1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2016 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 Linux NICS <linux.nics@intel.com>
24 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
27 *******************************************************************************/
29 #include <linux/types.h>
30 #include <linux/module.h>
31 #include <linux/pci.h>
32 #include <linux/netdevice.h>
33 #include <linux/vmalloc.h>
34 #include <linux/string.h>
36 #include <linux/interrupt.h>
38 #include <linux/tcp.h>
39 #include <linux/sctp.h>
40 #include <linux/pkt_sched.h>
41 #include <linux/ipv6.h>
42 #include <linux/slab.h>
43 #include <net/checksum.h>
44 #include <net/ip6_checksum.h>
45 #include <linux/etherdevice.h>
46 #include <linux/ethtool.h>
48 #include <linux/if_vlan.h>
49 #include <linux/if_macvlan.h>
50 #include <linux/if_bridge.h>
51 #include <linux/prefetch.h>
52 #include <linux/bpf.h>
53 #include <linux/bpf_trace.h>
54 #include <linux/atomic.h>
55 #include <scsi/fc/fc_fcoe.h>
56 #include <net/udp_tunnel.h>
57 #include <net/pkt_cls.h>
58 #include <net/tc_act/tc_gact.h>
59 #include <net/tc_act/tc_mirred.h>
60 #include <net/vxlan.h>
64 #include "ixgbe_common.h"
65 #include "ixgbe_dcb_82599.h"
66 #include "ixgbe_sriov.h"
67 #include "ixgbe_model.h"
69 char ixgbe_driver_name[] = "ixgbe";
70 static const char ixgbe_driver_string[] =
71 "Intel(R) 10 Gigabit PCI Express Network Driver";
73 char ixgbe_default_device_descr[] =
74 "Intel(R) 10 Gigabit Network Connection";
76 static char ixgbe_default_device_descr[] =
77 "Intel(R) 10 Gigabit Network Connection";
79 #define DRV_VERSION "5.1.0-k"
80 const char ixgbe_driver_version[] = DRV_VERSION;
81 static const char ixgbe_copyright[] =
82 "Copyright (c) 1999-2016 Intel Corporation.";
84 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";
86 static const struct ixgbe_info *ixgbe_info_tbl[] = {
87 [board_82598] = &ixgbe_82598_info,
88 [board_82599] = &ixgbe_82599_info,
89 [board_X540] = &ixgbe_X540_info,
90 [board_X550] = &ixgbe_X550_info,
91 [board_X550EM_x] = &ixgbe_X550EM_x_info,
92 [board_x550em_x_fw] = &ixgbe_x550em_x_fw_info,
93 [board_x550em_a] = &ixgbe_x550em_a_info,
94 [board_x550em_a_fw] = &ixgbe_x550em_a_fw_info,
97 /* ixgbe_pci_tbl - PCI Device ID Table
99 * Wildcard entries (PCI_ANY_ID) should come last
100 * Last entry must be all 0s
102 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
103 * Class, Class Mask, private data (not used) }
105 static const struct pci_device_id ixgbe_pci_tbl[] = {
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
117 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
118 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
119 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
120 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
121 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
122 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
123 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
124 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
125 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
126 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
127 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
128 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
129 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
130 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
131 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
132 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
133 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
134 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
135 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
136 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
137 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T1), board_X550},
138 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
139 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_XFI), board_X550EM_x},
140 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
141 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x},
142 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x},
143 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_1G_T), board_x550em_x_fw},
144 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR), board_x550em_a },
145 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR_L), board_x550em_a },
146 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP_N), board_x550em_a },
147 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII), board_x550em_a },
148 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII_L), board_x550em_a },
149 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_10G_T), board_x550em_a},
150 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP), board_x550em_a },
151 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T), board_x550em_a_fw },
152 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T_L), board_x550em_a_fw },
153 /* required last entry */
156 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
158 #ifdef CONFIG_IXGBE_DCA
159 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
161 static struct notifier_block dca_notifier = {
162 .notifier_call = ixgbe_notify_dca,
168 #ifdef CONFIG_PCI_IOV
169 static unsigned int max_vfs;
170 module_param(max_vfs, uint, 0);
171 MODULE_PARM_DESC(max_vfs,
172 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
173 #endif /* CONFIG_PCI_IOV */
175 static unsigned int allow_unsupported_sfp;
176 module_param(allow_unsupported_sfp, uint, 0);
177 MODULE_PARM_DESC(allow_unsupported_sfp,
178 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
180 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
181 static int debug = -1;
182 module_param(debug, int, 0);
183 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
185 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
186 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
187 MODULE_LICENSE("GPL");
188 MODULE_VERSION(DRV_VERSION);
190 static struct workqueue_struct *ixgbe_wq;
192 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
193 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *);
195 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
198 struct pci_dev *parent_dev;
199 struct pci_bus *parent_bus;
201 parent_bus = adapter->pdev->bus->parent;
205 parent_dev = parent_bus->self;
209 if (!pci_is_pcie(parent_dev))
212 pcie_capability_read_word(parent_dev, reg, value);
213 if (*value == IXGBE_FAILED_READ_CFG_WORD &&
214 ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
219 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
221 struct ixgbe_hw *hw = &adapter->hw;
225 hw->bus.type = ixgbe_bus_type_pci_express;
227 /* Get the negotiated link width and speed from PCI config space of the
228 * parent, as this device is behind a switch
230 err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
232 /* assume caller will handle error case */
236 hw->bus.width = ixgbe_convert_bus_width(link_status);
237 hw->bus.speed = ixgbe_convert_bus_speed(link_status);
243 * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
244 * @hw: hw specific details
246 * This function is used by probe to determine whether a device's PCI-Express
247 * bandwidth details should be gathered from the parent bus instead of from the
248 * device. Used to ensure that various locations all have the correct device ID
251 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
253 switch (hw->device_id) {
254 case IXGBE_DEV_ID_82599_SFP_SF_QP:
255 case IXGBE_DEV_ID_82599_QSFP_SF_QP:
262 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
265 struct ixgbe_hw *hw = &adapter->hw;
267 enum pci_bus_speed speed = PCI_SPEED_UNKNOWN;
268 enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN;
269 struct pci_dev *pdev;
271 /* Some devices are not connected over PCIe and thus do not negotiate
272 * speed. These devices do not have valid bus info, and thus any report
273 * we generate may not be correct.
275 if (hw->bus.type == ixgbe_bus_type_internal)
278 /* determine whether to use the parent device */
279 if (ixgbe_pcie_from_parent(&adapter->hw))
280 pdev = adapter->pdev->bus->parent->self;
282 pdev = adapter->pdev;
284 if (pcie_get_minimum_link(pdev, &speed, &width) ||
285 speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
286 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
291 case PCIE_SPEED_2_5GT:
292 /* 8b/10b encoding reduces max throughput by 20% */
295 case PCIE_SPEED_5_0GT:
296 /* 8b/10b encoding reduces max throughput by 20% */
299 case PCIE_SPEED_8_0GT:
300 /* 128b/130b encoding reduces throughput by less than 2% */
304 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
308 e_dev_info("PCI Express bandwidth of %dGT/s available\n",
310 e_dev_info("(Speed:%s, Width: x%d, Encoding Loss:%s)\n",
311 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" :
312 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" :
313 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" :
316 (speed == PCIE_SPEED_2_5GT ? "20%" :
317 speed == PCIE_SPEED_5_0GT ? "20%" :
318 speed == PCIE_SPEED_8_0GT ? "<2%" :
321 if (max_gts < expected_gts) {
322 e_dev_warn("This is not sufficient for optimal performance of this card.\n");
323 e_dev_warn("For optimal performance, at least %dGT/s of bandwidth is required.\n",
325 e_dev_warn("A slot with more lanes and/or higher speed is suggested.\n");
329 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
331 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
332 !test_bit(__IXGBE_REMOVING, &adapter->state) &&
333 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
334 queue_work(ixgbe_wq, &adapter->service_task);
337 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
339 struct ixgbe_adapter *adapter = hw->back;
344 e_dev_err("Adapter removed\n");
345 if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
346 ixgbe_service_event_schedule(adapter);
349 static void ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
353 /* The following check not only optimizes a bit by not
354 * performing a read on the status register when the
355 * register just read was a status register read that
356 * returned IXGBE_FAILED_READ_REG. It also blocks any
357 * potential recursion.
359 if (reg == IXGBE_STATUS) {
360 ixgbe_remove_adapter(hw);
363 value = ixgbe_read_reg(hw, IXGBE_STATUS);
364 if (value == IXGBE_FAILED_READ_REG)
365 ixgbe_remove_adapter(hw);
369 * ixgbe_read_reg - Read from device register
370 * @hw: hw specific details
371 * @reg: offset of register to read
373 * Returns : value read or IXGBE_FAILED_READ_REG if removed
375 * This function is used to read device registers. It checks for device
376 * removal by confirming any read that returns all ones by checking the
377 * status register value for all ones. This function avoids reading from
378 * the hardware if a removal was previously detected in which case it
379 * returns IXGBE_FAILED_READ_REG (all ones).
381 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
383 u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
386 if (ixgbe_removed(reg_addr))
387 return IXGBE_FAILED_READ_REG;
388 if (unlikely(hw->phy.nw_mng_if_sel &
389 IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE)) {
390 struct ixgbe_adapter *adapter;
393 for (i = 0; i < 200; ++i) {
394 value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY);
396 goto writes_completed;
397 if (value == IXGBE_FAILED_READ_REG) {
398 ixgbe_remove_adapter(hw);
399 return IXGBE_FAILED_READ_REG;
405 e_warn(hw, "register writes incomplete %08x\n", value);
409 value = readl(reg_addr + reg);
410 if (unlikely(value == IXGBE_FAILED_READ_REG))
411 ixgbe_check_remove(hw, reg);
415 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
419 pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
420 if (value == IXGBE_FAILED_READ_CFG_WORD) {
421 ixgbe_remove_adapter(hw);
427 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
429 struct ixgbe_adapter *adapter = hw->back;
432 if (ixgbe_removed(hw->hw_addr))
433 return IXGBE_FAILED_READ_CFG_WORD;
434 pci_read_config_word(adapter->pdev, reg, &value);
435 if (value == IXGBE_FAILED_READ_CFG_WORD &&
436 ixgbe_check_cfg_remove(hw, adapter->pdev))
437 return IXGBE_FAILED_READ_CFG_WORD;
441 #ifdef CONFIG_PCI_IOV
442 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
444 struct ixgbe_adapter *adapter = hw->back;
447 if (ixgbe_removed(hw->hw_addr))
448 return IXGBE_FAILED_READ_CFG_DWORD;
449 pci_read_config_dword(adapter->pdev, reg, &value);
450 if (value == IXGBE_FAILED_READ_CFG_DWORD &&
451 ixgbe_check_cfg_remove(hw, adapter->pdev))
452 return IXGBE_FAILED_READ_CFG_DWORD;
455 #endif /* CONFIG_PCI_IOV */
457 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
459 struct ixgbe_adapter *adapter = hw->back;
461 if (ixgbe_removed(hw->hw_addr))
463 pci_write_config_word(adapter->pdev, reg, value);
466 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
468 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
470 /* flush memory to make sure state is correct before next watchdog */
471 smp_mb__before_atomic();
472 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
475 struct ixgbe_reg_info {
480 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
482 /* General Registers */
483 {IXGBE_CTRL, "CTRL"},
484 {IXGBE_STATUS, "STATUS"},
485 {IXGBE_CTRL_EXT, "CTRL_EXT"},
487 /* Interrupt Registers */
488 {IXGBE_EICR, "EICR"},
491 {IXGBE_SRRCTL(0), "SRRCTL"},
492 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
493 {IXGBE_RDLEN(0), "RDLEN"},
494 {IXGBE_RDH(0), "RDH"},
495 {IXGBE_RDT(0), "RDT"},
496 {IXGBE_RXDCTL(0), "RXDCTL"},
497 {IXGBE_RDBAL(0), "RDBAL"},
498 {IXGBE_RDBAH(0), "RDBAH"},
501 {IXGBE_TDBAL(0), "TDBAL"},
502 {IXGBE_TDBAH(0), "TDBAH"},
503 {IXGBE_TDLEN(0), "TDLEN"},
504 {IXGBE_TDH(0), "TDH"},
505 {IXGBE_TDT(0), "TDT"},
506 {IXGBE_TXDCTL(0), "TXDCTL"},
508 /* List Terminator */
514 * ixgbe_regdump - register printout routine
516 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
522 switch (reginfo->ofs) {
523 case IXGBE_SRRCTL(0):
524 for (i = 0; i < 64; i++)
525 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
527 case IXGBE_DCA_RXCTRL(0):
528 for (i = 0; i < 64; i++)
529 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
532 for (i = 0; i < 64; i++)
533 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
536 for (i = 0; i < 64; i++)
537 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
540 for (i = 0; i < 64; i++)
541 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
543 case IXGBE_RXDCTL(0):
544 for (i = 0; i < 64; i++)
545 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
548 for (i = 0; i < 64; i++)
549 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
552 for (i = 0; i < 64; i++)
553 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
556 for (i = 0; i < 64; i++)
557 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
560 for (i = 0; i < 64; i++)
561 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
564 for (i = 0; i < 64; i++)
565 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
568 for (i = 0; i < 64; i++)
569 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
572 for (i = 0; i < 64; i++)
573 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
575 case IXGBE_TXDCTL(0):
576 for (i = 0; i < 64; i++)
577 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
580 pr_info("%-15s %08x\n",
581 reginfo->name, IXGBE_READ_REG(hw, reginfo->ofs));
591 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i, i + 7);
592 for (j = 0; j < 8; j++)
593 p += sprintf(p, " %08x", regs[i++]);
594 pr_err("%-15s%s\n", rname, buf);
599 static void ixgbe_print_buffer(struct ixgbe_ring *ring, int n)
601 struct ixgbe_tx_buffer *tx_buffer;
603 tx_buffer = &ring->tx_buffer_info[ring->next_to_clean];
604 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
605 n, ring->next_to_use, ring->next_to_clean,
606 (u64)dma_unmap_addr(tx_buffer, dma),
607 dma_unmap_len(tx_buffer, len),
608 tx_buffer->next_to_watch,
609 (u64)tx_buffer->time_stamp);
613 * ixgbe_dump - Print registers, tx-rings and rx-rings
615 static void ixgbe_dump(struct ixgbe_adapter *adapter)
617 struct net_device *netdev = adapter->netdev;
618 struct ixgbe_hw *hw = &adapter->hw;
619 struct ixgbe_reg_info *reginfo;
621 struct ixgbe_ring *ring;
622 struct ixgbe_tx_buffer *tx_buffer;
623 union ixgbe_adv_tx_desc *tx_desc;
624 struct my_u0 { u64 a; u64 b; } *u0;
625 struct ixgbe_ring *rx_ring;
626 union ixgbe_adv_rx_desc *rx_desc;
627 struct ixgbe_rx_buffer *rx_buffer_info;
630 if (!netif_msg_hw(adapter))
633 /* Print netdevice Info */
635 dev_info(&adapter->pdev->dev, "Net device Info\n");
636 pr_info("Device Name state "
638 pr_info("%-15s %016lX %016lX\n",
641 dev_trans_start(netdev));
644 /* Print Registers */
645 dev_info(&adapter->pdev->dev, "Register Dump\n");
646 pr_info(" Register Name Value\n");
647 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
648 reginfo->name; reginfo++) {
649 ixgbe_regdump(hw, reginfo);
652 /* Print TX Ring Summary */
653 if (!netdev || !netif_running(netdev))
656 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
657 pr_info(" %s %s %s %s\n",
658 "Queue [NTU] [NTC] [bi(ntc)->dma ]",
659 "leng", "ntw", "timestamp");
660 for (n = 0; n < adapter->num_tx_queues; n++) {
661 ring = adapter->tx_ring[n];
662 ixgbe_print_buffer(ring, n);
665 for (n = 0; n < adapter->num_xdp_queues; n++) {
666 ring = adapter->xdp_ring[n];
667 ixgbe_print_buffer(ring, n);
671 if (!netif_msg_tx_done(adapter))
672 goto rx_ring_summary;
674 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
676 /* Transmit Descriptor Formats
678 * 82598 Advanced Transmit Descriptor
679 * +--------------------------------------------------------------+
680 * 0 | Buffer Address [63:0] |
681 * +--------------------------------------------------------------+
682 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
683 * +--------------------------------------------------------------+
684 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
686 * 82598 Advanced Transmit Descriptor (Write-Back Format)
687 * +--------------------------------------------------------------+
689 * +--------------------------------------------------------------+
690 * 8 | RSV | STA | NXTSEQ |
691 * +--------------------------------------------------------------+
694 * 82599+ Advanced Transmit Descriptor
695 * +--------------------------------------------------------------+
696 * 0 | Buffer Address [63:0] |
697 * +--------------------------------------------------------------+
698 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN |
699 * +--------------------------------------------------------------+
700 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0
702 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
703 * +--------------------------------------------------------------+
705 * +--------------------------------------------------------------+
706 * 8 | RSV | STA | RSV |
707 * +--------------------------------------------------------------+
711 for (n = 0; n < adapter->num_tx_queues; n++) {
712 ring = adapter->tx_ring[n];
713 pr_info("------------------------------------\n");
714 pr_info("TX QUEUE INDEX = %d\n", ring->queue_index);
715 pr_info("------------------------------------\n");
716 pr_info("%s%s %s %s %s %s\n",
717 "T [desc] [address 63:0 ] ",
718 "[PlPOIdStDDt Ln] [bi->dma ] ",
719 "leng", "ntw", "timestamp", "bi->skb");
721 for (i = 0; ring->desc && (i < ring->count); i++) {
722 tx_desc = IXGBE_TX_DESC(ring, i);
723 tx_buffer = &ring->tx_buffer_info[i];
724 u0 = (struct my_u0 *)tx_desc;
725 if (dma_unmap_len(tx_buffer, len) > 0) {
726 const char *ring_desc;
728 if (i == ring->next_to_use &&
729 i == ring->next_to_clean)
730 ring_desc = " NTC/U";
731 else if (i == ring->next_to_use)
733 else if (i == ring->next_to_clean)
737 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p%s",
741 (u64)dma_unmap_addr(tx_buffer, dma),
742 dma_unmap_len(tx_buffer, len),
743 tx_buffer->next_to_watch,
744 (u64)tx_buffer->time_stamp,
748 if (netif_msg_pktdata(adapter) &&
750 print_hex_dump(KERN_INFO, "",
751 DUMP_PREFIX_ADDRESS, 16, 1,
752 tx_buffer->skb->data,
753 dma_unmap_len(tx_buffer, len),
759 /* Print RX Rings Summary */
761 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
762 pr_info("Queue [NTU] [NTC]\n");
763 for (n = 0; n < adapter->num_rx_queues; n++) {
764 rx_ring = adapter->rx_ring[n];
765 pr_info("%5d %5X %5X\n",
766 n, rx_ring->next_to_use, rx_ring->next_to_clean);
770 if (!netif_msg_rx_status(adapter))
773 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
775 /* Receive Descriptor Formats
777 * 82598 Advanced Receive Descriptor (Read) Format
779 * +-----------------------------------------------------+
780 * 0 | Packet Buffer Address [63:1] |A0/NSE|
781 * +----------------------------------------------+------+
782 * 8 | Header Buffer Address [63:1] | DD |
783 * +-----------------------------------------------------+
786 * 82598 Advanced Receive Descriptor (Write-Back) Format
788 * 63 48 47 32 31 30 21 20 16 15 4 3 0
789 * +------------------------------------------------------+
790 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS |
791 * | Packet | IP | | | | Type | Type |
792 * | Checksum | Ident | | | | | |
793 * +------------------------------------------------------+
794 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
795 * +------------------------------------------------------+
796 * 63 48 47 32 31 20 19 0
798 * 82599+ Advanced Receive Descriptor (Read) Format
800 * +-----------------------------------------------------+
801 * 0 | Packet Buffer Address [63:1] |A0/NSE|
802 * +----------------------------------------------+------+
803 * 8 | Header Buffer Address [63:1] | DD |
804 * +-----------------------------------------------------+
807 * 82599+ Advanced Receive Descriptor (Write-Back) Format
809 * 63 48 47 32 31 30 21 20 17 16 4 3 0
810 * +------------------------------------------------------+
811 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS |
