 * (queue & 0x3)));
869 ivar |= (msix_vector << (8 * (queue & 0x3)));
870 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
872 case ixgbe_mac_82599EB:
875 case ixgbe_mac_X550EM_x:
876 case ixgbe_mac_x550em_a:
877 if (direction == -1) {
879 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
880 index = ((queue & 1) * 8);
881 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
882 ivar &= ~(0xFF << index);
883 ivar |= (msix_vector << index);
884 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
887 /* tx or rx causes */
888 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
889 index = ((16 * (queue & 1)) + (8 * direction));
890 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
891 ivar &= ~(0xFF << index);
892 ivar |= (msix_vector << index);
893 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
901 void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
906 switch (adapter->hw.mac.type) {
907 case ixgbe_mac_82598EB:
908 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
909 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
911 case ixgbe_mac_82599EB:
914 case ixgbe_mac_X550EM_x:
915 case ixgbe_mac_x550em_a:
916 mask = (qmask & 0xFFFFFFFF);
917 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
918 mask = (qmask >> 32);
919 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
926 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
928 struct ixgbe_hw *hw = &adapter->hw;
929 struct ixgbe_hw_stats *hwstats = &adapter->stats;
933 if ((hw->fc.current_mode != ixgbe_fc_full) &&
934 (hw->fc.current_mode != ixgbe_fc_rx_pause))
937 switch (hw->mac.type) {
938 case ixgbe_mac_82598EB:
939 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
942 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
944 hwstats->lxoffrxc += data;
946 /* refill credits (no tx hang) if we received xoff */
950 for (i = 0; i < adapter->num_tx_queues; i++)
951 clear_bit(__IXGBE_HANG_CHECK_ARMED,
952 &adapter->tx_ring[i]->state);
954 for (i = 0; i < adapter->num_xdp_queues; i++)
955 clear_bit(__IXGBE_HANG_CHECK_ARMED,
956 &adapter->xdp_ring[i]->state);
959 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
961 struct ixgbe_hw *hw = &adapter->hw;
962 struct ixgbe_hw_stats *hwstats = &adapter->stats;
966 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
968 if (adapter->ixgbe_ieee_pfc)
969 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
971 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
972 ixgbe_update_xoff_rx_lfc(adapter);
976 /* update stats for each tc, only valid with PFC enabled */
977 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
980 switch (hw->mac.type) {
981 case ixgbe_mac_82598EB:
982 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
985 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
987 hwstats->pxoffrxc[i] += pxoffrxc;
988 /* Get the TC for given UP */
989 tc = netdev_get_prio_tc_map(adapter->netdev, i);
990 xoff[tc] += pxoffrxc;
993 /* disarm tx queues that have received xoff frames */
994 for (i = 0; i < adapter->num_tx_queues; i++) {
995 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
997 tc = tx_ring->dcb_tc;
999 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1002 for (i = 0; i < adapter->num_xdp_queues; i++) {
1003 struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
1005 tc = xdp_ring->dcb_tc;
1007 clear_bit(__IXGBE_HANG_CHECK_ARMED, &xdp_ring->state);
1011 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1013 return ring->stats.packets;
1016 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1018 unsigned int head, tail;
1020 head = ring->next_to_clean;
1021 tail = ring->next_to_use;
1023 return ((head <= tail) ? tail : tail + ring->count) - head;
1026 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1028 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1029 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1030 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1032 clear_check_for_tx_hang(tx_ring);
1035 * Check for a hung queue, but be thorough. This verifies
1036 * that a transmit has been completed since the previous
1037 * check AND there is at least one packet pending. The
1038 * ARMED bit is set to indicate a potential hang. The
1039 * bit is cleared if a pause frame is received to remove
1040 * false hang detection due to PFC or 802.3x frames. By
1041 * requiring this to fail twice we avoid races with
1042 * pfc clearing the ARMED bit and conditions where we
1043 * run the check_tx_hang logic with a transmit completion
1044 * pending but without time to complete it yet.
1046 if (tx_done_old == tx_done && tx_pending)
1047 /* make sure it is true for two checks in a row */
1048 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1050 /* update completed stats and continue */
1051 tx_ring->tx_stats.tx_done_old = tx_done;
1052 /* reset the countdown */
1053 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1059 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1060 * @adapter: driver private struct
1062 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1065 /* Do the reset outside of interrupt context */
1066 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1067 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
1068 e_warn(drv, "initiating reset due to tx timeout\n");
1069 ixgbe_service_event_schedule(adapter);
1074 * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1075 * @netdev: network interface device structure
1076 * @queue_index: Tx queue to set
1077 * @maxrate: desired maximum transmit bitrate
1079 static int ixgbe_tx_maxrate(struct net_device *netdev,
1080 int queue_index, u32 maxrate)
1082 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1083 struct ixgbe_hw *hw = &adapter->hw;
1084 u32 bcnrc_val = ixgbe_link_mbps(adapter);
1089 /* Calculate the rate factor values to set */
1090 bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT;
1091 bcnrc_val /= maxrate;
1093 /* clear everything but the rate factor */
1094 bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK |
1095 IXGBE_RTTBCNRC_RF_DEC_MASK;
1097 /* enable the rate scheduler */
1098 bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA;
1100 IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index);
1101 IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val);
1107 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1108 * @q_vector: structure containing interrupt and ring information
1109 * @tx_ring: tx ring to clean
1110 * @napi_budget: Used to determine if we are in netpoll
1112 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1113 struct ixgbe_ring *tx_ring, int napi_budget)
1115 struct ixgbe_adapter *adapter = q_vector->adapter;
1116 struct ixgbe_tx_buffer *tx_buffer;
1117 union ixgbe_adv_tx_desc *tx_desc;
1118 unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
1119 unsigned int budget = q_vector->tx.work_limit;
1120 unsigned int i = tx_ring->next_to_clean;
1122 if (test_bit(__IXGBE_DOWN, &adapter->state))
1125 tx_buffer = &tx_ring->tx_buffer_info[i];
1126 tx_desc = IXGBE_TX_DESC(tx_ring, i);
1127 i -= tx_ring->count;
1130 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1132 /* if next_to_watch is not set then there is no work pending */
1136 /* prevent any other reads prior to eop_desc */
1139 /* if DD is not set pending work has not been completed */
1140 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1143 /* clear next_to_watch to prevent false hangs */
1144 tx_buffer->next_to_watch = NULL;
1146 /* update the statistics for this packet */
1147 total_bytes += tx_buffer->bytecount;
1148 total_packets += tx_buffer->gso_segs;
1149 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
1153 if (ring_is_xdp(tx_ring))
1154 xdp_return_frame(tx_buffer->xdpf);
1156 napi_consume_skb(tx_buffer->skb, napi_budget);
1158 /* unmap skb header data */
1159 dma_unmap_single(tx_ring->dev,
1160 dma_unmap_addr(tx_buffer, dma),
1161 dma_unmap_len(tx_buffer, len),
1164 /* clear tx_buffer data */
1165 dma_unmap_len_set(tx_buffer, len, 0);
1167 /* unmap remaining buffers */
1168 while (tx_desc != eop_desc) {
1173 i -= tx_ring->count;
1174 tx_buffer = tx_ring->tx_buffer_info;
1175 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1178 /* unmap any remaining paged data */
1179 if (dma_unmap_len(tx_buffer, len)) {
1180 dma_unmap_page(tx_ring->dev,
1181 dma_unmap_addr(tx_buffer, dma),
1182 dma_unmap_len(tx_buffer, len),
1184 dma_unmap_len_set(tx_buffer, len, 0);
1188 /* move us one more past the eop_desc for start of next pkt */
1193 i -= tx_ring->count;
1194 tx_buffer = tx_ring->tx_buffer_info;
1195 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1198 /* issue prefetch for next Tx descriptor */
1201 /* update budget accounting */
1203 } while (likely(budget));
1205 i += tx_ring->count;
1206 tx_ring->next_to_clean = i;
1207 u64_stats_update_begin(&tx_ring->syncp);
1208 tx_ring->stats.bytes += total_bytes;
1209 tx_ring->stats.packets += total_packets;
1210 u64_stats_update_end(&tx_ring->syncp);
1211 q_vector->tx.total_bytes += total_bytes;
1212 q_vector->tx.total_packets += total_packets;
1213 adapter->tx_ipsec += total_ipsec;
1215 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1216 /* schedule immediate reset if we believe we hung */
1217 struct ixgbe_hw *hw = &adapter->hw;
1218 e_err(drv, "Detected Tx Unit Hang %s\n"
1220 " TDH, TDT <%x>, <%x>\n"
1221 " next_to_use <%x>\n"
1222 " next_to_clean <%x>\n"
1223 "tx_buffer_info[next_to_clean]\n"
1224 " time_stamp <%lx>\n"
1226 ring_is_xdp(tx_ring) ? "(XDP)" : "",
1227 tx_ring->queue_index,
1228 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1229 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1230 tx_ring->next_to_use, i,
1231 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1233 if (!ring_is_xdp(tx_ring))
1234 netif_stop_subqueue(tx_ring->netdev,
1235 tx_ring->queue_index);
1238 "tx hang %d detected on queue %d, resetting adapter\n",
1239 adapter->tx_timeout_count + 1, tx_ring->queue_index);
1241 /* schedule immediate reset if we believe we hung */
1242 ixgbe_tx_timeout_reset(adapter);
1244 /* the adapter is about to reset, no point in enabling stuff */
1248 if (ring_is_xdp(tx_ring))
1251 netdev_tx_completed_queue(txring_txq(tx_ring),
1252 total_packets, total_bytes);
1254 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1255 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1256 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1257 /* Make sure that anybody stopping the queue after this
1258 * sees the new next_to_clean.
1261 if (__netif_subqueue_stopped(tx_ring->netdev,
1262 tx_ring->queue_index)
1263 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1264 netif_wake_subqueue(tx_ring->netdev,
1265 tx_ring->queue_index);
1266 ++tx_ring->tx_stats.restart_queue;
1273 #ifdef CONFIG_IXGBE_DCA
1274 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1275 struct ixgbe_ring *tx_ring,
1278 struct ixgbe_hw *hw = &adapter->hw;
1282 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1283 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1285 switch (hw->mac.type) {
1286 case ixgbe_mac_82598EB:
1287 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1289 case ixgbe_mac_82599EB:
1290 case ixgbe_mac_X540:
1291 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1292 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1295 /* for unknown hardware do not write register */
1300 * We can enable relaxed ordering for reads, but not writes when
1301 * DCA is enabled. This is due to a known issue in some chipsets
1302 * which will cause the DCA tag to be cleared.
1304 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1305 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1306 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1308 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1311 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1312 struct ixgbe_ring *rx_ring,
1315 struct ixgbe_hw *hw = &adapter->hw;
1317 u8 reg_idx = rx_ring->reg_idx;
1319 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1320 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1322 switch (hw->mac.type) {
1323 case ixgbe_mac_82599EB:
1324 case ixgbe_mac_X540:
1325 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1332 * We can enable relaxed ordering for reads, but not writes when
1333 * DCA is enabled. This is due to a known issue in some chipsets
1334 * which will cause the DCA tag to be cleared.
1336 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1337 IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1338 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1340 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1343 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1345 struct ixgbe_adapter *adapter = q_vector->adapter;
1346 struct ixgbe_ring *ring;
1347 int cpu = get_cpu();
1349 if (q_vector->cpu == cpu)
1352 ixgbe_for_each_ring(ring, q_vector->tx)
1353 ixgbe_update_tx_dca(adapter, ring, cpu);
1355 ixgbe_for_each_ring(ring, q_vector->rx)
1356 ixgbe_update_rx_dca(adapter, ring, cpu);
1358 q_vector->cpu = cpu;
1363 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1367 /* always use CB2 mode, difference is masked in the CB driver */
1368 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1369 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1370 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1372 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1373 IXGBE_DCA_CTRL_DCA_DISABLE);
1375 for (i = 0; i < adapter->num_q_vectors; i++) {
1376 adapter->q_vector[i]->cpu = -1;
1377 ixgbe_update_dca(adapter->q_vector[i]);
1381 static int __ixgbe_notify_dca(struct device *dev, void *data)
1383 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1384 unsigned long event = *(unsigned long *)data;
1386 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1390 case DCA_PROVIDER_ADD:
1391 /* if we're already enabled, don't do it again */
1392 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1394 if (dca_add_requester(dev) == 0) {
1395 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1396 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1397 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1400 /* fall through - DCA is disabled. */
1401 case DCA_PROVIDER_REMOVE:
1402 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1403 dca_remove_requester(dev);
1404 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1405 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1406 IXGBE_DCA_CTRL_DCA_DISABLE);
1414 #endif /* CONFIG_IXGBE_DCA */
1416 #define IXGBE_RSS_L4_TYPES_MASK \
1417 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1418 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1419 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1420 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1422 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1423 union ixgbe_adv_rx_desc *rx_desc,
1424 struct sk_buff *skb)
1428 if (!(ring->netdev->features & NETIF_F_RXHASH))
1431 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1432 IXGBE_RXDADV_RSSTYPE_MASK;
1437 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1438 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1439 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1444 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1445 * @ring: structure containing ring specific data
1446 * @rx_desc: advanced rx descriptor
1448 * Returns : true if it is FCoE pkt
1450 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1451 union ixgbe_adv_rx_desc *rx_desc)
1453 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1455 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1456 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1457 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1458 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1461 #endif /* IXGBE_FCOE */
1463 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1464 * @ring: structure containing ring specific data
1465 * @rx_desc: current Rx descriptor being processed
1466 * @skb: skb currently being received and modified
1468 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1469 union ixgbe_adv_rx_desc *rx_desc,
1470 struct sk_buff *skb)
1472 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1473 bool encap_pkt = false;
1475 skb_checksum_none_assert(skb);
1477 /* Rx csum disabled */
1478 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1481 /* check for VXLAN and Geneve packets */
1482 if (pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) {
1484 skb->encapsulation = 1;
1487 /* if IP and error */
1488 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1489 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1490 ring->rx_stats.csum_err++;
1494 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1497 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1499 * 82599 errata, UDP frames with a 0 checksum can be marked as
1502 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1503 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1506 ring->rx_stats.csum_err++;
1510 /* It must be a TCP or UDP packet with a valid checksum */
1511 skb->ip_summed = CHECKSUM_UNNECESSARY;
1513 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1516 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1517 skb->ip_summed = CHECKSUM_NONE;
1520 /* If we checked the outer header let the stack know */
1521 skb->csum_level = 1;
1525 static inline unsigned int ixgbe_rx_offset(struct ixgbe_ring *rx_ring)
1527 return ring_uses_build_skb(rx_ring) ? IXGBE_SKB_PAD : 0;
1530 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1531 struct ixgbe_rx_buffer *bi)
1533 struct page *page = bi->page;
1536 /* since we are recycling buffers we should seldom need to alloc */
1540 /* alloc new page for storage */
1541 page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1542 if (unlikely(!page)) {
1543 rx_ring->rx_stats.alloc_rx_page_failed++;
1547 /* map page for use */
1548 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1549 ixgbe_rx_pg_size(rx_ring),
1554 * if mapping failed free memory back to system since
1555 * there isn't much point in holding memory we can't use
1557 if (dma_mapping_error(rx_ring->dev, dma)) {
1558 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1560 rx_ring->rx_stats.alloc_rx_page_failed++;
1566 bi->page_offset = ixgbe_rx_offset(rx_ring);
1567 page_ref_add(page, USHRT_MAX - 1);
1568 bi->pagecnt_bias = USHRT_MAX;
1569 rx_ring->rx_stats.alloc_rx_page++;
1575 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1576 * @rx_ring: ring to place buffers on
1577 * @cleaned_count: number of buffers to replace
1579 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1581 union ixgbe_adv_rx_desc *rx_desc;
1582 struct ixgbe_rx_buffer *bi;
1583 u16 i = rx_ring->next_to_use;
1590 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1591 bi = &rx_ring->rx_buffer_info[i];
1592 i -= rx_ring->count;
1594 bufsz = ixgbe_rx_bufsz(rx_ring);
1597 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1600 /* sync the buffer for use by the device */
1601 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1602 bi->page_offset, bufsz,
1606 * Refresh the desc even if buffer_addrs didn't change
1607 * because each write-back erases this info.
1609 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1615 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1616 bi = rx_ring->rx_buffer_info;
1617 i -= rx_ring->count;
1620 /* clear the length for the next_to_use descriptor */
1621 rx_desc->wb.upper.length = 0;
1624 } while (cleaned_count);
1626 i += rx_ring->count;
1628 if (rx_ring->next_to_use != i) {
1629 rx_ring->next_to_use = i;
1631 /* update next to alloc since we have filled the ring */
1632 rx_ring->next_to_alloc = i;
1634 /* Force memory writes to complete before letting h/w
1635 * know there are new descriptors to fetch. (Only
1636 * applicable for weak-ordered memory model archs,
1640 writel(i, rx_ring->tail);
1644 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1645 struct sk_buff *skb)
1647 u16 hdr_len = skb_headlen(skb);
1649 /* set gso_size to avoid messing up TCP MSS */
1650 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1651 IXGBE_CB(skb)->append_cnt);
1652 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1655 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1656 struct sk_buff *skb)
1658 /* if append_cnt is 0 then frame is not RSC */
1659 if (!IXGBE_CB(skb)->append_cnt)
1662 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1663 rx_ring->rx_stats.rsc_flush++;
1665 ixgbe_set_rsc_gso_size(rx_ring, skb);
1667 /* gso_size is computed using append_cnt so always clear it last */
1668 IXGBE_CB(skb)->append_cnt = 0;
1672 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1673 * @rx_ring: rx descriptor ring packet is being transacted on
1674 * @rx_desc: pointer to the EOP Rx descriptor
1675 * @skb: pointer to current skb being populated
1677 * This function checks the ring, descriptor, and packet information in
1678 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1679 * other fields within the skb.
1681 void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1682 union ixgbe_adv_rx_desc *rx_desc,
1683 struct sk_buff *skb)
1685 struct net_device *dev = rx_ring->netdev;
1686 u32 flags = rx_ring->q_vector->adapter->flags;
1688 ixgbe_update_rsc_stats(rx_ring, skb);
1690 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1692 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1694 if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1695 ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1697 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1698 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1699 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1700 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1703 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
1704 ixgbe_ipsec_rx(rx_ring, rx_desc, skb);
1706 /* record Rx queue, or update MACVLAN statistics */
1707 if (netif_is_ixgbe(dev))
1708 skb_record_rx_queue(skb, rx_ring->queue_index);
1710 macvlan_count_rx(netdev_priv(dev), skb->len + ETH_HLEN, true,
1713 skb->protocol = eth_type_trans(skb, dev);
1716 void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1717 struct sk_buff *skb)
1719 napi_gro_receive(&q_vector->napi, skb);
1723 * ixgbe_is_non_eop - process handling of non-EOP buffers
1724 * @rx_ring: Rx ring being processed
1725 * @rx_desc: Rx descriptor for current buffer
1726 * @skb: Current socket buffer containing buffer in progress
1728 * This function updates next to clean. If the buffer is an EOP buffer
1729 * this function exits returning false, otherwise it will place the
1730 * sk_buff in the next buffer to be chained and return true indicating
1731 * that this is in fact a non-EOP buffer.
1733 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1734 union ixgbe_adv_rx_desc *rx_desc,
1735 struct sk_buff *skb)
1737 u32 ntc = rx_ring->next_to_clean + 1;
1739 /* fetch, update, and store next to clean */
1740 ntc = (ntc < rx_ring->count) ? ntc : 0;
1741 rx_ring->next_to_clean = ntc;
1743 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1745 /* update RSC append count if present */
1746 if (ring_is_rsc_enabled(rx_ring)) {
1747 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1748 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1750 if (unlikely(rsc_enabled)) {
1751 u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1753 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1754 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1756 /* update ntc based on RSC value */
1757 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1758 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1759 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1763 /* if we are the last buffer then there is nothing else to do */
1764 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1767 /* place skb in next buffer to be received */
1768 rx_ring->rx_buffer_info[ntc].skb = skb;
1769 rx_ring->rx_stats.non_eop_descs++;
1775 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1776 * @rx_ring: rx descriptor ring packet is being transacted on
1777 * @skb: pointer to current skb being adjusted
1779 * This function is an ixgbe specific version of __pskb_pull_tail. The
1780 * main difference between this version and the original function is that
1781 * this function can make several assumptions about the state of things
1782 * that allow for significant optimizations versus the standard function.
1783 * As a result we can do things like drop a frag and maintain an accurate
1784 * truesize for the skb.
1786 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1787 struct sk_buff *skb)
1789 skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1791 unsigned int pull_len;
1794 * it is valid to use page_address instead of kmap since we are
1795 * working with pages allocated out of the lomem pool per
1796 * alloc_page(GFP_ATOMIC)
1798 va = skb_frag_address(frag);
1801 * we need the header to contain the greater of either ETH_HLEN or
1802 * 60 bytes if the skb->len is less than 60 for skb_pad.
1804 pull_len = eth_get_headlen(skb->dev, va, IXGBE_RX_HDR_SIZE);
1806 /* align pull length to size of long to optimize memcpy performance */
1807 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1809 /* update all of the pointers */
1810 skb_frag_size_sub(frag, pull_len);
1811 skb_frag_off_add(frag, pull_len);
1812 skb->data_len -= pull_len;
1813 skb->tail += pull_len;
1817 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1818 * @rx_ring: rx descriptor ring packet is being transacted on
1819 * @skb: pointer to current skb being updated
1821 * This function provides a basic DMA sync up for the first fragment of an
1822 * skb. The reason for doing this is that the first fragment cannot be
1823 * unmapped until we have reached the end of packet descriptor for a buffer
1826 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1827 struct sk_buff *skb)
1829 if (ring_uses_build_skb(rx_ring)) {
1830 unsigned long offset = (unsigned long)(skb->data) & ~PAGE_MASK;
1832 dma_sync_single_range_for_cpu(rx_ring->dev,
1838 skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1840 dma_sync_single_range_for_cpu(rx_ring->dev,
1843 skb_frag_size(frag),
1847 /* If the page was released, just unmap it. */
1848 if (unlikely(IXGBE_CB(skb)->page_released)) {
1849 dma_unmap_page_attrs(rx_ring->dev, IXGBE_CB(skb)->dma,
1850 ixgbe_rx_pg_size(rx_ring),
1857 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1858 * @rx_ring: rx descriptor ring packet is being transacted on
1859 * @rx_desc: pointer to the EOP Rx descriptor
1860 * @skb: pointer to current skb being fixed
1862 * Check if the skb is valid in the XDP case it will be an error pointer.
1863 * Return true in this case to abort processing and advance to next
1866 * Check for corrupted packet headers caused by senders on the local L2
1867 * embedded NIC switch not setting up their Tx Descriptors right. These
1868 * should be very rare.
1870 * Also address the case where we are pulling data in on pages only
1871 * and as such no data is present in the skb header.
1873 * In addition if skb is not at least 60 bytes we need to pad it so that
1874 * it is large enough to qualify as a valid Ethernet frame.
1876 * Returns true if an error was encountered and skb was freed.
1878 bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1879 union ixgbe_adv_rx_desc *rx_desc,
1880 struct sk_buff *skb)
1882 struct net_device *netdev = rx_ring->netdev;
1884 /* XDP packets use error pointer so abort at this point */
1888 /* Verify netdev is present, and that packet does not have any
1889 * errors that would be unacceptable to the netdev.
1892 (unlikely(ixgbe_test_staterr(rx_desc,
1893 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1894 !(netdev->features & NETIF_F_RXALL)))) {
1895 dev_kfree_skb_any(skb);
1899 /* place header in linear portion of buffer */
1900 if (!skb_headlen(skb))
1901 ixgbe_pull_tail(rx_ring, skb);
1904 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1905 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1909 /* if eth_skb_pad returns an error the skb was freed */
1910 if (eth_skb_pad(skb))
1917 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1918 * @rx_ring: rx descriptor ring to store buffers on
1919 * @old_buff: donor buffer to have page reused
1921 * Synchronizes page for reuse by the adapter
1923 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1924 struct ixgbe_rx_buffer *old_buff)
1926 struct ixgbe_rx_buffer *new_buff;
1927 u16 nta = rx_ring->next_to_alloc;
1929 new_buff = &rx_ring->rx_buffer_info[nta];
1931 /* update, and store next to alloc */
1933 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1935 /* Transfer page from old buffer to new buffer.
1936 * Move each member individually to avoid possible store
1937 * forwarding stalls and unnecessary copy of skb.
1939 new_buff->dma = old_buff->dma;
1940 new_buff->page = old_buff->page;
1941 new_buff->page_offset = old_buff->page_offset;
1942 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
1945 static inline bool ixgbe_page_is_reserved(struct page *page)
1947 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1950 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer)
1952 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1953 struct page *page = rx_buffer->page;
1955 /* avoid re-using remote pages */
1956 if (unlikely(ixgbe_page_is_reserved(page)))
1959 #if (PAGE_SIZE < 8192)
1960 /* if we are only owner of page we can reuse it */
1961 if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
1964 /* The last offset is a bit aggressive in that we assume the
1965 * worst case of FCoE being enabled and using a 3K buffer.
1966 * However this should have minimal impact as the 1K extra is
1967 * still less than one buffer in size.
1969 #define IXGBE_LAST_OFFSET \
1970 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1971 if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1975 /* If we have drained the page fragment pool we need to update
1976 * the pagecnt_bias and page count so that we fully restock the
1977 * number of references the driver holds.
1979 if (unlikely(pagecnt_bias == 1)) {
1980 page_ref_add(page, USHRT_MAX - 1);
1981 rx_buffer->pagecnt_bias = USHRT_MAX;
1988 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1989 * @rx_ring: rx descriptor ring to transact packets on
1990 * @rx_buffer: buffer containing page to add
1991 * @skb: sk_buff to place the data into
1992 * @size: size of data in rx_buffer
1994 * This function will add the data contained in rx_buffer->page to the skb.
1995 * This is done either through a direct copy if the data in the buffer is
1996 * less than the skb header size, otherwise it will just attach the page as
1997 * a frag to the skb.
1999 * The function will then update the page offset if necessary and return
2000 * true if the buffer can be reused by the adapter.
2002 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
2003 struct ixgbe_rx_buffer *rx_buffer,
2004 struct sk_buff *skb,
2007 #if (PAGE_SIZE < 8192)
2008 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2010 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2011 SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2012 SKB_DATA_ALIGN(size);
2014 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2015 rx_buffer->page_offset, size, truesize);
2016 #if (PAGE_SIZE < 8192)
2017 rx_buffer->page_offset ^= truesize;
2019 rx_buffer->page_offset += truesize;
2023 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2024 union ixgbe_adv_rx_desc *rx_desc,
2025 struct sk_buff **skb,
2026 const unsigned int size)
2028 struct ixgbe_rx_buffer *rx_buffer;
2030 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2031 prefetchw(rx_buffer->page);
2032 *skb = rx_buffer->skb;
2034 /* Delay unmapping of the first packet. It carries the header
2035 * information, HW may still access the header after the writeback.
2036 * Only unmap it when EOP is reached
2038 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2043 ixgbe_dma_sync_frag(rx_ring, *skb);
2046 /* we are reusing so sync this buffer for CPU use */
2047 dma_sync_single_range_for_cpu(rx_ring->dev,
2049 rx_buffer->page_offset,
2053 rx_buffer->pagecnt_bias--;
2058 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2059 struct ixgbe_rx_buffer *rx_buffer,
2060 struct sk_buff *skb)
2062 if (ixgbe_can_reuse_rx_page(rx_buffer)) {
2063 /* hand second half of page back to the ring */
2064 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2066 if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2067 /* the page has been released from the ring */
2068 IXGBE_CB(skb)->page_released = true;
2070 /* we are not reusing the buffer so unmap it */
2071 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2072 ixgbe_rx_pg_size(rx_ring),
2076 __page_frag_cache_drain(rx_buffer->page,
2077 rx_buffer->pagecnt_bias);
2080 /* clear contents of rx_buffer */
2081 rx_buffer->page = NULL;
2082 rx_buffer->skb = NULL;
2085 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2086 struct ixgbe_rx_buffer *rx_buffer,
2087 struct xdp_buff *xdp,
2088 union ixgbe_adv_rx_desc *rx_desc)
2090 unsigned int size = xdp->data_end - xdp->data;
2091 #if (PAGE_SIZE < 8192)
2092 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2094 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2095 xdp->data_hard_start);
2097 struct sk_buff *skb;
2099 /* prefetch first cache line of first page */
2100 prefetch(xdp->data);
2101 #if L1_CACHE_BYTES < 128
2102 prefetch(xdp->data + L1_CACHE_BYTES);
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 prefetch(xdp->data_meta);
2167 #if L1_CACHE_BYTES < 128
2168 prefetch(xdp->data_meta + L1_CACHE_BYTES);
2171 /* build an skb to around the page buffer */
2172 skb = build_skb(xdp->data_hard_start, truesize);
2176 /* update pointers within the skb to store the data */
2177 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2178 __skb_put(skb, xdp->data_end - xdp->data);
2180 skb_metadata_set(skb, metasize);
2182 /* record DMA address if this is the start of a chain of buffers */
2183 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2184 IXGBE_CB(skb)->dma = rx_buffer->dma;
2186 /* update buffer offset */
2187 #if (PAGE_SIZE < 8192)
2188 rx_buffer->page_offset ^= truesize;
2190 rx_buffer->page_offset += truesize;
2196 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2197 struct ixgbe_ring *rx_ring,
2198 struct xdp_buff *xdp)
2200 int err, result = IXGBE_XDP_PASS;
2201 struct bpf_prog *xdp_prog;
2202 struct xdp_frame *xdpf;
2206 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2211 prefetchw(xdp->data_hard_start); /* xdp_frame write */
2213 act = bpf_prog_run_xdp(xdp_prog, xdp);
2218 xdpf = convert_to_xdp_frame(xdp);
2219 if (unlikely(!xdpf)) {
2220 result = IXGBE_XDP_CONSUMED;
2223 result = ixgbe_xmit_xdp_ring(adapter, xdpf);
2226 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2228 result = IXGBE_XDP_REDIR;
2230 result = IXGBE_XDP_CONSUMED;
2233 bpf_warn_invalid_xdp_action(act);
2236 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2237 /* fallthrough -- handle aborts by dropping packet */
2239 result = IXGBE_XDP_CONSUMED;
2244 return ERR_PTR(-result);
2247 static unsigned int ixgbe_rx_frame_truesize(struct ixgbe_ring *rx_ring,
2250 unsigned int truesize;
2252 #if (PAGE_SIZE < 8192)
2253 truesize = ixgbe_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
2255 truesize = ring_uses_build_skb(rx_ring) ?