812 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type |
813 * |/ Flow Dir Flt ID | | | | | |
814 * +------------------------------------------------------+
815 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
816 * +------------------------------------------------------+
817 * 63 48 47 32 31 20 19 0
820 for (n = 0; n < adapter->num_rx_queues; n++) {
821 rx_ring = adapter->rx_ring[n];
822 pr_info("------------------------------------\n");
823 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
824 pr_info("------------------------------------\n");
826 "R [desc] [ PktBuf A0] ",
827 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ",
828 "<-- Adv Rx Read format");
830 "RWB[desc] [PcsmIpSHl PtRs] ",
831 "[vl er S cks ln] ---------------- [bi->skb ] ",
832 "<-- Adv Rx Write-Back format");
834 for (i = 0; i < rx_ring->count; i++) {
835 const char *ring_desc;
837 if (i == rx_ring->next_to_use)
839 else if (i == rx_ring->next_to_clean)
844 rx_buffer_info = &rx_ring->rx_buffer_info[i];
845 rx_desc = IXGBE_RX_DESC(rx_ring, i);
846 u0 = (struct my_u0 *)rx_desc;
847 if (rx_desc->wb.upper.length) {
848 /* Descriptor Done */
849 pr_info("RWB[0x%03X] %016llX %016llX ---------------- %p%s\n",
856 pr_info("R [0x%03X] %016llX %016llX %016llX %p%s\n",
860 (u64)rx_buffer_info->dma,
864 if (netif_msg_pktdata(adapter) &&
865 rx_buffer_info->dma) {
866 print_hex_dump(KERN_INFO, "",
867 DUMP_PREFIX_ADDRESS, 16, 1,
868 page_address(rx_buffer_info->page) +
869 rx_buffer_info->page_offset,
870 ixgbe_rx_bufsz(rx_ring), true);
877 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
881 /* Let firmware take over control of h/w */
882 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
883 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
884 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
887 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
891 /* Let firmware know the driver has taken over */
892 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
893 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
894 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
898 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
899 * @adapter: pointer to adapter struct
900 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
901 * @queue: queue to map the corresponding interrupt to
902 * @msix_vector: the vector to map to the corresponding queue
905 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
906 u8 queue, u8 msix_vector)
909 struct ixgbe_hw *hw = &adapter->hw;
910 switch (hw->mac.type) {
911 case ixgbe_mac_82598EB:
912 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
915 index = (((direction * 64) + queue) >> 2) & 0x1F;
916 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
917 ivar &= ~(0xFF << (8 * (queue & 0x3)));
918 ivar |= (msix_vector << (8 * (queue & 0x3)));
919 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
921 case ixgbe_mac_82599EB:
924 case ixgbe_mac_X550EM_x:
925 case ixgbe_mac_x550em_a:
926 if (direction == -1) {
928 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
929 index = ((queue & 1) * 8);
930 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
931 ivar &= ~(0xFF << index);
932 ivar |= (msix_vector << index);
933 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
936 /* tx or rx causes */
937 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
938 index = ((16 * (queue & 1)) + (8 * direction));
939 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
940 ivar &= ~(0xFF << index);
941 ivar |= (msix_vector << index);
942 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
950 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
955 switch (adapter->hw.mac.type) {
956 case ixgbe_mac_82598EB:
957 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
958 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
960 case ixgbe_mac_82599EB:
963 case ixgbe_mac_X550EM_x:
964 case ixgbe_mac_x550em_a:
965 mask = (qmask & 0xFFFFFFFF);
966 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
967 mask = (qmask >> 32);
968 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
975 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
977 struct ixgbe_hw *hw = &adapter->hw;
978 struct ixgbe_hw_stats *hwstats = &adapter->stats;
982 if ((hw->fc.current_mode != ixgbe_fc_full) &&
983 (hw->fc.current_mode != ixgbe_fc_rx_pause))
986 switch (hw->mac.type) {
987 case ixgbe_mac_82598EB:
988 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
991 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
993 hwstats->lxoffrxc += data;
995 /* refill credits (no tx hang) if we received xoff */
999 for (i = 0; i < adapter->num_tx_queues; i++)
1000 clear_bit(__IXGBE_HANG_CHECK_ARMED,
1001 &adapter->tx_ring[i]->state);
1003 for (i = 0; i < adapter->num_xdp_queues; i++)
1004 clear_bit(__IXGBE_HANG_CHECK_ARMED,
1005 &adapter->xdp_ring[i]->state);
1008 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
1010 struct ixgbe_hw *hw = &adapter->hw;
1011 struct ixgbe_hw_stats *hwstats = &adapter->stats;
1015 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
1017 if (adapter->ixgbe_ieee_pfc)
1018 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
1020 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
1021 ixgbe_update_xoff_rx_lfc(adapter);
1025 /* update stats for each tc, only valid with PFC enabled */
1026 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
1029 switch (hw->mac.type) {
1030 case ixgbe_mac_82598EB:
1031 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
1034 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
1036 hwstats->pxoffrxc[i] += pxoffrxc;
1037 /* Get the TC for given UP */
1038 tc = netdev_get_prio_tc_map(adapter->netdev, i);
1039 xoff[tc] += pxoffrxc;
1042 /* disarm tx queues that have received xoff frames */
1043 for (i = 0; i < adapter->num_tx_queues; i++) {
1044 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
1046 tc = tx_ring->dcb_tc;
1048 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1051 for (i = 0; i < adapter->num_xdp_queues; i++) {
1052 struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
1054 tc = xdp_ring->dcb_tc;
1056 clear_bit(__IXGBE_HANG_CHECK_ARMED, &xdp_ring->state);
1060 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1062 return ring->stats.packets;
1065 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1067 struct ixgbe_adapter *adapter;
1068 struct ixgbe_hw *hw;
1071 if (ring->l2_accel_priv)
1072 adapter = ring->l2_accel_priv->real_adapter;
1074 adapter = netdev_priv(ring->netdev);
1077 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
1078 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
1081 return (head < tail) ?
1082 tail - head : (tail + ring->count - head);
1087 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1089 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1090 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1091 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1093 clear_check_for_tx_hang(tx_ring);
1096 * Check for a hung queue, but be thorough. This verifies
1097 * that a transmit has been completed since the previous
1098 * check AND there is at least one packet pending. The
1099 * ARMED bit is set to indicate a potential hang. The
1100 * bit is cleared if a pause frame is received to remove
1101 * false hang detection due to PFC or 802.3x frames. By
1102 * requiring this to fail twice we avoid races with
1103 * pfc clearing the ARMED bit and conditions where we
1104 * run the check_tx_hang logic with a transmit completion
1105 * pending but without time to complete it yet.
1107 if (tx_done_old == tx_done && tx_pending)
1108 /* make sure it is true for two checks in a row */
1109 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1111 /* update completed stats and continue */
1112 tx_ring->tx_stats.tx_done_old = tx_done;
1113 /* reset the countdown */
1114 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1120 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1121 * @adapter: driver private struct
1123 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1126 /* Do the reset outside of interrupt context */
1127 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1128 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
1129 e_warn(drv, "initiating reset due to tx timeout\n");
1130 ixgbe_service_event_schedule(adapter);
1135 * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1137 static int ixgbe_tx_maxrate(struct net_device *netdev,
1138 int queue_index, u32 maxrate)
1140 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1141 struct ixgbe_hw *hw = &adapter->hw;
1142 u32 bcnrc_val = ixgbe_link_mbps(adapter);
1147 /* Calculate the rate factor values to set */
1148 bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT;
1149 bcnrc_val /= maxrate;
1151 /* clear everything but the rate factor */
1152 bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK |
1153 IXGBE_RTTBCNRC_RF_DEC_MASK;
1155 /* enable the rate scheduler */
1156 bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA;
1158 IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index);
1159 IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val);
1165 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1166 * @q_vector: structure containing interrupt and ring information
1167 * @tx_ring: tx ring to clean
1168 * @napi_budget: Used to determine if we are in netpoll
1170 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1171 struct ixgbe_ring *tx_ring, int napi_budget)
1173 struct ixgbe_adapter *adapter = q_vector->adapter;
1174 struct ixgbe_tx_buffer *tx_buffer;
1175 union ixgbe_adv_tx_desc *tx_desc;
1176 unsigned int total_bytes = 0, total_packets = 0;
1177 unsigned int budget = q_vector->tx.work_limit;
1178 unsigned int i = tx_ring->next_to_clean;
1180 if (test_bit(__IXGBE_DOWN, &adapter->state))
1183 tx_buffer = &tx_ring->tx_buffer_info[i];
1184 tx_desc = IXGBE_TX_DESC(tx_ring, i);
1185 i -= tx_ring->count;
1188 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1190 /* if next_to_watch is not set then there is no work pending */
1194 /* prevent any other reads prior to eop_desc */
1197 /* if DD is not set pending work has not been completed */
1198 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1201 /* clear next_to_watch to prevent false hangs */
1202 tx_buffer->next_to_watch = NULL;
1204 /* update the statistics for this packet */
1205 total_bytes += tx_buffer->bytecount;
1206 total_packets += tx_buffer->gso_segs;
1209 if (ring_is_xdp(tx_ring))
1210 page_frag_free(tx_buffer->data);
1212 napi_consume_skb(tx_buffer->skb, napi_budget);
1214 /* unmap skb header data */
1215 dma_unmap_single(tx_ring->dev,
1216 dma_unmap_addr(tx_buffer, dma),
1217 dma_unmap_len(tx_buffer, len),
1220 /* clear tx_buffer data */
1221 dma_unmap_len_set(tx_buffer, len, 0);
1223 /* unmap remaining buffers */
1224 while (tx_desc != eop_desc) {
1229 i -= tx_ring->count;
1230 tx_buffer = tx_ring->tx_buffer_info;
1231 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1234 /* unmap any remaining paged data */
1235 if (dma_unmap_len(tx_buffer, len)) {
1236 dma_unmap_page(tx_ring->dev,
1237 dma_unmap_addr(tx_buffer, dma),
1238 dma_unmap_len(tx_buffer, len),
1240 dma_unmap_len_set(tx_buffer, len, 0);
1244 /* move us one more past the eop_desc for start of next pkt */
1249 i -= tx_ring->count;
1250 tx_buffer = tx_ring->tx_buffer_info;
1251 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1254 /* issue prefetch for next Tx descriptor */
1257 /* update budget accounting */
1259 } while (likely(budget));
1261 i += tx_ring->count;
1262 tx_ring->next_to_clean = i;
1263 u64_stats_update_begin(&tx_ring->syncp);
1264 tx_ring->stats.bytes += total_bytes;
1265 tx_ring->stats.packets += total_packets;
1266 u64_stats_update_end(&tx_ring->syncp);
1267 q_vector->tx.total_bytes += total_bytes;
1268 q_vector->tx.total_packets += total_packets;
1270 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1271 /* schedule immediate reset if we believe we hung */
1272 struct ixgbe_hw *hw = &adapter->hw;
1273 e_err(drv, "Detected Tx Unit Hang %s\n"
1275 " TDH, TDT <%x>, <%x>\n"
1276 " next_to_use <%x>\n"
1277 " next_to_clean <%x>\n"
1278 "tx_buffer_info[next_to_clean]\n"
1279 " time_stamp <%lx>\n"
1281 ring_is_xdp(tx_ring) ? "(XDP)" : "",
1282 tx_ring->queue_index,
1283 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1284 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1285 tx_ring->next_to_use, i,
1286 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1288 if (!ring_is_xdp(tx_ring))
1289 netif_stop_subqueue(tx_ring->netdev,
1290 tx_ring->queue_index);
1293 "tx hang %d detected on queue %d, resetting adapter\n",
1294 adapter->tx_timeout_count + 1, tx_ring->queue_index);
1296 /* schedule immediate reset if we believe we hung */
1297 ixgbe_tx_timeout_reset(adapter);
1299 /* the adapter is about to reset, no point in enabling stuff */
1303 if (ring_is_xdp(tx_ring))
1306 netdev_tx_completed_queue(txring_txq(tx_ring),
1307 total_packets, total_bytes);
1309 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1310 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1311 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1312 /* Make sure that anybody stopping the queue after this
1313 * sees the new next_to_clean.
1316 if (__netif_subqueue_stopped(tx_ring->netdev,
1317 tx_ring->queue_index)
1318 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1319 netif_wake_subqueue(tx_ring->netdev,
1320 tx_ring->queue_index);
1321 ++tx_ring->tx_stats.restart_queue;
1328 #ifdef CONFIG_IXGBE_DCA
1329 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1330 struct ixgbe_ring *tx_ring,
1333 struct ixgbe_hw *hw = &adapter->hw;
1337 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1338 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1340 switch (hw->mac.type) {
1341 case ixgbe_mac_82598EB:
1342 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1344 case ixgbe_mac_82599EB:
1345 case ixgbe_mac_X540:
1346 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1347 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1350 /* for unknown hardware do not write register */
1355 * We can enable relaxed ordering for reads, but not writes when
1356 * DCA is enabled. This is due to a known issue in some chipsets
1357 * which will cause the DCA tag to be cleared.
1359 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1360 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1361 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1363 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1366 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1367 struct ixgbe_ring *rx_ring,
1370 struct ixgbe_hw *hw = &adapter->hw;
1372 u8 reg_idx = rx_ring->reg_idx;
1374 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1375 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1377 switch (hw->mac.type) {
1378 case ixgbe_mac_82599EB:
1379 case ixgbe_mac_X540:
1380 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1387 * We can enable relaxed ordering for reads, but not writes when
1388 * DCA is enabled. This is due to a known issue in some chipsets
1389 * which will cause the DCA tag to be cleared.
1391 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1392 IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1393 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1395 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1398 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1400 struct ixgbe_adapter *adapter = q_vector->adapter;
1401 struct ixgbe_ring *ring;
1402 int cpu = get_cpu();
1404 if (q_vector->cpu == cpu)
1407 ixgbe_for_each_ring(ring, q_vector->tx)
1408 ixgbe_update_tx_dca(adapter, ring, cpu);
1410 ixgbe_for_each_ring(ring, q_vector->rx)
1411 ixgbe_update_rx_dca(adapter, ring, cpu);
1413 q_vector->cpu = cpu;
1418 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1422 /* always use CB2 mode, difference is masked in the CB driver */
1423 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1424 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1425 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1427 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1428 IXGBE_DCA_CTRL_DCA_DISABLE);
1430 for (i = 0; i < adapter->num_q_vectors; i++) {
1431 adapter->q_vector[i]->cpu = -1;
1432 ixgbe_update_dca(adapter->q_vector[i]);
1436 static int __ixgbe_notify_dca(struct device *dev, void *data)
1438 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1439 unsigned long event = *(unsigned long *)data;
1441 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1445 case DCA_PROVIDER_ADD:
1446 /* if we're already enabled, don't do it again */
1447 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1449 if (dca_add_requester(dev) == 0) {
1450 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1451 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1452 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1455 /* fall through - DCA is disabled. */
1456 case DCA_PROVIDER_REMOVE:
1457 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1458 dca_remove_requester(dev);
1459 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1460 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1461 IXGBE_DCA_CTRL_DCA_DISABLE);
1469 #endif /* CONFIG_IXGBE_DCA */
1471 #define IXGBE_RSS_L4_TYPES_MASK \
1472 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1473 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1474 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1475 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1477 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1478 union ixgbe_adv_rx_desc *rx_desc,
1479 struct sk_buff *skb)
1483 if (!(ring->netdev->features & NETIF_F_RXHASH))
1486 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1487 IXGBE_RXDADV_RSSTYPE_MASK;
1492 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1493 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1494 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1499 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1500 * @ring: structure containing ring specific data
1501 * @rx_desc: advanced rx descriptor
1503 * Returns : true if it is FCoE pkt
1505 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1506 union ixgbe_adv_rx_desc *rx_desc)
1508 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1510 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1511 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1512 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1513 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1516 #endif /* IXGBE_FCOE */
1518 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1519 * @ring: structure containing ring specific data
1520 * @rx_desc: current Rx descriptor being processed
1521 * @skb: skb currently being received and modified
1523 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1524 union ixgbe_adv_rx_desc *rx_desc,
1525 struct sk_buff *skb)
1527 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1528 bool encap_pkt = false;
1530 skb_checksum_none_assert(skb);
1532 /* Rx csum disabled */
1533 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1536 /* check for VXLAN and Geneve packets */
1537 if (pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) {
1539 skb->encapsulation = 1;
1542 /* if IP and error */
1543 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1544 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1545 ring->rx_stats.csum_err++;
1549 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1552 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1554 * 82599 errata, UDP frames with a 0 checksum can be marked as
1557 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1558 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1561 ring->rx_stats.csum_err++;
1565 /* It must be a TCP or UDP packet with a valid checksum */
1566 skb->ip_summed = CHECKSUM_UNNECESSARY;
1568 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1571 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1572 skb->ip_summed = CHECKSUM_NONE;
1575 /* If we checked the outer header let the stack know */
1576 skb->csum_level = 1;
1580 static inline unsigned int ixgbe_rx_offset(struct ixgbe_ring *rx_ring)
1582 return ring_uses_build_skb(rx_ring) ? IXGBE_SKB_PAD : 0;
1585 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1586 struct ixgbe_rx_buffer *bi)
1588 struct page *page = bi->page;
1591 /* since we are recycling buffers we should seldom need to alloc */
1595 /* alloc new page for storage */
1596 page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1597 if (unlikely(!page)) {
1598 rx_ring->rx_stats.alloc_rx_page_failed++;
1602 /* map page for use */
1603 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1604 ixgbe_rx_pg_size(rx_ring),
1609 * if mapping failed free memory back to system since
1610 * there isn't much point in holding memory we can't use
1612 if (dma_mapping_error(rx_ring->dev, dma)) {
1613 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1615 rx_ring->rx_stats.alloc_rx_page_failed++;
1621 bi->page_offset = ixgbe_rx_offset(rx_ring);
1622 bi->pagecnt_bias = 1;
1623 rx_ring->rx_stats.alloc_rx_page++;
1629 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1630 * @rx_ring: ring to place buffers on
1631 * @cleaned_count: number of buffers to replace
1633 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1635 union ixgbe_adv_rx_desc *rx_desc;
1636 struct ixgbe_rx_buffer *bi;
1637 u16 i = rx_ring->next_to_use;
1644 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1645 bi = &rx_ring->rx_buffer_info[i];
1646 i -= rx_ring->count;
1648 bufsz = ixgbe_rx_bufsz(rx_ring);
1651 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1654 /* sync the buffer for use by the device */
1655 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1656 bi->page_offset, bufsz,
1660 * Refresh the desc even if buffer_addrs didn't change
1661 * because each write-back erases this info.
1663 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1669 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1670 bi = rx_ring->rx_buffer_info;
1671 i -= rx_ring->count;
1674 /* clear the length for the next_to_use descriptor */
1675 rx_desc->wb.upper.length = 0;
1678 } while (cleaned_count);
1680 i += rx_ring->count;
1682 if (rx_ring->next_to_use != i) {
1683 rx_ring->next_to_use = i;
1685 /* update next to alloc since we have filled the ring */
1686 rx_ring->next_to_alloc = i;
1688 /* Force memory writes to complete before letting h/w
1689 * know there are new descriptors to fetch. (Only
1690 * applicable for weak-ordered memory model archs,
1694 writel(i, rx_ring->tail);
1698 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1699 struct sk_buff *skb)
1701 u16 hdr_len = skb_headlen(skb);
1703 /* set gso_size to avoid messing up TCP MSS */
1704 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1705 IXGBE_CB(skb)->append_cnt);
1706 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1709 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1710 struct sk_buff *skb)
1712 /* if append_cnt is 0 then frame is not RSC */
1713 if (!IXGBE_CB(skb)->append_cnt)
1716 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1717 rx_ring->rx_stats.rsc_flush++;
1719 ixgbe_set_rsc_gso_size(rx_ring, skb);
1721 /* gso_size is computed using append_cnt so always clear it last */
1722 IXGBE_CB(skb)->append_cnt = 0;
1726 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1727 * @rx_ring: rx descriptor ring packet is being transacted on
1728 * @rx_desc: pointer to the EOP Rx descriptor
1729 * @skb: pointer to current skb being populated
1731 * This function checks the ring, descriptor, and packet information in
1732 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1733 * other fields within the skb.
1735 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1736 union ixgbe_adv_rx_desc *rx_desc,
1737 struct sk_buff *skb)
1739 struct net_device *dev = rx_ring->netdev;
1740 u32 flags = rx_ring->q_vector->adapter->flags;
1742 ixgbe_update_rsc_stats(rx_ring, skb);
1744 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1746 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1748 if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1749 ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1751 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1752 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1753 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1754 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1757 skb_record_rx_queue(skb, rx_ring->queue_index);
1759 skb->protocol = eth_type_trans(skb, dev);
1762 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1763 struct sk_buff *skb)
1765 napi_gro_receive(&q_vector->napi, skb);
1769 * ixgbe_is_non_eop - process handling of non-EOP buffers
1770 * @rx_ring: Rx ring being processed
1771 * @rx_desc: Rx descriptor for current buffer
1772 * @skb: Current socket buffer containing buffer in progress
1774 * This function updates next to clean. If the buffer is an EOP buffer
1775 * this function exits returning false, otherwise it will place the
1776 * sk_buff in the next buffer to be chained and return true indicating
1777 * that this is in fact a non-EOP buffer.
1779 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1780 union ixgbe_adv_rx_desc *rx_desc,
1781 struct sk_buff *skb)
1783 u32 ntc = rx_ring->next_to_clean + 1;
1785 /* fetch, update, and store next to clean */
1786 ntc = (ntc < rx_ring->count) ? ntc : 0;
1787 rx_ring->next_to_clean = ntc;
1789 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1791 /* update RSC append count if present */
1792 if (ring_is_rsc_enabled(rx_ring)) {
1793 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1794 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1796 if (unlikely(rsc_enabled)) {
1797 u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1799 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1800 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1802 /* update ntc based on RSC value */
1803 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1804 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1805 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1809 /* if we are the last buffer then there is nothing else to do */
1810 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1813 /* place skb in next buffer to be received */
1814 rx_ring->rx_buffer_info[ntc].skb = skb;
1815 rx_ring->rx_stats.non_eop_descs++;
1821 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1822 * @rx_ring: rx descriptor ring packet is being transacted on
1823 * @skb: pointer to current skb being adjusted
1825 * This function is an ixgbe specific version of __pskb_pull_tail. The
1826 * main difference between this version and the original function is that
1827 * this function can make several assumptions about the state of things
1828 * that allow for significant optimizations versus the standard function.
1829 * As a result we can do things like drop a frag and maintain an accurate
1830 * truesize for the skb.
1832 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1833 struct sk_buff *skb)
1835 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1837 unsigned int pull_len;
1840 * it is valid to use page_address instead of kmap since we are
1841 * working with pages allocated out of the lomem pool per
1842 * alloc_page(GFP_ATOMIC)
1844 va = skb_frag_address(frag);
1847 * we need the header to contain the greater of either ETH_HLEN or
1848 * 60 bytes if the skb->len is less than 60 for skb_pad.
1850 pull_len = eth_get_headlen(va, IXGBE_RX_HDR_SIZE);
1852 /* align pull length to size of long to optimize memcpy performance */
1853 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1855 /* update all of the pointers */
1856 skb_frag_size_sub(frag, pull_len);
1857 frag->page_offset += pull_len;
1858 skb->data_len -= pull_len;
1859 skb->tail += pull_len;
1863 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1864 * @rx_ring: rx descriptor ring packet is being transacted on
1865 * @skb: pointer to current skb being updated
1867 * This function provides a basic DMA sync up for the first fragment of an
1868 * skb. The reason for doing this is that the first fragment cannot be
1869 * unmapped until we have reached the end of packet descriptor for a buffer
1872 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1873 struct sk_buff *skb)
1875 /* if the page was released unmap it, else just sync our portion */
1876 if (unlikely(IXGBE_CB(skb)->page_released)) {
1877 dma_unmap_page_attrs(rx_ring->dev, IXGBE_CB(skb)->dma,
1878 ixgbe_rx_pg_size(rx_ring),
1882 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1884 dma_sync_single_range_for_cpu(rx_ring->dev,
1887 skb_frag_size(frag),
1893 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1894 * @rx_ring: rx descriptor ring packet is being transacted on
1895 * @rx_desc: pointer to the EOP Rx descriptor
1896 * @skb: pointer to current skb being fixed
1898 * Check if the skb is valid in the XDP case it will be an error pointer.
1899 * Return true in this case to abort processing and advance to next
1902 * Check for corrupted packet headers caused by senders on the local L2
1903 * embedded NIC switch not setting up their Tx Descriptors right. These
1904 * should be very rare.
1906 * Also address the case where we are pulling data in on pages only
1907 * and as such no data is present in the skb header.
1909 * In addition if skb is not at least 60 bytes we need to pad it so that
1910 * it is large enough to qualify as a valid Ethernet frame.
1912 * Returns true if an error was encountered and skb was freed.
1914 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1915 union ixgbe_adv_rx_desc *rx_desc,
1916 struct sk_buff *skb)
1918 struct net_device *netdev = rx_ring->netdev;
1920 /* XDP packets use error pointer so abort at this point */
1924 /* verify that the packet does not have any known errors */
1925 if (unlikely(ixgbe_test_staterr(rx_desc,
1926 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1927 !(netdev->features & NETIF_F_RXALL))) {
1928 dev_kfree_skb_any(skb);
1932 /* place header in linear portion of buffer */
1933 if (!skb_headlen(skb))
1934 ixgbe_pull_tail(rx_ring, skb);
1937 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1938 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1942 /* if eth_skb_pad returns an error the skb was freed */
1943 if (eth_skb_pad(skb))
1950 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1951 * @rx_ring: rx descriptor ring to store buffers on
1952 * @old_buff: donor buffer to have page reused
1954 * Synchronizes page for reuse by the adapter
1956 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1957 struct ixgbe_rx_buffer *old_buff)
1959 struct ixgbe_rx_buffer *new_buff;
1960 u16 nta = rx_ring->next_to_alloc;
1962 new_buff = &rx_ring->rx_buffer_info[nta];
1964 /* update, and store next to alloc */
1966 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1968 /* Transfer page from old buffer to new buffer.
1969 * Move each member individually to avoid possible store
1970 * forwarding stalls and unnecessary copy of skb.
1972 new_buff->dma = old_buff->dma;
1973 new_buff->page = old_buff->page;
1974 new_buff->page_offset = old_buff->page_offset;
1975 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
1978 static inline bool ixgbe_page_is_reserved(struct page *page)
1980 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1983 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer)
1985 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1986 struct page *page = rx_buffer->page;
1988 /* avoid re-using remote pages */
1989 if (unlikely(ixgbe_page_is_reserved(page)))
1992 #if (PAGE_SIZE < 8192)
1993 /* if we are only owner of page we can reuse it */
1994 if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
1997 /* The last offset is a bit aggressive in that we assume the
1998 * worst case of FCoE being enabled and using a 3K buffer.
1999 * However this should have minimal impact as the 1K extra is
2000 * still less than one buffer in size.
2002 #define IXGBE_LAST_OFFSET \
2003 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
2004 if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
2008 /* If we have drained the page fragment pool we need to update
2009 * the pagecnt_bias and page count so that we fully restock the
2010 * number of references the driver holds.
2012 if (unlikely(!pagecnt_bias)) {
2013 page_ref_add(page, USHRT_MAX);
2014 rx_buffer->pagecnt_bias = USHRT_MAX;
2021 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
2022 * @rx_ring: rx descriptor ring to transact packets on
2023 * @rx_buffer: buffer containing page to add
2024 * @rx_desc: descriptor containing length of buffer written by hardware
2025 * @skb: sk_buff to place the data into
2027 * This function will add the data contained in rx_buffer->page to the skb.
2028 * This is done either through a direct copy if the data in the buffer is
2029 * less than the skb header size, otherwise it will just attach the page as
2030 * a frag to the skb.
2032 * The function will then update the page offset if necessary and return
2033 * true if the buffer can be reused by the adapter.
2035 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
2036 struct ixgbe_rx_buffer *rx_buffer,
2037 struct sk_buff *skb,
2040 #if (PAGE_SIZE < 8192)
2041 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2043 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2044 SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2045 SKB_DATA_ALIGN(size);
2047 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2048 rx_buffer->page_offset, size, truesize);
2049 #if (PAGE_SIZE < 8192)
2050 rx_buffer->page_offset ^= truesize;
2052 rx_buffer->page_offset += truesize;
2056 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2057 union ixgbe_adv_rx_desc *rx_desc,
2058 struct sk_buff **skb,
2059 const unsigned int size)
2061 struct ixgbe_rx_buffer *rx_buffer;
2063 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2064 prefetchw(rx_buffer->page);
2065 *skb = rx_buffer->skb;
2067 /* Delay unmapping of the first packet. It carries the header
2068 * information, HW may still access the header after the writeback.
2069 * Only unmap it when EOP is reached
2071 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2076 ixgbe_dma_sync_frag(rx_ring, *skb);
2079 /* we are reusing so sync this buffer for CPU use */
2080 dma_sync_single_range_for_cpu(rx_ring->dev,
2082 rx_buffer->page_offset,
2086 rx_buffer->pagecnt_bias--;
2091 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2092 struct ixgbe_rx_buffer *rx_buffer,
2093 struct sk_buff *skb)
2095 if (ixgbe_can_reuse_rx_page(rx_buffer)) {
2096 /* hand second half of page back to the ring */
2097 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2099 if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2100 /* the page has been released from the ring */
2101 IXGBE_CB(skb)->page_released = true;
2103 /* we are not reusing the buffer so unmap it */
2104 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2105 ixgbe_rx_pg_size(rx_ring),
2109 __page_frag_cache_drain(rx_buffer->page,
2110 rx_buffer->pagecnt_bias);
2113 /* clear contents of rx_buffer */
2114 rx_buffer->page = NULL;
2115 rx_buffer->skb = NULL;
2118 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2119 struct ixgbe_rx_buffer *rx_buffer,
2120 struct xdp_buff *xdp,
2121 union ixgbe_adv_rx_desc *rx_desc)
2123 unsigned int size = xdp->data_end - xdp->data;
2124 #if (PAGE_SIZE < 8192)
2125 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2127 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2128 xdp->data_hard_start);
2130 struct sk_buff *skb;
2132 /* prefetch first cache line of first page */
2133 prefetch(xdp->data);
2134 #if L1_CACHE_BYTES < 128
2135 prefetch(xdp->data + L1_CACHE_BYTES);
2137 /* Note, we get here by enabling legacy-rx via:
2139 * ethtool --set-priv-flags <dev> legacy-rx on
2141 * In this mode, we currently get 0 extra XDP headroom as
2142 * opposed to having legacy-rx off, where we process XDP
2143 * packets going to stack via ixgbe_build_skb(). The latter
2144 * provides us currently with 192 bytes of headroom.
2146 * For ixgbe_construct_skb() mode it means that the
2147 * xdp->data_meta will always point to xdp->data, since
2148 * the helper cannot expand the head. Should this ever
2149 * change in future for legacy-rx mode on, then lets also
2150 * add xdp->data_meta handling here.
2153 /* allocate a skb to store the frags */
2154 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2158 if (size > IXGBE_RX_HDR_SIZE) {
2159 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2160 IXGBE_CB(skb)->dma = rx_buffer->dma;
2162 skb_add_rx_frag(skb, 0, rx_buffer->page,
2163 xdp->data - page_address(rx_buffer->page),
2165 #if (PAGE_SIZE < 8192)
2166 rx_buffer->page_offset ^= truesize;
2168 rx_buffer->page_offset += truesize;
2171 memcpy(__skb_put(skb, size),
2172 xdp->data, ALIGN(size, sizeof(long)));
2173 rx_buffer->pagecnt_bias++;
2179 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2180 struct ixgbe_rx_buffer *rx_buffer,
2181 struct xdp_buff *xdp,
2182 union ixgbe_adv_rx_desc *rx_desc)
2184 unsigned int metasize = xdp->data - xdp->data_meta;
2185 #if (PAGE_SIZE < 8192)
2186 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2188 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2189 SKB_DATA_ALIGN(xdp->data_end -
2190 xdp->data_hard_start);
2192 struct sk_buff *skb;
2194 /* Prefetch first cache line of first page. If xdp->data_meta
2195 * is unused, this points extactly as xdp->data, otherwise we
2196 * likely have a consumer accessing first few bytes of meta
2197 * data, and then actual data.
2199 prefetch(xdp->data_meta);
2200 #if L1_CACHE_BYTES < 128
2201 prefetch(xdp->data_meta + L1_CACHE_BYTES);
2204 /* build an skb to around the page buffer */
2205 skb = build_skb(xdp->data_hard_start, truesize);
2209 /* update pointers within the skb to store the data */
2210 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2211 __skb_put(skb, xdp->data_end - xdp->data);
2213 skb_metadata_set(skb, metasize);
2215 /* record DMA address if this is the start of a chain of buffers */
2216 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2217 IXGBE_CB(skb)->dma = rx_buffer->dma;
2219 /* update buffer offset */
2220 #if (PAGE_SIZE < 8192)
2221 rx_buffer->page_offset ^= truesize;
2223 rx_buffer->page_offset += truesize;
2229 #define IXGBE_XDP_PASS 0
2230 #define IXGBE_XDP_CONSUMED 1
2231 #define IXGBE_XDP_TX 2
2233 static int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
2234 struct xdp_buff *xdp);
2236 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2237 struct ixgbe_ring *rx_ring,
2238 struct xdp_buff *xdp)
2240 int err, result = IXGBE_XDP_PASS;
2241 struct bpf_prog *xdp_prog;
2245 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2250 act = bpf_prog_run_xdp(xdp_prog, xdp);
2255 result = ixgbe_xmit_xdp_ring(adapter, xdp);
2258 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2260 result = IXGBE_XDP_TX;
2262 result = IXGBE_XDP_CONSUMED;
2265 bpf_warn_invalid_xdp_action(act);
2268 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2269 /* fallthrough -- handle aborts by dropping packet */
2271 result = IXGBE_XDP_CONSUMED;
2276 return ERR_PTR(-result);
2279 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2280 struct ixgbe_rx_buffer *rx_buffer,
2283 #if (PAGE_SIZE < 8192)
2284 unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2286 rx_buffer->page_offset ^= truesize;
2288 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2289 SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2290 SKB_DATA_ALIGN(size);
2292 rx_buffer->page_offset += truesize;
2297 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2298 * @q_vector: structure containing interrupt and ring information
2299 * @rx_ring: rx descriptor ring to transact packets on
2300 * @budget: Total limit on number of packets to process
2302 * This function provides a "bounce buffer" approach to Rx interrupt
2303 * processing. The advantage to this is that on systems that have
2304 * expensive overhead for IOMMU access this provides a means of avoiding
2305 * it by maintaining the mapping of the page to the syste.
2307 * Returns amount of work completed
2309 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2310 struct ixgbe_ring *rx_ring,
2313 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2314 struct ixgbe_adapter *adapter = q_vector->adapter;
2317 unsigned int mss = 0;
2318 #endif /* IXGBE_FCOE */
2319 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2320 bool xdp_xmit = false;
2321 struct xdp_buff xdp;
2323 xdp.rxq = &rx_ring->xdp_rxq;
2325 while (likely(total_rx_packets < budget)) {
2326 union ixgbe_adv_rx_desc *rx_desc;
2327 struct ixgbe_rx_buffer *rx_buffer;
2328 struct sk_buff *skb;
2331 /* return some buffers to hardware, one at a time is too slow */
2332 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2333 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2337 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2338 size = le16_to_cpu(rx_desc->wb.upper.length);
2342 /* This memory barrier is needed to keep us from reading
2343 * any other fields out of the rx_desc until we know the
2344 * descriptor has been written back
2348 rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size);
2350 /* retrieve a buffer from the ring */
2352 xdp.data = page_address(rx_buffer->page) +
2353 rx_buffer->page_offset;
2354 xdp.data_meta = xdp.data;
2355 xdp.data_hard_start = xdp.data -
2356 ixgbe_rx_offset(rx_ring);
2357 xdp.data_end = xdp.data + size;
2359 skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2363 if (PTR_ERR(skb) == -IXGBE_XDP_TX) {
2365 ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2367 rx_buffer->pagecnt_bias++;
2370 total_rx_bytes += size;
2372 ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2373 } else if (ring_uses_build_skb(rx_ring)) {
2374 skb = ixgbe_build_skb(rx_ring, rx_buffer,
2377 skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2381 /* exit if we failed to retrieve a buffer */
2383 rx_ring->rx_stats.alloc_rx_buff_failed++;
2384 rx_buffer->pagecnt_bias++;
2388 ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb);
2391 /* place incomplete frames back on ring for completion */
2392 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2395 /* verify the packet layout is correct */
2396 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2399 /* probably a little skewed due to removing CRC */
2400 total_rx_bytes += skb->len;
2402 /* populate checksum, timestamp, VLAN, and protocol */
2403 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2406 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2407 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2408 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2409 /* include DDPed FCoE data */
2410 if (ddp_bytes > 0) {
2412 mss = rx_ring->netdev->mtu -
2413 sizeof(struct fcoe_hdr) -
2414 sizeof(struct fc_frame_header) -
2415 sizeof(struct fcoe_crc_eof);
2419 total_rx_bytes += ddp_bytes;
2420 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2424 dev_kfree_skb_any(skb);
2429 #endif /* IXGBE_FCOE */
2430 ixgbe_rx_skb(q_vector, skb);
2432 /* update budget accounting */
2437 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2439 /* Force memory writes to complete before letting h/w
2440 * know there are new descriptors to fetch.