2256 SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) +
2257 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
2258 SKB_DATA_ALIGN(size);
2263 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2264 struct ixgbe_rx_buffer *rx_buffer,
2267 unsigned int truesize = ixgbe_rx_frame_truesize(rx_ring, size);
2268 #if (PAGE_SIZE < 8192)
2269 rx_buffer->page_offset ^= truesize;
2271 rx_buffer->page_offset += truesize;
2276 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2277 * @q_vector: structure containing interrupt and ring information
2278 * @rx_ring: rx descriptor ring to transact packets on
2279 * @budget: Total limit on number of packets to process
2281 * This function provides a "bounce buffer" approach to Rx interrupt
2282 * processing. The advantage to this is that on systems that have
2283 * expensive overhead for IOMMU access this provides a means of avoiding
2284 * it by maintaining the mapping of the page to the syste.
2286 * Returns amount of work completed
2288 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2289 struct ixgbe_ring *rx_ring,
2292 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2293 struct ixgbe_adapter *adapter = q_vector->adapter;
2296 unsigned int mss = 0;
2297 #endif /* IXGBE_FCOE */
2298 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2299 unsigned int xdp_xmit = 0;
2300 struct xdp_buff xdp;
2302 xdp.rxq = &rx_ring->xdp_rxq;
2304 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
2305 #if (PAGE_SIZE < 8192)
2306 xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, 0);
2309 while (likely(total_rx_packets < budget)) {
2310 union ixgbe_adv_rx_desc *rx_desc;
2311 struct ixgbe_rx_buffer *rx_buffer;
2312 struct sk_buff *skb;
2315 /* return some buffers to hardware, one at a time is too slow */
2316 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2317 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2321 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2322 size = le16_to_cpu(rx_desc->wb.upper.length);
2326 /* This memory barrier is needed to keep us from reading
2327 * any other fields out of the rx_desc until we know the
2328 * descriptor has been written back
2332 rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size);
2334 /* retrieve a buffer from the ring */
2336 xdp.data = page_address(rx_buffer->page) +
2337 rx_buffer->page_offset;
2338 xdp.data_meta = xdp.data;
2339 xdp.data_hard_start = xdp.data -
2340 ixgbe_rx_offset(rx_ring);
2341 xdp.data_end = xdp.data + size;
2342 #if (PAGE_SIZE > 4096)
2343 /* At larger PAGE_SIZE, frame_sz depend on len size */
2344 xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, size);
2346 skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2350 unsigned int xdp_res = -PTR_ERR(skb);
2352 if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2353 xdp_xmit |= xdp_res;
2354 ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2356 rx_buffer->pagecnt_bias++;
2359 total_rx_bytes += size;
2361 ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2362 } else if (ring_uses_build_skb(rx_ring)) {
2363 skb = ixgbe_build_skb(rx_ring, rx_buffer,
2366 skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2370 /* exit if we failed to retrieve a buffer */
2372 rx_ring->rx_stats.alloc_rx_buff_failed++;
2373 rx_buffer->pagecnt_bias++;
2377 ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb);
2380 /* place incomplete frames back on ring for completion */
2381 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2384 /* verify the packet layout is correct */
2385 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2388 /* probably a little skewed due to removing CRC */
2389 total_rx_bytes += skb->len;
2391 /* populate checksum, timestamp, VLAN, and protocol */
2392 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2395 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2396 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2397 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2398 /* include DDPed FCoE data */
2399 if (ddp_bytes > 0) {
2401 mss = rx_ring->netdev->mtu -
2402 sizeof(struct fcoe_hdr) -
2403 sizeof(struct fc_frame_header) -
2404 sizeof(struct fcoe_crc_eof);
2408 total_rx_bytes += ddp_bytes;
2409 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2413 dev_kfree_skb_any(skb);
2418 #endif /* IXGBE_FCOE */
2419 ixgbe_rx_skb(q_vector, skb);
2421 /* update budget accounting */
2425 if (xdp_xmit & IXGBE_XDP_REDIR)
2428 if (xdp_xmit & IXGBE_XDP_TX) {
2429 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2431 /* Force memory writes to complete before letting h/w
2432 * know there are new descriptors to fetch.
2435 writel(ring->next_to_use, ring->tail);
2438 u64_stats_update_begin(&rx_ring->syncp);
2439 rx_ring->stats.packets += total_rx_packets;
2440 rx_ring->stats.bytes += total_rx_bytes;
2441 u64_stats_update_end(&rx_ring->syncp);
2442 q_vector->rx.total_packets += total_rx_packets;
2443 q_vector->rx.total_bytes += total_rx_bytes;
2445 return total_rx_packets;
2449 * ixgbe_configure_msix - Configure MSI-X hardware
2450 * @adapter: board private structure
2452 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2455 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2457 struct ixgbe_q_vector *q_vector;
2461 /* Populate MSIX to EITR Select */
2462 if (adapter->num_vfs > 32) {
2463 u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2464 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2468 * Populate the IVAR table and set the ITR values to the
2469 * corresponding register.
2471 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2472 struct ixgbe_ring *ring;
2473 q_vector = adapter->q_vector[v_idx];
2475 ixgbe_for_each_ring(ring, q_vector->rx)
2476 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2478 ixgbe_for_each_ring(ring, q_vector->tx)
2479 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2481 ixgbe_write_eitr(q_vector);
2484 switch (adapter->hw.mac.type) {
2485 case ixgbe_mac_82598EB:
2486 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2489 case ixgbe_mac_82599EB:
2490 case ixgbe_mac_X540:
2491 case ixgbe_mac_X550:
2492 case ixgbe_mac_X550EM_x:
2493 case ixgbe_mac_x550em_a:
2494 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2499 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2501 /* set up to autoclear timer, and the vectors */
2502 mask = IXGBE_EIMS_ENABLE_MASK;
2503 mask &= ~(IXGBE_EIMS_OTHER |
2504 IXGBE_EIMS_MAILBOX |
2507 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2511 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2512 * @q_vector: structure containing interrupt and ring information
2513 * @ring_container: structure containing ring performance data
2515 * Stores a new ITR value based on packets and byte
2516 * counts during the last interrupt. The advantage of per interrupt
2517 * computation is faster updates and more accurate ITR for the current
2518 * traffic pattern. Constants in this function were computed
2519 * based on theoretical maximum wire speed and thresholds were set based
2520 * on testing data as well as attempting to minimize response time
2521 * while increasing bulk throughput.
2523 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2524 struct ixgbe_ring_container *ring_container)
2526 unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2527 IXGBE_ITR_ADAPTIVE_LATENCY;
2528 unsigned int avg_wire_size, packets, bytes;
2529 unsigned long next_update = jiffies;
2531 /* If we don't have any rings just leave ourselves set for maximum
2532 * possible latency so we take ourselves out of the equation.
2534 if (!ring_container->ring)
2537 /* If we didn't update within up to 1 - 2 jiffies we can assume
2538 * that either packets are coming in so slow there hasn't been
2539 * any work, or that there is so much work that NAPI is dealing
2540 * with interrupt moderation and we don't need to do anything.
2542 if (time_after(next_update, ring_container->next_update))
2545 packets = ring_container->total_packets;
2547 /* We have no packets to actually measure against. This means
2548 * either one of the other queues on this vector is active or
2549 * we are a Tx queue doing TSO with too high of an interrupt rate.
2551 * When this occurs just tick up our delay by the minimum value
2552 * and hope that this extra delay will prevent us from being called
2553 * without any work on our queue.
2556 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2557 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2558 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2559 itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2563 bytes = ring_container->total_bytes;
2565 /* If packets are less than 4 or bytes are less than 9000 assume
2566 * insufficient data to use bulk rate limiting approach. We are
2567 * likely latency driven.
2569 if (packets < 4 && bytes < 9000) {
2570 itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2571 goto adjust_by_size;
2574 /* Between 4 and 48 we can assume that our current interrupt delay
2575 * is only slightly too low. As such we should increase it by a small
2579 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2580 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2581 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2585 /* Between 48 and 96 is our "goldilocks" zone where we are working
2586 * out "just right". Just report that our current ITR is good for us.
2589 itr = q_vector->itr >> 2;
2593 /* If packet count is 96 or greater we are likely looking at a slight
2594 * overrun of the delay we want. Try halving our delay to see if that
2595 * will cut the number of packets in half per interrupt.
2597 if (packets < 256) {
2598 itr = q_vector->itr >> 3;
2599 if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2600 itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2604 /* The paths below assume we are dealing with a bulk ITR since number
2605 * of packets is 256 or greater. We are just going to have to compute
2606 * a value and try to bring the count under control, though for smaller
2607 * packet sizes there isn't much we can do as NAPI polling will likely
2608 * be kicking in sooner rather than later.
2610 itr = IXGBE_ITR_ADAPTIVE_BULK;
2613 /* If packet counts are 256 or greater we can assume we have a gross
2614 * overestimation of what the rate should be. Instead of trying to fine
2615 * tune it just use the formula below to try and dial in an exact value
2616 * give the current packet size of the frame.
2618 avg_wire_size = bytes / packets;
2620 /* The following is a crude approximation of:
2621 * wmem_default / (size + overhead) = desired_pkts_per_int
2622 * rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2623 * (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2625 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2626 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2629 * (170 * (size + 24)) / (size + 640) = ITR
2631 * We first do some math on the packet size and then finally bitshift
2632 * by 8 after rounding up. We also have to account for PCIe link speed
2633 * difference as ITR scales based on this.
2635 if (avg_wire_size <= 60) {
2636 /* Start at 50k ints/sec */
2637 avg_wire_size = 5120;
2638 } else if (avg_wire_size <= 316) {
2639 /* 50K ints/sec to 16K ints/sec */
2640 avg_wire_size *= 40;
2641 avg_wire_size += 2720;
2642 } else if (avg_wire_size <= 1084) {
2643 /* 16K ints/sec to 9.2K ints/sec */
2644 avg_wire_size *= 15;
2645 avg_wire_size += 11452;
2646 } else if (avg_wire_size < 1968) {
2647 /* 9.2K ints/sec to 8K ints/sec */
2649 avg_wire_size += 22420;
2651 /* plateau at a limit of 8K ints/sec */
2652 avg_wire_size = 32256;
2655 /* If we are in low latency mode half our delay which doubles the rate
2656 * to somewhere between 100K to 16K ints/sec
2658 if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2659 avg_wire_size >>= 1;
2661 /* Resultant value is 256 times larger than it needs to be. This
2662 * gives us room to adjust the value as needed to either increase
2663 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2665 * Use addition as we have already recorded the new latency flag
2666 * for the ITR value.
2668 switch (q_vector->adapter->link_speed) {
2669 case IXGBE_LINK_SPEED_10GB_FULL:
2670 case IXGBE_LINK_SPEED_100_FULL:
2672 itr += DIV_ROUND_UP(avg_wire_size,
2673 IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2674 IXGBE_ITR_ADAPTIVE_MIN_INC;
2676 case IXGBE_LINK_SPEED_2_5GB_FULL:
2677 case IXGBE_LINK_SPEED_1GB_FULL:
2678 case IXGBE_LINK_SPEED_10_FULL:
2679 if (avg_wire_size > 8064)
2680 avg_wire_size = 8064;
2681 itr += DIV_ROUND_UP(avg_wire_size,
2682 IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2683 IXGBE_ITR_ADAPTIVE_MIN_INC;
2688 /* write back value */
2689 ring_container->itr = itr;
2691 /* next update should occur within next jiffy */
2692 ring_container->next_update = next_update + 1;
2694 ring_container->total_bytes = 0;
2695 ring_container->total_packets = 0;
2699 * ixgbe_write_eitr - write EITR register in hardware specific way
2700 * @q_vector: structure containing interrupt and ring information
2702 * This function is made to be called by ethtool and by the driver
2703 * when it needs to update EITR registers at runtime. Hardware
2704 * specific quirks/differences are taken care of here.
2706 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2708 struct ixgbe_adapter *adapter = q_vector->adapter;
2709 struct ixgbe_hw *hw = &adapter->hw;
2710 int v_idx = q_vector->v_idx;
2711 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2713 switch (adapter->hw.mac.type) {
2714 case ixgbe_mac_82598EB:
2715 /* must write high and low 16 bits to reset counter */
2716 itr_reg |= (itr_reg << 16);
2718 case ixgbe_mac_82599EB:
2719 case ixgbe_mac_X540:
2720 case ixgbe_mac_X550:
2721 case ixgbe_mac_X550EM_x:
2722 case ixgbe_mac_x550em_a:
2724 * set the WDIS bit to not clear the timer bits and cause an
2725 * immediate assertion of the interrupt
2727 itr_reg |= IXGBE_EITR_CNT_WDIS;
2732 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2735 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2739 ixgbe_update_itr(q_vector, &q_vector->tx);
2740 ixgbe_update_itr(q_vector, &q_vector->rx);
2742 /* use the smallest value of new ITR delay calculations */
2743 new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2745 /* Clear latency flag if set, shift into correct position */
2746 new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2749 if (new_itr != q_vector->itr) {
2750 /* save the algorithm value here */
2751 q_vector->itr = new_itr;
2753 ixgbe_write_eitr(q_vector);
2758 * ixgbe_check_overtemp_subtask - check for over temperature
2759 * @adapter: pointer to adapter
2761 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2763 struct ixgbe_hw *hw = &adapter->hw;
2764 u32 eicr = adapter->interrupt_event;
2767 if (test_bit(__IXGBE_DOWN, &adapter->state))
2770 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2773 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2775 switch (hw->device_id) {
2776 case IXGBE_DEV_ID_82599_T3_LOM:
2778 * Since the warning interrupt is for both ports
2779 * we don't have to check if:
2780 * - This interrupt wasn't for our port.
2781 * - We may have missed the interrupt so always have to
2782 * check if we got a LSC
2784 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2785 !(eicr & IXGBE_EICR_LSC))
2788 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2790 bool link_up = false;
2792 hw->mac.ops.check_link(hw, &speed, &link_up, false);
2798 /* Check if this is not due to overtemp */
2799 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2803 case IXGBE_DEV_ID_X550EM_A_1G_T:
2804 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2805 rc = hw->phy.ops.check_overtemp(hw);
2806 if (rc != IXGBE_ERR_OVERTEMP)
2810 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2812 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2816 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2818 adapter->interrupt_event = 0;
2821 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2823 struct ixgbe_hw *hw = &adapter->hw;
2825 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2826 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2827 e_crit(probe, "Fan has stopped, replace the adapter\n");
2828 /* write to clear the interrupt */
2829 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2833 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2835 struct ixgbe_hw *hw = &adapter->hw;
2837 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2840 switch (adapter->hw.mac.type) {
2841 case ixgbe_mac_82599EB:
2843 * Need to check link state so complete overtemp check
2846 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2847 (eicr & IXGBE_EICR_LSC)) &&
2848 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2849 adapter->interrupt_event = eicr;
2850 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2851 ixgbe_service_event_schedule(adapter);
2855 case ixgbe_mac_x550em_a:
2856 if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2857 adapter->interrupt_event = eicr;
2858 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2859 ixgbe_service_event_schedule(adapter);
2860 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2861 IXGBE_EICR_GPI_SDP0_X550EM_a);
2862 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2863 IXGBE_EICR_GPI_SDP0_X550EM_a);
2866 case ixgbe_mac_X550:
2867 case ixgbe_mac_X540:
2868 if (!(eicr & IXGBE_EICR_TS))
2875 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2878 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2880 switch (hw->mac.type) {
2881 case ixgbe_mac_82598EB:
2882 if (hw->phy.type == ixgbe_phy_nl)
2885 case ixgbe_mac_82599EB:
2886 case ixgbe_mac_X550EM_x:
2887 case ixgbe_mac_x550em_a:
2888 switch (hw->mac.ops.get_media_type(hw)) {
2889 case ixgbe_media_type_fiber:
2890 case ixgbe_media_type_fiber_qsfp:
2900 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2902 struct ixgbe_hw *hw = &adapter->hw;
2903 u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2905 if (!ixgbe_is_sfp(hw))
2908 /* Later MAC's use different SDP */
2909 if (hw->mac.type >= ixgbe_mac_X540)
2910 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2912 if (eicr & eicr_mask) {
2913 /* Clear the interrupt */
2914 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2915 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2916 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2917 adapter->sfp_poll_time = 0;
2918 ixgbe_service_event_schedule(adapter);
2922 if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2923 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2924 /* Clear the interrupt */
2925 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2926 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2927 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2928 ixgbe_service_event_schedule(adapter);
2933 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2935 struct ixgbe_hw *hw = &adapter->hw;
2938 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2939 adapter->link_check_timeout = jiffies;
2940 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2941 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2942 IXGBE_WRITE_FLUSH(hw);
2943 ixgbe_service_event_schedule(adapter);
2947 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2951 struct ixgbe_hw *hw = &adapter->hw;
2953 switch (hw->mac.type) {
2954 case ixgbe_mac_82598EB:
2955 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2956 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2958 case ixgbe_mac_82599EB:
2959 case ixgbe_mac_X540:
2960 case ixgbe_mac_X550:
2961 case ixgbe_mac_X550EM_x:
2962 case ixgbe_mac_x550em_a:
2963 mask = (qmask & 0xFFFFFFFF);
2965 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2966 mask = (qmask >> 32);
2968 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2973 /* skip the flush */
2976 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2980 struct ixgbe_hw *hw = &adapter->hw;
2982 switch (hw->mac.type) {
2983 case ixgbe_mac_82598EB:
2984 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2985 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2987 case ixgbe_mac_82599EB:
2988 case ixgbe_mac_X540:
2989 case ixgbe_mac_X550:
2990 case ixgbe_mac_X550EM_x:
2991 case ixgbe_mac_x550em_a:
2992 mask = (qmask & 0xFFFFFFFF);
2994 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2995 mask = (qmask >> 32);
2997 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
3002 /* skip the flush */
3006 * ixgbe_irq_enable - Enable default interrupt generation settings
3007 * @adapter: board private structure
3008 * @queues: enable irqs for queues
3009 * @flush: flush register write
3011 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
3014 struct ixgbe_hw *hw = &adapter->hw;
3015 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
3017 /* don't reenable LSC while waiting for link */
3018 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
3019 mask &= ~IXGBE_EIMS_LSC;
3021 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3022 switch (adapter->hw.mac.type) {
3023 case ixgbe_mac_82599EB:
3024 mask |= IXGBE_EIMS_GPI_SDP0(hw);
3026 case ixgbe_mac_X540:
3027 case ixgbe_mac_X550:
3028 case ixgbe_mac_X550EM_x:
3029 case ixgbe_mac_x550em_a:
3030 mask |= IXGBE_EIMS_TS;
3035 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3036 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3037 switch (adapter->hw.mac.type) {
3038 case ixgbe_mac_82599EB:
3039 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3040 mask |= IXGBE_EIMS_GPI_SDP2(hw);
3042 case ixgbe_mac_X540:
3043 case ixgbe_mac_X550:
3044 case ixgbe_mac_X550EM_x:
3045 case ixgbe_mac_x550em_a:
3046 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3047 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3048 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3049 mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3050 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3051 mask |= IXGBE_EICR_GPI_SDP0_X540;
3052 mask |= IXGBE_EIMS_ECC;
3053 mask |= IXGBE_EIMS_MAILBOX;
3059 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3060 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3061 mask |= IXGBE_EIMS_FLOW_DIR;
3063 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3065 ixgbe_irq_enable_queues(adapter, ~0);
3067 IXGBE_WRITE_FLUSH(&adapter->hw);
3070 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3072 struct ixgbe_adapter *adapter = data;
3073 struct ixgbe_hw *hw = &adapter->hw;
3077 * Workaround for Silicon errata. Use clear-by-write instead
3078 * of clear-by-read. Reading with EICS will return the
3079 * interrupt causes without clearing, which later be done
3080 * with the write to EICR.
3082 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3084 /* The lower 16bits of the EICR register are for the queue interrupts
3085 * which should be masked here in order to not accidentally clear them if
3086 * the bits are high when ixgbe_msix_other is called. There is a race
3087 * condition otherwise which results in possible performance loss
3088 * especially if the ixgbe_msix_other interrupt is triggering
3089 * consistently (as it would when PPS is turned on for the X540 device)
3093 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3095 if (eicr & IXGBE_EICR_LSC)
3096 ixgbe_check_lsc(adapter);
3098 if (eicr & IXGBE_EICR_MAILBOX)
3099 ixgbe_msg_task(adapter);
3101 switch (hw->mac.type) {
3102 case ixgbe_mac_82599EB:
3103 case ixgbe_mac_X540:
3104 case ixgbe_mac_X550:
3105 case ixgbe_mac_X550EM_x:
3106 case ixgbe_mac_x550em_a:
3107 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3108 (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3109 adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3110 ixgbe_service_event_schedule(adapter);
3111 IXGBE_WRITE_REG(hw, IXGBE_EICR,
3112 IXGBE_EICR_GPI_SDP0_X540);
3114 if (eicr & IXGBE_EICR_ECC) {
3115 e_info(link, "Received ECC Err, initiating reset\n");
3116 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3117 ixgbe_service_event_schedule(adapter);
3118 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3120 /* Handle Flow Director Full threshold interrupt */
3121 if (eicr & IXGBE_EICR_FLOW_DIR) {
3122 int reinit_count = 0;
3124 for (i = 0; i < adapter->num_tx_queues; i++) {
3125 struct ixgbe_ring *ring = adapter->tx_ring[i];
3126 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3131 /* no more flow director interrupts until after init */
3132 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3133 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3134 ixgbe_service_event_schedule(adapter);
3137 ixgbe_check_sfp_event(adapter, eicr);
3138 ixgbe_check_overtemp_event(adapter, eicr);
3144 ixgbe_check_fan_failure(adapter, eicr);
3146 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3147 ixgbe_ptp_check_pps_event(adapter);
3149 /* re-enable the original interrupt state, no lsc, no queues */
3150 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3151 ixgbe_irq_enable(adapter, false, false);
3156 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3158 struct ixgbe_q_vector *q_vector = data;
3160 /* EIAM disabled interrupts (on this vector) for us */
3162 if (q_vector->rx.ring || q_vector->tx.ring)
3163 napi_schedule_irqoff(&q_vector->napi);
3169 * ixgbe_poll - NAPI Rx polling callback
3170 * @napi: structure for representing this polling device
3171 * @budget: how many packets driver is allowed to clean
3173 * This function is used for legacy and MSI, NAPI mode
3175 int ixgbe_poll(struct napi_struct *napi, int budget)
3177 struct ixgbe_q_vector *q_vector =
3178 container_of(napi, struct ixgbe_q_vector, napi);
3179 struct ixgbe_adapter *adapter = q_vector->adapter;
3180 struct ixgbe_ring *ring;
3181 int per_ring_budget, work_done = 0;
3182 bool clean_complete = true;
3184 #ifdef CONFIG_IXGBE_DCA
3185 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3186 ixgbe_update_dca(q_vector);
3189 ixgbe_for_each_ring(ring, q_vector->tx) {
3190 bool wd = ring->xsk_umem ?
3191 ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3192 ixgbe_clean_tx_irq(q_vector, ring, budget);
3195 clean_complete = false;
3198 /* Exit if we are called by netpoll */
3202 /* attempt to distribute budget to each queue fairly, but don't allow
3203 * the budget to go below 1 because we'll exit polling */
3204 if (q_vector->rx.count > 1)
3205 per_ring_budget = max(budget/q_vector->rx.count, 1);
3207 per_ring_budget = budget;
3209 ixgbe_for_each_ring(ring, q_vector->rx) {
3210 int cleaned = ring->xsk_umem ?
3211 ixgbe_clean_rx_irq_zc(q_vector, ring,
3213 ixgbe_clean_rx_irq(q_vector, ring,
3216 work_done += cleaned;
3217 if (cleaned >= per_ring_budget)
3218 clean_complete = false;
3221 /* If all work not completed, return budget and keep polling */
3222 if (!clean_complete)
3225 /* all work done, exit the polling mode */
3226 if (likely(napi_complete_done(napi, work_done))) {
3227 if (adapter->rx_itr_setting & 1)
3228 ixgbe_set_itr(q_vector);
3229 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3230 ixgbe_irq_enable_queues(adapter,
3231 BIT_ULL(q_vector->v_idx));
3234 return min(work_done, budget - 1);
3238 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3239 * @adapter: board private structure
3241 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3242 * interrupts from the kernel.
3244 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3246 struct net_device *netdev = adapter->netdev;
3247 unsigned int ri = 0, ti = 0;
3250 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3251 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3252 struct msix_entry *entry = &adapter->msix_entries[vector];
3254 if (q_vector->tx.ring && q_vector->rx.ring) {
3255 snprintf(q_vector->name, sizeof(q_vector->name),
3256 "%s-TxRx-%u", netdev->name, ri++);
3258 } else if (q_vector->rx.ring) {
3259 snprintf(q_vector->name, sizeof(q_vector->name),
3260 "%s-rx-%u", netdev->name, ri++);
3261 } else if (q_vector->tx.ring) {
3262 snprintf(q_vector->name, sizeof(q_vector->name),
3263 "%s-tx-%u", netdev->name, ti++);
3265 /* skip this unused q_vector */
3268 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3269 q_vector->name, q_vector);
3271 e_err(probe, "request_irq failed for MSIX interrupt "
3272 "Error: %d\n", err);
3273 goto free_queue_irqs;
3275 /* If Flow Director is enabled, set interrupt affinity */
3276 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3277 /* assign the mask for this irq */
3278 irq_set_affinity_hint(entry->vector,
3279 &q_vector->affinity_mask);
3283 err = request_irq(adapter->msix_entries[vector].vector,
3284 ixgbe_msix_other, 0, netdev->name, adapter);
3286 e_err(probe, "request_irq for msix_other failed: %d\n", err);
3287 goto free_queue_irqs;
3295 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3297 free_irq(adapter->msix_entries[vector].vector,
3298 adapter->q_vector[vector]);
3300 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3301 pci_disable_msix(adapter->pdev);
3302 kfree(adapter->msix_entries);
3303 adapter->msix_entries = NULL;
3308 * ixgbe_intr - legacy mode Interrupt Handler
3309 * @irq: interrupt number
3310 * @data: pointer to a network interface device structure
3312 static irqreturn_t ixgbe_intr(int irq, void *data)
3314 struct ixgbe_adapter *adapter = data;
3315 struct ixgbe_hw *hw = &adapter->hw;
3316 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3320 * Workaround for silicon errata #26 on 82598. Mask the interrupt
3321 * before the read of EICR.
3323 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3325 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3326 * therefore no explicit interrupt disable is necessary */
3327 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3330 * shared interrupt alert!
3331 * make sure interrupts are enabled because the read will
3332 * have disabled interrupts due to EIAM
3333 * finish the workaround of silicon errata on 82598. Unmask
3334 * the interrupt that we masked before the EICR read.
3336 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3337 ixgbe_irq_enable(adapter, true, true);
3338 return IRQ_NONE; /* Not our interrupt */
3341 if (eicr & IXGBE_EICR_LSC)
3342 ixgbe_check_lsc(adapter);
3344 switch (hw->mac.type) {
3345 case ixgbe_mac_82599EB:
3346 ixgbe_check_sfp_event(adapter, eicr);
3348 case ixgbe_mac_X540:
3349 case ixgbe_mac_X550:
3350 case ixgbe_mac_X550EM_x:
3351 case ixgbe_mac_x550em_a:
3352 if (eicr & IXGBE_EICR_ECC) {
3353 e_info(link, "Received ECC Err, initiating reset\n");
3354 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3355 ixgbe_service_event_schedule(adapter);
3356 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3358 ixgbe_check_overtemp_event(adapter, eicr);
3364 ixgbe_check_fan_failure(adapter, eicr);
3365 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3366 ixgbe_ptp_check_pps_event(adapter);
3368 /* would disable interrupts here but EIAM disabled it */
3369 napi_schedule_irqoff(&q_vector->napi);
3372 * re-enable link(maybe) and non-queue interrupts, no flush.
3373 * ixgbe_poll will re-enable the queue interrupts
3375 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3376 ixgbe_irq_enable(adapter, false, false);
3382 * ixgbe_request_irq - initialize interrupts
3383 * @adapter: board private structure
3385 * Attempts to configure interrupts using the best available
3386 * capabilities of the hardware and kernel.
3388 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3390 struct net_device *netdev = adapter->netdev;
3393 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3394 err = ixgbe_request_msix_irqs(adapter);
3395 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3396 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3397 netdev->name, adapter);
3399 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3400 netdev->name, adapter);
3403 e_err(probe, "request_irq failed, Error %d\n", err);
3408 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3412 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3413 free_irq(adapter->pdev->irq, adapter);
3417 if (!adapter->msix_entries)
3420 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3421 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3422 struct msix_entry *entry = &adapter->msix_entries[vector];
3424 /* free only the irqs that were actually requested */
3425 if (!q_vector->rx.ring && !q_vector->tx.ring)
3428 /* clear the affinity_mask in the IRQ descriptor */
3429 irq_set_affinity_hint(entry->vector, NULL);
3431 free_irq(entry->vector, q_vector);
3434 free_irq(adapter->msix_entries[vector].vector, adapter);
3438 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3439 * @adapter: board private structure
3441 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3443 switch (adapter->hw.mac.type) {
3444 case ixgbe_mac_82598EB:
3445 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3447 case ixgbe_mac_82599EB:
3448 case ixgbe_mac_X540:
3449 case ixgbe_mac_X550:
3450 case ixgbe_mac_X550EM_x:
3451 case ixgbe_mac_x550em_a:
3452 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3453 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3454 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3459 IXGBE_WRITE_FLUSH(&adapter->hw);
3460 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3463 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3464 synchronize_irq(adapter->msix_entries[vector].vector);
3466 synchronize_irq(adapter->msix_entries[vector++].vector);
3468 synchronize_irq(adapter->pdev->irq);
3473 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3474 * @adapter: board private structure
3477 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3479 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3481 ixgbe_write_eitr(q_vector);
3483 ixgbe_set_ivar(adapter, 0, 0, 0);
3484 ixgbe_set_ivar(adapter, 1, 0, 0);
3486 e_info(hw, "Legacy interrupt IVAR setup done\n");
3490 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3491 * @adapter: board private structure
3492 * @ring: structure containing ring specific data
3494 * Configure the Tx descriptor ring after a reset.