2443 writel(ring->next_to_use, ring->tail);
2448 u64_stats_update_begin(&rx_ring->syncp);
2449 rx_ring->stats.packets += total_rx_packets;
2450 rx_ring->stats.bytes += total_rx_bytes;
2451 u64_stats_update_end(&rx_ring->syncp);
2452 q_vector->rx.total_packets += total_rx_packets;
2453 q_vector->rx.total_bytes += total_rx_bytes;
2455 return total_rx_packets;
2459 * ixgbe_configure_msix - Configure MSI-X hardware
2460 * @adapter: board private structure
2462 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2465 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2467 struct ixgbe_q_vector *q_vector;
2471 /* Populate MSIX to EITR Select */
2472 if (adapter->num_vfs > 32) {
2473 u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2474 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2478 * Populate the IVAR table and set the ITR values to the
2479 * corresponding register.
2481 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2482 struct ixgbe_ring *ring;
2483 q_vector = adapter->q_vector[v_idx];
2485 ixgbe_for_each_ring(ring, q_vector->rx)
2486 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2488 ixgbe_for_each_ring(ring, q_vector->tx)
2489 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2491 ixgbe_write_eitr(q_vector);
2494 switch (adapter->hw.mac.type) {
2495 case ixgbe_mac_82598EB:
2496 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2499 case ixgbe_mac_82599EB:
2500 case ixgbe_mac_X540:
2501 case ixgbe_mac_X550:
2502 case ixgbe_mac_X550EM_x:
2503 case ixgbe_mac_x550em_a:
2504 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2509 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2511 /* set up to autoclear timer, and the vectors */
2512 mask = IXGBE_EIMS_ENABLE_MASK;
2513 mask &= ~(IXGBE_EIMS_OTHER |
2514 IXGBE_EIMS_MAILBOX |
2517 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2521 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2522 * @q_vector: structure containing interrupt and ring information
2523 * @ring_container: structure containing ring performance data
2525 * Stores a new ITR value based on packets and byte
2526 * counts during the last interrupt. The advantage of per interrupt
2527 * computation is faster updates and more accurate ITR for the current
2528 * traffic pattern. Constants in this function were computed
2529 * based on theoretical maximum wire speed and thresholds were set based
2530 * on testing data as well as attempting to minimize response time
2531 * while increasing bulk throughput.
2533 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2534 struct ixgbe_ring_container *ring_container)
2536 unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2537 IXGBE_ITR_ADAPTIVE_LATENCY;
2538 unsigned int avg_wire_size, packets, bytes;
2539 unsigned long next_update = jiffies;
2541 /* If we don't have any rings just leave ourselves set for maximum
2542 * possible latency so we take ourselves out of the equation.
2544 if (!ring_container->ring)
2547 /* If we didn't update within up to 1 - 2 jiffies we can assume
2548 * that either packets are coming in so slow there hasn't been
2549 * any work, or that there is so much work that NAPI is dealing
2550 * with interrupt moderation and we don't need to do anything.
2552 if (time_after(next_update, ring_container->next_update))
2555 packets = ring_container->total_packets;
2557 /* We have no packets to actually measure against. This means
2558 * either one of the other queues on this vector is active or
2559 * we are a Tx queue doing TSO with too high of an interrupt rate.
2561 * When this occurs just tick up our delay by the minimum value
2562 * and hope that this extra delay will prevent us from being called
2563 * without any work on our queue.
2566 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2567 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2568 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2569 itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2573 bytes = ring_container->total_bytes;
2575 /* If packets are less than 4 or bytes are less than 9000 assume
2576 * insufficient data to use bulk rate limiting approach. We are
2577 * likely latency driven.
2579 if (packets < 4 && bytes < 9000) {
2580 itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2581 goto adjust_by_size;
2584 /* Between 4 and 48 we can assume that our current interrupt delay
2585 * is only slightly too low. As such we should increase it by a small
2589 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2590 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2591 itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2595 /* Between 48 and 96 is our "goldilocks" zone where we are working
2596 * out "just right". Just report that our current ITR is good for us.
2599 itr = q_vector->itr >> 2;
2603 /* If packet count is 96 or greater we are likely looking at a slight
2604 * overrun of the delay we want. Try halving our delay to see if that
2605 * will cut the number of packets in half per interrupt.
2607 if (packets < 256) {
2608 itr = q_vector->itr >> 3;
2609 if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2610 itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2614 /* The paths below assume we are dealing with a bulk ITR since number
2615 * of packets is 256 or greater. We are just going to have to compute
2616 * a value and try to bring the count under control, though for smaller
2617 * packet sizes there isn't much we can do as NAPI polling will likely
2618 * be kicking in sooner rather than later.
2620 itr = IXGBE_ITR_ADAPTIVE_BULK;
2623 /* If packet counts are 256 or greater we can assume we have a gross
2624 * overestimation of what the rate should be. Instead of trying to fine
2625 * tune it just use the formula below to try and dial in an exact value
2626 * give the current packet size of the frame.
2628 avg_wire_size = bytes / packets;
2630 /* The following is a crude approximation of:
2631 * wmem_default / (size + overhead) = desired_pkts_per_int
2632 * rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2633 * (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2635 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2636 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2639 * (170 * (size + 24)) / (size + 640) = ITR
2641 * We first do some math on the packet size and then finally bitshift
2642 * by 8 after rounding up. We also have to account for PCIe link speed
2643 * difference as ITR scales based on this.
2645 if (avg_wire_size <= 60) {
2646 /* Start at 50k ints/sec */
2647 avg_wire_size = 5120;
2648 } else if (avg_wire_size <= 316) {
2649 /* 50K ints/sec to 16K ints/sec */
2650 avg_wire_size *= 40;
2651 avg_wire_size += 2720;
2652 } else if (avg_wire_size <= 1084) {
2653 /* 16K ints/sec to 9.2K ints/sec */
2654 avg_wire_size *= 15;
2655 avg_wire_size += 11452;
2656 } else if (avg_wire_size <= 1980) {
2657 /* 9.2K ints/sec to 8K ints/sec */
2659 avg_wire_size += 22420;
2661 /* plateau at a limit of 8K ints/sec */
2662 avg_wire_size = 32256;
2665 /* If we are in low latency mode half our delay which doubles the rate
2666 * to somewhere between 100K to 16K ints/sec
2668 if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2669 avg_wire_size >>= 1;
2671 /* Resultant value is 256 times larger than it needs to be. This
2672 * gives us room to adjust the value as needed to either increase
2673 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2675 * Use addition as we have already recorded the new latency flag
2676 * for the ITR value.
2678 switch (q_vector->adapter->link_speed) {
2679 case IXGBE_LINK_SPEED_10GB_FULL:
2680 case IXGBE_LINK_SPEED_100_FULL:
2682 itr += DIV_ROUND_UP(avg_wire_size,
2683 IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2684 IXGBE_ITR_ADAPTIVE_MIN_INC;
2686 case IXGBE_LINK_SPEED_2_5GB_FULL:
2687 case IXGBE_LINK_SPEED_1GB_FULL:
2688 case IXGBE_LINK_SPEED_10_FULL:
2689 itr += DIV_ROUND_UP(avg_wire_size,
2690 IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2691 IXGBE_ITR_ADAPTIVE_MIN_INC;
2696 /* write back value */
2697 ring_container->itr = itr;
2699 /* next update should occur within next jiffy */
2700 ring_container->next_update = next_update + 1;
2702 ring_container->total_bytes = 0;
2703 ring_container->total_packets = 0;
2707 * ixgbe_write_eitr - write EITR register in hardware specific way
2708 * @q_vector: structure containing interrupt and ring information
2710 * This function is made to be called by ethtool and by the driver
2711 * when it needs to update EITR registers at runtime. Hardware
2712 * specific quirks/differences are taken care of here.
2714 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2716 struct ixgbe_adapter *adapter = q_vector->adapter;
2717 struct ixgbe_hw *hw = &adapter->hw;
2718 int v_idx = q_vector->v_idx;
2719 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2721 switch (adapter->hw.mac.type) {
2722 case ixgbe_mac_82598EB:
2723 /* must write high and low 16 bits to reset counter */
2724 itr_reg |= (itr_reg << 16);
2726 case ixgbe_mac_82599EB:
2727 case ixgbe_mac_X540:
2728 case ixgbe_mac_X550:
2729 case ixgbe_mac_X550EM_x:
2730 case ixgbe_mac_x550em_a:
2732 * set the WDIS bit to not clear the timer bits and cause an
2733 * immediate assertion of the interrupt
2735 itr_reg |= IXGBE_EITR_CNT_WDIS;
2740 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2743 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2747 ixgbe_update_itr(q_vector, &q_vector->tx);
2748 ixgbe_update_itr(q_vector, &q_vector->rx);
2750 /* use the smallest value of new ITR delay calculations */
2751 new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2753 /* Clear latency flag if set, shift into correct position */
2754 new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2757 if (new_itr != q_vector->itr) {
2758 /* save the algorithm value here */
2759 q_vector->itr = new_itr;
2761 ixgbe_write_eitr(q_vector);
2766 * ixgbe_check_overtemp_subtask - check for over temperature
2767 * @adapter: pointer to adapter
2769 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2771 struct ixgbe_hw *hw = &adapter->hw;
2772 u32 eicr = adapter->interrupt_event;
2775 if (test_bit(__IXGBE_DOWN, &adapter->state))
2778 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2781 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2783 switch (hw->device_id) {
2784 case IXGBE_DEV_ID_82599_T3_LOM:
2786 * Since the warning interrupt is for both ports
2787 * we don't have to check if:
2788 * - This interrupt wasn't for our port.
2789 * - We may have missed the interrupt so always have to
2790 * check if we got a LSC
2792 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2793 !(eicr & IXGBE_EICR_LSC))
2796 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2798 bool link_up = false;
2800 hw->mac.ops.check_link(hw, &speed, &link_up, false);
2806 /* Check if this is not due to overtemp */
2807 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2811 case IXGBE_DEV_ID_X550EM_A_1G_T:
2812 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2813 rc = hw->phy.ops.check_overtemp(hw);
2814 if (rc != IXGBE_ERR_OVERTEMP)
2818 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2820 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2824 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2826 adapter->interrupt_event = 0;
2829 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2831 struct ixgbe_hw *hw = &adapter->hw;
2833 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2834 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2835 e_crit(probe, "Fan has stopped, replace the adapter\n");
2836 /* write to clear the interrupt */
2837 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2841 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2843 struct ixgbe_hw *hw = &adapter->hw;
2845 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2848 switch (adapter->hw.mac.type) {
2849 case ixgbe_mac_82599EB:
2851 * Need to check link state so complete overtemp check
2854 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2855 (eicr & IXGBE_EICR_LSC)) &&
2856 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2857 adapter->interrupt_event = eicr;
2858 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2859 ixgbe_service_event_schedule(adapter);
2863 case ixgbe_mac_x550em_a:
2864 if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2865 adapter->interrupt_event = eicr;
2866 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2867 ixgbe_service_event_schedule(adapter);
2868 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2869 IXGBE_EICR_GPI_SDP0_X550EM_a);
2870 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2871 IXGBE_EICR_GPI_SDP0_X550EM_a);
2874 case ixgbe_mac_X550:
2875 case ixgbe_mac_X540:
2876 if (!(eicr & IXGBE_EICR_TS))
2883 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2886 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2888 switch (hw->mac.type) {
2889 case ixgbe_mac_82598EB:
2890 if (hw->phy.type == ixgbe_phy_nl)
2893 case ixgbe_mac_82599EB:
2894 case ixgbe_mac_X550EM_x:
2895 case ixgbe_mac_x550em_a:
2896 switch (hw->mac.ops.get_media_type(hw)) {
2897 case ixgbe_media_type_fiber:
2898 case ixgbe_media_type_fiber_qsfp:
2908 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2910 struct ixgbe_hw *hw = &adapter->hw;
2911 u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2913 if (!ixgbe_is_sfp(hw))
2916 /* Later MAC's use different SDP */
2917 if (hw->mac.type >= ixgbe_mac_X540)
2918 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2920 if (eicr & eicr_mask) {
2921 /* Clear the interrupt */
2922 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2923 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2924 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2925 adapter->sfp_poll_time = 0;
2926 ixgbe_service_event_schedule(adapter);
2930 if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2931 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2932 /* Clear the interrupt */
2933 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2934 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2935 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2936 ixgbe_service_event_schedule(adapter);
2941 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2943 struct ixgbe_hw *hw = &adapter->hw;
2946 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2947 adapter->link_check_timeout = jiffies;
2948 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2949 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2950 IXGBE_WRITE_FLUSH(hw);
2951 ixgbe_service_event_schedule(adapter);
2955 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2959 struct ixgbe_hw *hw = &adapter->hw;
2961 switch (hw->mac.type) {
2962 case ixgbe_mac_82598EB:
2963 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2964 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2966 case ixgbe_mac_82599EB:
2967 case ixgbe_mac_X540:
2968 case ixgbe_mac_X550:
2969 case ixgbe_mac_X550EM_x:
2970 case ixgbe_mac_x550em_a:
2971 mask = (qmask & 0xFFFFFFFF);
2973 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2974 mask = (qmask >> 32);
2976 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2981 /* skip the flush */
2984 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2988 struct ixgbe_hw *hw = &adapter->hw;
2990 switch (hw->mac.type) {
2991 case ixgbe_mac_82598EB:
2992 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2993 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2995 case ixgbe_mac_82599EB:
2996 case ixgbe_mac_X540:
2997 case ixgbe_mac_X550:
2998 case ixgbe_mac_X550EM_x:
2999 case ixgbe_mac_x550em_a:
3000 mask = (qmask & 0xFFFFFFFF);
3002 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
3003 mask = (qmask >> 32);
3005 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
3010 /* skip the flush */
3014 * ixgbe_irq_enable - Enable default interrupt generation settings
3015 * @adapter: board private structure
3017 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
3020 struct ixgbe_hw *hw = &adapter->hw;
3021 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
3023 /* don't reenable LSC while waiting for link */
3024 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
3025 mask &= ~IXGBE_EIMS_LSC;
3027 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3028 switch (adapter->hw.mac.type) {
3029 case ixgbe_mac_82599EB:
3030 mask |= IXGBE_EIMS_GPI_SDP0(hw);
3032 case ixgbe_mac_X540:
3033 case ixgbe_mac_X550:
3034 case ixgbe_mac_X550EM_x:
3035 case ixgbe_mac_x550em_a:
3036 mask |= IXGBE_EIMS_TS;
3041 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3042 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3043 switch (adapter->hw.mac.type) {
3044 case ixgbe_mac_82599EB:
3045 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3046 mask |= IXGBE_EIMS_GPI_SDP2(hw);
3048 case ixgbe_mac_X540:
3049 case ixgbe_mac_X550:
3050 case ixgbe_mac_X550EM_x:
3051 case ixgbe_mac_x550em_a:
3052 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3053 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3054 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3055 mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3056 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3057 mask |= IXGBE_EICR_GPI_SDP0_X540;
3058 mask |= IXGBE_EIMS_ECC;
3059 mask |= IXGBE_EIMS_MAILBOX;
3065 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3066 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3067 mask |= IXGBE_EIMS_FLOW_DIR;
3069 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3071 ixgbe_irq_enable_queues(adapter, ~0);
3073 IXGBE_WRITE_FLUSH(&adapter->hw);
3076 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3078 struct ixgbe_adapter *adapter = data;
3079 struct ixgbe_hw *hw = &adapter->hw;
3083 * Workaround for Silicon errata. Use clear-by-write instead
3084 * of clear-by-read. Reading with EICS will return the
3085 * interrupt causes without clearing, which later be done
3086 * with the write to EICR.
3088 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3090 /* The lower 16bits of the EICR register are for the queue interrupts
3091 * which should be masked here in order to not accidentally clear them if
3092 * the bits are high when ixgbe_msix_other is called. There is a race
3093 * condition otherwise which results in possible performance loss
3094 * especially if the ixgbe_msix_other interrupt is triggering
3095 * consistently (as it would when PPS is turned on for the X540 device)
3099 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3101 if (eicr & IXGBE_EICR_LSC)
3102 ixgbe_check_lsc(adapter);
3104 if (eicr & IXGBE_EICR_MAILBOX)
3105 ixgbe_msg_task(adapter);
3107 switch (hw->mac.type) {
3108 case ixgbe_mac_82599EB:
3109 case ixgbe_mac_X540:
3110 case ixgbe_mac_X550:
3111 case ixgbe_mac_X550EM_x:
3112 case ixgbe_mac_x550em_a:
3113 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3114 (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3115 adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3116 ixgbe_service_event_schedule(adapter);
3117 IXGBE_WRITE_REG(hw, IXGBE_EICR,
3118 IXGBE_EICR_GPI_SDP0_X540);
3120 if (eicr & IXGBE_EICR_ECC) {
3121 e_info(link, "Received ECC Err, initiating reset\n");
3122 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3123 ixgbe_service_event_schedule(adapter);
3124 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3126 /* Handle Flow Director Full threshold interrupt */
3127 if (eicr & IXGBE_EICR_FLOW_DIR) {
3128 int reinit_count = 0;
3130 for (i = 0; i < adapter->num_tx_queues; i++) {
3131 struct ixgbe_ring *ring = adapter->tx_ring[i];
3132 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3137 /* no more flow director interrupts until after init */
3138 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3139 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3140 ixgbe_service_event_schedule(adapter);
3143 ixgbe_check_sfp_event(adapter, eicr);
3144 ixgbe_check_overtemp_event(adapter, eicr);
3150 ixgbe_check_fan_failure(adapter, eicr);
3152 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3153 ixgbe_ptp_check_pps_event(adapter);
3155 /* re-enable the original interrupt state, no lsc, no queues */
3156 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3157 ixgbe_irq_enable(adapter, false, false);
3162 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3164 struct ixgbe_q_vector *q_vector = data;
3166 /* EIAM disabled interrupts (on this vector) for us */
3168 if (q_vector->rx.ring || q_vector->tx.ring)
3169 napi_schedule_irqoff(&q_vector->napi);
3175 * ixgbe_poll - NAPI Rx polling callback
3176 * @napi: structure for representing this polling device
3177 * @budget: how many packets driver is allowed to clean
3179 * This function is used for legacy and MSI, NAPI mode
3181 int ixgbe_poll(struct napi_struct *napi, int budget)
3183 struct ixgbe_q_vector *q_vector =
3184 container_of(napi, struct ixgbe_q_vector, napi);
3185 struct ixgbe_adapter *adapter = q_vector->adapter;
3186 struct ixgbe_ring *ring;
3187 int per_ring_budget, work_done = 0;
3188 bool clean_complete = true;
3190 #ifdef CONFIG_IXGBE_DCA
3191 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3192 ixgbe_update_dca(q_vector);
3195 ixgbe_for_each_ring(ring, q_vector->tx) {
3196 if (!ixgbe_clean_tx_irq(q_vector, ring, budget))
3197 clean_complete = false;
3200 /* Exit if we are called by netpoll */
3204 /* attempt to distribute budget to each queue fairly, but don't allow
3205 * the budget to go below 1 because we'll exit polling */
3206 if (q_vector->rx.count > 1)
3207 per_ring_budget = max(budget/q_vector->rx.count, 1);
3209 per_ring_budget = budget;
3211 ixgbe_for_each_ring(ring, q_vector->rx) {
3212 int cleaned = ixgbe_clean_rx_irq(q_vector, ring,
3215 work_done += cleaned;
3216 if (cleaned >= per_ring_budget)
3217 clean_complete = false;
3220 /* If all work not completed, return budget and keep polling */
3221 if (!clean_complete)
3224 /* all work done, exit the polling mode */
3225 napi_complete_done(napi, work_done);
3226 if (adapter->rx_itr_setting & 1)
3227 ixgbe_set_itr(q_vector);
3228 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3229 ixgbe_irq_enable_queues(adapter, BIT_ULL(q_vector->v_idx));
3231 return min(work_done, budget - 1);
3235 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3236 * @adapter: board private structure
3238 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3239 * interrupts from the kernel.