3496 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3497 struct ixgbe_ring *ring)
3499 struct ixgbe_hw *hw = &adapter->hw;
3500 u64 tdba = ring->dma;
3502 u32 txdctl = IXGBE_TXDCTL_ENABLE;
3503 u8 reg_idx = ring->reg_idx;
3505 ring->xsk_umem = NULL;
3506 if (ring_is_xdp(ring))
3507 ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
3509 /* disable queue to avoid issues while updating state */
3510 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3511 IXGBE_WRITE_FLUSH(hw);
3513 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3514 (tdba & DMA_BIT_MASK(32)));
3515 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3516 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3517 ring->count * sizeof(union ixgbe_adv_tx_desc));
3518 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3519 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3520 ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3523 * set WTHRESH to encourage burst writeback, it should not be set
3524 * higher than 1 when:
3525 * - ITR is 0 as it could cause false TX hangs
3526 * - ITR is set to > 100k int/sec and BQL is enabled
3528 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3529 * to or less than the number of on chip descriptors, which is
3532 if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3533 txdctl |= 1u << 16; /* WTHRESH = 1 */
3535 txdctl |= 8u << 16; /* WTHRESH = 8 */
3538 * Setting PTHRESH to 32 both improves performance
3539 * and avoids a TX hang with DFP enabled
3541 txdctl |= (1u << 8) | /* HTHRESH = 1 */
3542 32; /* PTHRESH = 32 */
3544 /* reinitialize flowdirector state */
3545 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3546 ring->atr_sample_rate = adapter->atr_sample_rate;
3547 ring->atr_count = 0;
3548 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3550 ring->atr_sample_rate = 0;
3553 /* initialize XPS */
3554 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3555 struct ixgbe_q_vector *q_vector = ring->q_vector;
3558 netif_set_xps_queue(ring->netdev,
3559 &q_vector->affinity_mask,
3563 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3565 /* reinitialize tx_buffer_info */
3566 memset(ring->tx_buffer_info, 0,
3567 sizeof(struct ixgbe_tx_buffer) * ring->count);
3570 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3572 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3573 if (hw->mac.type == ixgbe_mac_82598EB &&
3574 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3577 /* poll to verify queue is enabled */
3579 usleep_range(1000, 2000);
3580 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3581 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3583 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3586 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3588 struct ixgbe_hw *hw = &adapter->hw;
3590 u8 tcs = adapter->hw_tcs;
3592 if (hw->mac.type == ixgbe_mac_82598EB)
3595 /* disable the arbiter while setting MTQC */
3596 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3597 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3598 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3600 /* set transmit pool layout */
3601 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3602 mtqc = IXGBE_MTQC_VT_ENA;
3604 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3606 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3607 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3608 IXGBE_82599_VMDQ_4Q_MASK)
3609 mtqc |= IXGBE_MTQC_32VF;
3611 mtqc |= IXGBE_MTQC_64VF;
3614 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3615 } else if (tcs > 1) {
3616 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3618 u8 max_txq = adapter->num_tx_queues +
3619 adapter->num_xdp_queues;
3621 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3623 mtqc = IXGBE_MTQC_64Q_1PB;
3627 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3629 /* Enable Security TX Buffer IFG for multiple pb */
3631 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3632 sectx |= IXGBE_SECTX_DCB;
3633 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3636 /* re-enable the arbiter */
3637 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3638 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3642 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3643 * @adapter: board private structure
3645 * Configure the Tx unit of the MAC after a reset.
3647 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3649 struct ixgbe_hw *hw = &adapter->hw;
3653 ixgbe_setup_mtqc(adapter);
3655 if (hw->mac.type != ixgbe_mac_82598EB) {
3656 /* DMATXCTL.EN must be before Tx queues are enabled */
3657 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3658 dmatxctl |= IXGBE_DMATXCTL_TE;
3659 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3662 /* Setup the HW Tx Head and Tail descriptor pointers */
3663 for (i = 0; i < adapter->num_tx_queues; i++)
3664 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3665 for (i = 0; i < adapter->num_xdp_queues; i++)
3666 ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3669 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3670 struct ixgbe_ring *ring)
3672 struct ixgbe_hw *hw = &adapter->hw;
3673 u8 reg_idx = ring->reg_idx;
3674 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3676 srrctl |= IXGBE_SRRCTL_DROP_EN;
3678 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3681 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3682 struct ixgbe_ring *ring)
3684 struct ixgbe_hw *hw = &adapter->hw;
3685 u8 reg_idx = ring->reg_idx;
3686 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3688 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3690 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3693 #ifdef CONFIG_IXGBE_DCB
3694 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3696 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3700 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3702 if (adapter->ixgbe_ieee_pfc)
3703 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3706 * We should set the drop enable bit if:
3709 * Number of Rx queues > 1 and flow control is disabled
3711 * This allows us to avoid head of line blocking for security
3712 * and performance reasons.
3714 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3715 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3716 for (i = 0; i < adapter->num_rx_queues; i++)
3717 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3719 for (i = 0; i < adapter->num_rx_queues; i++)
3720 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3724 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3726 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3727 struct ixgbe_ring *rx_ring)
3729 struct ixgbe_hw *hw = &adapter->hw;
3731 u8 reg_idx = rx_ring->reg_idx;
3733 if (hw->mac.type == ixgbe_mac_82598EB) {
3734 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3737 * if VMDq is not active we must program one srrctl register
3738 * per RSS queue since we have enabled RDRXCTL.MVMEN
3743 /* configure header buffer length, needed for RSC */
3744 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3746 /* configure the packet buffer length */
3747 if (rx_ring->xsk_umem) {
3748 u32 xsk_buf_len = rx_ring->xsk_umem->chunk_size_nohr -
3749 XDP_PACKET_HEADROOM;
3751 /* If the MAC support setting RXDCTL.RLPML, the
3752 * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3753 * RXDCTL.RLPML is set to the actual UMEM buffer
3754 * size. If not, then we are stuck with a 1k buffer
3755 * size resolution. In this case frames larger than
3756 * the UMEM buffer size viewed in a 1k resolution will
3759 if (hw->mac.type != ixgbe_mac_82599EB)
3760 srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3762 srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3763 } else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3764 srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3766 srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3769 /* configure descriptor type */
3770 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3772 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3776 * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3777 * @adapter: device handle
3779 * - 82598/82599/X540: 128
3780 * - X550(non-SRIOV mode): 512
3781 * - X550(SRIOV mode): 64
3783 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3785 if (adapter->hw.mac.type < ixgbe_mac_X550)
3787 else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3794 * ixgbe_store_key - Write the RSS key to HW
3795 * @adapter: device handle
3797 * Write the RSS key stored in adapter.rss_key to HW.
3799 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3801 struct ixgbe_hw *hw = &adapter->hw;
3804 for (i = 0; i < 10; i++)
3805 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3809 * ixgbe_init_rss_key - Initialize adapter RSS key
3810 * @adapter: device handle
3812 * Allocates and initializes the RSS key if it is not allocated.
3814 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3818 if (!adapter->rss_key) {
3819 rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3820 if (unlikely(!rss_key))
3823 netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3824 adapter->rss_key = rss_key;
3831 * ixgbe_store_reta - Write the RETA table to HW
3832 * @adapter: device handle
3834 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3836 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3838 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3839 struct ixgbe_hw *hw = &adapter->hw;
3842 u8 *indir_tbl = adapter->rss_indir_tbl;
3844 /* Fill out the redirection table as follows:
3845 * - 82598: 8 bit wide entries containing pair of 4 bit RSS
3847 * - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3848 * - X550: 8 bit wide entries containing 6 bit RSS index
3850 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3851 indices_multi = 0x11;
3853 indices_multi = 0x1;
3855 /* Write redirection table to HW */
3856 for (i = 0; i < reta_entries; i++) {
3857 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3860 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3862 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3870 * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3871 * @adapter: device handle
3873 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3875 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3877 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3878 struct ixgbe_hw *hw = &adapter->hw;
3881 /* Write redirection table to HW */
3882 for (i = 0; i < reta_entries; i++) {
3883 u16 pool = adapter->num_rx_pools;
3885 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3891 IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3897 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3900 u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3901 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3903 /* Program table for at least 4 queues w/ SR-IOV so that VFs can
3904 * make full use of any rings they may have. We will use the
3905 * PSRTYPE register to control how many rings we use within the PF.
3907 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3910 /* Fill out hash function seeds */
3911 ixgbe_store_key(adapter);
3913 /* Fill out redirection table */
3914 memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3916 for (i = 0, j = 0; i < reta_entries; i++, j++) {
3920 adapter->rss_indir_tbl[i] = j;
3923 ixgbe_store_reta(adapter);
3926 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3928 struct ixgbe_hw *hw = &adapter->hw;
3929 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3932 /* Fill out hash function seeds */
3933 for (i = 0; i < 10; i++) {
3934 u16 pool = adapter->num_rx_pools;
3938 IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3939 *(adapter->rss_key + i));
3942 /* Fill out the redirection table */
3943 for (i = 0, j = 0; i < 64; i++, j++) {
3947 adapter->rss_indir_tbl[i] = j;
3950 ixgbe_store_vfreta(adapter);
3953 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3955 struct ixgbe_hw *hw = &adapter->hw;
3956 u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3959 /* Disable indicating checksum in descriptor, enables RSS hash */
3960 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3961 rxcsum |= IXGBE_RXCSUM_PCSD;
3962 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3964 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3965 if (adapter->ring_feature[RING_F_RSS].mask)
3966 mrqc = IXGBE_MRQC_RSSEN;
3968 u8 tcs = adapter->hw_tcs;
3970 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3972 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3974 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3975 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3976 IXGBE_82599_VMDQ_4Q_MASK)
3977 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3979 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3981 /* Enable L3/L4 for Tx Switched packets only for X550,
3982 * older devices do not support this feature
3984 if (hw->mac.type >= ixgbe_mac_X550)
3985 mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3988 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3990 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3992 mrqc = IXGBE_MRQC_RSSEN;
3996 /* Perform hash on these packet types */
3997 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3998 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3999 IXGBE_MRQC_RSS_FIELD_IPV6 |
4000 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
4002 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
4003 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
4004 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
4005 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
4007 if ((hw->mac.type >= ixgbe_mac_X550) &&
4008 (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
4009 u16 pool = adapter->num_rx_pools;
4011 /* Enable VF RSS mode */
4012 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
4013 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4015 /* Setup RSS through the VF registers */
4016 ixgbe_setup_vfreta(adapter);
4017 vfmrqc = IXGBE_MRQC_RSSEN;
4018 vfmrqc |= rss_field;
4022 IXGBE_PFVFMRQC(VMDQ_P(pool)),
4025 ixgbe_setup_reta(adapter);
4027 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4032 * ixgbe_configure_rscctl - enable RSC for the indicated ring
4033 * @adapter: address of board private structure
4034 * @ring: structure containing ring specific data
4036 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4037 struct ixgbe_ring *ring)
4039 struct ixgbe_hw *hw = &adapter->hw;
4041 u8 reg_idx = ring->reg_idx;
4043 if (!ring_is_rsc_enabled(ring))
4046 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4047 rscctrl |= IXGBE_RSCCTL_RSCEN;
4049 * we must limit the number of descriptors so that the
4050 * total size of max desc * buf_len is not greater
4053 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4054 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4057 #define IXGBE_MAX_RX_DESC_POLL 10
4058 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4059 struct ixgbe_ring *ring)
4061 struct ixgbe_hw *hw = &adapter->hw;
4062 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4064 u8 reg_idx = ring->reg_idx;
4066 if (ixgbe_removed(hw->hw_addr))
4068 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4069 if (hw->mac.type == ixgbe_mac_82598EB &&
4070 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4074 usleep_range(1000, 2000);
4075 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4076 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4079 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4080 "the polling period\n", reg_idx);
4084 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4085 struct ixgbe_ring *ring)
4087 struct ixgbe_hw *hw = &adapter->hw;
4088 union ixgbe_adv_rx_desc *rx_desc;
4089 u64 rdba = ring->dma;
4091 u8 reg_idx = ring->reg_idx;
4093 xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4094 ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
4095 if (ring->xsk_umem) {
4096 ring->zca.free = ixgbe_zca_free;
4097 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4102 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4103 MEM_TYPE_PAGE_SHARED, NULL));
4106 /* disable queue to avoid use of these values while updating state */
4107 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4108 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4110 /* write value back with RXDCTL.ENABLE bit cleared */
4111 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4112 IXGBE_WRITE_FLUSH(hw);
4114 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4115 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4116 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4117 ring->count * sizeof(union ixgbe_adv_rx_desc));
4118 /* Force flushing of IXGBE_RDLEN to prevent MDD */
4119 IXGBE_WRITE_FLUSH(hw);
4121 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4122 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4123 ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4125 ixgbe_configure_srrctl(adapter, ring);
4126 ixgbe_configure_rscctl(adapter, ring);
4128 if (hw->mac.type == ixgbe_mac_82598EB) {
4130 * enable cache line friendly hardware writes:
4131 * PTHRESH=32 descriptors (half the internal cache),
4132 * this also removes ugly rx_no_buffer_count increment
4133 * HTHRESH=4 descriptors (to minimize latency on fetch)
4134 * WTHRESH=8 burst writeback up to two cache lines
4136 rxdctl &= ~0x3FFFFF;
4138 #if (PAGE_SIZE < 8192)
4139 /* RXDCTL.RLPML does not work on 82599 */
4140 } else if (hw->mac.type != ixgbe_mac_82599EB) {
4141 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4142 IXGBE_RXDCTL_RLPML_EN);
4144 /* Limit the maximum frame size so we don't overrun the skb.
4145 * This can happen in SRIOV mode when the MTU of the VF is
4146 * higher than the MTU of the PF.
4148 if (ring_uses_build_skb(ring) &&
4149 !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4150 rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4151 IXGBE_RXDCTL_RLPML_EN;
4155 if (ring->xsk_umem && hw->mac.type != ixgbe_mac_82599EB) {
4156 u32 xsk_buf_len = ring->xsk_umem->chunk_size_nohr -
4157 XDP_PACKET_HEADROOM;
4159 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4160 IXGBE_RXDCTL_RLPML_EN);
4161 rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4163 ring->rx_buf_len = xsk_buf_len;
4166 /* initialize rx_buffer_info */
4167 memset(ring->rx_buffer_info, 0,
4168 sizeof(struct ixgbe_rx_buffer) * ring->count);
4170 /* initialize Rx descriptor 0 */
4171 rx_desc = IXGBE_RX_DESC(ring, 0);
4172 rx_desc->wb.upper.length = 0;
4174 /* enable receive descriptor ring */
4175 rxdctl |= IXGBE_RXDCTL_ENABLE;
4176 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4178 ixgbe_rx_desc_queue_enable(adapter, ring);
4180 ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4182 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4185 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4187 struct ixgbe_hw *hw = &adapter->hw;
4188 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4189 u16 pool = adapter->num_rx_pools;
4191 /* PSRTYPE must be initialized in non 82598 adapters */
4192 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4193 IXGBE_PSRTYPE_UDPHDR |
4194 IXGBE_PSRTYPE_IPV4HDR |
4195 IXGBE_PSRTYPE_L2HDR |
4196 IXGBE_PSRTYPE_IPV6HDR;
4198 if (hw->mac.type == ixgbe_mac_82598EB)
4202 psrtype |= 2u << 29;
4204 psrtype |= 1u << 29;
4207 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4210 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4212 struct ixgbe_hw *hw = &adapter->hw;
4213 u16 pool = adapter->num_rx_pools;
4214 u32 reg_offset, vf_shift, vmolr;
4215 u32 gcr_ext, vmdctl;
4218 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4221 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4222 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4223 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4224 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4225 vmdctl |= IXGBE_VT_CTL_REPLEN;
4226 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4228 /* accept untagged packets until a vlan tag is
4229 * specifically set for the VMDQ queue/pool
4231 vmolr = IXGBE_VMOLR_AUPE;
4233 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4235 vf_shift = VMDQ_P(0) % 32;
4236 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4238 /* Enable only the PF's pool for Tx/Rx */
4239 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4240 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4241 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4242 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4243 if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4244 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4246 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4247 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4249 /* clear VLAN promisc flag so VFTA will be updated if necessary */
4250 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4253 * Set up VF register offsets for selected VT Mode,
4254 * i.e. 32 or 64 VFs for SR-IOV
4256 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4257 case IXGBE_82599_VMDQ_8Q_MASK:
4258 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4260 case IXGBE_82599_VMDQ_4Q_MASK:
4261 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4264 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4268 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4270 for (i = 0; i < adapter->num_vfs; i++) {
4271 /* configure spoof checking */
4272 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4273 adapter->vfinfo[i].spoofchk_enabled);
4275 /* Enable/Disable RSS query feature */
4276 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4277 adapter->vfinfo[i].rss_query_enabled);
4281 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4283 struct ixgbe_hw *hw = &adapter->hw;
4284 struct net_device *netdev = adapter->netdev;
4285 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4286 struct ixgbe_ring *rx_ring;
4291 /* adjust max frame to be able to do baby jumbo for FCoE */
4292 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4293 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4294 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4296 #endif /* IXGBE_FCOE */
4298 /* adjust max frame to be at least the size of a standard frame */
4299 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4300 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4302 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4303 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4304 mhadd &= ~IXGBE_MHADD_MFS_MASK;
4305 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4307 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4310 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4311 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4312 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4313 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4316 * Setup the HW Rx Head and Tail Descriptor Pointers and
4317 * the Base and Length of the Rx Descriptor Ring
4319 for (i = 0; i < adapter->num_rx_queues; i++) {
4320 rx_ring = adapter->rx_ring[i];
4322 clear_ring_rsc_enabled(rx_ring);
4323 clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4324 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4326 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4327 set_ring_rsc_enabled(rx_ring);
4329 if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4330 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4332 if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4335 set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4337 #if (PAGE_SIZE < 8192)
4338 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4339 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4341 if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4342 (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4343 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4348 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4350 struct ixgbe_hw *hw = &adapter->hw;
4351 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4353 switch (hw->mac.type) {
4354 case ixgbe_mac_82598EB:
4356 * For VMDq support of different descriptor types or
4357 * buffer sizes through the use of multiple SRRCTL
4358 * registers, RDRXCTL.MVMEN must be set to 1
4360 * also, the manual doesn't mention it clearly but DCA hints
4361 * will only use queue 0's tags unless this bit is set. Side
4362 * effects of setting this bit are only that SRRCTL must be
4363 * fully programmed [0..15]
4365 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4367 case ixgbe_mac_X550:
4368 case ixgbe_mac_X550EM_x:
4369 case ixgbe_mac_x550em_a:
4370 if (adapter->num_vfs)
4371 rdrxctl |= IXGBE_RDRXCTL_PSP;
4373 case ixgbe_mac_82599EB:
4374 case ixgbe_mac_X540:
4375 /* Disable RSC for ACK packets */
4376 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4377 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4378 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4379 /* hardware requires some bits to be set by default */
4380 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4381 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4384 /* We should do nothing since we don't know this hardware */
4388 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4392 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4393 * @adapter: board private structure
4395 * Configure the Rx unit of the MAC after a reset.
4397 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4399 struct ixgbe_hw *hw = &adapter->hw;
4403 /* disable receives while setting up the descriptors */
4404 hw->mac.ops.disable_rx(hw);
4406 ixgbe_setup_psrtype(adapter);
4407 ixgbe_setup_rdrxctl(adapter);
4410 rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4411 rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4412 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4413 rfctl |= IXGBE_RFCTL_RSC_DIS;
4415 /* disable NFS filtering */
4416 rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4417 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4419 /* Program registers for the distribution of queues */
4420 ixgbe_setup_mrqc(adapter);
4422 /* set_rx_buffer_len must be called before ring initialization */
4423 ixgbe_set_rx_buffer_len(adapter);
4426 * Setup the HW Rx Head and Tail Descriptor Pointers and
4427 * the Base and Length of the Rx Descriptor Ring
4429 for (i = 0; i < adapter->num_rx_queues; i++)
4430 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4432 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4433 /* disable drop enable for 82598 parts */
4434 if (hw->mac.type == ixgbe_mac_82598EB)
4435 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4437 /* enable all receives */
4438 rxctrl |= IXGBE_RXCTRL_RXEN;
4439 hw->mac.ops.enable_rx_dma(hw, rxctrl);
4442 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4443 __be16 proto, u16 vid)
4445 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4446 struct ixgbe_hw *hw = &adapter->hw;
4448 /* add VID to filter table */
4449 if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4450 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4452 set_bit(vid, adapter->active_vlans);
4457 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4462 /* short cut the special case */
4466 /* Search for the vlan id in the VLVF entries */
4467 for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4468 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4469 if ((vlvf & VLAN_VID_MASK) == vlan)
4476 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4478 struct ixgbe_hw *hw = &adapter->hw;
4482 idx = ixgbe_find_vlvf_entry(hw, vid);
4486 /* See if any other pools are set for this VLAN filter
4487 * entry other than the PF.
4489 word = idx * 2 + (VMDQ_P(0) / 32);
4490 bits = ~BIT(VMDQ_P(0) % 32);
4491 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4493 /* Disable the filter so this falls into the default pool. */
4494 if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4495 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4496 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4497 IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4501 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4502 __be16 proto, u16 vid)
4504 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4505 struct ixgbe_hw *hw = &adapter->hw;
4507 /* remove VID from filter table */
4508 if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4509 hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4511 clear_bit(vid, adapter->active_vlans);
4517 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4518 * @adapter: driver data
4520 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4522 struct ixgbe_hw *hw = &adapter->hw;
4526 switch (hw->mac.type) {
4527 case ixgbe_mac_82598EB:
4528 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4529 vlnctrl &= ~IXGBE_VLNCTRL_VME;
4530 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4532 case ixgbe_mac_82599EB:
4533 case ixgbe_mac_X540:
4534 case ixgbe_mac_X550:
4535 case ixgbe_mac_X550EM_x:
4536 case ixgbe_mac_x550em_a:
4537 for (i = 0; i < adapter->num_rx_queues; i++) {
4538 struct ixgbe_ring *ring = adapter->rx_ring[i];
4540 if (!netif_is_ixgbe(ring->netdev))
4544 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4545 vlnctrl &= ~IXGBE_RXDCTL_VME;
4546 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4555 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4556 * @adapter: driver data
4558 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4560 struct ixgbe_hw *hw = &adapter->hw;
4564 switch (hw->mac.type) {
4565 case ixgbe_mac_82598EB:
4566 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4567 vlnctrl |= IXGBE_VLNCTRL_VME;
4568 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4570 case ixgbe_mac_82599EB:
4571 case ixgbe_mac_X540:
4572 case ixgbe_mac_X550:
4573 case ixgbe_mac_X550EM_x:
4574 case ixgbe_mac_x550em_a:
4575 for (i = 0; i < adapter->num_rx_queues; i++) {
4576 struct ixgbe_ring *ring = adapter->rx_ring[i];
4578 if (!netif_is_ixgbe(ring->netdev))
4582 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4583 vlnctrl |= IXGBE_RXDCTL_VME;
4584 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4592 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4594 struct ixgbe_hw *hw = &adapter->hw;
4597 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4599 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4600 /* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4601 vlnctrl |= IXGBE_VLNCTRL_VFE;
4602 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4604 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4605 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4609 /* Nothing to do for 82598 */
4610 if (hw->mac.type == ixgbe_mac_82598EB)
4613 /* We are already in VLAN promisc, nothing to do */
4614 if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4617 /* Set flag so we don't redo unnecessary work */
4618 adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4620 /* Add PF to all active pools */
4621 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4622 u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4623 u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4625 vlvfb |= BIT(VMDQ_P(0) % 32);
4626 IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4629 /* Set all bits in the VLAN filter table array */
4630 for (i = hw->mac.vft_size; i--;)
4631 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4634 #define VFTA_BLOCK_SIZE 8
4635 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4637 struct ixgbe_hw *hw = &adapter->hw;
4638 u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4639 u32 vid_start = vfta_offset * 32;
4640 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4641 u32 i, vid, word, bits;
4643 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4644 u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4646 /* pull VLAN ID from VLVF */
4647 vid = vlvf & VLAN_VID_MASK;
4649 /* only concern outselves with a certain range */
4650 if (vid < vid_start || vid >= vid_end)
4654 /* record VLAN ID in VFTA */
4655 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4657 /* if PF is part of this then continue */
4658 if (test_bit(vid, adapter->active_vlans))
4662 /* remove PF from the pool */
4663 word = i * 2 + VMDQ_P(0) / 32;
4664 bits = ~BIT(VMDQ_P(0) % 32);
4665 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4666 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4669 /* extract values from active_vlans and write back to VFTA */
4670 for (i = VFTA_BLOCK_SIZE; i--;) {
4671 vid = (vfta_offset + i) * 32;
4672 word = vid / BITS_PER_LONG;
4673 bits = vid % BITS_PER_LONG;
4675 vfta[i] |= adapter->active_vlans[word] >> bits;
4677 IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4681 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4683 struct ixgbe_hw *hw = &adapter->hw;
4686 /* Set VLAN filtering to enabled */
4687 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4688 vlnctrl |= IXGBE_VLNCTRL_VFE;
4689 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4691 if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4692 hw->mac.type == ixgbe_mac_82598EB)
4695 /* We are not in VLAN promisc, nothing to do */
4696 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4699 /* Set flag so we don't redo unnecessary work */
4700 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4702 for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4703 ixgbe_scrub_vfta(adapter, i);
4706 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4710 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4712 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4713 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4717 * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4718 * @netdev: network interface device structure
4720 * Writes multicast address list to the MTA hash table.
4721 * Returns: -ENOMEM on failure
4722 * 0 on no addresses written
4723 * X on writing X addresses to MTA
4725 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4727 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4728 struct ixgbe_hw *hw = &adapter->hw;
4730 if (!netif_running(netdev))
4733 if (hw->mac.ops.update_mc_addr_list)
4734 hw->mac.ops.update_mc_addr_list(hw, netdev);
4738 #ifdef CONFIG_PCI_IOV
4739 ixgbe_restore_vf_multicasts(adapter);
4742 return netdev_mc_count(netdev);
4745 #ifdef CONFIG_PCI_IOV
4746 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4748 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4749 struct ixgbe_hw *hw = &adapter->hw;
4752 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4753 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4755 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4756 hw->mac.ops.set_rar(hw, i,
4761 hw->mac.ops.clear_rar(hw, i);
4766 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4768 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4769 struct ixgbe_hw *hw = &adapter->hw;
4772 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4773 if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4776 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4778 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4779 hw->mac.ops.set_rar(hw, i,
4784 hw->mac.ops.clear_rar(hw, i);
4788 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4790 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4791 struct ixgbe_hw *hw = &adapter->hw;
4794 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4795 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4796 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4799 ixgbe_sync_mac_table(adapter);
4802 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4804 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4805 struct ixgbe_hw *hw = &adapter->hw;
4808 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4809 /* do not count default RAR as available */
4810 if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4813 /* only count unused and addresses that belong to us */
4814 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4815 if (mac_table->pool != pool)
4825 /* this function destroys the first RAR entry */
4826 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4828 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4829 struct ixgbe_hw *hw = &adapter->hw;
4831 memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4832 mac_table->pool = VMDQ_P(0);
4834 mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4836 hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4840 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4841 const u8 *addr, u16 pool)
4843 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4844 struct ixgbe_hw *hw = &adapter->hw;
4847 if (is_zero_ether_addr(addr))
4850 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4851 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4854 ether_addr_copy(mac_table->addr, addr);
4855 mac_table->pool = pool;
4857 mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4858 IXGBE_MAC_STATE_IN_USE;
4860 ixgbe_sync_mac_table(adapter);
4868 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4869 const u8 *addr, u16 pool)
4871 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4872 struct ixgbe_hw *hw = &adapter->hw;
4875 if (is_zero_ether_addr(addr))
4878 /* search table for addr, if found clear IN_USE flag and sync */
4879 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4880 /* we can only delete an entry if it is in use */
4881 if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4883 /* we only care about entries that belong to the given pool */
4884 if (mac_table->pool != pool)
4886 /* we only care about a specific MAC address */
4887 if (!ether_addr_equal(addr, mac_table->addr))
4890 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4891 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4893 ixgbe_sync_mac_table(adapter);
4901 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4903 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4906 ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4908 return min_t(int, ret, 0);
4911 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4913 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4915 ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4921 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4922 * @netdev: network interface device structure
4924 * The set_rx_method entry point is called whenever the unicast/multicast
4925 * address list or the network interface flags are updated. This routine is
4926 * responsible for configuring the hardware for proper unicast, multicast and
4929 void ixgbe_set_rx_mode(struct net_device *netdev)
4931 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4932 struct ixgbe_hw *hw = &adapter->hw;
4933 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4934 netdev_features_t features = netdev->features;
4937 /* Check for Promiscuous and All Multicast modes */
4938 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4940 /* set all bits that we expect to always be set */
4941 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4942 fctrl |= IXGBE_FCTRL_BAM;
4943 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4944 fctrl |= IXGBE_FCTRL_PMCF;
4946 /* clear the bits we are changing the status of */
4947 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4948 if (netdev->flags & IFF_PROMISC) {
4949 hw->addr_ctrl.user_set_promisc = true;
4950 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4951 vmolr |= IXGBE_VMOLR_MPE;
4952 features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4954 if (netdev->flags & IFF_ALLMULTI) {
4955 fctrl |= IXGBE_FCTRL_MPE;
4956 vmolr |= IXGBE_VMOLR_MPE;
4958 hw->addr_ctrl.user_set_promisc = false;
4962 * Write addresses to available RAR registers, if there is not
4963 * sufficient space to store all the addresses then enable
4964 * unicast promiscuous mode
4966 if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4967 fctrl |= IXGBE_FCTRL_UPE;
4968 vmolr |= IXGBE_VMOLR_ROPE;
4971 /* Write addresses to the MTA, if the attempt fails
4972 * then we should just turn on promiscuous mode so
4973 * that we can at least receive multicast traffic
4975 count = ixgbe_write_mc_addr_list(netdev);
4977 fctrl |= IXGBE_FCTRL_MPE;
4978 vmolr |= IXGBE_VMOLR_MPE;
4980 vmolr |= IXGBE_VMOLR_ROMPE;
4983 if (hw->mac.type != ixgbe_mac_82598EB) {
4984 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4985 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4987 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4990 /* This is useful for sniffing bad packets. */
4991 if (features & NETIF_F_RXALL) {
4992 /* UPE and MPE will be handled by normal PROMISC logic
4993 * in e1000e_set_rx_mode */
4994 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4995 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4996 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4998 fctrl &= ~(IXGBE_FCTRL_DPF);
4999 /* NOTE: VLAN filtering is disabled by setting PROMISC */
5002 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5004 if (features & NETIF_F_HW_VLAN_CTAG_RX)
5005 ixgbe_vlan_strip_enable(adapter);
5007 ixgbe_vlan_strip_disable(adapter);
5009 if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
5010 ixgbe_vlan_promisc_disable(adapter);
5012 ixgbe_vlan_promisc_enable(adapter);
5015 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
5019 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
5020 napi_enable(&adapter->q_vector[q_idx]->napi);
5023 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
5027 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
5028 napi_disable(&adapter->q_vector[q_idx]->napi);
5031 static void ixgbe_clear_udp_tunnel_port(struct ixgbe_adapter *adapter, u32 mask)
5033 struct ixgbe_hw *hw = &adapter->hw;
5036 if (!(adapter->flags & (IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE |
5037 IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)))
5040 vxlanctrl = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) & ~mask;
5041 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, vxlanctrl);
5043 if (mask & IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK)
5044 adapter->vxlan_port = 0;
5046 if (mask & IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK)
5047 adapter->geneve_port = 0;
5050 #ifdef CONFIG_IXGBE_DCB
5052 * ixgbe_configure_dcb - Configure DCB hardware
5053 * @adapter: ixgbe adapter struct
5055 * This is called by the driver on open to configure the DCB hardware.
5056 * This is also called by the gennetlink interface when reconfiguring
5059 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5061 struct ixgbe_hw *hw = &adapter->hw;
5062 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5064 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5065 if (hw->mac.type == ixgbe_mac_82598EB)
5066 netif_set_gso_max_size(adapter->netdev, 65536);
5070 if (hw->mac.type == ixgbe_mac_82598EB)
5071 netif_set_gso_max_size(adapter->netdev, 32768);
5074 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5075 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5078 /* reconfigure the hardware */
5079 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5080 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5082 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5084 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5085 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5086 ixgbe_dcb_hw_ets(&adapter->hw,
5087 adapter->ixgbe_ieee_ets,
5089 ixgbe_dcb_hw_pfc_config(&adapter->hw,
5090 adapter->ixgbe_ieee_pfc->pfc_en,
5091 adapter->ixgbe_ieee_ets->prio_tc);
5094 /* Enable RSS Hash per TC */
5095 if (hw->mac.type != ixgbe_mac_82598EB) {
5097 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5104 /* write msb to all 8 TCs in one write */
5105 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5110 /* Additional bittime to account for IXGBE framing */
5111 #define IXGBE_ETH_FRAMING 20
5114 * ixgbe_hpbthresh - calculate high water mark for flow control
5116 * @adapter: board private structure to calculate for
5117 * @pb: packet buffer to calculate
5119 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5121 struct ixgbe_hw *hw = &adapter->hw;
5122 struct net_device *dev = adapter->netdev;
5123 int link, tc, kb, marker;
5126 /* Calculate max LAN frame size */
5127 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5130 /* FCoE traffic class uses FCOE jumbo frames */
5131 if ((dev->features & NETIF_F_FCOE_MTU) &&
5132 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5133 (pb == ixgbe_fcoe_get_tc(adapter)))
5134 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5137 /* Calculate delay value for device */
5138 switch (hw->mac.type) {
5139 case ixgbe_mac_X540:
5140 case ixgbe_mac_X550:
5141 case ixgbe_mac_X550EM_x:
5142 case ixgbe_mac_x550em_a:
5143 dv_id = IXGBE_DV_X540(link, tc);
5146 dv_id = IXGBE_DV(link, tc);
5150 /* Loopback switch introduces additional latency */
5151 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5152 dv_id += IXGBE_B2BT(tc);
5154 /* Delay value is calculated in bit times convert to KB */
5155 kb = IXGBE_BT2KB(dv_id);
5156 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5158 marker = rx_pba - kb;
5160 /* It is possible that the packet buffer is not large enough
5161 * to provide required headroom. In this case throw an error
5162 * to user and a do the best we can.
5165 e_warn(drv, "Packet Buffer(%i) can not provide enough"
5166 "headroom to support flow control."
5167 "Decrease MTU or number of traffic classes\n", pb);
5175 * ixgbe_lpbthresh - calculate low water mark for for flow control
5177 * @adapter: board private structure to calculate for
5178 * @pb: packet buffer to calculate
5180 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5182 struct ixgbe_hw *hw = &adapter->hw;
5183 struct net_device *dev = adapter->netdev;
5187 /* Calculate max LAN frame size */
5188 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5191 /* FCoE traffic class uses FCOE jumbo frames */
5192 if ((dev->features & NETIF_F_FCOE_MTU) &&
5193 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5194 (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5195 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5198 /* Calculate delay value for device */
5199 switch (hw->mac.type) {
5200 case ixgbe_mac_X540:
5201 case ixgbe_mac_X550:
5202 case ixgbe_mac_X550EM_x:
5203 case ixgbe_mac_x550em_a:
5204 dv_id = IXGBE_LOW_DV_X540(tc);
5207 dv_id = IXGBE_LOW_DV(tc);
5211 /* Delay value is calculated in bit times convert to KB */
5212 return IXGBE_BT2KB(dv_id);
5216 * ixgbe_pbthresh_setup - calculate and setup high low water marks
5218 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5220 struct ixgbe_hw *hw = &adapter->hw;
5221 int num_tc = adapter->hw_tcs;
5227 for (i = 0; i < num_tc; i++) {
5228 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5229 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5231 /* Low water marks must not be larger than high water marks */
5232 if (hw->fc.low_water[i] > hw->fc.high_water[i])
5233 hw->fc.low_water[i] = 0;
5236 for (; i < MAX_TRAFFIC_CLASS; i++)
5237 hw->fc.high_water[i] = 0;
5240 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5242 struct ixgbe_hw *hw = &adapter->hw;
5244 u8 tc = adapter->hw_tcs;
5246 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5247 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5248 hdrm = 32 << adapter->fdir_pballoc;
5252 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5253 ixgbe_pbthresh_setup(adapter);
5256 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5258 struct ixgbe_hw *hw = &adapter->hw;
5259 struct hlist_node *node2;
5260 struct ixgbe_fdir_filter *filter;
5263 spin_lock(&adapter->fdir_perfect_lock);
5265 if (!hlist_empty(&adapter->fdir_filter_list))
5266 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5268 hlist_for_each_entry_safe(filter, node2,
5269 &adapter->fdir_filter_list, fdir_node) {
5270 if (filter->action == IXGBE_FDIR_DROP_QUEUE) {
5271 queue = IXGBE_FDIR_DROP_QUEUE;
5273 u32 ring = ethtool_get_flow_spec_ring(filter->action);
5274 u8 vf = ethtool_get_flow_spec_ring_vf(filter->action);
5276 if (!vf && (ring >= adapter->num_rx_queues)) {
5277 e_err(drv, "FDIR restore failed without VF, ring: %u\n",
5281 ((vf > adapter->num_vfs) ||
5282 ring >= adapter->num_rx_queues_per_pool)) {
5283 e_err(drv, "FDIR restore failed with VF, vf: %hhu, ring: %u\n",
5288 /* Map the ring onto the absolute queue index */
5290 queue = adapter->rx_ring[ring]->reg_idx;
5293 adapter->num_rx_queues_per_pool) + ring;
5296 ixgbe_fdir_write_perfect_filter_82599(hw,
5297 &filter->filter, filter->sw_idx, queue);
5300 spin_unlock(&adapter->fdir_perfect_lock);
5304 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5305 * @rx_ring: ring to free buffers from
5307 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5309 u16 i = rx_ring->next_to_clean;
5310 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5312 if (rx_ring->xsk_umem) {
5313 ixgbe_xsk_clean_rx_ring(rx_ring);
5317 /* Free all the Rx ring sk_buffs */
5318 while (i != rx_ring->next_to_alloc) {
5319 if (rx_buffer->skb) {
5320 struct sk_buff *skb = rx_buffer->skb;
5321 if (IXGBE_CB(skb)->page_released)
5322 dma_unmap_page_attrs(rx_ring->dev,
5324 ixgbe_rx_pg_size(rx_ring),
5330 /* Invalidate cache lines that may have been written to by
5331 * device so that we avoid corrupting memory.
5333 dma_sync_single_range_for_cpu(rx_ring->dev,
5335 rx_buffer->page_offset,
5336 ixgbe_rx_bufsz(rx_ring),
5339 /* free resources associated with mapping */
5340 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5341 ixgbe_rx_pg_size(rx_ring),
5344 __page_frag_cache_drain(rx_buffer->page,
5345 rx_buffer->pagecnt_bias);
5349 if (i == rx_ring->count) {
5351 rx_buffer = rx_ring->rx_buffer_info;
5356 rx_ring->next_to_alloc = 0;
5357 rx_ring->next_to_clean = 0;
5358 rx_ring->next_to_use = 0;
5361 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5362 struct ixgbe_fwd_adapter *accel)
5364 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5365 int num_tc = netdev_get_num_tc(adapter->netdev);
5366 struct net_device *vdev = accel->netdev;
5369 baseq = accel->pool * adapter->num_rx_queues_per_pool;
5370 netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5371 accel->pool, adapter->num_rx_pools,
5372 baseq, baseq + adapter->num_rx_queues_per_pool);
5374 accel->rx_base_queue = baseq;
5375 accel->tx_base_queue = baseq;
5377 /* record configuration for macvlan interface in vdev */
5378 for (i = 0; i < num_tc; i++)
5379 netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5380 i, rss_i, baseq + (rss_i * i));
5382 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5383 adapter->rx_ring[baseq + i]->netdev = vdev;
5385 /* Guarantee all rings are updated before we update the
5386 * MAC address filter.
5390 /* ixgbe_add_mac_filter will return an index if it succeeds, so we
5391 * need to only treat it as an error value if it is negative.
5393 err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5394 VMDQ_P(accel->pool));
5398 /* if we cannot add the MAC rule then disable the offload */
5399 macvlan_release_l2fw_offload(vdev);
5401 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5402 adapter->rx_ring[baseq + i]->netdev = NULL;
5404 netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5406 /* unbind the queues and drop the subordinate channel config */
5407 netdev_unbind_sb_channel(adapter->netdev, vdev);
5408 netdev_set_sb_channel(vdev, 0);
5410 clear_bit(accel->pool, adapter->fwd_bitmask);
5416 static int ixgbe_macvlan_up(struct net_device *vdev, void *data)
5418 struct ixgbe_adapter *adapter = data;
5419 struct ixgbe_fwd_adapter *accel;
5421 if (!netif_is_macvlan(vdev))
5424 accel = macvlan_accel_priv(vdev);
5428 ixgbe_fwd_ring_up(adapter, accel);
5433 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5435 netdev_walk_all_upper_dev_rcu(adapter->netdev,
5436 ixgbe_macvlan_up, adapter);
5439 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5441 struct ixgbe_hw *hw = &adapter->hw;
5443 ixgbe_configure_pb(adapter);
5444 #ifdef CONFIG_IXGBE_DCB
5445 ixgbe_configure_dcb(adapter);
5448 * We must restore virtualization before VLANs or else
5449 * the VLVF registers will not be populated
5451 ixgbe_configure_virtualization(adapter);
5453 ixgbe_set_rx_mode(adapter->netdev);
5454 ixgbe_restore_vlan(adapter);
5455 ixgbe_ipsec_restore(adapter);
5457 switch (hw->mac.type) {
5458 case ixgbe_mac_82599EB:
5459 case ixgbe_mac_X540:
5460 hw->mac.ops.disable_rx_buff(hw);
5466 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5467 ixgbe_init_fdir_signature_82599(&adapter->hw,
5468 adapter->fdir_pballoc);
5469 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5470 ixgbe_init_fdir_perfect_82599(&adapter->hw,
5471 adapter->fdir_pballoc);
5472 ixgbe_fdir_filter_restore(adapter);
5475 switch (hw->mac.type) {
5476 case ixgbe_mac_82599EB:
5477 case ixgbe_mac_X540:
5478 hw->mac.ops.enable_rx_buff(hw);
5484 #ifdef CONFIG_IXGBE_DCA
5486 if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5487 ixgbe_setup_dca(adapter);
5488 #endif /* CONFIG_IXGBE_DCA */
5491 /* configure FCoE L2 filters, redirection table, and Rx control */
5492 ixgbe_configure_fcoe(adapter);
5494 #endif /* IXGBE_FCOE */
5495 ixgbe_configure_tx(adapter);
5496 ixgbe_configure_rx(adapter);
5497 ixgbe_configure_dfwd(adapter);
5501 * ixgbe_sfp_link_config - set up SFP+ link
5502 * @adapter: pointer to private adapter struct
5504 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5507 * We are assuming the worst case scenario here, and that
5508 * is that an SFP was inserted/removed after the reset
5509 * but before SFP detection was enabled. As such the best
5510 * solution is to just start searching as soon as we start
5512 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5513 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5515 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5516 adapter->sfp_poll_time = 0;
5520 * ixgbe_non_sfp_link_config - set up non-SFP+ link
5521 * @hw: pointer to private hardware struct
5523 * Returns 0 on success, negative on failure
5525 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5528 bool autoneg, link_up = false;
5529 int ret = IXGBE_ERR_LINK_SETUP;
5531 if (hw->mac.ops.check_link)
5532 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5537 speed = hw->phy.autoneg_advertised;
5538 if ((!speed) && (hw->mac.ops.get_link_capabilities))
5539 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5544 if (hw->mac.ops.setup_link)
5545 ret = hw->mac.ops.setup_link(hw, speed, link_up);
5550 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5552 struct ixgbe_hw *hw = &adapter->hw;
5555 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5556 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5558 gpie |= IXGBE_GPIE_EIAME;
5560 * use EIAM to auto-mask when MSI-X interrupt is asserted
5561 * this saves a register write for every interrupt
5563 switch (hw->mac.type) {
5564 case ixgbe_mac_82598EB:
5565 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5567 case ixgbe_mac_82599EB:
5568 case ixgbe_mac_X540:
5569 case ixgbe_mac_X550:
5570 case ixgbe_mac_X550EM_x:
5571 case ixgbe_mac_x550em_a:
5573 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5574 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5578 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
5579 * specifically only auto mask tx and rx interrupts */
5580 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5583 /* XXX: to interrupt immediately for EICS writes, enable this */
5584 /* gpie |= IXGBE_GPIE_EIMEN; */
5586 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5587 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5589 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5590 case IXGBE_82599_VMDQ_8Q_MASK:
5591 gpie |= IXGBE_GPIE_VTMODE_16;
5593 case IXGBE_82599_VMDQ_4Q_MASK:
5594 gpie |= IXGBE_GPIE_VTMODE_32;
5597 gpie |= IXGBE_GPIE_VTMODE_64;
5602 /* Enable Thermal over heat sensor interrupt */
5603 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5604 switch (adapter->hw.mac.type) {
5605 case ixgbe_mac_82599EB:
5606 gpie |= IXGBE_SDP0_GPIEN_8259X;
5613 /* Enable fan failure interrupt */
5614 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5615 gpie |= IXGBE_SDP1_GPIEN(hw);
5617 switch (hw->mac.type) {
5618 case ixgbe_mac_82599EB:
5619 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5621 case ixgbe_mac_X550EM_x:
5622 case ixgbe_mac_x550em_a:
5623 gpie |= IXGBE_SDP0_GPIEN_X540;
5629 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5632 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5634 struct ixgbe_hw *hw = &adapter->hw;
5638 ixgbe_get_hw_control(adapter);
5639 ixgbe_setup_gpie(adapter);
5641 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5642 ixgbe_configure_msix(adapter);
5644 ixgbe_configure_msi_and_legacy(adapter);
5646 /* enable the optics for 82599 SFP+ fiber */
5647 if (hw->mac.ops.enable_tx_laser)
5648 hw->mac.ops.enable_tx_laser(hw);
5650 if (hw->phy.ops.set_phy_power)
5651 hw->phy.ops.set_phy_power(hw, true);
5653 smp_mb__before_atomic();
5654 clear_bit(__IXGBE_DOWN, &adapter->state);
5655 ixgbe_napi_enable_all(adapter);
5657 if (ixgbe_is_sfp(hw)) {
5658 ixgbe_sfp_link_config(adapter);
5660 err = ixgbe_non_sfp_link_config(hw);
5662 e_err(probe, "link_config FAILED %d\n", err);
5665 /* clear any pending interrupts, may auto mask */
5666 IXGBE_READ_REG(hw, IXGBE_EICR);
5667 ixgbe_irq_enable(adapter, true, true);
5670 * If this adapter has a fan, check to see if we had a failure
5671 * before we enabled the interrupt.
5673 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5674 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5675 if (esdp & IXGBE_ESDP_SDP1)
5676 e_crit(drv, "Fan has stopped, replace the adapter\n");
5679 /* bring the link up in the watchdog, this could race with our first
5680 * link up interrupt but shouldn't be a problem */
5681 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5682 adapter->link_check_timeout = jiffies;
5683 mod_timer(&adapter->service_timer, jiffies);
5685 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5686 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5687 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5688 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5691 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5693 WARN_ON(in_interrupt());
5694 /* put off any impending NetWatchDogTimeout */
5695 netif_trans_update(adapter->netdev);
5697 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5698 usleep_range(1000, 2000);
5699 if (adapter->hw.phy.type == ixgbe_phy_fw)
5700 ixgbe_watchdog_link_is_down(adapter);
5701 ixgbe_down(adapter);
5703 * If SR-IOV enabled then wait a bit before bringing the adapter
5704 * back up to give the VFs time to respond to the reset. The
5705 * two second wait is based upon the watchdog timer cycle in
5708 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5711 clear_bit(__IXGBE_RESETTING, &adapter->state);
5714 void ixgbe_up(struct ixgbe_adapter *adapter)
5716 /* hardware has been reset, we need to reload some things */
5717 ixgbe_configure(adapter);
5719 ixgbe_up_complete(adapter);
5722 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5726 pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5728 switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5729 case IXGBE_PCIDEVCTRL2_17_34s:
5730 case IXGBE_PCIDEVCTRL2_4_8s:
5731 /* For now we cap the upper limit on delay to 2 seconds
5732 * as we end up going up to 34 seconds of delay in worst
5733 * case timeout value.
5735 case IXGBE_PCIDEVCTRL2_1_2s:
5736 return 2000000ul; /* 2.0 s */
5737 case IXGBE_PCIDEVCTRL2_260_520ms:
5738 return 520000ul; /* 520 ms */
5739 case IXGBE_PCIDEVCTRL2_65_130ms:
5740 return 130000ul; /* 130 ms */
5741 case IXGBE_PCIDEVCTRL2_16_32ms:
5742 return 32000ul; /* 32 ms */
5743 case IXGBE_PCIDEVCTRL2_1_2ms:
5744 return 2000ul; /* 2 ms */
5745 case IXGBE_PCIDEVCTRL2_50_100us:
5746 return 100ul; /* 100 us */
5747 case IXGBE_PCIDEVCTRL2_16_32ms_def:
5748 return 32000ul; /* 32 ms */
5753 /* We shouldn't need to hit this path, but just in case default as
5754 * though completion timeout is not supported and support 32ms.
5759 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5761 unsigned long wait_delay, delay_interval;
5762 struct ixgbe_hw *hw = &adapter->hw;
5766 /* disable receives */
5767 hw->mac.ops.disable_rx(hw);
5769 if (ixgbe_removed(hw->hw_addr))
5772 /* disable all enabled Rx queues */
5773 for (i = 0; i < adapter->num_rx_queues; i++) {
5774 struct ixgbe_ring *ring = adapter->rx_ring[i];
5775 u8 reg_idx = ring->reg_idx;
5777 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5778 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5779 rxdctl |= IXGBE_RXDCTL_SWFLSH;
5781 /* write value back with RXDCTL.ENABLE bit cleared */
5782 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5785 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5786 if (hw->mac.type == ixgbe_mac_82598EB &&
5787 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5790 /* Determine our minimum delay interval. We will increase this value
5791 * with each subsequent test. This way if the device returns quickly
5792 * we should spend as little time as possible waiting, however as
5793 * the time increases we will wait for larger periods of time.
5795 * The trick here is that we increase the interval using the
5796 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5797 * of that wait is that it totals up to 100x whatever interval we
5798 * choose. Since our minimum wait is 100us we can just divide the
5799 * total timeout by 100 to get our minimum delay interval.
5801 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5803 wait_loop = IXGBE_MAX_RX_DESC_POLL;
5804 wait_delay = delay_interval;
5806 while (wait_loop--) {
5807 usleep_range(wait_delay, wait_delay + 10);
5808 wait_delay += delay_interval * 2;
5811 /* OR together the reading of all the active RXDCTL registers,
5812 * and then test the result. We need the disable to complete
5813 * before we start freeing the memory and invalidating the
5816 for (i = 0; i < adapter->num_rx_queues; i++) {
5817 struct ixgbe_ring *ring = adapter->rx_ring[i];
5818 u8 reg_idx = ring->reg_idx;
5820 rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5823 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5828 "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5831 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5833 unsigned long wait_delay, delay_interval;
5834 struct ixgbe_hw *hw = &adapter->hw;
5838 if (ixgbe_removed(hw->hw_addr))
5841 /* disable all enabled Tx queues */
5842 for (i = 0; i < adapter->num_tx_queues; i++) {
5843 struct ixgbe_ring *ring = adapter->tx_ring[i];
5844 u8 reg_idx = ring->reg_idx;
5846 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5849 /* disable all enabled XDP Tx queues */
5850 for (i = 0; i < adapter->num_xdp_queues; i++) {
5851 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5852 u8 reg_idx = ring->reg_idx;
5854 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5857 /* If the link is not up there shouldn't be much in the way of
5858 * pending transactions. Those that are left will be flushed out
5859 * when the reset logic goes through the flush sequence to clean out
5860 * the pending Tx transactions.
5862 if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5863 goto dma_engine_disable;
5865 /* Determine our minimum delay interval. We will increase this value
5866 * with each subsequent test. This way if the device returns quickly
5867 * we should spend as little time as possible waiting, however as
5868 * the time increases we will wait for larger periods of time.
5870 * The trick here is that we increase the interval using the
5871 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5872 * of that wait is that it totals up to 100x whatever interval we
5873 * choose. Since our minimum wait is 100us we can just divide the
5874 * total timeout by 100 to get our minimum delay interval.
5876 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5878 wait_loop = IXGBE_MAX_RX_DESC_POLL;
5879 wait_delay = delay_interval;
5881 while (wait_loop--) {
5882 usleep_range(wait_delay, wait_delay + 10);
5883 wait_delay += delay_interval * 2;
5886 /* OR together the reading of all the active TXDCTL registers,
5887 * and then test the result. We need the disable to complete
5888 * before we start freeing the memory and invalidating the
5891 for (i = 0; i < adapter->num_tx_queues; i++) {
5892 struct ixgbe_ring *ring = adapter->tx_ring[i];
5893 u8 reg_idx = ring->reg_idx;
5895 txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5897 for (i = 0; i < adapter->num_xdp_queues; i++) {
5898 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5899 u8 reg_idx = ring->reg_idx;
5901 txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5904 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5905 goto dma_engine_disable;
5909 "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5912 /* Disable the Tx DMA engine on 82599 and later MAC */
5913 switch (hw->mac.type) {
5914 case ixgbe_mac_82599EB:
5915 case ixgbe_mac_X540:
5916 case ixgbe_mac_X550:
5917 case ixgbe_mac_X550EM_x:
5918 case ixgbe_mac_x550em_a:
5919 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5920 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5921 ~IXGBE_DMATXCTL_TE));
5928 void ixgbe_reset(struct ixgbe_adapter *adapter)
5930 struct ixgbe_hw *hw = &adapter->hw;
5931 struct net_device *netdev = adapter->netdev;
5934 if (ixgbe_removed(hw->hw_addr))
5936 /* lock SFP init bit to prevent race conditions with the watchdog */
5937 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5938 usleep_range(1000, 2000);
5940 /* clear all SFP and link config related flags while holding SFP_INIT */
5941 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5942 IXGBE_FLAG2_SFP_NEEDS_RESET);
5943 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5945 err = hw->mac.ops.init_hw(hw);
5948 case IXGBE_ERR_SFP_NOT_PRESENT:
5949 case IXGBE_ERR_SFP_NOT_SUPPORTED:
5951 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5952 e_dev_err("master disable timed out\n");
5954 case IXGBE_ERR_EEPROM_VERSION:
5955 /* We are running on a pre-production device, log a warning */
5956 e_dev_warn("This device is a pre-production adapter/LOM. "
5957 "Please be aware there may be issues associated with "
5958 "your hardware. If you are experiencing problems "
5959 "please contact your Intel or hardware "
5960 "representative who provided you with this "
5964 e_dev_err("Hardware Error: %d\n", err);
5967 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5969 /* flush entries out of MAC table */
5970 ixgbe_flush_sw_mac_table(adapter);
5971 __dev_uc_unsync(netdev, NULL);
5973 /* do not flush user set addresses */
5974 ixgbe_mac_set_default_filter(adapter);
5976 /* update SAN MAC vmdq pool selection */
5977 if (hw->mac.san_mac_rar_index)
5978 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5980 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5981 ixgbe_ptp_reset(adapter);
5983 if (hw->phy.ops.set_phy_power) {
5984 if (!netif_running(adapter->netdev) && !adapter->wol)
5985 hw->phy.ops.set_phy_power(hw, false);
5987 hw->phy.ops.set_phy_power(hw, true);
5992 * ixgbe_clean_tx_ring - Free Tx Buffers
5993 * @tx_ring: ring to be cleaned
5995 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5997 u16 i = tx_ring->next_to_clean;
5998 struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
6000 if (tx_ring->xsk_umem) {
6001 ixgbe_xsk_clean_tx_ring(tx_ring);
6005 while (i != tx_ring->next_to_use) {
6006 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
6008 /* Free all the Tx ring sk_buffs */
6009 if (ring_is_xdp(tx_ring))
6010 xdp_return_frame(tx_buffer->xdpf);
6012 dev_kfree_skb_any(tx_buffer->skb);
6014 /* unmap skb header data */
6015 dma_unmap_single(tx_ring->dev,
6016 dma_unmap_addr(tx_buffer, dma),
6017 dma_unmap_len(tx_buffer, len),
6020 /* check for eop_desc to determine the end of the packet */
6021 eop_desc = tx_buffer->next_to_watch;
6022 tx_desc = IXGBE_TX_DESC(tx_ring, i);
6024 /* unmap remaining buffers */
6025 while (tx_desc != eop_desc) {
6029 if (unlikely(i == tx_ring->count)) {
6031 tx_buffer = tx_ring->tx_buffer_info;
6032 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6035 /* unmap any remaining paged data */
6036 if (dma_unmap_len(tx_buffer, len))
6037 dma_unmap_page(tx_ring->dev,
6038 dma_unmap_addr(tx_buffer, dma),
6039 dma_unmap_len(tx_buffer, len),
6043 /* move us one more past the eop_desc for start of next pkt */
6046 if (unlikely(i == tx_ring->count)) {
6048 tx_buffer = tx_ring->tx_buffer_info;
6052 /* reset BQL for queue */
6053 if (!ring_is_xdp(tx_ring))
6054 netdev_tx_reset_queue(txring_txq(tx_ring));
6057 /* reset next_to_use and next_to_clean */
6058 tx_ring->next_to_use = 0;
6059 tx_ring->next_to_clean = 0;
6063 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6064 * @adapter: board private structure
6066 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6070 for (i = 0; i < adapter->num_rx_queues; i++)
6071 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6075 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6076 * @adapter: board private structure
6078 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6082 for (i = 0; i < adapter->num_tx_queues; i++)
6083 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6084 for (i = 0; i < adapter->num_xdp_queues; i++)
6085 ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6088 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6090 struct hlist_node *node2;
6091 struct ixgbe_fdir_filter *filter;
6093 spin_lock(&adapter->fdir_perfect_lock);
6095 hlist_for_each_entry_safe(filter, node2,
6096 &adapter->fdir_filter_list, fdir_node) {
6097 hlist_del(&filter->fdir_node);
6100 adapter->fdir_filter_count = 0;
6102 spin_unlock(&adapter->fdir_perfect_lock);
6105 void ixgbe_down(struct ixgbe_adapter *adapter)
6107 struct net_device *netdev = adapter->netdev;
6108 struct ixgbe_hw *hw = &adapter->hw;
6111 /* signal that we are down to the interrupt handler */
6112 if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6113 return; /* do nothing if already down */
6115 /* Shut off incoming Tx traffic */
6116 netif_tx_stop_all_queues(netdev);
6118 /* call carrier off first to avoid false dev_watchdog timeouts */
6119 netif_carrier_off(netdev);
6120 netif_tx_disable(netdev);
6123 ixgbe_disable_rx(adapter);
6125 /* synchronize_rcu() needed for pending XDP buffers to drain */
6126 if (adapter->xdp_ring[0])
6129 ixgbe_irq_disable(adapter);
6131 ixgbe_napi_disable_all(adapter);
6133 clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6134 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6135 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6137 del_timer_sync(&adapter->service_timer);
6139 if (adapter->num_vfs) {
6140 /* Clear EITR Select mapping */
6141 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6143 /* Mark all the VFs as inactive */
6144 for (i = 0 ; i < adapter->num_vfs; i++)
6145 adapter->vfinfo[i].clear_to_send = false;
6147 /* ping all the active vfs to let them know we are going down */
6148 ixgbe_ping_all_vfs(adapter);
6150 /* Disable all VFTE/VFRE TX/RX */
6151 ixgbe_disable_tx_rx(adapter);
6154 /* disable transmits in the hardware now that interrupts are off */
6155 ixgbe_disable_tx(adapter);
6157 if (!pci_channel_offline(adapter->pdev))
6158 ixgbe_reset(adapter);
6160 /* power down the optics for 82599 SFP+ fiber */
6161 if (hw->mac.ops.disable_tx_laser)
6162 hw->mac.ops.disable_tx_laser(hw);
6164 ixgbe_clean_all_tx_rings(adapter);
6165 ixgbe_clean_all_rx_rings(adapter);
6169 * ixgbe_eee_capable - helper function to determine EEE support on X550
6170 * @adapter: board private structure
6172 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6174 struct ixgbe_hw *hw = &adapter->hw;
6176 switch (hw->device_id) {
6177 case IXGBE_DEV_ID_X550EM_A_1G_T:
6178 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6179 if (!hw->phy.eee_speeds_supported)
6181 adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6182 if (!hw->phy.eee_speeds_advertised)
6184 adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6187 adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6188 adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6194 * ixgbe_tx_timeout - Respond to a Tx Hang
6195 * @netdev: network interface device structure
6197 static void ixgbe_tx_timeout(struct net_device *netdev, unsigned int txqueue)
6199 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6201 /* Do the reset outside of interrupt context */
6202 ixgbe_tx_timeout_reset(adapter);
6205 #ifdef CONFIG_IXGBE_DCB
6206 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6208 struct ixgbe_hw *hw = &adapter->hw;
6209 struct tc_configuration *tc;
6212 switch (hw->mac.type) {
6213 case ixgbe_mac_82598EB:
6214 case ixgbe_mac_82599EB:
6215 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6216 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6218 case ixgbe_mac_X540:
6219 case ixgbe_mac_X550:
6220 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6221 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6223 case ixgbe_mac_X550EM_x:
6224 case ixgbe_mac_x550em_a:
6226 adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6227 adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6231 /* Configure DCB traffic classes */
6232 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6233 tc = &adapter->dcb_cfg.tc_config[j];
6234 tc->path[DCB_TX_CONFIG].bwg_id = 0;
6235 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6236 tc->path[DCB_RX_CONFIG].bwg_id = 0;
6237 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6238 tc->dcb_pfc = pfc_disabled;
6241 /* Initialize default user to priority mapping, UPx->TC0 */
6242 tc = &adapter->dcb_cfg.tc_config[0];
6243 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6244 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6246 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6247 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6248 adapter->dcb_cfg.pfc_mode_enable = false;
6249 adapter->dcb_set_bitmap = 0x00;
6250 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6251 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6252 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6253 sizeof(adapter->temp_dcb_cfg));
6258 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6259 * @adapter: board private structure to initialize
6260 * @ii: pointer to ixgbe_info for device
6262 * ixgbe_sw_init initializes the Adapter private data structure.
6263 * Fields are initialized based on PCI device information and
6264 * OS network device settings (MTU size).