3241 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3243 struct net_device *netdev = adapter->netdev;
3244 unsigned int ri = 0, ti = 0;
3247 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3248 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3249 struct msix_entry *entry = &adapter->msix_entries[vector];
3251 if (q_vector->tx.ring && q_vector->rx.ring) {
3252 snprintf(q_vector->name, sizeof(q_vector->name),
3253 "%s-TxRx-%u", netdev->name, ri++);
3255 } else if (q_vector->rx.ring) {
3256 snprintf(q_vector->name, sizeof(q_vector->name),
3257 "%s-rx-%u", netdev->name, ri++);
3258 } else if (q_vector->tx.ring) {
3259 snprintf(q_vector->name, sizeof(q_vector->name),
3260 "%s-tx-%u", netdev->name, ti++);
3262 /* skip this unused q_vector */
3265 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3266 q_vector->name, q_vector);
3268 e_err(probe, "request_irq failed for MSIX interrupt "
3269 "Error: %d\n", err);
3270 goto free_queue_irqs;
3272 /* If Flow Director is enabled, set interrupt affinity */
3273 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3274 /* assign the mask for this irq */
3275 irq_set_affinity_hint(entry->vector,
3276 &q_vector->affinity_mask);
3280 err = request_irq(adapter->msix_entries[vector].vector,
3281 ixgbe_msix_other, 0, netdev->name, adapter);
3283 e_err(probe, "request_irq for msix_other failed: %d\n", err);
3284 goto free_queue_irqs;
3292 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3294 free_irq(adapter->msix_entries[vector].vector,
3295 adapter->q_vector[vector]);
3297 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3298 pci_disable_msix(adapter->pdev);
3299 kfree(adapter->msix_entries);
3300 adapter->msix_entries = NULL;
3305 * ixgbe_intr - legacy mode Interrupt Handler
3306 * @irq: interrupt number
3307 * @data: pointer to a network interface device structure
3309 static irqreturn_t ixgbe_intr(int irq, void *data)
3311 struct ixgbe_adapter *adapter = data;
3312 struct ixgbe_hw *hw = &adapter->hw;
3313 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3317 * Workaround for silicon errata #26 on 82598. Mask the interrupt
3318 * before the read of EICR.
3320 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3322 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3323 * therefore no explicit interrupt disable is necessary */
3324 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3327 * shared interrupt alert!
3328 * make sure interrupts are enabled because the read will
3329 * have disabled interrupts due to EIAM
3330 * finish the workaround of silicon errata on 82598. Unmask
3331 * the interrupt that we masked before the EICR read.
3333 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3334 ixgbe_irq_enable(adapter, true, true);
3335 return IRQ_NONE; /* Not our interrupt */
3338 if (eicr & IXGBE_EICR_LSC)
3339 ixgbe_check_lsc(adapter);
3341 switch (hw->mac.type) {
3342 case ixgbe_mac_82599EB:
3343 ixgbe_check_sfp_event(adapter, eicr);
3345 case ixgbe_mac_X540:
3346 case ixgbe_mac_X550:
3347 case ixgbe_mac_X550EM_x:
3348 case ixgbe_mac_x550em_a:
3349 if (eicr & IXGBE_EICR_ECC) {
3350 e_info(link, "Received ECC Err, initiating reset\n");
3351 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3352 ixgbe_service_event_schedule(adapter);
3353 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3355 ixgbe_check_overtemp_event(adapter, eicr);
3361 ixgbe_check_fan_failure(adapter, eicr);
3362 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3363 ixgbe_ptp_check_pps_event(adapter);
3365 /* would disable interrupts here but EIAM disabled it */
3366 napi_schedule_irqoff(&q_vector->napi);
3369 * re-enable link(maybe) and non-queue interrupts, no flush.
3370 * ixgbe_poll will re-enable the queue interrupts
3372 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3373 ixgbe_irq_enable(adapter, false, false);
3379 * ixgbe_request_irq - initialize interrupts
3380 * @adapter: board private structure
3382 * Attempts to configure interrupts using the best available
3383 * capabilities of the hardware and kernel.
3385 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3387 struct net_device *netdev = adapter->netdev;
3390 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3391 err = ixgbe_request_msix_irqs(adapter);
3392 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3393 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3394 netdev->name, adapter);
3396 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3397 netdev->name, adapter);
3400 e_err(probe, "request_irq failed, Error %d\n", err);
3405 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3409 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3410 free_irq(adapter->pdev->irq, adapter);
3414 if (!adapter->msix_entries)
3417 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3418 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3419 struct msix_entry *entry = &adapter->msix_entries[vector];
3421 /* free only the irqs that were actually requested */
3422 if (!q_vector->rx.ring && !q_vector->tx.ring)
3425 /* clear the affinity_mask in the IRQ descriptor */
3426 irq_set_affinity_hint(entry->vector, NULL);
3428 free_irq(entry->vector, q_vector);
3431 free_irq(adapter->msix_entries[vector].vector, adapter);
3435 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3436 * @adapter: board private structure
3438 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3440 switch (adapter->hw.mac.type) {
3441 case ixgbe_mac_82598EB:
3442 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3444 case ixgbe_mac_82599EB:
3445 case ixgbe_mac_X540:
3446 case ixgbe_mac_X550:
3447 case ixgbe_mac_X550EM_x:
3448 case ixgbe_mac_x550em_a:
3449 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3450 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3451 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3456 IXGBE_WRITE_FLUSH(&adapter->hw);
3457 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3460 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3461 synchronize_irq(adapter->msix_entries[vector].vector);
3463 synchronize_irq(adapter->msix_entries[vector++].vector);
3465 synchronize_irq(adapter->pdev->irq);
3470 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3473 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3475 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3477 ixgbe_write_eitr(q_vector);
3479 ixgbe_set_ivar(adapter, 0, 0, 0);
3480 ixgbe_set_ivar(adapter, 1, 0, 0);
3482 e_info(hw, "Legacy interrupt IVAR setup done\n");
3486 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3487 * @adapter: board private structure
3488 * @ring: structure containing ring specific data
3490 * Configure the Tx descriptor ring after a reset.
3492 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3493 struct ixgbe_ring *ring)
3495 struct ixgbe_hw *hw = &adapter->hw;
3496 u64 tdba = ring->dma;
3498 u32 txdctl = IXGBE_TXDCTL_ENABLE;
3499 u8 reg_idx = ring->reg_idx;
3501 /* disable queue to avoid issues while updating state */
3502 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3503 IXGBE_WRITE_FLUSH(hw);
3505 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3506 (tdba & DMA_BIT_MASK(32)));
3507 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3508 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3509 ring->count * sizeof(union ixgbe_adv_tx_desc));
3510 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3511 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3512 ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3515 * set WTHRESH to encourage burst writeback, it should not be set
3516 * higher than 1 when:
3517 * - ITR is 0 as it could cause false TX hangs
3518 * - ITR is set to > 100k int/sec and BQL is enabled
3520 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3521 * to or less than the number of on chip descriptors, which is
3524 if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3525 txdctl |= 1u << 16; /* WTHRESH = 1 */
3527 txdctl |= 8u << 16; /* WTHRESH = 8 */
3530 * Setting PTHRESH to 32 both improves performance
3531 * and avoids a TX hang with DFP enabled
3533 txdctl |= (1u << 8) | /* HTHRESH = 1 */
3534 32; /* PTHRESH = 32 */
3536 /* reinitialize flowdirector state */
3537 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3538 ring->atr_sample_rate = adapter->atr_sample_rate;
3539 ring->atr_count = 0;
3540 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3542 ring->atr_sample_rate = 0;
3545 /* initialize XPS */
3546 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3547 struct ixgbe_q_vector *q_vector = ring->q_vector;
3550 netif_set_xps_queue(ring->netdev,
3551 &q_vector->affinity_mask,
3555 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3557 /* reinitialize tx_buffer_info */
3558 memset(ring->tx_buffer_info, 0,
3559 sizeof(struct ixgbe_tx_buffer) * ring->count);
3562 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3564 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3565 if (hw->mac.type == ixgbe_mac_82598EB &&
3566 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3569 /* poll to verify queue is enabled */
3571 usleep_range(1000, 2000);
3572 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3573 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3575 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3578 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3580 struct ixgbe_hw *hw = &adapter->hw;
3582 u8 tcs = netdev_get_num_tc(adapter->netdev);
3584 if (hw->mac.type == ixgbe_mac_82598EB)
3587 /* disable the arbiter while setting MTQC */
3588 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3589 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3590 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3592 /* set transmit pool layout */
3593 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3594 mtqc = IXGBE_MTQC_VT_ENA;
3596 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3598 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3599 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3600 IXGBE_82599_VMDQ_4Q_MASK)
3601 mtqc |= IXGBE_MTQC_32VF;
3603 mtqc |= IXGBE_MTQC_64VF;
3606 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3608 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3610 mtqc = IXGBE_MTQC_64Q_1PB;
3613 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3615 /* Enable Security TX Buffer IFG for multiple pb */
3617 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3618 sectx |= IXGBE_SECTX_DCB;
3619 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3622 /* re-enable the arbiter */
3623 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3624 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3628 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3629 * @adapter: board private structure
3631 * Configure the Tx unit of the MAC after a reset.
3633 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3635 struct ixgbe_hw *hw = &adapter->hw;
3639 ixgbe_setup_mtqc(adapter);
3641 if (hw->mac.type != ixgbe_mac_82598EB) {
3642 /* DMATXCTL.EN must be before Tx queues are enabled */
3643 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3644 dmatxctl |= IXGBE_DMATXCTL_TE;
3645 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3648 /* Setup the HW Tx Head and Tail descriptor pointers */
3649 for (i = 0; i < adapter->num_tx_queues; i++)
3650 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3651 for (i = 0; i < adapter->num_xdp_queues; i++)
3652 ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3655 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3656 struct ixgbe_ring *ring)
3658 struct ixgbe_hw *hw = &adapter->hw;
3659 u8 reg_idx = ring->reg_idx;
3660 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3662 srrctl |= IXGBE_SRRCTL_DROP_EN;
3664 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3667 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3668 struct ixgbe_ring *ring)
3670 struct ixgbe_hw *hw = &adapter->hw;
3671 u8 reg_idx = ring->reg_idx;
3672 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3674 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3676 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3679 #ifdef CONFIG_IXGBE_DCB
3680 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3682 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3686 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3688 if (adapter->ixgbe_ieee_pfc)
3689 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3692 * We should set the drop enable bit if:
3695 * Number of Rx queues > 1 and flow control is disabled
3697 * This allows us to avoid head of line blocking for security
3698 * and performance reasons.
3700 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3701 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3702 for (i = 0; i < adapter->num_rx_queues; i++)
3703 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3705 for (i = 0; i < adapter->num_rx_queues; i++)
3706 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3710 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3712 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3713 struct ixgbe_ring *rx_ring)
3715 struct ixgbe_hw *hw = &adapter->hw;
3717 u8 reg_idx = rx_ring->reg_idx;
3719 if (hw->mac.type == ixgbe_mac_82598EB) {
3720 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3723 * if VMDq is not active we must program one srrctl register
3724 * per RSS queue since we have enabled RDRXCTL.MVMEN
3729 /* configure header buffer length, needed for RSC */
3730 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3732 /* configure the packet buffer length */
3733 if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state))
3734 srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3736 srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3738 /* configure descriptor type */
3739 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3741 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3745 * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3746 * @adapter: device handle
3748 * - 82598/82599/X540: 128
3749 * - X550(non-SRIOV mode): 512
3750 * - X550(SRIOV mode): 64
3752 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3754 if (adapter->hw.mac.type < ixgbe_mac_X550)
3756 else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3763 * ixgbe_store_key - Write the RSS key to HW
3764 * @adapter: device handle
3766 * Write the RSS key stored in adapter.rss_key to HW.
3768 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3770 struct ixgbe_hw *hw = &adapter->hw;
3773 for (i = 0; i < 10; i++)
3774 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3778 * ixgbe_init_rss_key - Initialize adapter RSS key
3779 * @adapter: device handle
3781 * Allocates and initializes the RSS key if it is not allocated.
3783 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3787 if (!adapter->rss_key) {
3788 rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3789 if (unlikely(!rss_key))
3792 netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3793 adapter->rss_key = rss_key;
3800 * ixgbe_store_reta - Write the RETA table to HW
3801 * @adapter: device handle
3803 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3805 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3807 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3808 struct ixgbe_hw *hw = &adapter->hw;
3811 u8 *indir_tbl = adapter->rss_indir_tbl;
3813 /* Fill out the redirection table as follows:
3814 * - 82598: 8 bit wide entries containing pair of 4 bit RSS
3816 * - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3817 * - X550: 8 bit wide entries containing 6 bit RSS index
3819 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3820 indices_multi = 0x11;
3822 indices_multi = 0x1;
3824 /* Write redirection table to HW */
3825 for (i = 0; i < reta_entries; i++) {
3826 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3829 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3831 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3839 * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3840 * @adapter: device handle
3842 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3844 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3846 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3847 struct ixgbe_hw *hw = &adapter->hw;
3849 unsigned int pf_pool = adapter->num_vfs;
3851 /* Write redirection table to HW */
3852 for (i = 0; i < reta_entries; i++) {
3853 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3855 IXGBE_WRITE_REG(hw, IXGBE_PFVFRETA(i >> 2, pf_pool),
3862 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3865 u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3866 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3868 /* Program table for at least 4 queues w/ SR-IOV so that VFs can
3869 * make full use of any rings they may have. We will use the
3870 * PSRTYPE register to control how many rings we use within the PF.
3872 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3875 /* Fill out hash function seeds */
3876 ixgbe_store_key(adapter);
3878 /* Fill out redirection table */
3879 memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3881 for (i = 0, j = 0; i < reta_entries; i++, j++) {
3885 adapter->rss_indir_tbl[i] = j;
3888 ixgbe_store_reta(adapter);
3891 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3893 struct ixgbe_hw *hw = &adapter->hw;
3894 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3895 unsigned int pf_pool = adapter->num_vfs;
3898 /* Fill out hash function seeds */
3899 for (i = 0; i < 10; i++)
3900 IXGBE_WRITE_REG(hw, IXGBE_PFVFRSSRK(i, pf_pool),
3901 *(adapter->rss_key + i));
3903 /* Fill out the redirection table */
3904 for (i = 0, j = 0; i < 64; i++, j++) {
3908 adapter->rss_indir_tbl[i] = j;
3911 ixgbe_store_vfreta(adapter);
3914 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3916 struct ixgbe_hw *hw = &adapter->hw;
3917 u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3920 /* Disable indicating checksum in descriptor, enables RSS hash */
3921 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3922 rxcsum |= IXGBE_RXCSUM_PCSD;
3923 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3925 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3926 if (adapter->ring_feature[RING_F_RSS].mask)
3927 mrqc = IXGBE_MRQC_RSSEN;
3929 u8 tcs = netdev_get_num_tc(adapter->netdev);
3931 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3933 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3935 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3936 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3937 IXGBE_82599_VMDQ_4Q_MASK)
3938 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3940 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3942 /* Enable L3/L4 for Tx Switched packets */
3943 mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3946 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3948 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3950 mrqc = IXGBE_MRQC_RSSEN;
3954 /* Perform hash on these packet types */
3955 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3956 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3957 IXGBE_MRQC_RSS_FIELD_IPV6 |
3958 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3960 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3961 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3962 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3963 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3965 if ((hw->mac.type >= ixgbe_mac_X550) &&
3966 (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3967 unsigned int pf_pool = adapter->num_vfs;
3969 /* Enable VF RSS mode */
3970 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3971 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3973 /* Setup RSS through the VF registers */
3974 ixgbe_setup_vfreta(adapter);
3975 vfmrqc = IXGBE_MRQC_RSSEN;
3976 vfmrqc |= rss_field;
3977 IXGBE_WRITE_REG(hw, IXGBE_PFVFMRQC(pf_pool), vfmrqc);
3979 ixgbe_setup_reta(adapter);
3981 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3986 * ixgbe_configure_rscctl - enable RSC for the indicated ring
3987 * @adapter: address of board private structure
3988 * @index: index of ring to set
3990 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3991 struct ixgbe_ring *ring)
3993 struct ixgbe_hw *hw = &adapter->hw;
3995 u8 reg_idx = ring->reg_idx;
3997 if (!ring_is_rsc_enabled(ring))
4000 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4001 rscctrl |= IXGBE_RSCCTL_RSCEN;
4003 * we must limit the number of descriptors so that the
4004 * total size of max desc * buf_len is not greater
4007 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4008 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4011 #define IXGBE_MAX_RX_DESC_POLL 10
4012 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4013 struct ixgbe_ring *ring)
4015 struct ixgbe_hw *hw = &adapter->hw;
4016 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4018 u8 reg_idx = ring->reg_idx;
4020 if (ixgbe_removed(hw->hw_addr))
4022 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4023 if (hw->mac.type == ixgbe_mac_82598EB &&
4024 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4028 usleep_range(1000, 2000);
4029 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4030 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4033 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4034 "the polling period\n", reg_idx);
4038 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
4039 struct ixgbe_ring *ring)
4041 struct ixgbe_hw *hw = &adapter->hw;
4042 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4044 u8 reg_idx = ring->reg_idx;
4046 if (ixgbe_removed(hw->hw_addr))
4048 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4049 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4051 /* write value back with RXDCTL.ENABLE bit cleared */
4052 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4054 if (hw->mac.type == ixgbe_mac_82598EB &&
4055 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4058 /* the hardware may take up to 100us to really disable the rx queue */
4061 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4062 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
4065 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
4066 "the polling period\n", reg_idx);
4070 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4071 struct ixgbe_ring *ring)
4073 struct ixgbe_hw *hw = &adapter->hw;
4074 union ixgbe_adv_rx_desc *rx_desc;
4075 u64 rdba = ring->dma;
4077 u8 reg_idx = ring->reg_idx;
4079 /* disable queue to avoid issues while updating state */
4080 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4081 ixgbe_disable_rx_queue(adapter, ring);
4083 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4084 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4085 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4086 ring->count * sizeof(union ixgbe_adv_rx_desc));
4087 /* Force flushing of IXGBE_RDLEN to prevent MDD */
4088 IXGBE_WRITE_FLUSH(hw);
4090 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4091 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4092 ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4094 ixgbe_configure_srrctl(adapter, ring);
4095 ixgbe_configure_rscctl(adapter, ring);
4097 if (hw->mac.type == ixgbe_mac_82598EB) {
4099 * enable cache line friendly hardware writes:
4100 * PTHRESH=32 descriptors (half the internal cache),
4101 * this also removes ugly rx_no_buffer_count increment
4102 * HTHRESH=4 descriptors (to minimize latency on fetch)
4103 * WTHRESH=8 burst writeback up to two cache lines
4105 rxdctl &= ~0x3FFFFF;
4107 #if (PAGE_SIZE < 8192)
4109 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4110 IXGBE_RXDCTL_RLPML_EN);
4112 /* Limit the maximum frame size so we don't overrun the skb */
4113 if (ring_uses_build_skb(ring) &&
4114 !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4115 rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4116 IXGBE_RXDCTL_RLPML_EN;
4120 /* initialize rx_buffer_info */
4121 memset(ring->rx_buffer_info, 0,
4122 sizeof(struct ixgbe_rx_buffer) * ring->count);
4124 /* initialize Rx descriptor 0 */
4125 rx_desc = IXGBE_RX_DESC(ring, 0);
4126 rx_desc->wb.upper.length = 0;
4128 /* enable receive descriptor ring */
4129 rxdctl |= IXGBE_RXDCTL_ENABLE;
4130 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4132 ixgbe_rx_desc_queue_enable(adapter, ring);
4133 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4136 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4138 struct ixgbe_hw *hw = &adapter->hw;
4139 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4142 /* PSRTYPE must be initialized in non 82598 adapters */
4143 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4144 IXGBE_PSRTYPE_UDPHDR |
4145 IXGBE_PSRTYPE_IPV4HDR |
4146 IXGBE_PSRTYPE_L2HDR |
4147 IXGBE_PSRTYPE_IPV6HDR;
4149 if (hw->mac.type == ixgbe_mac_82598EB)
4153 psrtype |= 2u << 29;
4155 psrtype |= 1u << 29;
4157 for_each_set_bit(pool, &adapter->fwd_bitmask, 32)
4158 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4161 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4163 struct ixgbe_hw *hw = &adapter->hw;
4164 u32 reg_offset, vf_shift;
4165 u32 gcr_ext, vmdctl;
4168 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4171 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4172 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4173 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4174 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4175 vmdctl |= IXGBE_VT_CTL_REPLEN;
4176 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4178 vf_shift = VMDQ_P(0) % 32;
4179 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4181 /* Enable only the PF's pool for Tx/Rx */
4182 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4183 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4184 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4185 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4186 if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4187 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4189 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4190 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4192 /* clear VLAN promisc flag so VFTA will be updated if necessary */
4193 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4196 * Set up VF register offsets for selected VT Mode,
4197 * i.e. 32 or 64 VFs for SR-IOV
4199 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4200 case IXGBE_82599_VMDQ_8Q_MASK:
4201 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4203 case IXGBE_82599_VMDQ_4Q_MASK:
4204 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4207 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4211 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4213 for (i = 0; i < adapter->num_vfs; i++) {
4214 /* configure spoof checking */
4215 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4216 adapter->vfinfo[i].spoofchk_enabled);
4218 /* Enable/Disable RSS query feature */
4219 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4220 adapter->vfinfo[i].rss_query_enabled);
4224 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4226 struct ixgbe_hw *hw = &adapter->hw;
4227 struct net_device *netdev = adapter->netdev;
4228 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4229 struct ixgbe_ring *rx_ring;
4234 /* adjust max frame to be able to do baby jumbo for FCoE */
4235 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4236 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4237 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4239 #endif /* IXGBE_FCOE */
4241 /* adjust max frame to be at least the size of a standard frame */
4242 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4243 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4245 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4246 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4247 mhadd &= ~IXGBE_MHADD_MFS_MASK;
4248 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4250 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4253 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4254 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4255 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4256 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4259 * Setup the HW Rx Head and Tail Descriptor Pointers and
4260 * the Base and Length of the Rx Descriptor Ring
4262 for (i = 0; i < adapter->num_rx_queues; i++) {
4263 rx_ring = adapter->rx_ring[i];
4265 clear_ring_rsc_enabled(rx_ring);
4266 clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4267 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4269 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4270 set_ring_rsc_enabled(rx_ring);
4272 if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4273 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4275 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4276 if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4279 set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4281 #if (PAGE_SIZE < 8192)
4282 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4283 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4285 if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4286 (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4287 set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4292 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4294 struct ixgbe_hw *hw = &adapter->hw;
4295 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4297 switch (hw->mac.type) {
4298 case ixgbe_mac_82598EB:
4300 * For VMDq support of different descriptor types or
4301 * buffer sizes through the use of multiple SRRCTL
4302 * registers, RDRXCTL.MVMEN must be set to 1
4304 * also, the manual doesn't mention it clearly but DCA hints
4305 * will only use queue 0's tags unless this bit is set. Side
4306 * effects of setting this bit are only that SRRCTL must be
4307 * fully programmed [0..15]
4309 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4311 case ixgbe_mac_X550:
4312 case ixgbe_mac_X550EM_x:
4313 case ixgbe_mac_x550em_a:
4314 if (adapter->num_vfs)
4315 rdrxctl |= IXGBE_RDRXCTL_PSP;
4317 case ixgbe_mac_82599EB:
4318 case ixgbe_mac_X540:
4319 /* Disable RSC for ACK packets */
4320 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4321 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4322 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4323 /* hardware requires some bits to be set by default */
4324 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4325 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4328 /* We should do nothing since we don't know this hardware */
4332 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4336 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4337 * @adapter: board private structure
4339 * Configure the Rx unit of the MAC after a reset.
4341 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4343 struct ixgbe_hw *hw = &adapter->hw;
4347 /* disable receives while setting up the descriptors */
4348 hw->mac.ops.disable_rx(hw);
4350 ixgbe_setup_psrtype(adapter);
4351 ixgbe_setup_rdrxctl(adapter);
4354 rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4355 rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4356 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4357 rfctl |= IXGBE_RFCTL_RSC_DIS;
4359 /* disable NFS filtering */
4360 rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4361 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4363 /* Program registers for the distribution of queues */
4364 ixgbe_setup_mrqc(adapter);
4366 /* set_rx_buffer_len must be called before ring initialization */
4367 ixgbe_set_rx_buffer_len(adapter);
4370 * Setup the HW Rx Head and Tail Descriptor Pointers and
4371 * the Base and Length of the Rx Descriptor Ring
4373 for (i = 0; i < adapter->num_rx_queues; i++)
4374 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4376 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4377 /* disable drop enable for 82598 parts */
4378 if (hw->mac.type == ixgbe_mac_82598EB)
4379 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4381 /* enable all receives */
4382 rxctrl |= IXGBE_RXCTRL_RXEN;
4383 hw->mac.ops.enable_rx_dma(hw, rxctrl);
4386 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4387 __be16 proto, u16 vid)
4389 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4390 struct ixgbe_hw *hw = &adapter->hw;
4392 /* add VID to filter table */
4393 if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4394 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4396 set_bit(vid, adapter->active_vlans);
4401 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4406 /* short cut the special case */
4410 /* Search for the vlan id in the VLVF entries */
4411 for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4412 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4413 if ((vlvf & VLAN_VID_MASK) == vlan)
4420 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4422 struct ixgbe_hw *hw = &adapter->hw;
4426 idx = ixgbe_find_vlvf_entry(hw, vid);
4430 /* See if any other pools are set for this VLAN filter
4431 * entry other than the PF.
4433 word = idx * 2 + (VMDQ_P(0) / 32);
4434 bits = ~BIT(VMDQ_P(0) % 32);
4435 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4437 /* Disable the filter so this falls into the default pool. */
4438 if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4439 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4440 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4441 IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4445 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4446 __be16 proto, u16 vid)
4448 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4449 struct ixgbe_hw *hw = &adapter->hw;
4451 /* remove VID from filter table */
4452 if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4453 hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4455 clear_bit(vid, adapter->active_vlans);
4461 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4462 * @adapter: driver data
4464 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4466 struct ixgbe_hw *hw = &adapter->hw;
4470 switch (hw->mac.type) {
4471 case ixgbe_mac_82598EB:
4472 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4473 vlnctrl &= ~IXGBE_VLNCTRL_VME;
4474 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4476 case ixgbe_mac_82599EB:
4477 case ixgbe_mac_X540:
4478 case ixgbe_mac_X550:
4479 case ixgbe_mac_X550EM_x:
4480 case ixgbe_mac_x550em_a:
4481 for (i = 0; i < adapter->num_rx_queues; i++) {
4482 struct ixgbe_ring *ring = adapter->rx_ring[i];
4484 if (ring->l2_accel_priv)
4487 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4488 vlnctrl &= ~IXGBE_RXDCTL_VME;
4489 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4498 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4499 * @adapter: driver data
4501 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4503 struct ixgbe_hw *hw = &adapter->hw;
4507 switch (hw->mac.type) {
4508 case ixgbe_mac_82598EB:
4509 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4510 vlnctrl |= IXGBE_VLNCTRL_VME;
4511 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4513 case ixgbe_mac_82599EB:
4514 case ixgbe_mac_X540:
4515 case ixgbe_mac_X550:
4516 case ixgbe_mac_X550EM_x:
4517 case ixgbe_mac_x550em_a:
4518 for (i = 0; i < adapter->num_rx_queues; i++) {
4519 struct ixgbe_ring *ring = adapter->rx_ring[i];
4521 if (ring->l2_accel_priv)
4524 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4525 vlnctrl |= IXGBE_RXDCTL_VME;
4526 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4534 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4536 struct ixgbe_hw *hw = &adapter->hw;
4539 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4541 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4542 /* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4543 vlnctrl |= IXGBE_VLNCTRL_VFE;
4544 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4546 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4547 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4551 /* Nothing to do for 82598 */
4552 if (hw->mac.type == ixgbe_mac_82598EB)
4555 /* We are already in VLAN promisc, nothing to do */
4556 if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4559 /* Set flag so we don't redo unnecessary work */
4560 adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4562 /* Add PF to all active pools */
4563 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4564 u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4565 u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4567 vlvfb |= BIT(VMDQ_P(0) % 32);
4568 IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4571 /* Set all bits in the VLAN filter table array */
4572 for (i = hw->mac.vft_size; i--;)
4573 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4576 #define VFTA_BLOCK_SIZE 8
4577 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4579 struct ixgbe_hw *hw = &adapter->hw;
4580 u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4581 u32 vid_start = vfta_offset * 32;
4582 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4583 u32 i, vid, word, bits;
4585 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4586 u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4588 /* pull VLAN ID from VLVF */
4589 vid = vlvf & VLAN_VID_MASK;
4591 /* only concern outselves with a certain range */
4592 if (vid < vid_start || vid >= vid_end)
4596 /* record VLAN ID in VFTA */
4597 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4599 /* if PF is part of this then continue */
4600 if (test_bit(vid, adapter->active_vlans))
4604 /* remove PF from the pool */
4605 word = i * 2 + VMDQ_P(0) / 32;
4606 bits = ~BIT(VMDQ_P(0) % 32);
4607 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4608 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4611 /* extract values from active_vlans and write back to VFTA */
4612 for (i = VFTA_BLOCK_SIZE; i--;) {
4613 vid = (vfta_offset + i) * 32;
4614 word = vid / BITS_PER_LONG;
4615 bits = vid % BITS_PER_LONG;
4617 vfta[i] |= adapter->active_vlans[word] >> bits;
4619 IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4623 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4625 struct ixgbe_hw *hw = &adapter->hw;
4628 /* Set VLAN filtering to enabled */
4629 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4630 vlnctrl |= IXGBE_VLNCTRL_VFE;
4631 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4633 if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4634 hw->mac.type == ixgbe_mac_82598EB)
4637 /* We are not in VLAN promisc, nothing to do */
4638 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4641 /* Set flag so we don't redo unnecessary work */
4642 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4644 for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4645 ixgbe_scrub_vfta(adapter, i);
4648 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4652 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4654 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4655 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4659 * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4660 * @netdev: network interface device structure
4662 * Writes multicast address list to the MTA hash table.
4663 * Returns: -ENOMEM on failure
4664 * 0 on no addresses written
4665 * X on writing X addresses to MTA
4667 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4669 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4670 struct ixgbe_hw *hw = &adapter->hw;
4672 if (!netif_running(netdev))
4675 if (hw->mac.ops.update_mc_addr_list)
4676 hw->mac.ops.update_mc_addr_list(hw, netdev);
4680 #ifdef CONFIG_PCI_IOV
4681 ixgbe_restore_vf_multicasts(adapter);
4684 return netdev_mc_count(netdev);
4687 #ifdef CONFIG_PCI_IOV
4688 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4690 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4691 struct ixgbe_hw *hw = &adapter->hw;
4694 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4695 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4697 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4698 hw->mac.ops.set_rar(hw, i,
4703 hw->mac.ops.clear_rar(hw, i);
4708 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4710 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4711 struct ixgbe_hw *hw = &adapter->hw;
4714 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4715 if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4718 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4720 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4721 hw->mac.ops.set_rar(hw, i,
4726 hw->mac.ops.clear_rar(hw, i);
4730 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4732 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4733 struct ixgbe_hw *hw = &adapter->hw;
4736 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4737 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4738 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4741 ixgbe_sync_mac_table(adapter);
4744 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4746 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4747 struct ixgbe_hw *hw = &adapter->hw;
4750 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4751 /* do not count default RAR as available */
4752 if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4755 /* only count unused and addresses that belong to us */
4756 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4757 if (mac_table->pool != pool)
4767 /* this function destroys the first RAR entry */
4768 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4770 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4771 struct ixgbe_hw *hw = &adapter->hw;
4773 memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4774 mac_table->pool = VMDQ_P(0);
4776 mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4778 hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4782 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4783 const u8 *addr, u16 pool)
4785 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4786 struct ixgbe_hw *hw = &adapter->hw;
4789 if (is_zero_ether_addr(addr))
4792 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4793 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4796 ether_addr_copy(mac_table->addr, addr);
4797 mac_table->pool = pool;
4799 mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4800 IXGBE_MAC_STATE_IN_USE;
4802 ixgbe_sync_mac_table(adapter);
4810 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4811 const u8 *addr, u16 pool)
4813 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4814 struct ixgbe_hw *hw = &adapter->hw;
4817 if (is_zero_ether_addr(addr))
4820 /* search table for addr, if found clear IN_USE flag and sync */
4821 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4822 /* we can only delete an entry if it is in use */
4823 if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4825 /* we only care about entries that belong to the given pool */
4826 if (mac_table->pool != pool)
4828 /* we only care about a specific MAC address */
4829 if (!ether_addr_equal(addr, mac_table->addr))
4832 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4833 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4835 ixgbe_sync_mac_table(adapter);
4843 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
4844 * @netdev: network interface device structure
4846 * Writes unicast address list to the RAR table.
4847 * Returns: -ENOMEM on failure/insufficient address space
4848 * 0 on no addresses written
4849 * X on writing X addresses to the RAR table
4851 static int ixgbe_write_uc_addr_list(struct net_device *netdev, int vfn)
4853 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4856 /* return ENOMEM indicating insufficient memory for addresses */
4857 if (netdev_uc_count(netdev) > ixgbe_available_rars(adapter, vfn))
4860 if (!netdev_uc_empty(netdev)) {
4861 struct netdev_hw_addr *ha;
4862 netdev_for_each_uc_addr(ha, netdev) {
4863 ixgbe_del_mac_filter(adapter, ha->addr, vfn);
4864 ixgbe_add_mac_filter(adapter, ha->addr, vfn);
4871 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4873 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4876 ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4878 return min_t(int, ret, 0);
4881 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4883 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4885 ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4891 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4892 * @netdev: network interface device structure
4894 * The set_rx_method entry point is called whenever the unicast/multicast
4895 * address list or the network interface flags are updated. This routine is
4896 * responsible for configuring the hardware for proper unicast, multicast and
4899 void ixgbe_set_rx_mode(struct net_device *netdev)
4901 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4902 struct ixgbe_hw *hw = &adapter->hw;
4903 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4904 netdev_features_t features = netdev->features;
4907 /* Check for Promiscuous and All Multicast modes */
4908 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4910 /* set all bits that we expect to always be set */
4911 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4912 fctrl |= IXGBE_FCTRL_BAM;
4913 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4914 fctrl |= IXGBE_FCTRL_PMCF;
4916 /* clear the bits we are changing the status of */
4917 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4918 if (netdev->flags & IFF_PROMISC) {
4919 hw->addr_ctrl.user_set_promisc = true;
4920 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4921 vmolr |= IXGBE_VMOLR_MPE;
4922 features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4924 if (netdev->flags & IFF_ALLMULTI) {
4925 fctrl |= IXGBE_FCTRL_MPE;
4926 vmolr |= IXGBE_VMOLR_MPE;
4928 hw->addr_ctrl.user_set_promisc = false;
4932 * Write addresses to available RAR registers, if there is not
4933 * sufficient space to store all the addresses then enable
4934 * unicast promiscuous mode
4936 if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4937 fctrl |= IXGBE_FCTRL_UPE;
4938 vmolr |= IXGBE_VMOLR_ROPE;
4941 /* Write addresses to the MTA, if the attempt fails
4942 * then we should just turn on promiscuous mode so
4943 * that we can at least receive multicast traffic
4945 count = ixgbe_write_mc_addr_list(netdev);
4947 fctrl |= IXGBE_FCTRL_MPE;
4948 vmolr |= IXGBE_VMOLR_MPE;
4950 vmolr |= IXGBE_VMOLR_ROMPE;
4953 if (hw->mac.type != ixgbe_mac_82598EB) {
4954 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4955 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4957 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4960 /* This is useful for sniffing bad packets. */
4961 if (features & NETIF_F_RXALL) {
4962 /* UPE and MPE will be handled by normal PROMISC logic
4963 * in e1000e_set_rx_mode */
4964 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4965 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4966 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4968 fctrl &= ~(IXGBE_FCTRL_DPF);
4969 /* NOTE: VLAN filtering is disabled by setting PROMISC */
4972 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4974 if (features & NETIF_F_HW_VLAN_CTAG_RX)
4975 ixgbe_vlan_strip_enable(adapter);
4977 ixgbe_vlan_strip_disable(adapter);
4979 if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4980 ixgbe_vlan_promisc_disable(adapter);
4982 ixgbe_vlan_promisc_enable(adapter);
4985 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4989 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4990 napi_enable(&adapter->q_vector[q_idx]->napi);
4993 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4997 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4998 napi_disable(&adapter->q_vector[q_idx]->napi);
5001 static void ixgbe_clear_udp_tunnel_port(struct ixgbe_adapter *adapter, u32 mask)
5003 struct ixgbe_hw *hw = &adapter->hw;
5006 if (!(adapter->flags & (IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE |
5007 IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)))
5010 vxlanctrl = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) & ~mask;
5011 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, vxlanctrl);
5013 if (mask & IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK)
5014 adapter->vxlan_port = 0;
5016 if (mask & IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK)
5017 adapter->geneve_port = 0;
5020 #ifdef CONFIG_IXGBE_DCB
5022 * ixgbe_configure_dcb - Configure DCB hardware
5023 * @adapter: ixgbe adapter struct
5025 * This is called by the driver on open to configure the DCB hardware.