6266 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6267 const struct ixgbe_info *ii)
6269 struct ixgbe_hw *hw = &adapter->hw;
6270 struct pci_dev *pdev = adapter->pdev;
6271 unsigned int rss, fdir;
6275 /* PCI config space info */
6277 hw->vendor_id = pdev->vendor;
6278 hw->device_id = pdev->device;
6279 hw->revision_id = pdev->revision;
6280 hw->subsystem_vendor_id = pdev->subsystem_vendor;
6281 hw->subsystem_device_id = pdev->subsystem_device;
6283 /* get_invariants needs the device IDs */
6284 ii->get_invariants(hw);
6286 /* Set common capability flags and settings */
6287 rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6288 adapter->ring_feature[RING_F_RSS].limit = rss;
6289 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6290 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6291 adapter->atr_sample_rate = 20;
6292 fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6293 adapter->ring_feature[RING_F_FDIR].limit = fdir;
6294 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6295 adapter->ring_feature[RING_F_VMDQ].limit = 1;
6296 #ifdef CONFIG_IXGBE_DCA
6297 adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6299 #ifdef CONFIG_IXGBE_DCB
6300 adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6301 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6304 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6305 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6306 #ifdef CONFIG_IXGBE_DCB
6307 /* Default traffic class to use for FCoE */
6308 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6309 #endif /* CONFIG_IXGBE_DCB */
6310 #endif /* IXGBE_FCOE */
6312 /* initialize static ixgbe jump table entries */
6313 adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6315 if (!adapter->jump_tables[0])
6317 adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6319 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6320 adapter->jump_tables[i] = NULL;
6322 adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6323 sizeof(struct ixgbe_mac_addr),
6325 if (!adapter->mac_table)
6328 if (ixgbe_init_rss_key(adapter))
6331 adapter->af_xdp_zc_qps = bitmap_zalloc(MAX_XDP_QUEUES, GFP_KERNEL);
6332 if (!adapter->af_xdp_zc_qps)
6335 /* Set MAC specific capability flags and exceptions */
6336 switch (hw->mac.type) {
6337 case ixgbe_mac_82598EB:
6338 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6340 if (hw->device_id == IXGBE_DEV_ID_82598AT)
6341 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6343 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6344 adapter->ring_feature[RING_F_FDIR].limit = 0;
6345 adapter->atr_sample_rate = 0;
6346 adapter->fdir_pballoc = 0;
6348 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6349 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6350 #ifdef CONFIG_IXGBE_DCB
6351 adapter->fcoe.up = 0;
6352 #endif /* IXGBE_DCB */
6353 #endif /* IXGBE_FCOE */
6355 case ixgbe_mac_82599EB:
6356 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6357 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6359 case ixgbe_mac_X540:
6360 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6361 if (fwsm & IXGBE_FWSM_TS_ENABLED)
6362 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6364 case ixgbe_mac_x550em_a:
6365 adapter->flags |= IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE;
6366 switch (hw->device_id) {
6367 case IXGBE_DEV_ID_X550EM_A_1G_T:
6368 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6369 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6375 case ixgbe_mac_X550EM_x:
6376 #ifdef CONFIG_IXGBE_DCB
6377 adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6380 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6381 #ifdef CONFIG_IXGBE_DCB
6382 adapter->fcoe.up = 0;
6383 #endif /* IXGBE_DCB */
6384 #endif /* IXGBE_FCOE */
6386 case ixgbe_mac_X550:
6387 if (hw->mac.type == ixgbe_mac_X550)
6388 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6389 #ifdef CONFIG_IXGBE_DCA
6390 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6392 adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
6399 /* FCoE support exists, always init the FCoE lock */
6400 spin_lock_init(&adapter->fcoe.lock);
6403 /* n-tuple support exists, always init our spinlock */
6404 spin_lock_init(&adapter->fdir_perfect_lock);
6406 #ifdef CONFIG_IXGBE_DCB
6407 ixgbe_init_dcb(adapter);
6409 ixgbe_init_ipsec_offload(adapter);
6411 /* default flow control settings */
6412 hw->fc.requested_mode = ixgbe_fc_full;
6413 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
6414 ixgbe_pbthresh_setup(adapter);
6415 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6416 hw->fc.send_xon = true;
6417 hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6419 #ifdef CONFIG_PCI_IOV
6421 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6423 /* assign number of SR-IOV VFs */
6424 if (hw->mac.type != ixgbe_mac_82598EB) {
6425 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6427 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6430 #endif /* CONFIG_PCI_IOV */
6432 /* enable itr by default in dynamic mode */
6433 adapter->rx_itr_setting = 1;
6434 adapter->tx_itr_setting = 1;
6436 /* set default ring sizes */
6437 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6438 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6440 /* set default work limits */
6441 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6443 /* initialize eeprom parameters */
6444 if (ixgbe_init_eeprom_params_generic(hw)) {
6445 e_dev_err("EEPROM initialization failed\n");
6449 /* PF holds first pool slot */
6450 set_bit(0, adapter->fwd_bitmask);
6451 set_bit(__IXGBE_DOWN, &adapter->state);
6457 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6458 * @tx_ring: tx descriptor ring (for a specific queue) to setup
6460 * Return 0 on success, negative on failure
6462 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6464 struct device *dev = tx_ring->dev;
6465 int orig_node = dev_to_node(dev);
6466 int ring_node = NUMA_NO_NODE;
6469 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6471 if (tx_ring->q_vector)
6472 ring_node = tx_ring->q_vector->numa_node;
6474 tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6475 if (!tx_ring->tx_buffer_info)
6476 tx_ring->tx_buffer_info = vmalloc(size);
6477 if (!tx_ring->tx_buffer_info)
6480 /* round up to nearest 4K */
6481 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6482 tx_ring->size = ALIGN(tx_ring->size, 4096);
6484 set_dev_node(dev, ring_node);
6485 tx_ring->desc = dma_alloc_coherent(dev,
6489 set_dev_node(dev, orig_node);
6491 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6492 &tx_ring->dma, GFP_KERNEL);
6496 tx_ring->next_to_use = 0;
6497 tx_ring->next_to_clean = 0;
6501 vfree(tx_ring->tx_buffer_info);
6502 tx_ring->tx_buffer_info = NULL;
6503 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6508 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6509 * @adapter: board private structure
6511 * If this function returns with an error, then it's possible one or
6512 * more of the rings is populated (while the rest are not). It is the
6513 * callers duty to clean those orphaned rings.
6515 * Return 0 on success, negative on failure
6517 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6519 int i, j = 0, err = 0;
6521 for (i = 0; i < adapter->num_tx_queues; i++) {
6522 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6526 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6529 for (j = 0; j < adapter->num_xdp_queues; j++) {
6530 err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6534 e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6540 /* rewind the index freeing the rings as we go */
6542 ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6544 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6549 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6550 * @adapter: pointer to ixgbe_adapter
6551 * @rx_ring: rx descriptor ring (for a specific queue) to setup
6553 * Returns 0 on success, negative on failure
6555 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6556 struct ixgbe_ring *rx_ring)
6558 struct device *dev = rx_ring->dev;
6559 int orig_node = dev_to_node(dev);
6560 int ring_node = NUMA_NO_NODE;
6563 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6565 if (rx_ring->q_vector)
6566 ring_node = rx_ring->q_vector->numa_node;
6568 rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6569 if (!rx_ring->rx_buffer_info)
6570 rx_ring->rx_buffer_info = vmalloc(size);
6571 if (!rx_ring->rx_buffer_info)
6574 /* Round up to nearest 4K */
6575 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6576 rx_ring->size = ALIGN(rx_ring->size, 4096);
6578 set_dev_node(dev, ring_node);
6579 rx_ring->desc = dma_alloc_coherent(dev,
6583 set_dev_node(dev, orig_node);
6585 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6586 &rx_ring->dma, GFP_KERNEL);
6590 rx_ring->next_to_clean = 0;
6591 rx_ring->next_to_use = 0;
6593 /* XDP RX-queue info */
6594 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6595 rx_ring->queue_index) < 0)
6598 rx_ring->xdp_prog = adapter->xdp_prog;
6602 vfree(rx_ring->rx_buffer_info);
6603 rx_ring->rx_buffer_info = NULL;
6604 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6609 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6610 * @adapter: board private structure
6612 * If this function returns with an error, then it's possible one or
6613 * more of the rings is populated (while the rest are not). It is the
6614 * callers duty to clean those orphaned rings.
6616 * Return 0 on success, negative on failure
6618 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6622 for (i = 0; i < adapter->num_rx_queues; i++) {
6623 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6627 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6632 err = ixgbe_setup_fcoe_ddp_resources(adapter);
6637 /* rewind the index freeing the rings as we go */
6639 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6644 * ixgbe_free_tx_resources - Free Tx Resources per Queue
6645 * @tx_ring: Tx descriptor ring for a specific queue
6647 * Free all transmit software resources
6649 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6651 ixgbe_clean_tx_ring(tx_ring);
6653 vfree(tx_ring->tx_buffer_info);
6654 tx_ring->tx_buffer_info = NULL;
6656 /* if not set, then don't free */
6660 dma_free_coherent(tx_ring->dev, tx_ring->size,
6661 tx_ring->desc, tx_ring->dma);
6663 tx_ring->desc = NULL;
6667 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6668 * @adapter: board private structure
6670 * Free all transmit software resources
6672 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6676 for (i = 0; i < adapter->num_tx_queues; i++)
6677 if (adapter->tx_ring[i]->desc)
6678 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6679 for (i = 0; i < adapter->num_xdp_queues; i++)
6680 if (adapter->xdp_ring[i]->desc)
6681 ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6685 * ixgbe_free_rx_resources - Free Rx Resources
6686 * @rx_ring: ring to clean the resources from
6688 * Free all receive software resources
6690 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6692 ixgbe_clean_rx_ring(rx_ring);
6694 rx_ring->xdp_prog = NULL;
6695 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6696 vfree(rx_ring->rx_buffer_info);
6697 rx_ring->rx_buffer_info = NULL;
6699 /* if not set, then don't free */
6703 dma_free_coherent(rx_ring->dev, rx_ring->size,
6704 rx_ring->desc, rx_ring->dma);
6706 rx_ring->desc = NULL;
6710 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6711 * @adapter: board private structure
6713 * Free all receive software resources
6715 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6720 ixgbe_free_fcoe_ddp_resources(adapter);
6723 for (i = 0; i < adapter->num_rx_queues; i++)
6724 if (adapter->rx_ring[i]->desc)
6725 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6729 * ixgbe_change_mtu - Change the Maximum Transfer Unit
6730 * @netdev: network interface device structure
6731 * @new_mtu: new value for maximum frame size
6733 * Returns 0 on success, negative on failure
6735 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6737 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6739 if (adapter->xdp_prog) {
6740 int new_frame_size = new_mtu + ETH_HLEN + ETH_FCS_LEN +
6744 for (i = 0; i < adapter->num_rx_queues; i++) {
6745 struct ixgbe_ring *ring = adapter->rx_ring[i];
6747 if (new_frame_size > ixgbe_rx_bufsz(ring)) {
6748 e_warn(probe, "Requested MTU size is not supported with XDP\n");
6755 * For 82599EB we cannot allow legacy VFs to enable their receive
6756 * paths when MTU greater than 1500 is configured. So display a
6757 * warning that legacy VFs will be disabled.
6759 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6760 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6761 (new_mtu > ETH_DATA_LEN))
6762 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6764 netdev_dbg(netdev, "changing MTU from %d to %d\n",
6765 netdev->mtu, new_mtu);
6767 /* must set new MTU before calling down or up */
6768 netdev->mtu = new_mtu;
6770 if (netif_running(netdev))
6771 ixgbe_reinit_locked(adapter);
6777 * ixgbe_open - Called when a network interface is made active
6778 * @netdev: network interface device structure
6780 * Returns 0 on success, negative value on failure
6782 * The open entry point is called when a network interface is made
6783 * active by the system (IFF_UP). At this point all resources needed
6784 * for transmit and receive operations are allocated, the interrupt
6785 * handler is registered with the OS, the watchdog timer is started,
6786 * and the stack is notified that the interface is ready.
6788 int ixgbe_open(struct net_device *netdev)
6790 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6791 struct ixgbe_hw *hw = &adapter->hw;
6794 /* disallow open during test */
6795 if (test_bit(__IXGBE_TESTING, &adapter->state))
6798 netif_carrier_off(netdev);
6800 /* allocate transmit descriptors */
6801 err = ixgbe_setup_all_tx_resources(adapter);
6805 /* allocate receive descriptors */
6806 err = ixgbe_setup_all_rx_resources(adapter);
6810 ixgbe_configure(adapter);
6812 err = ixgbe_request_irq(adapter);
6816 /* Notify the stack of the actual queue counts. */
6817 queues = adapter->num_tx_queues;
6818 err = netif_set_real_num_tx_queues(netdev, queues);
6820 goto err_set_queues;
6822 queues = adapter->num_rx_queues;
6823 err = netif_set_real_num_rx_queues(netdev, queues);
6825 goto err_set_queues;
6827 ixgbe_ptp_init(adapter);
6829 ixgbe_up_complete(adapter);
6831 ixgbe_clear_udp_tunnel_port(adapter, IXGBE_VXLANCTRL_ALL_UDPPORT_MASK);
6832 udp_tunnel_get_rx_info(netdev);
6837 ixgbe_free_irq(adapter);
6839 ixgbe_free_all_rx_resources(adapter);
6840 if (hw->phy.ops.set_phy_power && !adapter->wol)
6841 hw->phy.ops.set_phy_power(&adapter->hw, false);
6843 ixgbe_free_all_tx_resources(adapter);
6845 ixgbe_reset(adapter);
6850 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6852 ixgbe_ptp_suspend(adapter);
6854 if (adapter->hw.phy.ops.enter_lplu) {
6855 adapter->hw.phy.reset_disable = true;
6856 ixgbe_down(adapter);
6857 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6858 adapter->hw.phy.reset_disable = false;
6860 ixgbe_down(adapter);
6863 ixgbe_free_irq(adapter);
6865 ixgbe_free_all_tx_resources(adapter);
6866 ixgbe_free_all_rx_resources(adapter);
6870 * ixgbe_close - Disables a network interface
6871 * @netdev: network interface device structure
6873 * Returns 0, this is not allowed to fail
6875 * The close entry point is called when an interface is de-activated
6876 * by the OS. The hardware is still under the drivers control, but
6877 * needs to be disabled. A global MAC reset is issued to stop the
6878 * hardware, and all transmit and receive resources are freed.
6880 int ixgbe_close(struct net_device *netdev)
6882 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6884 ixgbe_ptp_stop(adapter);
6886 if (netif_device_present(netdev))
6887 ixgbe_close_suspend(adapter);
6889 ixgbe_fdir_filter_exit(adapter);
6891 ixgbe_release_hw_control(adapter);
6897 static int ixgbe_resume(struct pci_dev *pdev)
6899 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6900 struct net_device *netdev = adapter->netdev;
6903 adapter->hw.hw_addr = adapter->io_addr;
6904 pci_set_power_state(pdev, PCI_D0);
6905 pci_restore_state(pdev);
6907 * pci_restore_state clears dev->state_saved so call
6908 * pci_save_state to restore it.
6910 pci_save_state(pdev);
6912 err = pci_enable_device_mem(pdev);
6914 e_dev_err("Cannot enable PCI device from suspend\n");
6917 smp_mb__before_atomic();
6918 clear_bit(__IXGBE_DISABLED, &adapter->state);
6919 pci_set_master(pdev);
6921 pci_wake_from_d3(pdev, false);
6923 ixgbe_reset(adapter);
6925 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6928 err = ixgbe_init_interrupt_scheme(adapter);
6929 if (!err && netif_running(netdev))
6930 err = ixgbe_open(netdev);
6934 netif_device_attach(netdev);
6939 #endif /* CONFIG_PM */
6941 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6943 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6944 struct net_device *netdev = adapter->netdev;
6945 struct ixgbe_hw *hw = &adapter->hw;
6947 u32 wufc = adapter->wol;
6953 netif_device_detach(netdev);
6955 if (netif_running(netdev))
6956 ixgbe_close_suspend(adapter);
6958 ixgbe_clear_interrupt_scheme(adapter);
6962 retval = pci_save_state(pdev);
6967 if (hw->mac.ops.stop_link_on_d3)
6968 hw->mac.ops.stop_link_on_d3(hw);
6973 ixgbe_set_rx_mode(netdev);
6975 /* enable the optics for 82599 SFP+ fiber as we can WoL */
6976 if (hw->mac.ops.enable_tx_laser)
6977 hw->mac.ops.enable_tx_laser(hw);
6979 /* enable the reception of multicast packets */
6980 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6981 fctrl |= IXGBE_FCTRL_MPE;
6982 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6984 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6985 ctrl |= IXGBE_CTRL_GIO_DIS;
6986 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6988 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6990 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6991 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6994 switch (hw->mac.type) {
6995 case ixgbe_mac_82598EB:
6996 pci_wake_from_d3(pdev, false);
6998 case ixgbe_mac_82599EB:
6999 case ixgbe_mac_X540:
7000 case ixgbe_mac_X550:
7001 case ixgbe_mac_X550EM_x:
7002 case ixgbe_mac_x550em_a:
7003 pci_wake_from_d3(pdev, !!wufc);
7009 *enable_wake = !!wufc;
7010 if (hw->phy.ops.set_phy_power && !*enable_wake)
7011 hw->phy.ops.set_phy_power(hw, false);
7013 ixgbe_release_hw_control(adapter);
7015 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
7016 pci_disable_device(pdev);
7022 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
7027 retval = __ixgbe_shutdown(pdev, &wake);
7032 pci_prepare_to_sleep(pdev);
7034 pci_wake_from_d3(pdev, false);
7035 pci_set_power_state(pdev, PCI_D3hot);
7040 #endif /* CONFIG_PM */
7042 static void ixgbe_shutdown(struct pci_dev *pdev)
7046 __ixgbe_shutdown(pdev, &wake);
7048 if (system_state == SYSTEM_POWER_OFF) {
7049 pci_wake_from_d3(pdev, wake);
7050 pci_set_power_state(pdev, PCI_D3hot);
7055 * ixgbe_update_stats - Update the board statistics counters.
7056 * @adapter: board private structure
7058 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7060 struct net_device *netdev = adapter->netdev;
7061 struct ixgbe_hw *hw = &adapter->hw;
7062 struct ixgbe_hw_stats *hwstats = &adapter->stats;
7064 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7065 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7066 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7067 u64 alloc_rx_page = 0;
7068 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7070 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7071 test_bit(__IXGBE_RESETTING, &adapter->state))
7074 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7077 for (i = 0; i < adapter->num_rx_queues; i++) {
7078 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7079 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7081 adapter->rsc_total_count = rsc_count;
7082 adapter->rsc_total_flush = rsc_flush;
7085 for (i = 0; i < adapter->num_rx_queues; i++) {
7086 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
7087 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7088 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7089 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7090 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7091 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7092 bytes += rx_ring->stats.bytes;
7093 packets += rx_ring->stats.packets;
7095 adapter->non_eop_descs = non_eop_descs;
7096 adapter->alloc_rx_page = alloc_rx_page;
7097 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7098 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7099 adapter->hw_csum_rx_error = hw_csum_rx_error;
7100 netdev->stats.rx_bytes = bytes;
7101 netdev->stats.rx_packets = packets;
7105 /* gather some stats to the adapter struct that are per queue */
7106 for (i = 0; i < adapter->num_tx_queues; i++) {
7107 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7108 restart_queue += tx_ring->tx_stats.restart_queue;
7109 tx_busy += tx_ring->tx_stats.tx_busy;
7110 bytes += tx_ring->stats.bytes;
7111 packets += tx_ring->stats.packets;
7113 for (i = 0; i < adapter->num_xdp_queues; i++) {
7114 struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
7116 restart_queue += xdp_ring->tx_stats.restart_queue;
7117 tx_busy += xdp_ring->tx_stats.tx_busy;
7118 bytes += xdp_ring->stats.bytes;
7119 packets += xdp_ring->stats.packets;
7121 adapter->restart_queue = restart_queue;
7122 adapter->tx_busy = tx_busy;
7123 netdev->stats.tx_bytes = bytes;
7124 netdev->stats.tx_packets = packets;
7126 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7128 /* 8 register reads */
7129 for (i = 0; i < 8; i++) {
7130 /* for packet buffers not used, the register should read 0 */
7131 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7133 hwstats->mpc[i] += mpc;
7134 total_mpc += hwstats->mpc[i];
7135 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7136 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7137 switch (hw->mac.type) {
7138 case ixgbe_mac_82598EB:
7139 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7140 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7141 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7142 hwstats->pxonrxc[i] +=
7143 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7145 case ixgbe_mac_82599EB:
7146 case ixgbe_mac_X540:
7147 case ixgbe_mac_X550:
7148 case ixgbe_mac_X550EM_x:
7149 case ixgbe_mac_x550em_a:
7150 hwstats->pxonrxc[i] +=
7151 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7158 /*16 register reads */
7159 for (i = 0; i < 16; i++) {
7160 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7161 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7162 if ((hw->mac.type == ixgbe_mac_82599EB) ||
7163 (hw->mac.type == ixgbe_mac_X540) ||
7164 (hw->mac.type == ixgbe_mac_X550) ||
7165 (hw->mac.type == ixgbe_mac_X550EM_x) ||
7166 (hw->mac.type == ixgbe_mac_x550em_a)) {
7167 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7168 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7169 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7170 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7174 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7175 /* work around hardware counting issue */
7176 hwstats->gprc -= missed_rx;
7178 ixgbe_update_xoff_received(adapter);
7180 /* 82598 hardware only has a 32 bit counter in the high register */
7181 switch (hw->mac.type) {
7182 case ixgbe_mac_82598EB:
7183 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7184 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7185 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7186 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7188 case ixgbe_mac_X540:
7189 case ixgbe_mac_X550:
7190 case ixgbe_mac_X550EM_x:
7191 case ixgbe_mac_x550em_a:
7192 /* OS2BMC stats are X540 and later */
7193 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7194 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7195 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7196 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7198 case ixgbe_mac_82599EB:
7199 for (i = 0; i < 16; i++)
7200 adapter->hw_rx_no_dma_resources +=
7201 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7202 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7203 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7204 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7205 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7206 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7207 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7208 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7209 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7210 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7212 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7213 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7214 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7215 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7216 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7217 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7218 /* Add up per cpu counters for total ddp aloc fail */
7219 if (adapter->fcoe.ddp_pool) {
7220 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7221 struct ixgbe_fcoe_ddp_pool *ddp_pool;
7223 u64 noddp = 0, noddp_ext_buff = 0;
7224 for_each_possible_cpu(cpu) {
7225 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7226 noddp += ddp_pool->noddp;
7227 noddp_ext_buff += ddp_pool->noddp_ext_buff;
7229 hwstats->fcoe_noddp = noddp;
7230 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7232 #endif /* IXGBE_FCOE */
7237 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7238 hwstats->bprc += bprc;
7239 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7240 if (hw->mac.type == ixgbe_mac_82598EB)
7241 hwstats->mprc -= bprc;
7242 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7243 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7244 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7245 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7246 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7247 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7248 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7249 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7250 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7251 hwstats->lxontxc += lxon;
7252 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7253 hwstats->lxofftxc += lxoff;
7254 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7255 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7257 * 82598 errata - tx of flow control packets is included in tx counters
7259 xon_off_tot = lxon + lxoff;
7260 hwstats->gptc -= xon_off_tot;
7261 hwstats->mptc -= xon_off_tot;
7262 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7263 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7264 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7265 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7266 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7267 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7268 hwstats->ptc64 -= xon_off_tot;
7269 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7270 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7271 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7272 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7273 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7274 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7276 /* Fill out the OS statistics structure */
7277 netdev->stats.multicast = hwstats->mprc;
7280 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7281 netdev->stats.rx_dropped = 0;
7282 netdev->stats.rx_length_errors = hwstats->rlec;
7283 netdev->stats.rx_crc_errors = hwstats->crcerrs;
7284 netdev->stats.rx_missed_errors = total_mpc;
7288 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7289 * @adapter: pointer to the device adapter structure
7291 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7293 struct ixgbe_hw *hw = &adapter->hw;
7296 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7299 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7301 /* if interface is down do nothing */
7302 if (test_bit(__IXGBE_DOWN, &adapter->state))
7305 /* do nothing if we are not using signature filters */
7306 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7309 adapter->fdir_overflow++;
7311 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7312 for (i = 0; i < adapter->num_tx_queues; i++)
7313 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7314 &(adapter->tx_ring[i]->state));
7315 for (i = 0; i < adapter->num_xdp_queues; i++)
7316 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7317 &adapter->xdp_ring[i]->state);
7318 /* re-enable flow director interrupts */
7319 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7321 e_err(probe, "failed to finish FDIR re-initialization, "
7322 "ignored adding FDIR ATR filters\n");
7327 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7328 * @adapter: pointer to the device adapter structure
7330 * This function serves two purposes. First it strobes the interrupt lines
7331 * in order to make certain interrupts are occurring. Secondly it sets the
7332 * bits needed to check for TX hangs. As a result we should immediately
7333 * determine if a hang has occurred.
7335 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7337 struct ixgbe_hw *hw = &adapter->hw;
7341 /* If we're down, removing or resetting, just bail */
7342 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7343 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7344 test_bit(__IXGBE_RESETTING, &adapter->state))
7347 /* Force detection of hung controller */
7348 if (netif_carrier_ok(adapter->netdev)) {
7349 for (i = 0; i < adapter->num_tx_queues; i++)
7350 set_check_for_tx_hang(adapter->tx_ring[i]);
7351 for (i = 0; i < adapter->num_xdp_queues; i++)
7352 set_check_for_tx_hang(adapter->xdp_ring[i]);
7355 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7357 * for legacy and MSI interrupts don't set any bits
7358 * that are enabled for EIAM, because this operation
7359 * would set *both* EIMS and EICS for any bit in EIAM
7361 IXGBE_WRITE_REG(hw, IXGBE_EICS,
7362 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7364 /* get one bit for every active tx/rx interrupt vector */
7365 for (i = 0; i < adapter->num_q_vectors; i++) {
7366 struct ixgbe_q_vector *qv = adapter->q_vector[i];
7367 if (qv->rx.ring || qv->tx.ring)
7372 /* Cause software interrupt to ensure rings are cleaned */
7373 ixgbe_irq_rearm_queues(adapter, eics);
7377 * ixgbe_watchdog_update_link - update the link status
7378 * @adapter: pointer to the device adapter structure
7380 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7382 struct ixgbe_hw *hw = &adapter->hw;
7383 u32 link_speed = adapter->link_speed;
7384 bool link_up = adapter->link_up;
7385 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7387 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7390 if (hw->mac.ops.check_link) {
7391 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7393 /* always assume link is up, if no check link function */
7394 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7398 if (adapter->ixgbe_ieee_pfc)
7399 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7401 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7402 hw->mac.ops.fc_enable(hw);
7403 ixgbe_set_rx_drop_en(adapter);
7407 time_after(jiffies, (adapter->link_check_timeout +
7408 IXGBE_TRY_LINK_TIMEOUT))) {
7409 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7410 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7411 IXGBE_WRITE_FLUSH(hw);
7414 adapter->link_up = link_up;
7415 adapter->link_speed = link_speed;
7418 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7420 #ifdef CONFIG_IXGBE_DCB
7421 struct net_device *netdev = adapter->netdev;
7422 struct dcb_app app = {
7423 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7428 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7429 up = dcb_ieee_getapp_mask(netdev, &app);
7431 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7436 * ixgbe_watchdog_link_is_up - update netif_carrier status and
7437 * print link up message
7438 * @adapter: pointer to the device adapter structure
7440 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7442 struct net_device *netdev = adapter->netdev;
7443 struct ixgbe_hw *hw = &adapter->hw;
7444 u32 link_speed = adapter->link_speed;
7445 const char *speed_str;
7446 bool flow_rx, flow_tx;
7448 /* only continue if link was previously down */
7449 if (netif_carrier_ok(netdev))
7452 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7454 switch (hw->mac.type) {
7455 case ixgbe_mac_82598EB: {
7456 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7457 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7458 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7459 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7462 case ixgbe_mac_X540:
7463 case ixgbe_mac_X550:
7464 case ixgbe_mac_X550EM_x:
7465 case ixgbe_mac_x550em_a:
7466 case ixgbe_mac_82599EB: {
7467 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7468 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7469 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7470 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7479 adapter->last_rx_ptp_check = jiffies;
7481 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7482 ixgbe_ptp_start_cyclecounter(adapter);
7484 switch (link_speed) {
7485 case IXGBE_LINK_SPEED_10GB_FULL:
7486 speed_str = "10 Gbps";
7488 case IXGBE_LINK_SPEED_5GB_FULL:
7489 speed_str = "5 Gbps";
7491 case IXGBE_LINK_SPEED_2_5GB_FULL:
7492 speed_str = "2.5 Gbps";
7494 case IXGBE_LINK_SPEED_1GB_FULL:
7495 speed_str = "1 Gbps";
7497 case IXGBE_LINK_SPEED_100_FULL:
7498 speed_str = "100 Mbps";
7500 case IXGBE_LINK_SPEED_10_FULL:
7501 speed_str = "10 Mbps";
7504 speed_str = "unknown speed";
7507 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7508 ((flow_rx && flow_tx) ? "RX/TX" :
7510 (flow_tx ? "TX" : "None"))));
7512 netif_carrier_on(netdev);
7513 ixgbe_check_vf_rate_limit(adapter);
7515 /* enable transmits */
7516 netif_tx_wake_all_queues(adapter->netdev);
7518 /* update the default user priority for VFs */
7519 ixgbe_update_default_up(adapter);
7521 /* ping all the active vfs to let them know link has changed */
7522 ixgbe_ping_all_vfs(adapter);
7526 * ixgbe_watchdog_link_is_down - update netif_carrier status and
7527 * print link down message
7528 * @adapter: pointer to the adapter structure
7530 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7532 struct net_device *netdev = adapter->netdev;
7533 struct ixgbe_hw *hw = &adapter->hw;
7535 adapter->link_up = false;
7536 adapter->link_speed = 0;
7538 /* only continue if link was up previously */
7539 if (!netif_carrier_ok(netdev))
7542 /* poll for SFP+ cable when link is down */
7543 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7544 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7546 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7547 ixgbe_ptp_start_cyclecounter(adapter);
7549 e_info(drv, "NIC Link is Down\n");
7550 netif_carrier_off(netdev);
7552 /* ping all the active vfs to let them know link has changed */
7553 ixgbe_ping_all_vfs(adapter);
7556 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7560 for (i = 0; i < adapter->num_tx_queues; i++) {
7561 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7563 if (tx_ring->next_to_use != tx_ring->next_to_clean)
7567 for (i = 0; i < adapter->num_xdp_queues; i++) {
7568 struct ixgbe_ring *ring = adapter->xdp_ring[i];
7570 if (ring->next_to_use != ring->next_to_clean)
7577 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7579 struct ixgbe_hw *hw = &adapter->hw;
7580 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7581 u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7585 if (!adapter->num_vfs)
7588 /* resetting the PF is only needed for MAC before X550 */
7589 if (hw->mac.type >= ixgbe_mac_X550)
7592 for (i = 0; i < adapter->num_vfs; i++) {
7593 for (j = 0; j < q_per_pool; j++) {
7596 h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7597 t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7608 * ixgbe_watchdog_flush_tx - flush queues on link down
7609 * @adapter: pointer to the device adapter structure
7611 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7613 if (!netif_carrier_ok(adapter->netdev)) {
7614 if (ixgbe_ring_tx_pending(adapter) ||
7615 ixgbe_vf_tx_pending(adapter)) {
7616 /* We've lost link, so the controller stops DMA,
7617 * but we've got queued Tx work that's never going
7618 * to get done, so reset controller to flush Tx.
7619 * (Do the reset outside of interrupt context).
7621 e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7622 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7627 #ifdef CONFIG_PCI_IOV
7628 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7630 struct ixgbe_hw *hw = &adapter->hw;
7631 struct pci_dev *pdev = adapter->pdev;
7635 if (!(netif_carrier_ok(adapter->netdev)))
7638 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7639 if (gpc) /* If incrementing then no need for the check below */
7641 /* Check to see if a bad DMA write target from an errant or
7642 * malicious VF has caused a PCIe error. If so then we can
7643 * issue a VFLR to the offending VF(s) and then resume without
7644 * requesting a full slot reset.
7650 /* check status reg for all VFs owned by this PF */
7651 for (vf = 0; vf < adapter->num_vfs; ++vf) {
7652 struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7657 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7658 if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7659 status_reg & PCI_STATUS_REC_MASTER_ABORT)
7664 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7668 /* Do not perform spoof check for 82598 or if not in IOV mode */
7669 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7670 adapter->num_vfs == 0)
7673 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7676 * ssvpc register is cleared on read, if zero then no
7677 * spoofed packets in the last interval.
7682 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7685 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7690 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7693 #endif /* CONFIG_PCI_IOV */
7697 * ixgbe_watchdog_subtask - check and bring link up
7698 * @adapter: pointer to the device adapter structure
7700 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7702 /* if interface is down, removing or resetting, do nothing */
7703 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7704 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7705 test_bit(__IXGBE_RESETTING, &adapter->state))
7708 ixgbe_watchdog_update_link(adapter);
7710 if (adapter->link_up)
7711 ixgbe_watchdog_link_is_up(adapter);
7713 ixgbe_watchdog_link_is_down(adapter);
7715 ixgbe_check_for_bad_vf(adapter);
7716 ixgbe_spoof_check(adapter);
7717 ixgbe_update_stats(adapter);
7719 ixgbe_watchdog_flush_tx(adapter);
7723 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7724 * @adapter: the ixgbe adapter structure
7726 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7728 struct ixgbe_hw *hw = &adapter->hw;
7731 /* not searching for SFP so there is nothing to do here */
7732 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7733 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7736 if (adapter->sfp_poll_time &&
7737 time_after(adapter->sfp_poll_time, jiffies))
7738 return; /* If not yet time to poll for SFP */
7740 /* someone else is in init, wait until next service event */
7741 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7744 adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7746 err = hw->phy.ops.identify_sfp(hw);
7747 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7750 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7751 /* If no cable is present, then we need to reset
7752 * the next time we find a good cable. */
7753 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7760 /* exit if reset not needed */
7761 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7764 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7767 * A module may be identified correctly, but the EEPROM may not have
7768 * support for that module. setup_sfp() will fail in that case, so
7769 * we should not allow that module to load.
7771 if (hw->mac.type == ixgbe_mac_82598EB)
7772 err = hw->phy.ops.reset(hw);
7774 err = hw->mac.ops.setup_sfp(hw);
7776 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7779 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7780 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7783 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7785 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7786 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7787 e_dev_err("failed to initialize because an unsupported "
7788 "SFP+ module type was detected.\n");
7789 e_dev_err("Reload the driver after installing a "
7790 "supported module.\n");
7791 unregister_netdev(adapter->netdev);
7796 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7797 * @adapter: the ixgbe adapter structure
7799 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7801 struct ixgbe_hw *hw = &adapter->hw;
7804 bool autoneg = false;
7806 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7809 /* someone else is in init, wait until next service event */
7810 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7813 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7815 hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7817 /* advertise highest capable link speed */
7818 if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7819 speed = IXGBE_LINK_SPEED_10GB_FULL;
7821 speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7822 IXGBE_LINK_SPEED_1GB_FULL);
7824 if (hw->mac.ops.setup_link)
7825 hw->mac.ops.setup_link(hw, speed, true);
7827 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7828 adapter->link_check_timeout = jiffies;
7829 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7833 * ixgbe_service_timer - Timer Call-back
7834 * @t: pointer to timer_list structure
7836 static void ixgbe_service_timer(struct timer_list *t)
7838 struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7839 unsigned long next_event_offset;
7841 /* poll faster when waiting for link */
7842 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7843 next_event_offset = HZ / 10;
7845 next_event_offset = HZ * 2;
7847 /* Reset the timer */
7848 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7850 ixgbe_service_event_schedule(adapter);
7853 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7855 struct ixgbe_hw *hw = &adapter->hw;
7858 if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7861 adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7863 if (!hw->phy.ops.handle_lasi)
7866 status = hw->phy.ops.handle_lasi(&adapter->hw);
7867 if (status != IXGBE_ERR_OVERTEMP)
7870 e_crit(drv, "%s\n", ixgbe_overheat_msg);
7873 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7875 if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7879 /* If we're already down, removing or resetting, just bail */
7880 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7881 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7882 test_bit(__IXGBE_RESETTING, &adapter->state)) {
7887 ixgbe_dump(adapter);
7888 netdev_err(adapter->netdev, "Reset adapter\n");
7889 adapter->tx_timeout_count++;
7891 ixgbe_reinit_locked(adapter);
7896 * ixgbe_check_fw_error - Check firmware for errors
7897 * @adapter: the adapter private structure
7899 * Check firmware errors in register FWSM
7901 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
7903 struct ixgbe_hw *hw = &adapter->hw;
7906 /* read fwsm.ext_err_ind register and log errors */
7907 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
7909 if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
7910 !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
7911 e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7914 if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
7915 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");
7923 * ixgbe_service_task - manages and runs subtasks
7924 * @work: pointer to work_struct containing our data
7926 static void ixgbe_service_task(struct work_struct *work)
7928 struct ixgbe_adapter *adapter = container_of(work,
7929 struct ixgbe_adapter,
7931 if (ixgbe_removed(adapter->hw.hw_addr)) {
7932 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7934 ixgbe_down(adapter);
7937 ixgbe_service_event_complete(adapter);
7940 if (ixgbe_check_fw_error(adapter)) {
7941 if (!test_bit(__IXGBE_DOWN, &adapter->state))
7942 unregister_netdev(adapter->netdev);
7943 ixgbe_service_event_complete(adapter);
7946 if (adapter->flags2 & IXGBE_FLAG2_UDP_TUN_REREG_NEEDED) {
7948 adapter->flags2 &= ~IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
7949 udp_tunnel_get_rx_info(adapter->netdev);
7952 ixgbe_reset_subtask(adapter);
7953 ixgbe_phy_interrupt_subtask(adapter);
7954 ixgbe_sfp_detection_subtask(adapter);
7955 ixgbe_sfp_link_config_subtask(adapter);
7956 ixgbe_check_overtemp_subtask(adapter);
7957 ixgbe_watchdog_subtask(adapter);
7958 ixgbe_fdir_reinit_subtask(adapter);
7959 ixgbe_check_hang_subtask(adapter);
7961 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7962 ixgbe_ptp_overflow_check(adapter);
7963 if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
7964 ixgbe_ptp_rx_hang(adapter);
7965 ixgbe_ptp_tx_hang(adapter);
7968 ixgbe_service_event_complete(adapter);
7971 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7972 struct ixgbe_tx_buffer *first,
7974 struct ixgbe_ipsec_tx_data *itd)
7976 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7977 struct sk_buff *skb = first->skb;
7988 u32 paylen, l4_offset;
7989 u32 fceof_saidx = 0;
7992 if (skb->ip_summed != CHECKSUM_PARTIAL)
7995 if (!skb_is_gso(skb))
7998 err = skb_cow_head(skb, 0);
8002 if (eth_p_mpls(first->protocol))
8003 ip.hdr = skb_inner_network_header(skb);
8005 ip.hdr = skb_network_header(skb);
8006 l4.hdr = skb_checksum_start(skb);
8008 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
8009 type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
8010 IXGBE_ADVTXD_TUCMD_L4T_UDP : IXGBE_ADVTXD_TUCMD_L4T_TCP;
8012 /* initialize outer IP header fields */
8013 if (ip.v4->version == 4) {
8014 unsigned char *csum_start = skb_checksum_start(skb);
8015 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
8016 int len = csum_start - trans_start;
8018 /* IP header will have to cancel out any data that
8019 * is not a part of the outer IP header, so set to
8020 * a reverse csum if needed, else init check to 0.
8022 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
8023 csum_fold(csum_partial(trans_start,
8025 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
8028 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8029 IXGBE_TX_FLAGS_CSUM |
8030 IXGBE_TX_FLAGS_IPV4;
8032 ip.v6->payload_len = 0;
8033 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8034 IXGBE_TX_FLAGS_CSUM;
8037 /* determine offset of inner transport header */
8038 l4_offset = l4.hdr - skb->data;
8040 /* remove payload length from inner checksum */
8041 paylen = skb->len - l4_offset;
8043 if (type_tucmd & IXGBE_ADVTXD_TUCMD_L4T_TCP) {
8044 /* compute length of segmentation header */
8045 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
8046 csum_replace_by_diff(&l4.tcp->check,
8047 (__force __wsum)htonl(paylen));
8049 /* compute length of segmentation header */
8050 *hdr_len = sizeof(*l4.udp) + l4_offset;
8051 csum_replace_by_diff(&l4.udp->check,
8052 (__force __wsum)htonl(paylen));
8055 /* update gso size and bytecount with header size */
8056 first->gso_segs = skb_shinfo(skb)->gso_segs;
8057 first->bytecount += (first->gso_segs - 1) * *hdr_len;
8059 /* mss_l4len_id: use 0 as index for TSO */
8060 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8061 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8063 fceof_saidx |= itd->sa_idx;
8064 type_tucmd |= itd->flags | itd->trailer_len;
8066 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8067 vlan_macip_lens = l4.hdr - ip.hdr;
8068 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8069 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8071 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8077 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb)
8079 unsigned int offset = 0;
8081 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
8083 return offset == skb_checksum_start_offset(skb);
8086 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8087 struct ixgbe_tx_buffer *first,
8088 struct ixgbe_ipsec_tx_data *itd)
8090 struct sk_buff *skb = first->skb;
8091 u32 vlan_macip_lens = 0;
8092 u32 fceof_saidx = 0;
8095 if (skb->ip_summed != CHECKSUM_PARTIAL) {
8097 if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8098 IXGBE_TX_FLAGS_CC)))
8103 switch (skb->csum_offset) {
8104 case offsetof(struct tcphdr, check):
8105 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8107 case offsetof(struct udphdr, check):
8109 case offsetof(struct sctphdr, checksum):
8110 /* validate that this is actually an SCTP request */
8111 if (((first->protocol == htons(ETH_P_IP)) &&
8112 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
8113 ((first->protocol == htons(ETH_P_IPV6)) &&
8114 ixgbe_ipv6_csum_is_sctp(skb))) {
8115 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8120 skb_checksum_help(skb);
8124 /* update TX checksum flag */
8125 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8126 vlan_macip_lens = skb_checksum_start_offset(skb) -
8127 skb_network_offset(skb);
8129 /* vlan_macip_lens: MACLEN, VLAN tag */
8130 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8131 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8133 fceof_saidx |= itd->sa_idx;
8134 type_tucmd |= itd->flags | itd->trailer_len;
8136 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8139 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8140 ((_flag <= _result) ? \
8141 ((u32)(_input & _flag) * (_result / _flag)) : \
8142 ((u32)(_input & _flag) / (_flag / _result)))
8144 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8146 /* set type for advanced descriptor with frame checksum insertion */
8147 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8148 IXGBE_ADVTXD_DCMD_DEXT |
8149 IXGBE_ADVTXD_DCMD_IFCS;
8151 /* set HW vlan bit if vlan is present */
8152 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8153 IXGBE_ADVTXD_DCMD_VLE);
8155 /* set segmentation enable bits for TSO/FSO */
8156 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8157 IXGBE_ADVTXD_DCMD_TSE);
8159 /* set timestamp bit if present */
8160 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8161 IXGBE_ADVTXD_MAC_TSTAMP);
8163 /* insert frame checksum */
8164 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8169 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8170 u32 tx_flags, unsigned int paylen)
8172 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8174 /* enable L4 checksum for TSO and TX checksum offload */
8175 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8176 IXGBE_TX_FLAGS_CSUM,
8177 IXGBE_ADVTXD_POPTS_TXSM);
8179 /* enable IPv4 checksum for TSO */
8180 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8181 IXGBE_TX_FLAGS_IPV4,
8182 IXGBE_ADVTXD_POPTS_IXSM);
8185 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8186 IXGBE_TX_FLAGS_IPSEC,
8187 IXGBE_ADVTXD_POPTS_IPSEC);
8190 * Check Context must be set if Tx switch is enabled, which it
8191 * always is for case where virtual functions are running
8193 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8197 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8200 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8202 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8204 /* Herbert's original patch had:
8205 * smp_mb__after_netif_stop_queue();
8206 * but since that doesn't exist yet, just open code it.
8210 /* We need to check again in a case another CPU has just
8211 * made room available.
8213 if (likely(ixgbe_desc_unused(tx_ring) < size))
8216 /* A reprieve! - use start_queue because it doesn't call schedule */
8217 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8218 ++tx_ring->tx_stats.restart_queue;
8222 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8224 if (likely(ixgbe_desc_unused(tx_ring) >= size))
8227 return __ixgbe_maybe_stop_tx(tx_ring, size);
8230 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8231 struct ixgbe_tx_buffer *first,
8234 struct sk_buff *skb = first->skb;
8235 struct ixgbe_tx_buffer *tx_buffer;
8236 union ixgbe_adv_tx_desc *tx_desc;
8239 unsigned int data_len, size;
8240 u32 tx_flags = first->tx_flags;
8241 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8242 u16 i = tx_ring->next_to_use;
8244 tx_desc = IXGBE_TX_DESC(tx_ring, i);
8246 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8248 size = skb_headlen(skb);
8249 data_len = skb->data_len;
8252 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8253 if (data_len < sizeof(struct fcoe_crc_eof)) {
8254 size -= sizeof(struct fcoe_crc_eof) - data_len;
8257 data_len -= sizeof(struct fcoe_crc_eof);
8262 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8266 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8267 if (dma_mapping_error(tx_ring->dev, dma))
8270 /* record length, and DMA address */
8271 dma_unmap_len_set(tx_buffer, len, size);
8272 dma_unmap_addr_set(tx_buffer, dma, dma);
8274 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8276 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8277 tx_desc->read.cmd_type_len =
8278 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8282 if (i == tx_ring->count) {
8283 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8286 tx_desc->read.olinfo_status = 0;
8288 dma += IXGBE_MAX_DATA_PER_TXD;
8289 size -= IXGBE_MAX_DATA_PER_TXD;
8291 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8294 if (likely(!data_len))
8297 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8301 if (i == tx_ring->count) {
8302 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8305 tx_desc->read.olinfo_status = 0;
8308 size = min_t(unsigned int, data_len, skb_frag_size(frag));
8310 size = skb_frag_size(frag);
8314 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8317 tx_buffer = &tx_ring->tx_buffer_info[i];
8320 /* write last descriptor with RS and EOP bits */
8321 cmd_type |= size | IXGBE_TXD_CMD;
8322 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8324 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8326 /* set the timestamp */
8327 first->time_stamp = jiffies;
8329 skb_tx_timestamp(skb);
8332 * Force memory writes to complete before letting h/w know there
8333 * are new descriptors to fetch. (Only applicable for weak-ordered
8334 * memory model archs, such as IA-64).
8336 * We also need this memory barrier to make certain all of the
8337 * status bits have been updated before next_to_watch is written.
8341 /* set next_to_watch value indicating a packet is present */
8342 first->next_to_watch = tx_desc;
8345 if (i == tx_ring->count)
8348 tx_ring->next_to_use = i;
8350 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8352 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
8353 writel(i, tx_ring->tail);
8358 dev_err(tx_ring->dev, "TX DMA map failed\n");
8360 /* clear dma mappings for failed tx_buffer_info map */
8362 tx_buffer = &tx_ring->tx_buffer_info[i];
8363 if (dma_unmap_len(tx_buffer, len))
8364 dma_unmap_page(tx_ring->dev,
8365 dma_unmap_addr(tx_buffer, dma),
8366 dma_unmap_len(tx_buffer, len),
8368 dma_unmap_len_set(tx_buffer, len, 0);
8369 if (tx_buffer == first)
8372 i += tx_ring->count;
8376 dev_kfree_skb_any(first->skb);
8379 tx_ring->next_to_use = i;
8384 static void ixgbe_atr(struct ixgbe_ring *ring,
8385 struct ixgbe_tx_buffer *first)
8387 struct ixgbe_q_vector *q_vector = ring->q_vector;
8388 union ixgbe_atr_hash_dword input = { .dword = 0 };
8389 union ixgbe_atr_hash_dword common = { .dword = 0 };
8391 unsigned char *network;
8393 struct ipv6hdr *ipv6;
8397 struct sk_buff *skb;
8401 /* if ring doesn't have a interrupt vector, cannot perform ATR */
8405 /* do nothing if sampling is disabled */
8406 if (!ring->atr_sample_rate)
8411 /* currently only IPv4/IPv6 with TCP is supported */
8412 if ((first->protocol != htons(ETH_P_IP)) &&
8413 (first->protocol != htons(ETH_P_IPV6)))
8416 /* snag network header to get L4 type and address */
8418 hdr.network = skb_network_header(skb);
8419 if (unlikely(hdr.network <= skb->data))
8421 if (skb->encapsulation &&
8422 first->protocol == htons(ETH_P_IP) &&
8423 hdr.ipv4->protocol == IPPROTO_UDP) {
8424 struct ixgbe_adapter *adapter = q_vector->adapter;
8426 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8430 /* verify the port is recognized as VXLAN */
8431 if (adapter->vxlan_port &&
8432 udp_hdr(skb)->dest == adapter->vxlan_port)
8433 hdr.network = skb_inner_network_header(skb);
8435 if (adapter->geneve_port &&
8436 udp_hdr(skb)->dest == adapter->geneve_port)
8437 hdr.network = skb_inner_network_header(skb);
8440 /* Make sure we have at least [minimum IPv4 header + TCP]
8441 * or [IPv6 header] bytes
8443 if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8446 /* Currently only IPv4/IPv6 with TCP is supported */
8447 switch (hdr.ipv4->version) {
8449 /* access ihl as u8 to avoid unaligned access on ia64 */
8450 hlen = (hdr.network[0] & 0x0F) << 2;
8451 l4_proto = hdr.ipv4->protocol;
8454 hlen = hdr.network - skb->data;
8455 l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8456 hlen -= hdr.network - skb->data;
8462 if (l4_proto != IPPROTO_TCP)
8465 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8466 hlen + sizeof(struct tcphdr)))
8469 th = (struct tcphdr *)(hdr.network + hlen);
8471 /* skip this packet since the socket is closing */
8475 /* sample on all syn packets or once every atr sample count */
8476 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8479 /* reset sample count */
8480 ring->atr_count = 0;
8482 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8485 * src and dst are inverted, think how the receiver sees them
8487 * The input is broken into two sections, a non-compressed section
8488 * containing vm_pool, vlan_id, and flow_type. The rest of the data
8489 * is XORed together and stored in the compressed dword.
8491 input.formatted.vlan_id = vlan_id;
8494 * since src port and flex bytes occupy the same word XOR them together
8495 * and write the value to source port portion of compressed dword
8497 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8498 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8500 common.port.src ^= th->dest ^ first->protocol;
8501 common.port.dst ^= th->source;
8503 switch (hdr.ipv4->version) {
8505 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8506 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8509 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8510 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8511 hdr.ipv6->saddr.s6_addr32[1] ^
8512 hdr.ipv6->saddr.s6_addr32[2] ^
8513 hdr.ipv6->saddr.s6_addr32[3] ^
8514 hdr.ipv6->daddr.s6_addr32[0] ^
8515 hdr.ipv6->daddr.s6_addr32[1] ^
8516 hdr.ipv6->daddr.s6_addr32[2] ^
8517 hdr.ipv6->daddr.s6_addr32[3];
8523 if (hdr.network != skb_network_header(skb))
8524 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8526 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
8527 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8528 input, common, ring->queue_index);
8532 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8533 struct net_device *sb_dev)
8535 struct ixgbe_adapter *adapter;
8536 struct ixgbe_ring_feature *f;
8540 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8541 struct net_device *vdev = sb_dev;
8543 txq = vdev->tc_to_txq[tc].offset;
8544 txq += reciprocal_scale(skb_get_hash(skb),
8545 vdev->tc_to_txq[tc].count);
8551 * only execute the code below if protocol is FCoE
8552 * or FIP and we have FCoE enabled on the adapter
8554 switch (vlan_get_protocol(skb)) {
8555 case htons(ETH_P_FCOE):
8556 case htons(ETH_P_FIP):
8557 adapter = netdev_priv(dev);
8559 if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8563 return netdev_pick_tx(dev, skb, sb_dev);
8566 f = &adapter->ring_feature[RING_F_FCOE];
8568 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8571 while (txq >= f->indices)
8574 return txq + f->offset;
8578 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8579 struct xdp_frame *xdpf)
8581 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8582 struct ixgbe_tx_buffer *tx_buffer;
8583 union ixgbe_adv_tx_desc *tx_desc;
8590 if (unlikely(!ixgbe_desc_unused(ring)))
8591 return IXGBE_XDP_CONSUMED;
8593 dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
8594 if (dma_mapping_error(ring->dev, dma))
8595 return IXGBE_XDP_CONSUMED;
8597 /* record the location of the first descriptor for this packet */
8598 tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8599 tx_buffer->bytecount = len;
8600 tx_buffer->gso_segs = 1;
8601 tx_buffer->protocol = 0;
8603 i = ring->next_to_use;
8604 tx_desc = IXGBE_TX_DESC(ring, i);
8606 dma_unmap_len_set(tx_buffer, len, len);
8607 dma_unmap_addr_set(tx_buffer, dma, dma);
8608 tx_buffer->xdpf = xdpf;
8610 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8612 /* put descriptor type bits */
8613 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8614 IXGBE_ADVTXD_DCMD_DEXT |
8615 IXGBE_ADVTXD_DCMD_IFCS;
8616 cmd_type |= len | IXGBE_TXD_CMD;
8617 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8618 tx_desc->read.olinfo_status =
8619 cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8621 /* Avoid any potential race with xdp_xmit and cleanup */
8624 /* set next_to_watch value indicating a packet is present */
8626 if (i == ring->count)
8629 tx_buffer->next_to_watch = tx_desc;
8630 ring->next_to_use = i;
8632 return IXGBE_XDP_TX;
8635 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8636 struct ixgbe_adapter *adapter,
8637 struct ixgbe_ring *tx_ring)
8639 struct ixgbe_tx_buffer *first;
8643 u16 count = TXD_USE_COUNT(skb_headlen(skb));
8644 struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8645 __be16 protocol = skb->protocol;
8649 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8650 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8651 * + 2 desc gap to keep tail from touching head,
8652 * + 1 desc for context descriptor,
8653 * otherwise try next time
8655 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8656 count += TXD_USE_COUNT(skb_frag_size(
8657 &skb_shinfo(skb)->frags[f]));
8659 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8660 tx_ring->tx_stats.tx_busy++;
8661 return NETDEV_TX_BUSY;
8664 /* record the location of the first descriptor for this packet */
8665 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8667 first->bytecount = skb->len;
8668 first->gso_segs = 1;
8670 /* if we have a HW VLAN tag being added default to the HW one */
8671 if (skb_vlan_tag_present(skb)) {
8672 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8673 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8674 /* else if it is a SW VLAN check the next protocol and store the tag */
8675 } else if (protocol == htons(ETH_P_8021Q)) {
8676 struct vlan_hdr *vhdr, _vhdr;
8677 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8681 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8682 IXGBE_TX_FLAGS_VLAN_SHIFT;
8683 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8685 protocol = vlan_get_protocol(skb);
8687 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8688 adapter->ptp_clock) {
8689 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
8690 !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8692 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8693 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8695 /* schedule check for Tx timestamp */
8696 adapter->ptp_tx_skb = skb_get(skb);
8697 adapter->ptp_tx_start = jiffies;
8698 schedule_work(&adapter->ptp_tx_work);
8700 adapter->tx_hwtstamp_skipped++;
8704 #ifdef CONFIG_PCI_IOV
8706 * Use the l2switch_enable flag - would be false if the DMA
8707 * Tx switch had been disabled.
8709 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8710 tx_flags |= IXGBE_TX_FLAGS_CC;
8713 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8714 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8715 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8716 (skb->priority != TC_PRIO_CONTROL))) {
8717 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8718 tx_flags |= (skb->priority & 0x7) <<
8719 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8720 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8721 struct vlan_ethhdr *vhdr;
8723 if (skb_cow_head(skb, 0))
8725 vhdr = (struct vlan_ethhdr *)skb->data;
8726 vhdr->h_vlan_TCI = htons(tx_flags >>
8727 IXGBE_TX_FLAGS_VLAN_SHIFT);
8729 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8733 /* record initial flags and protocol */
8734 first->tx_flags = tx_flags;
8735 first->protocol = protocol;
8738 /* setup tx offload for FCoE */
8739 if ((protocol == htons(ETH_P_FCOE)) &&
8740 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8741 tso = ixgbe_fso(tx_ring, first, &hdr_len);
8748 #endif /* IXGBE_FCOE */
8750 #ifdef CONFIG_IXGBE_IPSEC
8751 if (xfrm_offload(skb) &&
8752 !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8755 tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8759 ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8761 /* add the ATR filter if ATR is on */
8762 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8763 ixgbe_atr(tx_ring, first);
8767 #endif /* IXGBE_FCOE */
8768 if (ixgbe_tx_map(tx_ring, first, hdr_len))
8769 goto cleanup_tx_timestamp;
8771 return NETDEV_TX_OK;
8774 dev_kfree_skb_any(first->skb);
8776 cleanup_tx_timestamp:
8777 if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8778 dev_kfree_skb_any(adapter->ptp_tx_skb);
8779 adapter->ptp_tx_skb = NULL;
8780 cancel_work_sync(&adapter->ptp_tx_work);
8781 clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8784 return NETDEV_TX_OK;
8787 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8788 struct net_device *netdev,
8789 struct ixgbe_ring *ring)
8791 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8792 struct ixgbe_ring *tx_ring;
8795 * The minimum packet size for olinfo paylen is 17 so pad the skb
8796 * in order to meet this minimum size requirement.
8798 if (skb_put_padto(skb, 17))
8799 return NETDEV_TX_OK;
8801 tx_ring = ring ? ring : adapter->tx_ring[skb_get_queue_mapping(skb)];
8802 if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8803 return NETDEV_TX_BUSY;
8805 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8808 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8809 struct net_device *netdev)
8811 return __ixgbe_xmit_frame(skb, netdev, NULL);
8815 * ixgbe_set_mac - Change the Ethernet Address of the NIC
8816 * @netdev: network interface device structure
8817 * @p: pointer to an address structure
8819 * Returns 0 on success, negative on failure
8821 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8823 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8824 struct ixgbe_hw *hw = &adapter->hw;
8825 struct sockaddr *addr = p;
8827 if (!is_valid_ether_addr(addr->sa_data))
8828 return -EADDRNOTAVAIL;
8830 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8831 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8833 ixgbe_mac_set_default_filter(adapter);
8839 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8841 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8842 struct ixgbe_hw *hw = &adapter->hw;
8846 if (adapter->mii_bus) {
8849 if (devad != MDIO_DEVAD_NONE)
8850 regnum |= (devad << 16) | MII_ADDR_C45;
8852 return mdiobus_read(adapter->mii_bus, prtad, regnum);
8855 if (prtad != hw->phy.mdio.prtad)
8857 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8863 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8864 u16 addr, u16 value)
8866 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8867 struct ixgbe_hw *hw = &adapter->hw;
8869 if (adapter->mii_bus) {
8872 if (devad != MDIO_DEVAD_NONE)
8873 regnum |= (devad << 16) | MII_ADDR_C45;
8875 return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8878 if (prtad != hw->phy.mdio.prtad)
8880 return hw->phy.ops.write_reg(hw, addr, devad, value);
8883 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8885 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8889 return ixgbe_ptp_set_ts_config(adapter, req);
8891 return ixgbe_ptp_get_ts_config(adapter, req);
8893 if (!adapter->hw.phy.ops.read_reg)
8897 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8902 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8904 * @dev: network interface device structure
8906 * Returns non-zero on failure
8908 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8911 struct ixgbe_adapter *adapter = netdev_priv(dev);
8912 struct ixgbe_hw *hw = &adapter->hw;
8914 if (is_valid_ether_addr(hw->mac.san_addr)) {
8916 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8919 /* update SAN MAC vmdq pool selection */
8920 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8926 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8928 * @dev: network interface device structure
8930 * Returns non-zero on failure
8932 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8935 struct ixgbe_adapter *adapter = netdev_priv(dev);
8936 struct ixgbe_mac_info *mac = &adapter->hw.mac;
8938 if (is_valid_ether_addr(mac->san_addr)) {
8940 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8946 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8947 struct ixgbe_ring *ring)
8954 start = u64_stats_fetch_begin_irq(&ring->syncp);
8955 packets = ring->stats.packets;
8956 bytes = ring->stats.bytes;
8957 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8958 stats->tx_packets += packets;
8959 stats->tx_bytes += bytes;
8963 static void ixgbe_get_stats64(struct net_device *netdev,
8964 struct rtnl_link_stats64 *stats)
8966 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8970 for (i = 0; i < adapter->num_rx_queues; i++) {
8971 struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8977 start = u64_stats_fetch_begin_irq(&ring->syncp);
8978 packets = ring->stats.packets;
8979 bytes = ring->stats.bytes;
8980 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8981 stats->rx_packets += packets;
8982 stats->rx_bytes += bytes;
8986 for (i = 0; i < adapter->num_tx_queues; i++) {
8987 struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8989 ixgbe_get_ring_stats64(stats, ring);
8991 for (i = 0; i < adapter->num_xdp_queues; i++) {
8992 struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8994 ixgbe_get_ring_stats64(stats, ring);
8998 /* following stats updated by ixgbe_watchdog_task() */
8999 stats->multicast = netdev->stats.multicast;
9000 stats->rx_errors = netdev->stats.rx_errors;
9001 stats->rx_length_errors = netdev->stats.rx_length_errors;
9002 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
9003 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
9006 #ifdef CONFIG_IXGBE_DCB
9008 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
9009 * @adapter: pointer to ixgbe_adapter
9010 * @tc: number of traffic classes currently enabled
9012 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
9013 * 802.1Q priority maps to a packet buffer that exists.