5026 * This is also called by the gennetlink interface when reconfiguring
5029 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5031 struct ixgbe_hw *hw = &adapter->hw;
5032 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5034 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5035 if (hw->mac.type == ixgbe_mac_82598EB)
5036 netif_set_gso_max_size(adapter->netdev, 65536);
5040 if (hw->mac.type == ixgbe_mac_82598EB)
5041 netif_set_gso_max_size(adapter->netdev, 32768);
5044 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5045 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5048 /* reconfigure the hardware */
5049 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5050 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5052 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5054 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5055 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5056 ixgbe_dcb_hw_ets(&adapter->hw,
5057 adapter->ixgbe_ieee_ets,
5059 ixgbe_dcb_hw_pfc_config(&adapter->hw,
5060 adapter->ixgbe_ieee_pfc->pfc_en,
5061 adapter->ixgbe_ieee_ets->prio_tc);
5064 /* Enable RSS Hash per TC */
5065 if (hw->mac.type != ixgbe_mac_82598EB) {
5067 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5074 /* write msb to all 8 TCs in one write */
5075 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5080 /* Additional bittime to account for IXGBE framing */
5081 #define IXGBE_ETH_FRAMING 20
5084 * ixgbe_hpbthresh - calculate high water mark for flow control
5086 * @adapter: board private structure to calculate for
5087 * @pb: packet buffer to calculate
5089 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5091 struct ixgbe_hw *hw = &adapter->hw;
5092 struct net_device *dev = adapter->netdev;
5093 int link, tc, kb, marker;
5096 /* Calculate max LAN frame size */
5097 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5100 /* FCoE traffic class uses FCOE jumbo frames */
5101 if ((dev->features & NETIF_F_FCOE_MTU) &&
5102 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5103 (pb == ixgbe_fcoe_get_tc(adapter)))
5104 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5107 /* Calculate delay value for device */
5108 switch (hw->mac.type) {
5109 case ixgbe_mac_X540:
5110 case ixgbe_mac_X550:
5111 case ixgbe_mac_X550EM_x:
5112 case ixgbe_mac_x550em_a:
5113 dv_id = IXGBE_DV_X540(link, tc);
5116 dv_id = IXGBE_DV(link, tc);
5120 /* Loopback switch introduces additional latency */
5121 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5122 dv_id += IXGBE_B2BT(tc);
5124 /* Delay value is calculated in bit times convert to KB */
5125 kb = IXGBE_BT2KB(dv_id);
5126 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5128 marker = rx_pba - kb;
5130 /* It is possible that the packet buffer is not large enough
5131 * to provide required headroom. In this case throw an error
5132 * to user and a do the best we can.
5135 e_warn(drv, "Packet Buffer(%i) can not provide enough"
5136 "headroom to support flow control."
5137 "Decrease MTU or number of traffic classes\n", pb);
5145 * ixgbe_lpbthresh - calculate low water mark for for flow control
5147 * @adapter: board private structure to calculate for
5148 * @pb: packet buffer to calculate
5150 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5152 struct ixgbe_hw *hw = &adapter->hw;
5153 struct net_device *dev = adapter->netdev;
5157 /* Calculate max LAN frame size */
5158 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5161 /* FCoE traffic class uses FCOE jumbo frames */
5162 if ((dev->features & NETIF_F_FCOE_MTU) &&
5163 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5164 (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5165 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5168 /* Calculate delay value for device */
5169 switch (hw->mac.type) {
5170 case ixgbe_mac_X540:
5171 case ixgbe_mac_X550:
5172 case ixgbe_mac_X550EM_x:
5173 case ixgbe_mac_x550em_a:
5174 dv_id = IXGBE_LOW_DV_X540(tc);
5177 dv_id = IXGBE_LOW_DV(tc);
5181 /* Delay value is calculated in bit times convert to KB */
5182 return IXGBE_BT2KB(dv_id);
5186 * ixgbe_pbthresh_setup - calculate and setup high low water marks
5188 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5190 struct ixgbe_hw *hw = &adapter->hw;
5191 int num_tc = netdev_get_num_tc(adapter->netdev);
5197 for (i = 0; i < num_tc; i++) {
5198 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5199 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5201 /* Low water marks must not be larger than high water marks */
5202 if (hw->fc.low_water[i] > hw->fc.high_water[i])
5203 hw->fc.low_water[i] = 0;
5206 for (; i < MAX_TRAFFIC_CLASS; i++)
5207 hw->fc.high_water[i] = 0;
5210 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5212 struct ixgbe_hw *hw = &adapter->hw;
5214 u8 tc = netdev_get_num_tc(adapter->netdev);
5216 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5217 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5218 hdrm = 32 << adapter->fdir_pballoc;
5222 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5223 ixgbe_pbthresh_setup(adapter);
5226 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5228 struct ixgbe_hw *hw = &adapter->hw;
5229 struct hlist_node *node2;
5230 struct ixgbe_fdir_filter *filter;
5232 spin_lock(&adapter->fdir_perfect_lock);
5234 if (!hlist_empty(&adapter->fdir_filter_list))
5235 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5237 hlist_for_each_entry_safe(filter, node2,
5238 &adapter->fdir_filter_list, fdir_node) {
5239 ixgbe_fdir_write_perfect_filter_82599(hw,
5242 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
5243 IXGBE_FDIR_DROP_QUEUE :
5244 adapter->rx_ring[filter->action]->reg_idx);
5247 spin_unlock(&adapter->fdir_perfect_lock);
5250 static void ixgbe_macvlan_set_rx_mode(struct net_device *dev, unsigned int pool,
5251 struct ixgbe_adapter *adapter)
5253 struct ixgbe_hw *hw = &adapter->hw;
5256 /* No unicast promiscuous support for VMDQ devices. */
5257 vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(pool));
5258 vmolr |= (IXGBE_VMOLR_ROMPE | IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE);
5260 /* clear the affected bit */
5261 vmolr &= ~IXGBE_VMOLR_MPE;
5263 if (dev->flags & IFF_ALLMULTI) {
5264 vmolr |= IXGBE_VMOLR_MPE;
5266 vmolr |= IXGBE_VMOLR_ROMPE;
5267 hw->mac.ops.update_mc_addr_list(hw, dev);
5269 ixgbe_write_uc_addr_list(adapter->netdev, pool);
5270 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(pool), vmolr);
5273 static void ixgbe_fwd_psrtype(struct ixgbe_fwd_adapter *vadapter)
5275 struct ixgbe_adapter *adapter = vadapter->real_adapter;
5276 int rss_i = adapter->num_rx_queues_per_pool;
5277 struct ixgbe_hw *hw = &adapter->hw;
5278 u16 pool = vadapter->pool;
5279 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
5280 IXGBE_PSRTYPE_UDPHDR |
5281 IXGBE_PSRTYPE_IPV4HDR |
5282 IXGBE_PSRTYPE_L2HDR |
5283 IXGBE_PSRTYPE_IPV6HDR;
5285 if (hw->mac.type == ixgbe_mac_82598EB)
5289 psrtype |= 2u << 29;
5291 psrtype |= 1u << 29;
5293 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
5297 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5298 * @rx_ring: ring to free buffers from
5300 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5302 u16 i = rx_ring->next_to_clean;
5303 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5305 /* Free all the Rx ring sk_buffs */
5306 while (i != rx_ring->next_to_alloc) {
5307 if (rx_buffer->skb) {
5308 struct sk_buff *skb = rx_buffer->skb;
5309 if (IXGBE_CB(skb)->page_released)
5310 dma_unmap_page_attrs(rx_ring->dev,
5312 ixgbe_rx_pg_size(rx_ring),
5318 /* Invalidate cache lines that may have been written to by
5319 * device so that we avoid corrupting memory.
5321 dma_sync_single_range_for_cpu(rx_ring->dev,
5323 rx_buffer->page_offset,
5324 ixgbe_rx_bufsz(rx_ring),
5327 /* free resources associated with mapping */
5328 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5329 ixgbe_rx_pg_size(rx_ring),
5332 __page_frag_cache_drain(rx_buffer->page,
5333 rx_buffer->pagecnt_bias);
5337 if (i == rx_ring->count) {
5339 rx_buffer = rx_ring->rx_buffer_info;
5343 rx_ring->next_to_alloc = 0;
5344 rx_ring->next_to_clean = 0;
5345 rx_ring->next_to_use = 0;
5348 static void ixgbe_disable_fwd_ring(struct ixgbe_fwd_adapter *vadapter,
5349 struct ixgbe_ring *rx_ring)
5351 struct ixgbe_adapter *adapter = vadapter->real_adapter;
5352 int index = rx_ring->queue_index + vadapter->rx_base_queue;
5354 /* shutdown specific queue receive and wait for dma to settle */
5355 ixgbe_disable_rx_queue(adapter, rx_ring);
5356 usleep_range(10000, 20000);
5357 ixgbe_irq_disable_queues(adapter, BIT_ULL(index));
5358 ixgbe_clean_rx_ring(rx_ring);
5359 rx_ring->l2_accel_priv = NULL;
5362 static int ixgbe_fwd_ring_down(struct net_device *vdev,
5363 struct ixgbe_fwd_adapter *accel)
5365 struct ixgbe_adapter *adapter = accel->real_adapter;
5366 unsigned int rxbase = accel->rx_base_queue;
5367 unsigned int txbase = accel->tx_base_queue;
5370 netif_tx_stop_all_queues(vdev);
5372 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
5373 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
5374 adapter->rx_ring[rxbase + i]->netdev = adapter->netdev;
5377 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
5378 adapter->tx_ring[txbase + i]->l2_accel_priv = NULL;
5379 adapter->tx_ring[txbase + i]->netdev = adapter->netdev;
5386 static int ixgbe_fwd_ring_up(struct net_device *vdev,
5387 struct ixgbe_fwd_adapter *accel)
5389 struct ixgbe_adapter *adapter = accel->real_adapter;
5390 unsigned int rxbase, txbase, queues;
5391 int i, baseq, err = 0;
5393 if (!test_bit(accel->pool, &adapter->fwd_bitmask))
5396 baseq = accel->pool * adapter->num_rx_queues_per_pool;
5397 netdev_dbg(vdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
5398 accel->pool, adapter->num_rx_pools,
5399 baseq, baseq + adapter->num_rx_queues_per_pool,
5400 adapter->fwd_bitmask);
5402 accel->netdev = vdev;
5403 accel->rx_base_queue = rxbase = baseq;
5404 accel->tx_base_queue = txbase = baseq;
5406 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5407 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
5409 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
5410 adapter->rx_ring[rxbase + i]->netdev = vdev;
5411 adapter->rx_ring[rxbase + i]->l2_accel_priv = accel;
5412 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[rxbase + i]);
5415 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
5416 adapter->tx_ring[txbase + i]->netdev = vdev;
5417 adapter->tx_ring[txbase + i]->l2_accel_priv = accel;
5420 queues = min_t(unsigned int,
5421 adapter->num_rx_queues_per_pool, vdev->num_tx_queues);
5422 err = netif_set_real_num_tx_queues(vdev, queues);
5426 err = netif_set_real_num_rx_queues(vdev, queues);
5430 if (is_valid_ether_addr(vdev->dev_addr))
5431 ixgbe_add_mac_filter(adapter, vdev->dev_addr, accel->pool);
5433 ixgbe_fwd_psrtype(accel);
5434 ixgbe_macvlan_set_rx_mode(vdev, accel->pool, adapter);
5437 ixgbe_fwd_ring_down(vdev, accel);
5441 static int ixgbe_upper_dev_walk(struct net_device *upper, void *data)
5443 if (netif_is_macvlan(upper)) {
5444 struct macvlan_dev *dfwd = netdev_priv(upper);
5445 struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv;
5448 ixgbe_fwd_ring_up(upper, vadapter);
5454 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5456 netdev_walk_all_upper_dev_rcu(adapter->netdev,
5457 ixgbe_upper_dev_walk, NULL);
5460 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5462 struct ixgbe_hw *hw = &adapter->hw;
5464 ixgbe_configure_pb(adapter);
5465 #ifdef CONFIG_IXGBE_DCB
5466 ixgbe_configure_dcb(adapter);
5469 * We must restore virtualization before VLANs or else
5470 * the VLVF registers will not be populated
5472 ixgbe_configure_virtualization(adapter);
5474 ixgbe_set_rx_mode(adapter->netdev);
5475 ixgbe_restore_vlan(adapter);
5477 switch (hw->mac.type) {
5478 case ixgbe_mac_82599EB:
5479 case ixgbe_mac_X540:
5480 hw->mac.ops.disable_rx_buff(hw);
5486 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5487 ixgbe_init_fdir_signature_82599(&adapter->hw,
5488 adapter->fdir_pballoc);
5489 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5490 ixgbe_init_fdir_perfect_82599(&adapter->hw,
5491 adapter->fdir_pballoc);
5492 ixgbe_fdir_filter_restore(adapter);
5495 switch (hw->mac.type) {
5496 case ixgbe_mac_82599EB:
5497 case ixgbe_mac_X540:
5498 hw->mac.ops.enable_rx_buff(hw);
5504 #ifdef CONFIG_IXGBE_DCA
5506 if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5507 ixgbe_setup_dca(adapter);
5508 #endif /* CONFIG_IXGBE_DCA */
5511 /* configure FCoE L2 filters, redirection table, and Rx control */
5512 ixgbe_configure_fcoe(adapter);
5514 #endif /* IXGBE_FCOE */
5515 ixgbe_configure_tx(adapter);
5516 ixgbe_configure_rx(adapter);
5517 ixgbe_configure_dfwd(adapter);
5521 * ixgbe_sfp_link_config - set up SFP+ link
5522 * @adapter: pointer to private adapter struct
5524 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5527 * We are assuming the worst case scenario here, and that
5528 * is that an SFP was inserted/removed after the reset
5529 * but before SFP detection was enabled. As such the best
5530 * solution is to just start searching as soon as we start
5532 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5533 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5535 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5536 adapter->sfp_poll_time = 0;
5540 * ixgbe_non_sfp_link_config - set up non-SFP+ link
5541 * @hw: pointer to private hardware struct
5543 * Returns 0 on success, negative on failure
5545 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5548 bool autoneg, link_up = false;
5549 int ret = IXGBE_ERR_LINK_SETUP;
5551 if (hw->mac.ops.check_link)
5552 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5557 speed = hw->phy.autoneg_advertised;
5558 if ((!speed) && (hw->mac.ops.get_link_capabilities))
5559 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5564 if (hw->mac.ops.setup_link)
5565 ret = hw->mac.ops.setup_link(hw, speed, link_up);
5570 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5572 struct ixgbe_hw *hw = &adapter->hw;
5575 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5576 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5578 gpie |= IXGBE_GPIE_EIAME;
5580 * use EIAM to auto-mask when MSI-X interrupt is asserted
5581 * this saves a register write for every interrupt
5583 switch (hw->mac.type) {
5584 case ixgbe_mac_82598EB:
5585 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5587 case ixgbe_mac_82599EB:
5588 case ixgbe_mac_X540:
5589 case ixgbe_mac_X550:
5590 case ixgbe_mac_X550EM_x:
5591 case ixgbe_mac_x550em_a:
5593 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5594 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5598 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
5599 * specifically only auto mask tx and rx interrupts */
5600 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5603 /* XXX: to interrupt immediately for EICS writes, enable this */
5604 /* gpie |= IXGBE_GPIE_EIMEN; */
5606 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5607 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5609 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5610 case IXGBE_82599_VMDQ_8Q_MASK:
5611 gpie |= IXGBE_GPIE_VTMODE_16;
5613 case IXGBE_82599_VMDQ_4Q_MASK:
5614 gpie |= IXGBE_GPIE_VTMODE_32;
5617 gpie |= IXGBE_GPIE_VTMODE_64;
5622 /* Enable Thermal over heat sensor interrupt */
5623 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5624 switch (adapter->hw.mac.type) {
5625 case ixgbe_mac_82599EB:
5626 gpie |= IXGBE_SDP0_GPIEN_8259X;
5633 /* Enable fan failure interrupt */
5634 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5635 gpie |= IXGBE_SDP1_GPIEN(hw);
5637 switch (hw->mac.type) {
5638 case ixgbe_mac_82599EB:
5639 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5641 case ixgbe_mac_X550EM_x:
5642 case ixgbe_mac_x550em_a:
5643 gpie |= IXGBE_SDP0_GPIEN_X540;
5649 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5652 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5654 struct ixgbe_hw *hw = &adapter->hw;
5658 ixgbe_get_hw_control(adapter);
5659 ixgbe_setup_gpie(adapter);
5661 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5662 ixgbe_configure_msix(adapter);
5664 ixgbe_configure_msi_and_legacy(adapter);
5666 /* enable the optics for 82599 SFP+ fiber */
5667 if (hw->mac.ops.enable_tx_laser)
5668 hw->mac.ops.enable_tx_laser(hw);
5670 if (hw->phy.ops.set_phy_power)
5671 hw->phy.ops.set_phy_power(hw, true);
5673 smp_mb__before_atomic();
5674 clear_bit(__IXGBE_DOWN, &adapter->state);
5675 ixgbe_napi_enable_all(adapter);
5677 if (ixgbe_is_sfp(hw)) {
5678 ixgbe_sfp_link_config(adapter);
5680 err = ixgbe_non_sfp_link_config(hw);
5682 e_err(probe, "link_config FAILED %d\n", err);
5685 /* clear any pending interrupts, may auto mask */
5686 IXGBE_READ_REG(hw, IXGBE_EICR);
5687 ixgbe_irq_enable(adapter, true, true);
5690 * If this adapter has a fan, check to see if we had a failure
5691 * before we enabled the interrupt.