9015 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
9017 struct ixgbe_hw *hw = &adapter->hw;
9021 /* 82598 have a static priority to TC mapping that can not
9022 * be changed so no validation is needed.
9024 if (hw->mac.type == ixgbe_mac_82598EB)
9027 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
9030 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
9031 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
9033 /* If up2tc is out of bounds default to zero */
9035 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
9039 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
9045 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
9046 * @adapter: Pointer to adapter struct
9048 * Populate the netdev user priority to tc map
9050 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
9052 struct net_device *dev = adapter->netdev;
9053 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9054 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9057 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9060 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9061 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9063 tc = ets->prio_tc[prio];
9065 netdev_set_prio_tc_map(dev, prio, tc);
9069 #endif /* CONFIG_IXGBE_DCB */
9070 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev, void *data)
9072 struct ixgbe_adapter *adapter = data;
9073 struct ixgbe_fwd_adapter *accel;
9076 /* we only care about macvlans... */
9077 if (!netif_is_macvlan(vdev))
9080 /* that have hardware offload enabled... */
9081 accel = macvlan_accel_priv(vdev);
9085 /* If we can relocate to a different bit do so */
9086 pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9087 if (pool < adapter->num_rx_pools) {
9088 set_bit(pool, adapter->fwd_bitmask);
9093 /* if we cannot find a free pool then disable the offload */
9094 netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9095 macvlan_release_l2fw_offload(vdev);
9097 /* unbind the queues and drop the subordinate channel config */
9098 netdev_unbind_sb_channel(adapter->netdev, vdev);
9099 netdev_set_sb_channel(vdev, 0);
9106 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9108 struct ixgbe_adapter *adapter = netdev_priv(dev);
9110 /* flush any stale bits out of the fwd bitmask */
9111 bitmap_clear(adapter->fwd_bitmask, 1, 63);
9113 /* walk through upper devices reassigning pools */
9114 netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9119 * ixgbe_setup_tc - configure net_device for multiple traffic classes
9121 * @dev: net device to configure
9122 * @tc: number of traffic classes to enable
9124 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9126 struct ixgbe_adapter *adapter = netdev_priv(dev);
9127 struct ixgbe_hw *hw = &adapter->hw;
9129 /* Hardware supports up to 8 traffic classes */
9130 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9133 if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9136 /* Hardware has to reinitialize queues and interrupts to
9137 * match packet buffer alignment. Unfortunately, the
9138 * hardware is not flexible enough to do this dynamically.
9140 if (netif_running(dev))
9143 ixgbe_reset(adapter);
9145 ixgbe_clear_interrupt_scheme(adapter);
9147 #ifdef CONFIG_IXGBE_DCB
9149 if (adapter->xdp_prog) {
9150 e_warn(probe, "DCB is not supported with XDP\n");
9152 ixgbe_init_interrupt_scheme(adapter);
9153 if (netif_running(dev))
9158 netdev_set_num_tc(dev, tc);
9159 ixgbe_set_prio_tc_map(adapter);
9161 adapter->hw_tcs = tc;
9162 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9164 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9165 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9166 adapter->hw.fc.requested_mode = ixgbe_fc_none;
9169 netdev_reset_tc(dev);
9171 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9172 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9174 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9175 adapter->hw_tcs = tc;
9177 adapter->temp_dcb_cfg.pfc_mode_enable = false;
9178 adapter->dcb_cfg.pfc_mode_enable = false;
9181 ixgbe_validate_rtr(adapter, tc);
9183 #endif /* CONFIG_IXGBE_DCB */
9184 ixgbe_init_interrupt_scheme(adapter);
9186 ixgbe_defrag_macvlan_pools(dev);
9188 if (netif_running(dev))
9189 return ixgbe_open(dev);
9194 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9195 struct tc_cls_u32_offload *cls)
9197 u32 hdl = cls->knode.handle;
9198 u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9199 u32 loc = cls->knode.handle & 0xfffff;
9201 struct ixgbe_jump_table *jump = NULL;
9203 if (loc > IXGBE_MAX_HW_ENTRIES)
9206 if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9209 /* Clear this filter in the link data it is associated with */
9210 if (uhtid != 0x800) {
9211 jump = adapter->jump_tables[uhtid];
9214 if (!test_bit(loc - 1, jump->child_loc_map))
9216 clear_bit(loc - 1, jump->child_loc_map);
9219 /* Check if the filter being deleted is a link */
9220 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9221 jump = adapter->jump_tables[i];
9222 if (jump && jump->link_hdl == hdl) {
9223 /* Delete filters in the hardware in the child hash
9224 * table associated with this link
9226 for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9227 if (!test_bit(j, jump->child_loc_map))
9229 spin_lock(&adapter->fdir_perfect_lock);
9230 err = ixgbe_update_ethtool_fdir_entry(adapter,
9233 spin_unlock(&adapter->fdir_perfect_lock);
9234 clear_bit(j, jump->child_loc_map);
9236 /* Remove resources for this link */
9240 adapter->jump_tables[i] = NULL;
9245 spin_lock(&adapter->fdir_perfect_lock);
9246 err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9247 spin_unlock(&adapter->fdir_perfect_lock);
9251 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9252 struct tc_cls_u32_offload *cls)
9254 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9256 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9259 /* This ixgbe devices do not support hash tables at the moment
9260 * so abort when given hash tables.
9262 if (cls->hnode.divisor > 0)
9265 set_bit(uhtid - 1, &adapter->tables);
9269 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9270 struct tc_cls_u32_offload *cls)
9272 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9274 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9277 clear_bit(uhtid - 1, &adapter->tables);
9281 #ifdef CONFIG_NET_CLS_ACT
9282 struct upper_walk_data {
9283 struct ixgbe_adapter *adapter;
9289 static int get_macvlan_queue(struct net_device *upper, void *_data)
9291 if (netif_is_macvlan(upper)) {
9292 struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9293 struct upper_walk_data *data = _data;
9294 struct ixgbe_adapter *adapter = data->adapter;
9295 int ifindex = data->ifindex;
9297 if (vadapter && upper->ifindex == ifindex) {
9298 data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9299 data->action = data->queue;
9307 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9308 u8 *queue, u64 *action)
9310 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9311 unsigned int num_vfs = adapter->num_vfs, vf;
9312 struct upper_walk_data data;
9313 struct net_device *upper;
9315 /* redirect to a SRIOV VF */
9316 for (vf = 0; vf < num_vfs; ++vf) {
9317 upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9318 if (upper->ifindex == ifindex) {
9319 *queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9321 *action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9326 /* redirect to a offloaded macvlan netdev */
9327 data.adapter = adapter;
9328 data.ifindex = ifindex;
9331 if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9332 get_macvlan_queue, &data)) {
9333 *action = data.action;
9334 *queue = data.queue;
9342 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9343 struct tcf_exts *exts, u64 *action, u8 *queue)
9345 const struct tc_action *a;
9348 if (!tcf_exts_has_actions(exts))
9351 tcf_exts_for_each_action(i, a, exts) {
9353 if (is_tcf_gact_shot(a)) {
9354 *action = IXGBE_FDIR_DROP_QUEUE;
9355 *queue = IXGBE_FDIR_DROP_QUEUE;
9359 /* Redirect to a VF or a offloaded macvlan */
9360 if (is_tcf_mirred_egress_redirect(a)) {
9361 struct net_device *dev = tcf_mirred_dev(a);
9365 return handle_redirect_action(adapter, dev->ifindex,
9375 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9376 struct tcf_exts *exts, u64 *action, u8 *queue)
9380 #endif /* CONFIG_NET_CLS_ACT */
9382 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9383 union ixgbe_atr_input *mask,
9384 struct tc_cls_u32_offload *cls,
9385 struct ixgbe_mat_field *field_ptr,
9386 struct ixgbe_nexthdr *nexthdr)
9390 bool found_entry = false, found_jump_field = false;
9392 for (i = 0; i < cls->knode.sel->nkeys; i++) {
9393 off = cls->knode.sel->keys[i].off;
9394 val = cls->knode.sel->keys[i].val;
9395 m = cls->knode.sel->keys[i].mask;
9397 for (j = 0; field_ptr[j].val; j++) {
9398 if (field_ptr[j].off == off) {
9399 field_ptr[j].val(input, mask, (__force u32)val,
9401 input->filter.formatted.flow_type |=
9408 if (nexthdr->off == cls->knode.sel->keys[i].off &&
9410 (__force u32)cls->knode.sel->keys[i].val &&
9412 (__force u32)cls->knode.sel->keys[i].mask)
9413 found_jump_field = true;
9419 if (nexthdr && !found_jump_field)
9425 mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9426 IXGBE_ATR_L4TYPE_MASK;
9428 if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9429 mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9434 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9435 struct tc_cls_u32_offload *cls)
9437 __be16 protocol = cls->common.protocol;
9438 u32 loc = cls->knode.handle & 0xfffff;
9439 struct ixgbe_hw *hw = &adapter->hw;
9440 struct ixgbe_mat_field *field_ptr;
9441 struct ixgbe_fdir_filter *input = NULL;
9442 union ixgbe_atr_input *mask = NULL;
9443 struct ixgbe_jump_table *jump = NULL;
9444 int i, err = -EINVAL;
9446 u32 uhtid, link_uhtid;
9448 uhtid = TC_U32_USERHTID(cls->knode.handle);
9449 link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9451 /* At the moment cls_u32 jumps to network layer and skips past
9452 * L2 headers. The canonical method to match L2 frames is to use
9453 * negative values. However this is error prone at best but really
9454 * just broken because there is no way to "know" what sort of hdr
9455 * is in front of the network layer. Fix cls_u32 to support L2
9456 * headers when needed.
9458 if (protocol != htons(ETH_P_IP))
9461 if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9462 e_err(drv, "Location out of range\n");
9466 /* cls u32 is a graph starting at root node 0x800. The driver tracks
9467 * links and also the fields used to advance the parser across each
9468 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9469 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9470 * To add support for new nodes update ixgbe_model.h parse structures
9471 * this function _should_ be generic try not to hardcode values here.
9473 if (uhtid == 0x800) {
9474 field_ptr = (adapter->jump_tables[0])->mat;
9476 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9478 if (!adapter->jump_tables[uhtid])
9480 field_ptr = (adapter->jump_tables[uhtid])->mat;
9486 /* At this point we know the field_ptr is valid and need to either
9487 * build cls_u32 link or attach filter. Because adding a link to
9488 * a handle that does not exist is invalid and the same for adding
9489 * rules to handles that don't exist.
9493 struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9495 if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9498 if (!test_bit(link_uhtid - 1, &adapter->tables))
9501 /* Multiple filters as links to the same hash table are not
9502 * supported. To add a new filter with the same next header
9503 * but different match/jump conditions, create a new hash table
9506 if (adapter->jump_tables[link_uhtid] &&
9507 (adapter->jump_tables[link_uhtid])->link_hdl) {
9508 e_err(drv, "Link filter exists for link: %x\n",
9513 for (i = 0; nexthdr[i].jump; i++) {
9514 if (nexthdr[i].o != cls->knode.sel->offoff ||
9515 nexthdr[i].s != cls->knode.sel->offshift ||
9517 (__force u32)cls->knode.sel->offmask)
9520 jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9523 input = kzalloc(sizeof(*input), GFP_KERNEL);
9528 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9533 jump->input = input;
9535 jump->link_hdl = cls->knode.handle;
9537 err = ixgbe_clsu32_build_input(input, mask, cls,
9538 field_ptr, &nexthdr[i]);
9540 jump->mat = nexthdr[i].jump;
9541 adapter->jump_tables[link_uhtid] = jump;
9552 input = kzalloc(sizeof(*input), GFP_KERNEL);
9555 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9561 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9562 if ((adapter->jump_tables[uhtid])->input)
9563 memcpy(input, (adapter->jump_tables[uhtid])->input,
9565 if ((adapter->jump_tables[uhtid])->mask)
9566 memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9569 /* Lookup in all child hash tables if this location is already
9570 * filled with a filter
9572 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9573 struct ixgbe_jump_table *link = adapter->jump_tables[i];
9575 if (link && (test_bit(loc - 1, link->child_loc_map))) {
9576 e_err(drv, "Filter exists in location: %x\n",
9583 err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9587 err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9592 input->sw_idx = loc;
9594 spin_lock(&adapter->fdir_perfect_lock);
9596 if (hlist_empty(&adapter->fdir_filter_list)) {
9597 memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9598 err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9600 goto err_out_w_lock;
9601 } else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9603 goto err_out_w_lock;
9606 ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9607 err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9608 input->sw_idx, queue);
9610 ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9611 spin_unlock(&adapter->fdir_perfect_lock);
9613 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9614 set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9619 spin_unlock(&adapter->fdir_perfect_lock);
9629 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9630 struct tc_cls_u32_offload *cls_u32)
9632 switch (cls_u32->command) {
9633 case TC_CLSU32_NEW_KNODE:
9634 case TC_CLSU32_REPLACE_KNODE:
9635 return ixgbe_configure_clsu32(adapter, cls_u32);
9636 case TC_CLSU32_DELETE_KNODE:
9637 return ixgbe_delete_clsu32(adapter, cls_u32);
9638 case TC_CLSU32_NEW_HNODE:
9639 case TC_CLSU32_REPLACE_HNODE:
9640 return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9641 case TC_CLSU32_DELETE_HNODE:
9642 return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9648 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9651 struct ixgbe_adapter *adapter = cb_priv;
9653 if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9657 case TC_SETUP_CLSU32:
9658 return ixgbe_setup_tc_cls_u32(adapter, type_data);
9664 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9665 struct tc_mqprio_qopt *mqprio)
9667 mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9668 return ixgbe_setup_tc(dev, mqprio->num_tc);
9671 static LIST_HEAD(ixgbe_block_cb_list);
9673 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9676 struct ixgbe_adapter *adapter = netdev_priv(dev);
9679 case TC_SETUP_BLOCK:
9680 return flow_block_cb_setup_simple(type_data,
9681 &ixgbe_block_cb_list,
9682 ixgbe_setup_tc_block_cb,
9683 adapter, adapter, true);
9684 case TC_SETUP_QDISC_MQPRIO:
9685 return ixgbe_setup_tc_mqprio(dev, type_data);
9691 #ifdef CONFIG_PCI_IOV
9692 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9694 struct net_device *netdev = adapter->netdev;
9697 ixgbe_setup_tc(netdev, adapter->hw_tcs);
9702 void ixgbe_do_reset(struct net_device *netdev)
9704 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9706 if (netif_running(netdev))
9707 ixgbe_reinit_locked(adapter);
9709 ixgbe_reset(adapter);
9712 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9713 netdev_features_t features)
9715 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9717 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9718 if (!(features & NETIF_F_RXCSUM))
9719 features &= ~NETIF_F_LRO;
9721 /* Turn off LRO if not RSC capable */
9722 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9723 features &= ~NETIF_F_LRO;
9725 if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9726 e_dev_err("LRO is not supported with XDP\n");
9727 features &= ~NETIF_F_LRO;
9733 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9735 int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9738 /* go back to full RSS if we're not running SR-IOV */
9739 if (!adapter->ring_feature[RING_F_VMDQ].offset)
9740 adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9741 IXGBE_FLAG_SRIOV_ENABLED);
9743 adapter->ring_feature[RING_F_RSS].limit = rss;
9744 adapter->ring_feature[RING_F_VMDQ].limit = 1;
9746 ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9749 static int ixgbe_set_features(struct net_device *netdev,
9750 netdev_features_t features)
9752 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9753 netdev_features_t changed = netdev->features ^ features;
9754 bool need_reset = false;
9756 /* Make sure RSC matches LRO, reset if change */
9757 if (!(features & NETIF_F_LRO)) {
9758 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9760 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9761 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9762 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9763 if (adapter->rx_itr_setting == 1 ||
9764 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9765 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9767 } else if ((changed ^ features) & NETIF_F_LRO) {
9768 e_info(probe, "rx-usecs set too low, "
9774 * Check if Flow Director n-tuple support or hw_tc support was
9775 * enabled or disabled. If the state changed, we need to reset.
9777 if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9778 /* turn off ATR, enable perfect filters and reset */
9779 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9782 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9783 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9785 /* turn off perfect filters, enable ATR and reset */
9786 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9789 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9791 /* We cannot enable ATR if SR-IOV is enabled */
9792 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9793 /* We cannot enable ATR if we have 2 or more tcs */
9794 (adapter->hw_tcs > 1) ||
9795 /* We cannot enable ATR if RSS is disabled */
9796 (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9797 /* A sample rate of 0 indicates ATR disabled */
9798 (!adapter->atr_sample_rate))
9799 ; /* do nothing not supported */
9800 else /* otherwise supported and set the flag */
9801 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9804 if (changed & NETIF_F_RXALL)
9807 netdev->features = features;
9809 if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) {
9810 if (features & NETIF_F_RXCSUM) {
9811 adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9813 u32 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9815 ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9819 if ((adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)) {
9820 if (features & NETIF_F_RXCSUM) {
9821 adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9823 u32 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9825 ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9829 if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9830 ixgbe_reset_l2fw_offload(adapter);
9831 else if (need_reset)
9832 ixgbe_do_reset(netdev);
9833 else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9834 NETIF_F_HW_VLAN_CTAG_FILTER))
9835 ixgbe_set_rx_mode(netdev);
9841 * ixgbe_add_udp_tunnel_port - Get notifications about adding UDP tunnel ports
9842 * @dev: The port's netdev
9843 * @ti: Tunnel endpoint information
9845 static void ixgbe_add_udp_tunnel_port(struct net_device *dev,
9846 struct udp_tunnel_info *ti)
9848 struct ixgbe_adapter *adapter = netdev_priv(dev);
9849 struct ixgbe_hw *hw = &adapter->hw;
9850 __be16 port = ti->port;
9854 if (ti->sa_family != AF_INET)
9858 case UDP_TUNNEL_TYPE_VXLAN:
9859 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9862 if (adapter->vxlan_port == port)
9865 if (adapter->vxlan_port) {
9867 "VXLAN port %d set, not adding port %d\n",
9868 ntohs(adapter->vxlan_port),
9873 adapter->vxlan_port = port;
9875 case UDP_TUNNEL_TYPE_GENEVE:
9876 if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9879 if (adapter->geneve_port == port)
9882 if (adapter->geneve_port) {
9884 "GENEVE port %d set, not adding port %d\n",
9885 ntohs(adapter->geneve_port),
9890 port_shift = IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT;
9891 adapter->geneve_port = port;
9897 reg = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) | ntohs(port) << port_shift;
9898 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, reg);
9902 * ixgbe_del_udp_tunnel_port - Get notifications about removing UDP tunnel ports
9903 * @dev: The port's netdev
9904 * @ti: Tunnel endpoint information
9906 static void ixgbe_del_udp_tunnel_port(struct net_device *dev,
9907 struct udp_tunnel_info *ti)
9909 struct ixgbe_adapter *adapter = netdev_priv(dev);
9912 if (ti->type != UDP_TUNNEL_TYPE_VXLAN &&
9913 ti->type != UDP_TUNNEL_TYPE_GENEVE)
9916 if (ti->sa_family != AF_INET)
9920 case UDP_TUNNEL_TYPE_VXLAN:
9921 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9924 if (adapter->vxlan_port != ti->port) {
9925 netdev_info(dev, "VXLAN port %d not found\n",
9930 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9932 case UDP_TUNNEL_TYPE_GENEVE:
9933 if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9936 if (adapter->geneve_port != ti->port) {
9937 netdev_info(dev, "GENEVE port %d not found\n",
9942 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9948 ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9949 adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9952 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9953 struct net_device *dev,
9954 const unsigned char *addr, u16 vid,
9956 struct netlink_ext_ack *extack)
9958 /* guarantee we can provide a unique filter for the unicast address */
9959 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9960 struct ixgbe_adapter *adapter = netdev_priv(dev);
9961 u16 pool = VMDQ_P(0);
9963 if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9967 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9971 * ixgbe_configure_bridge_mode - set various bridge modes
9972 * @adapter: the private structure
9973 * @mode: requested bridge mode
9975 * Configure some settings require for various bridge modes.
9977 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9980 struct ixgbe_hw *hw = &adapter->hw;
9981 unsigned int p, num_pools;
9985 case BRIDGE_MODE_VEPA:
9986 /* disable Tx loopback, rely on switch hairpin mode */
9987 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9989 /* must enable Rx switching replication to allow multicast
9990 * packet reception on all VFs, and to enable source address
9993 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9994 vmdctl |= IXGBE_VT_CTL_REPLEN;
9995 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9997 /* enable Rx source address pruning. Note, this requires
9998 * replication to be enabled or else it does nothing.
10000 num_pools = adapter->num_vfs + adapter->num_rx_pools;
10001 for (p = 0; p < num_pools; p++) {
10002 if (hw->mac.ops.set_source_address_pruning)
10003 hw->mac.ops.set_source_address_pruning(hw,
10008 case BRIDGE_MODE_VEB:
10009 /* enable Tx loopback for internal VF/PF communication */
10010 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
10011 IXGBE_PFDTXGSWC_VT_LBEN);
10013 /* disable Rx switching replication unless we have SR-IOV
10014 * virtual functions
10016 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
10017 if (!adapter->num_vfs)
10018 vmdctl &= ~IXGBE_VT_CTL_REPLEN;
10019 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
10021 /* disable Rx source address pruning, since we don't expect to
10022 * be receiving external loopback of our transmitted frames.
10024 num_pools = adapter->num_vfs + adapter->num_rx_pools;
10025 for (p = 0; p < num_pools; p++) {
10026 if (hw->mac.ops.set_source_address_pruning)
10027 hw->mac.ops.set_source_address_pruning(hw,
10036 adapter->bridge_mode = mode;
10038 e_info(drv, "enabling bridge mode: %s\n",
10039 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
10044 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
10045 struct nlmsghdr *nlh, u16 flags,
10046 struct netlink_ext_ack *extack)
10048 struct ixgbe_adapter *adapter = netdev_priv(dev);
10049 struct nlattr *attr, *br_spec;
10052 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10053 return -EOPNOTSUPP;
10055 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
10059 nla_for_each_nested(attr, br_spec, rem) {
10063 if (nla_type(attr) != IFLA_BRIDGE_MODE)
10066 if (nla_len(attr) < sizeof(mode))
10069 mode = nla_get_u16(attr);
10070 status = ixgbe_configure_bridge_mode(adapter, mode);
10080 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
10081 struct net_device *dev,
10082 u32 filter_mask, int nlflags)
10084 struct ixgbe_adapter *adapter = netdev_priv(dev);
10086 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10089 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
10090 adapter->bridge_mode, 0, 0, nlflags,
10091 filter_mask, NULL);
10094 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
10096 struct ixgbe_adapter *adapter = netdev_priv(pdev);
10097 struct ixgbe_fwd_adapter *accel;
10098 int tcs = adapter->hw_tcs ? : 1;
10101 if (adapter->xdp_prog) {
10102 e_warn(probe, "L2FW offload is not supported with XDP\n");
10103 return ERR_PTR(-EINVAL);
10106 /* The hardware supported by ixgbe only filters on the destination MAC
10107 * address. In order to avoid issues we only support offloading modes
10108 * where the hardware can actually provide the functionality.
10110 if (!macvlan_supports_dest_filter(vdev))
10111 return ERR_PTR(-EMEDIUMTYPE);
10113 /* We need to lock down the macvlan to be a single queue device so that
10114 * we can reuse the tc_to_txq field in the macvlan netdev to represent
10115 * the queue mapping to our netdev.
10117 if (netif_is_multiqueue(vdev))
10118 return ERR_PTR(-ERANGE);
10120 pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
10121 if (pool == adapter->num_rx_pools) {
10122 u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
10123 u16 reserved_pools;
10125 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
10126 adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
10127 adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
10128 return ERR_PTR(-EBUSY);
10130 /* Hardware has a limited number of available pools. Each VF,
10131 * and the PF require a pool. Check to ensure we don't
10132 * attempt to use more then the available number of pools.
10134 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
10135 return ERR_PTR(-EBUSY);
10137 /* Enable VMDq flag so device will be set in VM mode */
10138 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
10139 IXGBE_FLAG_SRIOV_ENABLED;
10141 /* Try to reserve as many queues per pool as possible,
10142 * we start with the configurations that support 4 queues
10143 * per pools, followed by 2, and then by just 1 per pool.
10145 if (used_pools < 32 && adapter->num_rx_pools < 16)
10146 reserved_pools = min_t(u16,
10148 16 - adapter->num_rx_pools);
10149 else if (adapter->num_rx_pools < 32)
10150 reserved_pools = min_t(u16,
10152 32 - adapter->num_rx_pools);
10154 reserved_pools = 64 - used_pools;
10157 if (!reserved_pools)
10158 return ERR_PTR(-EBUSY);
10160 adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
10162 /* Force reinit of ring allocation with VMDQ enabled */
10163 err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
10165 return ERR_PTR(err);
10167 if (pool >= adapter->num_rx_pools)
10168 return ERR_PTR(-ENOMEM);
10171 accel = kzalloc(sizeof(*accel), GFP_KERNEL);
10173 return ERR_PTR(-ENOMEM);
10175 set_bit(pool, adapter->fwd_bitmask);
10176 netdev_set_sb_channel(vdev, pool);
10177 accel->pool = pool;
10178 accel->netdev = vdev;
10180 if (!netif_running(pdev))
10183 err = ixgbe_fwd_ring_up(adapter, accel);
10185 return ERR_PTR(err);
10190 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10192 struct ixgbe_fwd_adapter *accel = priv;
10193 struct ixgbe_adapter *adapter = netdev_priv(pdev);
10194 unsigned int rxbase = accel->rx_base_queue;
10197 /* delete unicast filter associated with offloaded interface */
10198 ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10199 VMDQ_P(accel->pool));
10201 /* Allow remaining Rx packets to get flushed out of the
10202 * Rx FIFO before we drop the netdev for the ring.
10204 usleep_range(10000, 20000);
10206 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10207 struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10208 struct ixgbe_q_vector *qv = ring->q_vector;
10210 /* Make sure we aren't processing any packets and clear
10211 * netdev to shut down the ring.
10213 if (netif_running(adapter->netdev))
10214 napi_synchronize(&qv->napi);
10215 ring->netdev = NULL;
10218 /* unbind the queues and drop the subordinate channel config */
10219 netdev_unbind_sb_channel(pdev, accel->netdev);
10220 netdev_set_sb_channel(accel->netdev, 0);
10222 clear_bit(accel->pool, adapter->fwd_bitmask);
10226 #define IXGBE_MAX_MAC_HDR_LEN 127
10227 #define IXGBE_MAX_NETWORK_HDR_LEN 511
10229 static netdev_features_t
10230 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10231 netdev_features_t features)
10233 unsigned int network_hdr_len, mac_hdr_len;
10235 /* Make certain the headers can be described by a context descriptor */
10236 mac_hdr_len = skb_network_header(skb) - skb->data;
10237 if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10238 return features & ~(NETIF_F_HW_CSUM |
10240 NETIF_F_GSO_UDP_L4 |
10241 NETIF_F_HW_VLAN_CTAG_TX |
10245 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10246 if (unlikely(network_hdr_len > IXGBE_MAX_NETWORK_HDR_LEN))
10247 return features & ~(NETIF_F_HW_CSUM |
10249 NETIF_F_GSO_UDP_L4 |
10253 /* We can only support IPV4 TSO in tunnels if we can mangle the
10254 * inner IP ID field, so strip TSO if MANGLEID is not supported.
10255 * IPsec offoad sets skb->encapsulation but still can handle
10256 * the TSO, so it's the exception.
10258 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10259 #ifdef CONFIG_IXGBE_IPSEC
10260 if (!secpath_exists(skb))
10262 features &= ~NETIF_F_TSO;
10268 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10270 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10271 struct ixgbe_adapter *adapter = netdev_priv(dev);
10272 struct bpf_prog *old_prog;
10275 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10278 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10281 /* verify ixgbe ring attributes are sufficient for XDP */
10282 for (i = 0; i < adapter->num_rx_queues; i++) {
10283 struct ixgbe_ring *ring = adapter->rx_ring[i];
10285 if (ring_is_rsc_enabled(ring))
10288 if (frame_size > ixgbe_rx_bufsz(ring))
10292 if (nr_cpu_ids > MAX_XDP_QUEUES)
10295 old_prog = xchg(&adapter->xdp_prog, prog);
10296 need_reset = (!!prog != !!old_prog);
10298 /* If transitioning XDP modes reconfigure rings */
10303 /* Wait until ndo_xsk_wakeup completes. */
10305 err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10308 rcu_assign_pointer(adapter->xdp_prog, old_prog);
10312 for (i = 0; i < adapter->num_rx_queues; i++)
10313 (void)xchg(&adapter->rx_ring[i]->xdp_prog,
10314 adapter->xdp_prog);
10318 bpf_prog_put(old_prog);
10320 /* Kick start the NAPI context if there is an AF_XDP socket open
10321 * on that queue id. This so that receiving will start.
10323 if (need_reset && prog)
10324 for (i = 0; i < adapter->num_rx_queues; i++)
10325 if (adapter->xdp_ring[i]->xsk_umem)
10326 (void)ixgbe_xsk_wakeup(adapter->netdev, i,
10332 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10334 struct ixgbe_adapter *adapter = netdev_priv(dev);
10336 switch (xdp->command) {
10337 case XDP_SETUP_PROG:
10338 return ixgbe_xdp_setup(dev, xdp->prog);
10339 case XDP_QUERY_PROG:
10340 xdp->prog_id = adapter->xdp_prog ?
10341 adapter->xdp_prog->aux->id : 0;
10343 case XDP_SETUP_XSK_UMEM:
10344 return ixgbe_xsk_umem_setup(adapter, xdp->xsk.umem,
10345 xdp->xsk.queue_id);
10352 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10354 /* Force memory writes to complete before letting h/w know there
10355 * are new descriptors to fetch.
10358 writel(ring->next_to_use, ring->tail);
10361 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10362 struct xdp_frame **frames, u32 flags)
10364 struct ixgbe_adapter *adapter = netdev_priv(dev);
10365 struct ixgbe_ring *ring;
10369 if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10372 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10375 /* During program transitions its possible adapter->xdp_prog is assigned
10376 * but ring has not been configured yet. In this case simply abort xmit.
10378 ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10379 if (unlikely(!ring))
10382 if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10385 for (i = 0; i < n; i++) {
10386 struct xdp_frame *xdpf = frames[i];
10389 err = ixgbe_xmit_xdp_ring(adapter, xdpf);
10390 if (err != IXGBE_XDP_TX) {
10391 xdp_return_frame_rx_napi(xdpf);
10396 if (unlikely(flags & XDP_XMIT_FLUSH))
10397 ixgbe_xdp_ring_update_tail(ring);
10402 static const struct net_device_ops ixgbe_netdev_ops = {
10403 .ndo_open = ixgbe_open,
10404 .ndo_stop = ixgbe_close,
10405 .ndo_start_xmit = ixgbe_xmit_frame,
10406 .ndo_set_rx_mode = ixgbe_set_rx_mode,
10407 .ndo_validate_addr = eth_validate_addr,
10408 .ndo_set_mac_address = ixgbe_set_mac,
10409 .ndo_change_mtu = ixgbe_change_mtu,
10410 .ndo_tx_timeout = ixgbe_tx_timeout,
10411 .ndo_set_tx_maxrate = ixgbe_tx_maxrate,
10412 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
10413 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
10414 .ndo_do_ioctl = ixgbe_ioctl,
10415 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
10416 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
10417 .ndo_set_vf_rate = ixgbe_ndo_set_vf_bw,
10418 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
10419 .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10420 .ndo_set_vf_trust = ixgbe_ndo_set_vf_trust,
10421 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
10422 .ndo_get_stats64 = ixgbe_get_stats64,
10423 .ndo_setup_tc = __ixgbe_setup_tc,
10425 .ndo_select_queue = ixgbe_select_queue,
10426 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10427 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10428 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10429 .ndo_fcoe_enable = ixgbe_fcoe_enable,
10430 .ndo_fcoe_disable = ixgbe_fcoe_disable,
10431 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10432 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10433 #endif /* IXGBE_FCOE */
10434 .ndo_set_features = ixgbe_set_features,
10435 .ndo_fix_features = ixgbe_fix_features,
10436 .ndo_fdb_add = ixgbe_ndo_fdb_add,
10437 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
10438 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
10439 .ndo_dfwd_add_station = ixgbe_fwd_add,
10440 .ndo_dfwd_del_station = ixgbe_fwd_del,
10441 .ndo_udp_tunnel_add = ixgbe_add_udp_tunnel_port,
10442 .ndo_udp_tunnel_del = ixgbe_del_udp_tunnel_port,
10443 .ndo_features_check = ixgbe_features_check,
10444 .ndo_bpf = ixgbe_xdp,
10445 .ndo_xdp_xmit = ixgbe_xdp_xmit,
10446 .ndo_xsk_wakeup = ixgbe_xsk_wakeup,
10449 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10450 struct ixgbe_ring *tx_ring)
10452 unsigned long wait_delay, delay_interval;
10453 struct ixgbe_hw *hw = &adapter->hw;
10454 u8 reg_idx = tx_ring->reg_idx;
10458 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10460 /* delay mechanism from ixgbe_disable_tx */
10461 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10463 wait_loop = IXGBE_MAX_RX_DESC_POLL;
10464 wait_delay = delay_interval;
10466 while (wait_loop--) {
10467 usleep_range(wait_delay, wait_delay + 10);
10468 wait_delay += delay_interval * 2;
10469 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10471 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10475 e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10478 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10479 struct ixgbe_ring *tx_ring)
10481 set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10482 ixgbe_disable_txr_hw(adapter, tx_ring);
10485 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10486 struct ixgbe_ring *rx_ring)
10488 unsigned long wait_delay, delay_interval;
10489 struct ixgbe_hw *hw = &adapter->hw;
10490 u8 reg_idx = rx_ring->reg_idx;
10494 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10495 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10496 rxdctl |= IXGBE_RXDCTL_SWFLSH;
10498 /* write value back with RXDCTL.ENABLE bit cleared */
10499 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10501 /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10502 if (hw->mac.type == ixgbe_mac_82598EB &&
10503 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10506 /* delay mechanism from ixgbe_disable_rx */
10507 delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10509 wait_loop = IXGBE_MAX_RX_DESC_POLL;
10510 wait_delay = delay_interval;
10512 while (wait_loop--) {
10513 usleep_range(wait_delay, wait_delay + 10);
10514 wait_delay += delay_interval * 2;
10515 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10517 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10521 e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10524 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10526 memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10527 memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10530 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10532 memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10533 memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10537 * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10538 * @adapter: adapter structure
10539 * @ring: ring index
10541 * This function disables a certain Rx/Tx/XDP Tx ring. The function
10542 * assumes that the netdev is running.
10544 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10546 struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10548 rx_ring = adapter->rx_ring[ring];
10549 tx_ring = adapter->tx_ring[ring];
10550 xdp_ring = adapter->xdp_ring[ring];
10552 ixgbe_disable_txr(adapter, tx_ring);
10554 ixgbe_disable_txr(adapter, xdp_ring);
10555 ixgbe_disable_rxr_hw(adapter, rx_ring);
10560 /* Rx/Tx/XDP Tx share the same napi context. */
10561 napi_disable(&rx_ring->q_vector->napi);
10563 ixgbe_clean_tx_ring(tx_ring);
10565 ixgbe_clean_tx_ring(xdp_ring);
10566 ixgbe_clean_rx_ring(rx_ring);
10568 ixgbe_reset_txr_stats(tx_ring);
10570 ixgbe_reset_txr_stats(xdp_ring);
10571 ixgbe_reset_rxr_stats(rx_ring);
10575 * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10576 * @adapter: adapter structure
10577 * @ring: ring index
10579 * This function enables a certain Rx/Tx/XDP Tx ring. The function
10580 * assumes that the netdev is running.
10582 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10584 struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10586 rx_ring = adapter->rx_ring[ring];
10587 tx_ring = adapter->tx_ring[ring];
10588 xdp_ring = adapter->xdp_ring[ring];
10590 /* Rx/Tx/XDP Tx share the same napi context. */
10591 napi_enable(&rx_ring->q_vector->napi);
10593 ixgbe_configure_tx_ring(adapter, tx_ring);
10595 ixgbe_configure_tx_ring(adapter, xdp_ring);
10596 ixgbe_configure_rx_ring(adapter, rx_ring);
10598 clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10600 clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10604 * ixgbe_enumerate_functions - Get the number of ports this device has
10605 * @adapter: adapter structure
10607 * This function enumerates the phsyical functions co-located on a single slot,
10608 * in order to determine how many ports a device has. This is most useful in
10609 * determining the required GT/s of PCIe bandwidth necessary for optimal
10612 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10614 struct pci_dev *entry, *pdev = adapter->pdev;
10617 /* Some cards can not use the generic count PCIe functions method,
10618 * because they are behind a parent switch, so we hardcode these with
10619 * the correct number of functions.
10621 if (ixgbe_pcie_from_parent(&adapter->hw))
10624 list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10625 /* don't count virtual functions */
10626 if (entry->is_virtfn)
10629 /* When the devices on the bus don't all match our device ID,
10630 * we can't reliably determine the correct number of
10631 * functions. This can occur if a function has been direct
10632 * attached to a virtual machine using VT-d, for example. In
10633 * this case, simply return -1 to indicate this.
10635 if ((entry->vendor != pdev->vendor) ||
10636 (entry->device != pdev->device))
10646 * ixgbe_wol_supported - Check whether device supports WoL
10647 * @adapter: the adapter private structure
10648 * @device_id: the device ID
10649 * @subdevice_id: the subsystem device ID
10651 * This function is used by probe and ethtool to determine
10652 * which devices have WoL support
10655 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10658 struct ixgbe_hw *hw = &adapter->hw;
10659 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10661 /* WOL not supported on 82598 */
10662 if (hw->mac.type == ixgbe_mac_82598EB)
10665 /* check eeprom to see if WOL is enabled for X540 and newer */
10666 if (hw->mac.type >= ixgbe_mac_X540) {
10667 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10668 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10669 (hw->bus.func == 0)))
10673 /* WOL is determined based on device IDs for 82599 MACs */
10674 switch (device_id) {
10675 case IXGBE_DEV_ID_82599_SFP:
10676 /* Only these subdevices could supports WOL */
10677 switch (subdevice_id) {
10678 case IXGBE_SUBDEV_ID_82599_560FLR:
10679 case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10680 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10681 case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10682 /* only support first port */
10683 if (hw->bus.func != 0)
10686 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10687 case IXGBE_SUBDEV_ID_82599_SFP:
10688 case IXGBE_SUBDEV_ID_82599_RNDC:
10689 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10690 case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10691 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10692 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10696 case IXGBE_DEV_ID_82599EN_SFP:
10697 /* Only these subdevices support WOL */
10698 switch (subdevice_id) {
10699 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10703 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10704 /* All except this subdevice support WOL */
10705 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10708 case IXGBE_DEV_ID_82599_KX4:
10718 * ixgbe_set_fw_version - Set FW version
10719 * @adapter: the adapter private structure
10721 * This function is used by probe and ethtool to determine the FW version to
10722 * format to display. The FW version is taken from the EEPROM/NVM.
10724 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10726 struct ixgbe_hw *hw = &adapter->hw;
10727 struct ixgbe_nvm_version nvm_ver;
10729 ixgbe_get_oem_prod_version(hw, &nvm_ver);
10730 if (nvm_ver.oem_valid) {
10731 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10732 "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10733 nvm_ver.oem_release);
10737 ixgbe_get_etk_id(hw, &nvm_ver);
10738 ixgbe_get_orom_version(hw, &nvm_ver);
10740 if (nvm_ver.or_valid) {
10741 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10742 "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10743 nvm_ver.or_build, nvm_ver.or_patch);
10747 /* Set ETrack ID format */
10748 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10749 "0x%08x", nvm_ver.etk_id);
10753 * ixgbe_probe - Device Initialization Routine
10754 * @pdev: PCI device information struct
10755 * @ent: entry in ixgbe_pci_tbl
10757 * Returns 0 on success, negative on failure
10759 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10760 * The OS initialization, configuring of the adapter private structure,
10761 * and a hardware reset occur.
10763 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10765 struct net_device *netdev;
10766 struct ixgbe_adapter *adapter = NULL;
10767 struct ixgbe_hw *hw;
10768 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10769 int i, err, pci_using_dac, expected_gts;
10770 unsigned int indices = MAX_TX_QUEUES;
10771 u8 part_str[IXGBE_PBANUM_LENGTH];
10772 bool disable_dev = false;
10778 /* Catch broken hardware that put the wrong VF device ID in
10779 * the PCIe SR-IOV capability.
10781 if (pdev->is_virtfn) {
10782 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10783 pci_name(pdev), pdev->vendor, pdev->device);
10787 err = pci_enable_device_mem(pdev);
10791 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10794 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10796 dev_err(&pdev->dev,
10797 "No usable DMA configuration, aborting\n");
10803 err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10805 dev_err(&pdev->dev,
10806 "pci_request_selected_regions failed 0x%x\n", err);
10810 pci_enable_pcie_error_reporting(pdev);
10812 pci_set_master(pdev);
10813 pci_save_state(pdev);
10815 if (ii->mac == ixgbe_mac_82598EB) {
10816 #ifdef CONFIG_IXGBE_DCB
10817 /* 8 TC w/ 4 queues per TC */
10818 indices = 4 * MAX_TRAFFIC_CLASS;
10820 indices = IXGBE_MAX_RSS_INDICES;
10824 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10827 goto err_alloc_etherdev;
10830 SET_NETDEV_DEV(netdev, &pdev->dev);
10832 adapter = netdev_priv(netdev);
10834 adapter->netdev = netdev;
10835 adapter->pdev = pdev;
10837 hw->back = adapter;
10838 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10840 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10841 pci_resource_len(pdev, 0));
10842 adapter->io_addr = hw->hw_addr;
10843 if (!hw->hw_addr) {
10848 netdev->netdev_ops = &ixgbe_netdev_ops;
10849 ixgbe_set_ethtool_ops(netdev);
10850 netdev->watchdog_timeo = 5 * HZ;
10851 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10854 hw->mac.ops = *ii->mac_ops;
10855 hw->mac.type = ii->mac;
10856 hw->mvals = ii->mvals;
10858 hw->link.ops = *ii->link_ops;
10861 hw->eeprom.ops = *ii->eeprom_ops;
10862 eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10863 if (ixgbe_removed(hw->hw_addr)) {
10867 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10868 if (!(eec & BIT(8)))
10869 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10872 hw->phy.ops = *ii->phy_ops;
10873 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10874 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
10875 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10876 hw->phy.mdio.mmds = 0;
10877 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10878 hw->phy.mdio.dev = netdev;
10879 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10880 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10882 /* setup the private structure */
10883 err = ixgbe_sw_init(adapter, ii);
10887 /* Make sure the SWFW semaphore is in a valid state */
10888 if (hw->mac.ops.init_swfw_sync)
10889 hw->mac.ops.init_swfw_sync(hw);
10891 /* Make it possible the adapter to be woken up via WOL */
10892 switch (adapter->hw.mac.type) {
10893 case ixgbe_mac_82599EB:
10894 case ixgbe_mac_X540:
10895 case ixgbe_mac_X550:
10896 case ixgbe_mac_X550EM_x:
10897 case ixgbe_mac_x550em_a:
10898 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10905 * If there is a fan on this device and it has failed log the
10908 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10909 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10910 if (esdp & IXGBE_ESDP_SDP1)
10911 e_crit(probe, "Fan has stopped, replace the adapter\n");
10914 if (allow_unsupported_sfp)
10915 hw->allow_unsupported_sfp = allow_unsupported_sfp;
10917 /* reset_hw fills in the perm_addr as well */
10918 hw->phy.reset_if_overtemp = true;
10919 err = hw->mac.ops.reset_hw(hw);
10920 hw->phy.reset_if_overtemp = false;
10921 ixgbe_set_eee_capable(adapter);
10922 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10924 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10925 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10926 e_dev_err("Reload the driver after installing a supported module.\n");
10929 e_dev_err("HW Init failed: %d\n", err);
10933 #ifdef CONFIG_PCI_IOV
10934 /* SR-IOV not supported on the 82598 */
10935 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10938 ixgbe_init_mbx_params_pf(hw);
10939 hw->mbx.ops = ii->mbx_ops;
10940 pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10941 ixgbe_enable_sriov(adapter, max_vfs);
10945 netdev->features = NETIF_F_SG |
10952 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10953 NETIF_F_GSO_GRE_CSUM | \
10954 NETIF_F_GSO_IPXIP4 | \
10955 NETIF_F_GSO_IPXIP6 | \
10956 NETIF_F_GSO_UDP_TUNNEL | \
10957 NETIF_F_GSO_UDP_TUNNEL_CSUM)
10959 netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10960 netdev->features |= NETIF_F_GSO_PARTIAL |
10961 IXGBE_GSO_PARTIAL_FEATURES;
10963 if (hw->mac.type >= ixgbe_mac_82599EB)
10964 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
10966 #ifdef CONFIG_IXGBE_IPSEC
10967 #define IXGBE_ESP_FEATURES (NETIF_F_HW_ESP | \
10968 NETIF_F_HW_ESP_TX_CSUM | \
10971 if (adapter->ipsec)
10972 netdev->features |= IXGBE_ESP_FEATURES;
10974 /* copy netdev features into list of user selectable features */
10975 netdev->hw_features |= netdev->features |
10976 NETIF_F_HW_VLAN_CTAG_FILTER |
10977 NETIF_F_HW_VLAN_CTAG_RX |
10978 NETIF_F_HW_VLAN_CTAG_TX |
10980 NETIF_F_HW_L2FW_DOFFLOAD;
10982 if (hw->mac.type >= ixgbe_mac_82599EB)
10983 netdev->hw_features |= NETIF_F_NTUPLE |
10987 netdev->features |= NETIF_F_HIGHDMA;
10989 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10990 netdev->hw_enc_features |= netdev->vlan_features;
10991 netdev->mpls_features |= NETIF_F_SG |
10995 netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10997 /* set this bit last since it cannot be part of vlan_features */
10998 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10999 NETIF_F_HW_VLAN_CTAG_RX |
11000 NETIF_F_HW_VLAN_CTAG_TX;
11002 netdev->priv_flags |= IFF_UNICAST_FLT;
11003 netdev->priv_flags |= IFF_SUPP_NOFCS;
11005 /* MTU range: 68 - 9710 */
11006 netdev->min_mtu = ETH_MIN_MTU;
11007 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
11009 #ifdef CONFIG_IXGBE_DCB
11010 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
11011 netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
11015 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
11016 unsigned int fcoe_l;
11018 if (hw->mac.ops.get_device_caps) {
11019 hw->mac.ops.get_device_caps(hw, &device_caps);
11020 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
11021 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
11025 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
11026 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
11028 netdev->features |= NETIF_F_FSO |
11031 netdev->vlan_features |= NETIF_F_FSO |
11035 #endif /* IXGBE_FCOE */
11036 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
11037 netdev->hw_features |= NETIF_F_LRO;
11038 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
11039 netdev->features |= NETIF_F_LRO;
11041 if (ixgbe_check_fw_error(adapter)) {
11046 /* make sure the EEPROM is good */
11047 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
11048 e_dev_err("The EEPROM Checksum Is Not Valid\n");
11053 eth_platform_get_mac_address(&adapter->pdev->dev,
11054 adapter->hw.mac.perm_addr);
11056 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
11058 if (!is_valid_ether_addr(netdev->dev_addr)) {
11059 e_dev_err("invalid MAC address\n");
11064 /* Set hw->mac.addr to permanent MAC address */
11065 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
11066 ixgbe_mac_set_default_filter(adapter);
11068 timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
11070 if (ixgbe_removed(hw->hw_addr)) {
11074 INIT_WORK(&adapter->service_task, ixgbe_service_task);
11075 set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
11076 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
11078 err = ixgbe_init_interrupt_scheme(adapter);
11082 for (i = 0; i < adapter->num_rx_queues; i++)
11083 u64_stats_init(&adapter->rx_ring[i]->syncp);
11084 for (i = 0; i < adapter->num_tx_queues; i++)
11085 u64_stats_init(&adapter->tx_ring[i]->syncp);
11086 for (i = 0; i < adapter->num_xdp_queues; i++)
11087 u64_stats_init(&adapter->xdp_ring[i]->syncp);
11089 /* WOL not supported for all devices */
11091 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
11092 hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
11093 pdev->subsystem_device);
11094 if (hw->wol_enabled)
11095 adapter->wol = IXGBE_WUFC_MAG;
11097 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
11099 /* save off EEPROM version number */
11100 ixgbe_set_fw_version(adapter);
11102 /* pick up the PCI bus settings for reporting later */
11103 if (ixgbe_pcie_from_parent(hw))
11104 ixgbe_get_parent_bus_info(adapter);
11106 hw->mac.ops.get_bus_info(hw);
11108 /* calculate the expected PCIe bandwidth required for optimal
11109 * performance. Note that some older parts will never have enough
11110 * bandwidth due to being older generation PCIe parts. We clamp these
11111 * parts to ensure no warning is displayed if it can't be fixed.
11113 switch (hw->mac.type) {
11114 case ixgbe_mac_82598EB:
11115 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
11118 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
11122 /* don't check link if we failed to enumerate functions */
11123 if (expected_gts > 0)
11124 ixgbe_check_minimum_link(adapter, expected_gts);
11126 err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
11128 strlcpy(part_str, "Unknown", sizeof(part_str));
11129 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
11130 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
11131 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
11134 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
11135 hw->mac.type, hw->phy.type, part_str);
11137 e_dev_info("%pM\n", netdev->dev_addr);
11139 /* reset the hardware with the new settings */
11140 err = hw->mac.ops.start_hw(hw);
11141 if (err == IXGBE_ERR_EEPROM_VERSION) {
11142 /* We are running on a pre-production device, log a warning */
11143 e_dev_warn("This device is a pre-production adapter/LOM. "
11144 "Please be aware there may be issues associated "
11145 "with your hardware. If you are experiencing "
11146 "problems please contact your Intel or hardware "
11147 "representative who provided you with this "
11150 strcpy(netdev->name, "eth%d");
11151 pci_set_drvdata(pdev, adapter);
11152 err = register_netdev(netdev);
11157 /* power down the optics for 82599 SFP+ fiber */
11158 if (hw->mac.ops.disable_tx_laser)
11159 hw->mac.ops.disable_tx_laser(hw);
11161 /* carrier off reporting is important to ethtool even BEFORE open */
11162 netif_carrier_off(netdev);
11164 #ifdef CONFIG_IXGBE_DCA
11165 if (dca_add_requester(&pdev->dev) == 0) {
11166 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11167 ixgbe_setup_dca(adapter);
11170 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11171 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11172 for (i = 0; i < adapter->num_vfs; i++)
11173 ixgbe_vf_configuration(pdev, (i | 0x10000000));
11176 /* firmware requires driver version to be 0xFFFFFFFF
11177 * since os does not support feature
11179 if (hw->mac.ops.set_fw_drv_ver)
11180 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11181 sizeof(ixgbe_driver_version) - 1,
11182 ixgbe_driver_version);
11184 /* add san mac addr to netdev */
11185 ixgbe_add_sanmac_netdev(netdev);
11187 e_dev_info("%s\n", ixgbe_default_device_descr);
11189 #ifdef CONFIG_IXGBE_HWMON
11190 if (ixgbe_sysfs_init(adapter))
11191 e_err(probe, "failed to allocate sysfs resources\n");
11192 #endif /* CONFIG_IXGBE_HWMON */
11194 ixgbe_dbg_adapter_init(adapter);
11196 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11197 if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11198 hw->mac.ops.setup_link(hw,
11199 IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11202 ixgbe_mii_bus_init(hw);
11207 ixgbe_release_hw_control(adapter);
11208 ixgbe_clear_interrupt_scheme(adapter);
11210 ixgbe_disable_sriov(adapter);
11211 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11212 iounmap(adapter->io_addr);
11213 kfree(adapter->jump_tables[0]);
11214 kfree(adapter->mac_table);
11215 kfree(adapter->rss_key);
11216 bitmap_free(adapter->af_xdp_zc_qps);
11218 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11219 free_netdev(netdev);
11220 err_alloc_etherdev:
11221 pci_release_mem_regions(pdev);
11224 if (!adapter || disable_dev)
11225 pci_disable_device(pdev);
11230 * ixgbe_remove - Device Removal Routine
11231 * @pdev: PCI device information struct
11233 * ixgbe_remove is called by the PCI subsystem to alert the driver
11234 * that it should release a PCI device. The could be caused by a
11235 * Hot-Plug event, or because the driver is going to be removed from
11238 static void ixgbe_remove(struct pci_dev *pdev)
11240 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11241 struct net_device *netdev;
11245 /* if !adapter then we already cleaned up in probe */
11249 netdev = adapter->netdev;
11250 ixgbe_dbg_adapter_exit(adapter);
11252 set_bit(__IXGBE_REMOVING, &adapter->state);
11253 cancel_work_sync(&adapter->service_task);
11255 if (adapter->mii_bus)
11256 mdiobus_unregister(adapter->mii_bus);
11258 #ifdef CONFIG_IXGBE_DCA
11259 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11260 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11261 dca_remove_requester(&pdev->dev);
11262 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11263 IXGBE_DCA_CTRL_DCA_DISABLE);
11267 #ifdef CONFIG_IXGBE_HWMON
11268 ixgbe_sysfs_exit(adapter);
11269 #endif /* CONFIG_IXGBE_HWMON */
11271 /* remove the added san mac */
11272 ixgbe_del_sanmac_netdev(netdev);
11274 #ifdef CONFIG_PCI_IOV
11275 ixgbe_disable_sriov(adapter);
11277 if (netdev->reg_state == NETREG_REGISTERED)
11278 unregister_netdev(netdev);
11280 ixgbe_stop_ipsec_offload(adapter);
11281 ixgbe_clear_interrupt_scheme(adapter);
11283 ixgbe_release_hw_control(adapter);
11286 kfree(adapter->ixgbe_ieee_pfc);
11287 kfree(adapter->ixgbe_ieee_ets);
11290 iounmap(adapter->io_addr);
11291 pci_release_mem_regions(pdev);
11293 e_dev_info("complete\n");
11295 for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11296 if (adapter->jump_tables[i]) {
11297 kfree(adapter->jump_tables[i]->input);
11298 kfree(adapter->jump_tables[i]->mask);
11300 kfree(adapter->jump_tables[i]);
11303 kfree(adapter->mac_table);
11304 kfree(adapter->rss_key);
11305 bitmap_free(adapter->af_xdp_zc_qps);
11306 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11307 free_netdev(netdev);
11309 pci_disable_pcie_error_reporting(pdev);
11312 pci_disable_device(pdev);
11316 * ixgbe_io_error_detected - called when PCI error is detected
11317 * @pdev: Pointer to PCI device
11318 * @state: The current pci connection state
11320 * This function is called after a PCI bus error affecting
11321 * this device has been detected.
11323 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11324 pci_channel_state_t state)
11326 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11327 struct net_device *netdev = adapter->netdev;
11329 #ifdef CONFIG_PCI_IOV
11330 struct ixgbe_hw *hw = &adapter->hw;
11331 struct pci_dev *bdev, *vfdev;
11332 u32 dw0, dw1, dw2, dw3;
11334 u16 req_id, pf_func;
11336 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11337 adapter->num_vfs == 0)
11338 goto skip_bad_vf_detection;
11340 bdev = pdev->bus->self;
11341 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11342 bdev = bdev->bus->self;
11345 goto skip_bad_vf_detection;
11347 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11349 goto skip_bad_vf_detection;
11351 dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11352 dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11353 dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11354 dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11355 if (ixgbe_removed(hw->hw_addr))
11356 goto skip_bad_vf_detection;
11358 req_id = dw1 >> 16;
11359 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11360 if (!(req_id & 0x0080))
11361 goto skip_bad_vf_detection;
11363 pf_func = req_id & 0x01;
11364 if ((pf_func & 1) == (pdev->devfn & 1)) {
11365 unsigned int device_id;
11367 vf = (req_id & 0x7F) >> 1;
11368 e_dev_err("VF %d has caused a PCIe error\n", vf);
11369 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11370 "%8.8x\tdw3: %8.8x\n",
11371 dw0, dw1, dw2, dw3);
11372 switch (adapter->hw.mac.type) {
11373 case ixgbe_mac_82599EB:
11374 device_id = IXGBE_82599_VF_DEVICE_ID;
11376 case ixgbe_mac_X540:
11377 device_id = IXGBE_X540_VF_DEVICE_ID;
11379 case ixgbe_mac_X550:
11380 device_id = IXGBE_DEV_ID_X550_VF;
11382 case ixgbe_mac_X550EM_x:
11383 device_id = IXGBE_DEV_ID_X550EM_X_VF;
11385 case ixgbe_mac_x550em_a:
11386 device_id = IXGBE_DEV_ID_X550EM_A_VF;
11393 /* Find the pci device of the offending VF */
11394 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11396 if (vfdev->devfn == (req_id & 0xFF))
11398 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11402 * There's a slim chance the VF could have been hot plugged,
11403 * so if it is no longer present we don't need to issue the
11404 * VFLR. Just clean up the AER in that case.
11408 /* Free device reference count */
11409 pci_dev_put(vfdev);
11414 * Even though the error may have occurred on the other port
11415 * we still need to increment the vf error reference count for
11416 * both ports because the I/O resume function will be called
11417 * for both of them.
11419 adapter->vferr_refcount++;
11421 return PCI_ERS_RESULT_RECOVERED;
11423 skip_bad_vf_detection:
11424 #endif /* CONFIG_PCI_IOV */
11425 if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11426 return PCI_ERS_RESULT_DISCONNECT;
11428 if (!netif_device_present(netdev))
11429 return PCI_ERS_RESULT_DISCONNECT;
11432 netif_device_detach(netdev);
11434 if (netif_running(netdev))
11435 ixgbe_close_suspend(adapter);
11437 if (state == pci_channel_io_perm_failure) {
11439 return PCI_ERS_RESULT_DISCONNECT;
11442 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11443 pci_disable_device(pdev);
11446 /* Request a slot reset. */
11447 return PCI_ERS_RESULT_NEED_RESET;
11451 * ixgbe_io_slot_reset - called after the pci bus has been reset.
11452 * @pdev: Pointer to PCI device
11454 * Restart the card from scratch, as if from a cold-boot.
11456 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11458 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11459 pci_ers_result_t result;
11461 if (pci_enable_device_mem(pdev)) {
11462 e_err(probe, "Cannot re-enable PCI device after reset.\n");
11463 result = PCI_ERS_RESULT_DISCONNECT;
11465 smp_mb__before_atomic();
11466 clear_bit(__IXGBE_DISABLED, &adapter->state);
11467 adapter->hw.hw_addr = adapter->io_addr;
11468 pci_set_master(pdev);
11469 pci_restore_state(pdev);
11470 pci_save_state(pdev);
11472 pci_wake_from_d3(pdev, false);
11474 ixgbe_reset(adapter);
11475 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11476 result = PCI_ERS_RESULT_RECOVERED;
11483 * ixgbe_io_resume - called when traffic can start flowing again.
11484 * @pdev: Pointer to PCI device
11486 * This callback is called when the error recovery driver tells us that
11487 * its OK to resume normal operation.
11489 static void ixgbe_io_resume(struct pci_dev *pdev)
11491 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11492 struct net_device *netdev = adapter->netdev;
11494 #ifdef CONFIG_PCI_IOV
11495 if (adapter->vferr_refcount) {
11496 e_info(drv, "Resuming after VF err\n");
11497 adapter->vferr_refcount--;
11503 if (netif_running(netdev))
11504 ixgbe_open(netdev);
11506 netif_device_attach(netdev);
11510 static const struct pci_error_handlers ixgbe_err_handler = {
11511 .error_detected = ixgbe_io_error_detected,
11512 .slot_reset = ixgbe_io_slot_reset,
11513 .resume = ixgbe_io_resume,
11516 static struct pci_driver ixgbe_driver = {
11517 .name = ixgbe_driver_name,
11518 .id_table = ixgbe_pci_tbl,
11519 .probe = ixgbe_probe,
11520 .remove = ixgbe_remove,
11522 .suspend = ixgbe_suspend,
11523 .resume = ixgbe_resume,
11525 .shutdown = ixgbe_shutdown,
11526 .sriov_configure = ixgbe_pci_sriov_configure,
11527 .err_handler = &ixgbe_err_handler
11531 * ixgbe_init_module - Driver Registration Routine
11533 * ixgbe_init_module is the first routine called when the driver is
11534 * loaded. All it does is register with the PCI subsystem.
11536 static int __init ixgbe_init_module(void)
11539 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
11540 pr_info("%s\n", ixgbe_copyright);
11542 ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11544 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11550 ret = pci_register_driver(&ixgbe_driver);
11552 destroy_workqueue(ixgbe_wq);
11557 #ifdef CONFIG_IXGBE_DCA
11558 dca_register_notify(&dca_notifier);
11564 module_init(ixgbe_init_module);
11567 * ixgbe_exit_module - Driver Exit Cleanup Routine
11569 * ixgbe_exit_module is called just before the driver is removed
11572 static void __exit ixgbe_exit_module(void)
11574 #ifdef CONFIG_IXGBE_DCA
11575 dca_unregister_notify(&dca_notifier);
11577 pci_unregister_driver(&ixgbe_driver);
11581 destroy_workqueue(ixgbe_wq);
11586 #ifdef CONFIG_IXGBE_DCA
11587 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11592 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11593 __ixgbe_notify_dca);
11595 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11598 #endif /* CONFIG_IXGBE_DCA */
11600 module_exit(ixgbe_exit_module);