5693 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5694 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5695 if (esdp & IXGBE_ESDP_SDP1)
5696 e_crit(drv, "Fan has stopped, replace the adapter\n");
5699 /* bring the link up in the watchdog, this could race with our first
5700 * link up interrupt but shouldn't be a problem */
5701 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5702 adapter->link_check_timeout = jiffies;
5703 mod_timer(&adapter->service_timer, jiffies);
5705 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5706 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5707 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5708 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5711 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5713 WARN_ON(in_interrupt());
5714 /* put off any impending NetWatchDogTimeout */
5715 netif_trans_update(adapter->netdev);
5717 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5718 usleep_range(1000, 2000);
5719 if (adapter->hw.phy.type == ixgbe_phy_fw)
5720 ixgbe_watchdog_link_is_down(adapter);
5721 ixgbe_down(adapter);
5723 * If SR-IOV enabled then wait a bit before bringing the adapter
5724 * back up to give the VFs time to respond to the reset. The
5725 * two second wait is based upon the watchdog timer cycle in
5728 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5731 clear_bit(__IXGBE_RESETTING, &adapter->state);
5734 void ixgbe_up(struct ixgbe_adapter *adapter)
5736 /* hardware has been reset, we need to reload some things */
5737 ixgbe_configure(adapter);
5739 ixgbe_up_complete(adapter);
5742 void ixgbe_reset(struct ixgbe_adapter *adapter)
5744 struct ixgbe_hw *hw = &adapter->hw;
5745 struct net_device *netdev = adapter->netdev;
5748 if (ixgbe_removed(hw->hw_addr))
5750 /* lock SFP init bit to prevent race conditions with the watchdog */
5751 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5752 usleep_range(1000, 2000);
5754 /* clear all SFP and link config related flags while holding SFP_INIT */
5755 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5756 IXGBE_FLAG2_SFP_NEEDS_RESET);
5757 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5759 err = hw->mac.ops.init_hw(hw);
5762 case IXGBE_ERR_SFP_NOT_PRESENT:
5763 case IXGBE_ERR_SFP_NOT_SUPPORTED:
5765 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5766 e_dev_err("master disable timed out\n");
5768 case IXGBE_ERR_EEPROM_VERSION:
5769 /* We are running on a pre-production device, log a warning */
5770 e_dev_warn("This device is a pre-production adapter/LOM. "
5771 "Please be aware there may be issues associated with "
5772 "your hardware. If you are experiencing problems "
5773 "please contact your Intel or hardware "
5774 "representative who provided you with this "
5778 e_dev_err("Hardware Error: %d\n", err);
5781 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5783 /* flush entries out of MAC table */
5784 ixgbe_flush_sw_mac_table(adapter);
5785 __dev_uc_unsync(netdev, NULL);
5787 /* do not flush user set addresses */
5788 ixgbe_mac_set_default_filter(adapter);
5790 /* update SAN MAC vmdq pool selection */
5791 if (hw->mac.san_mac_rar_index)
5792 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5794 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5795 ixgbe_ptp_reset(adapter);
5797 if (hw->phy.ops.set_phy_power) {
5798 if (!netif_running(adapter->netdev) && !adapter->wol)
5799 hw->phy.ops.set_phy_power(hw, false);
5801 hw->phy.ops.set_phy_power(hw, true);
5806 * ixgbe_clean_tx_ring - Free Tx Buffers
5807 * @tx_ring: ring to be cleaned
5809 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5811 u16 i = tx_ring->next_to_clean;
5812 struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
5814 while (i != tx_ring->next_to_use) {
5815 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
5817 /* Free all the Tx ring sk_buffs */
5818 if (ring_is_xdp(tx_ring))
5819 page_frag_free(tx_buffer->data);
5821 dev_kfree_skb_any(tx_buffer->skb);
5823 /* unmap skb header data */
5824 dma_unmap_single(tx_ring->dev,
5825 dma_unmap_addr(tx_buffer, dma),
5826 dma_unmap_len(tx_buffer, len),
5829 /* check for eop_desc to determine the end of the packet */
5830 eop_desc = tx_buffer->next_to_watch;
5831 tx_desc = IXGBE_TX_DESC(tx_ring, i);
5833 /* unmap remaining buffers */
5834 while (tx_desc != eop_desc) {
5838 if (unlikely(i == tx_ring->count)) {
5840 tx_buffer = tx_ring->tx_buffer_info;
5841 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5844 /* unmap any remaining paged data */
5845 if (dma_unmap_len(tx_buffer, len))
5846 dma_unmap_page(tx_ring->dev,
5847 dma_unmap_addr(tx_buffer, dma),
5848 dma_unmap_len(tx_buffer, len),
5852 /* move us one more past the eop_desc for start of next pkt */
5855 if (unlikely(i == tx_ring->count)) {
5857 tx_buffer = tx_ring->tx_buffer_info;
5861 /* reset BQL for queue */
5862 if (!ring_is_xdp(tx_ring))
5863 netdev_tx_reset_queue(txring_txq(tx_ring));
5865 /* reset next_to_use and next_to_clean */
5866 tx_ring->next_to_use = 0;
5867 tx_ring->next_to_clean = 0;
5871 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
5872 * @adapter: board private structure
5874 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
5878 for (i = 0; i < adapter->num_rx_queues; i++)
5879 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
5883 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
5884 * @adapter: board private structure
5886 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
5890 for (i = 0; i < adapter->num_tx_queues; i++)
5891 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
5892 for (i = 0; i < adapter->num_xdp_queues; i++)
5893 ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
5896 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
5898 struct hlist_node *node2;
5899 struct ixgbe_fdir_filter *filter;
5901 spin_lock(&adapter->fdir_perfect_lock);
5903 hlist_for_each_entry_safe(filter, node2,
5904 &adapter->fdir_filter_list, fdir_node) {
5905 hlist_del(&filter->fdir_node);
5908 adapter->fdir_filter_count = 0;
5910 spin_unlock(&adapter->fdir_perfect_lock);
5913 static int ixgbe_disable_macvlan(struct net_device *upper, void *data)
5915 if (netif_is_macvlan(upper)) {
5916 struct macvlan_dev *vlan = netdev_priv(upper);
5918 if (vlan->fwd_priv) {
5919 netif_tx_stop_all_queues(upper);
5920 netif_carrier_off(upper);
5921 netif_tx_disable(upper);
5928 void ixgbe_down(struct ixgbe_adapter *adapter)
5930 struct net_device *netdev = adapter->netdev;
5931 struct ixgbe_hw *hw = &adapter->hw;
5934 /* signal that we are down to the interrupt handler */
5935 if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
5936 return; /* do nothing if already down */
5938 /* disable receives */
5939 hw->mac.ops.disable_rx(hw);
5941 /* disable all enabled rx queues */
5942 for (i = 0; i < adapter->num_rx_queues; i++)
5943 /* this call also flushes the previous write */
5944 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
5946 usleep_range(10000, 20000);
5948 /* synchronize_sched() needed for pending XDP buffers to drain */
5949 if (adapter->xdp_ring[0])
5950 synchronize_sched();
5951 netif_tx_stop_all_queues(netdev);
5953 /* call carrier off first to avoid false dev_watchdog timeouts */
5954 netif_carrier_off(netdev);
5955 netif_tx_disable(netdev);
5957 /* disable any upper devices */
5958 netdev_walk_all_upper_dev_rcu(adapter->netdev,
5959 ixgbe_disable_macvlan, NULL);
5961 ixgbe_irq_disable(adapter);
5963 ixgbe_napi_disable_all(adapter);
5965 clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
5966 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5967 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5969 del_timer_sync(&adapter->service_timer);
5971 if (adapter->num_vfs) {
5972 /* Clear EITR Select mapping */
5973 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
5975 /* Mark all the VFs as inactive */
5976 for (i = 0 ; i < adapter->num_vfs; i++)
5977 adapter->vfinfo[i].clear_to_send = false;
5979 /* ping all the active vfs to let them know we are going down */
5980 ixgbe_ping_all_vfs(adapter);
5982 /* Disable all VFTE/VFRE TX/RX */
5983 ixgbe_disable_tx_rx(adapter);
5986 /* disable transmits in the hardware now that interrupts are off */
5987 for (i = 0; i < adapter->num_tx_queues; i++) {
5988 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
5989 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5991 for (i = 0; i < adapter->num_xdp_queues; i++) {
5992 u8 reg_idx = adapter->xdp_ring[i]->reg_idx;
5994 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5997 /* Disable the Tx DMA engine on 82599 and later MAC */
5998 switch (hw->mac.type) {
5999 case ixgbe_mac_82599EB:
6000 case ixgbe_mac_X540:
6001 case ixgbe_mac_X550:
6002 case ixgbe_mac_X550EM_x:
6003 case ixgbe_mac_x550em_a:
6004 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
6005 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
6006 ~IXGBE_DMATXCTL_TE));
6012 if (!pci_channel_offline(adapter->pdev))
6013 ixgbe_reset(adapter);
6015 /* power down the optics for 82599 SFP+ fiber */
6016 if (hw->mac.ops.disable_tx_laser)
6017 hw->mac.ops.disable_tx_laser(hw);
6019 ixgbe_clean_all_tx_rings(adapter);
6020 ixgbe_clean_all_rx_rings(adapter);
6024 * ixgbe_eee_capable - helper function to determine EEE support on X550
6025 * @adapter: board private structure
6027 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6029 struct ixgbe_hw *hw = &adapter->hw;
6031 switch (hw->device_id) {
6032 case IXGBE_DEV_ID_X550EM_A_1G_T:
6033 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6034 if (!hw->phy.eee_speeds_supported)
6036 adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6037 if (!hw->phy.eee_speeds_advertised)
6039 adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6042 adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6043 adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6049 * ixgbe_tx_timeout - Respond to a Tx Hang
6050 * @netdev: network interface device structure
6052 static void ixgbe_tx_timeout(struct net_device *netdev)
6054 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6056 /* Do the reset outside of interrupt context */
6057 ixgbe_tx_timeout_reset(adapter);
6060 #ifdef CONFIG_IXGBE_DCB
6061 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6063 struct ixgbe_hw *hw = &adapter->hw;
6064 struct tc_configuration *tc;
6067 switch (hw->mac.type) {
6068 case ixgbe_mac_82598EB:
6069 case ixgbe_mac_82599EB:
6070 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6071 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6073 case ixgbe_mac_X540:
6074 case ixgbe_mac_X550:
6075 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6076 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6078 case ixgbe_mac_X550EM_x:
6079 case ixgbe_mac_x550em_a:
6081 adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6082 adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6086 /* Configure DCB traffic classes */
6087 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6088 tc = &adapter->dcb_cfg.tc_config[j];
6089 tc->path[DCB_TX_CONFIG].bwg_id = 0;
6090 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6091 tc->path[DCB_RX_CONFIG].bwg_id = 0;
6092 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6093 tc->dcb_pfc = pfc_disabled;
6096 /* Initialize default user to priority mapping, UPx->TC0 */
6097 tc = &adapter->dcb_cfg.tc_config[0];
6098 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6099 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6101 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6102 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6103 adapter->dcb_cfg.pfc_mode_enable = false;
6104 adapter->dcb_set_bitmap = 0x00;
6105 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6106 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6107 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6108 sizeof(adapter->temp_dcb_cfg));
6113 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6114 * @adapter: board private structure to initialize
6116 * ixgbe_sw_init initializes the Adapter private data structure.
6117 * Fields are initialized based on PCI device information and
6118 * OS network device settings (MTU size).
6120 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6121 const struct ixgbe_info *ii)
6123 struct ixgbe_hw *hw = &adapter->hw;
6124 struct pci_dev *pdev = adapter->pdev;
6125 unsigned int rss, fdir;
6129 /* PCI config space info */
6131 hw->vendor_id = pdev->vendor;
6132 hw->device_id = pdev->device;
6133 hw->revision_id = pdev->revision;
6134 hw->subsystem_vendor_id = pdev->subsystem_vendor;
6135 hw->subsystem_device_id = pdev->subsystem_device;
6137 /* get_invariants needs the device IDs */
6138 ii->get_invariants(hw);
6140 /* Set common capability flags and settings */
6141 rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6142 adapter->ring_feature[RING_F_RSS].limit = rss;
6143 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6144 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6145 adapter->atr_sample_rate = 20;
6146 fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6147 adapter->ring_feature[RING_F_FDIR].limit = fdir;
6148 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6149 #ifdef CONFIG_IXGBE_DCA
6150 adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6152 #ifdef CONFIG_IXGBE_DCB
6153 adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6154 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6157 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6158 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6159 #ifdef CONFIG_IXGBE_DCB
6160 /* Default traffic class to use for FCoE */
6161 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6162 #endif /* CONFIG_IXGBE_DCB */
6163 #endif /* IXGBE_FCOE */
6165 /* initialize static ixgbe jump table entries */
6166 adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6168 if (!adapter->jump_tables[0])
6170 adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6172 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6173 adapter->jump_tables[i] = NULL;
6175 adapter->mac_table = kzalloc(sizeof(struct ixgbe_mac_addr) *
6176 hw->mac.num_rar_entries,
6178 if (!adapter->mac_table)
6181 if (ixgbe_init_rss_key(adapter))
6184 /* Set MAC specific capability flags and exceptions */
6185 switch (hw->mac.type) {
6186 case ixgbe_mac_82598EB:
6187 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6189 if (hw->device_id == IXGBE_DEV_ID_82598AT)
6190 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6192 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6193 adapter->ring_feature[RING_F_FDIR].limit = 0;
6194 adapter->atr_sample_rate = 0;
6195 adapter->fdir_pballoc = 0;
6197 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6198 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6199 #ifdef CONFIG_IXGBE_DCB
6200 adapter->fcoe.up = 0;
6201 #endif /* IXGBE_DCB */
6202 #endif /* IXGBE_FCOE */
6204 case ixgbe_mac_82599EB:
6205 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6206 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6208 case ixgbe_mac_X540:
6209 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6210 if (fwsm & IXGBE_FWSM_TS_ENABLED)
6211 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6213 case ixgbe_mac_x550em_a:
6214 adapter->flags |= IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE;
6215 switch (hw->device_id) {
6216 case IXGBE_DEV_ID_X550EM_A_1G_T:
6217 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6218 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6224 case ixgbe_mac_X550EM_x:
6225 #ifdef CONFIG_IXGBE_DCB
6226 adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6229 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6230 #ifdef CONFIG_IXGBE_DCB
6231 adapter->fcoe.up = 0;
6232 #endif /* IXGBE_DCB */
6233 #endif /* IXGBE_FCOE */
6235 case ixgbe_mac_X550:
6236 if (hw->mac.type == ixgbe_mac_X550)
6237 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6238 #ifdef CONFIG_IXGBE_DCA
6239 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6241 adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
6248 /* FCoE support exists, always init the FCoE lock */
6249 spin_lock_init(&adapter->fcoe.lock);
6252 /* n-tuple support exists, always init our spinlock */
6253 spin_lock_init(&adapter->fdir_perfect_lock);
6255 #ifdef CONFIG_IXGBE_DCB
6256 ixgbe_init_dcb(adapter);
6259 /* default flow control settings */
6260 hw->fc.requested_mode = ixgbe_fc_full;
6261 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
6262 ixgbe_pbthresh_setup(adapter);
6263 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6264 hw->fc.send_xon = true;
6265 hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6267 #ifdef CONFIG_PCI_IOV
6269 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6271 /* assign number of SR-IOV VFs */
6272 if (hw->mac.type != ixgbe_mac_82598EB) {
6273 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6275 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6278 #endif /* CONFIG_PCI_IOV */
6280 /* enable itr by default in dynamic mode */
6281 adapter->rx_itr_setting = 1;
6282 adapter->tx_itr_setting = 1;
6284 /* set default ring sizes */
6285 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6286 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6288 /* set default work limits */
6289 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6291 /* initialize eeprom parameters */
6292 if (ixgbe_init_eeprom_params_generic(hw)) {
6293 e_dev_err("EEPROM initialization failed\n");
6297 /* PF holds first pool slot */
6298 set_bit(0, &adapter->fwd_bitmask);
6299 set_bit(__IXGBE_DOWN, &adapter->state);
6305 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6306 * @tx_ring: tx descriptor ring (for a specific queue) to setup
6308 * Return 0 on success, negative on failure
6310 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6312 struct device *dev = tx_ring->dev;
6313 int orig_node = dev_to_node(dev);
6317 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6319 if (tx_ring->q_vector)
6320 ring_node = tx_ring->q_vector->numa_node;
6322 tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6323 if (!tx_ring->tx_buffer_info)
6324 tx_ring->tx_buffer_info = vmalloc(size);
6325 if (!tx_ring->tx_buffer_info)
6328 /* round up to nearest 4K */
6329 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6330 tx_ring->size = ALIGN(tx_ring->size, 4096);
6332 set_dev_node(dev, ring_node);
6333 tx_ring->desc = dma_alloc_coherent(dev,
6337 set_dev_node(dev, orig_node);
6339 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6340 &tx_ring->dma, GFP_KERNEL);
6344 tx_ring->next_to_use = 0;
6345 tx_ring->next_to_clean = 0;
6349 vfree(tx_ring->tx_buffer_info);
6350 tx_ring->tx_buffer_info = NULL;
6351 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6356 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6357 * @adapter: board private structure
6359 * If this function returns with an error, then it's possible one or
6360 * more of the rings is populated (while the rest are not). It is the
6361 * callers duty to clean those orphaned rings.
6363 * Return 0 on success, negative on failure
6365 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6367 int i, j = 0, err = 0;
6369 for (i = 0; i < adapter->num_tx_queues; i++) {
6370 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6374 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6377 for (j = 0; j < adapter->num_xdp_queues; j++) {
6378 err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6382 e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6388 /* rewind the index freeing the rings as we go */
6390 ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6392 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6397 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6398 * @rx_ring: rx descriptor ring (for a specific queue) to setup
6400 * Returns 0 on success, negative on failure
6402 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6403 struct ixgbe_ring *rx_ring)
6405 struct device *dev = rx_ring->dev;
6406 int orig_node = dev_to_node(dev);
6410 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6412 if (rx_ring->q_vector)
6413 ring_node = rx_ring->q_vector->numa_node;
6415 rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6416 if (!rx_ring->rx_buffer_info)
6417 rx_ring->rx_buffer_info = vmalloc(size);
6418 if (!rx_ring->rx_buffer_info)
6421 /* Round up to nearest 4K */
6422 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6423 rx_ring->size = ALIGN(rx_ring->size, 4096);
6425 set_dev_node(dev, ring_node);
6426 rx_ring->desc = dma_alloc_coherent(dev,
6430 set_dev_node(dev, orig_node);
6432 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6433 &rx_ring->dma, GFP_KERNEL);
6437 rx_ring->next_to_clean = 0;
6438 rx_ring->next_to_use = 0;
6440 /* XDP RX-queue info */
6441 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6442 rx_ring->queue_index) < 0)
6445 rx_ring->xdp_prog = adapter->xdp_prog;
6449 vfree(rx_ring->rx_buffer_info);
6450 rx_ring->rx_buffer_info = NULL;
6451 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6456 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6457 * @adapter: board private structure
6459 * If this function returns with an error, then it's possible one or
6460 * more of the rings is populated (while the rest are not). It is the
6461 * callers duty to clean those orphaned rings.
6463 * Return 0 on success, negative on failure
6465 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6469 for (i = 0; i < adapter->num_rx_queues; i++) {
6470 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6474 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6479 err = ixgbe_setup_fcoe_ddp_resources(adapter);
6484 /* rewind the index freeing the rings as we go */
6486 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6491 * ixgbe_free_tx_resources - Free Tx Resources per Queue
6492 * @tx_ring: Tx descriptor ring for a specific queue
6494 * Free all transmit software resources
6496 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6498 ixgbe_clean_tx_ring(tx_ring);
6500 vfree(tx_ring->tx_buffer_info);
6501 tx_ring->tx_buffer_info = NULL;
6503 /* if not set, then don't free */
6507 dma_free_coherent(tx_ring->dev, tx_ring->size,
6508 tx_ring->desc, tx_ring->dma);
6510 tx_ring->desc = NULL;
6514 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6515 * @adapter: board private structure
6517 * Free all transmit software resources
6519 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6523 for (i = 0; i < adapter->num_tx_queues; i++)
6524 if (adapter->tx_ring[i]->desc)
6525 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6526 for (i = 0; i < adapter->num_xdp_queues; i++)
6527 if (adapter->xdp_ring[i]->desc)
6528 ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6532 * ixgbe_free_rx_resources - Free Rx Resources
6533 * @rx_ring: ring to clean the resources from
6535 * Free all receive software resources
6537 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6539 ixgbe_clean_rx_ring(rx_ring);
6541 rx_ring->xdp_prog = NULL;
6542 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6543 vfree(rx_ring->rx_buffer_info);
6544 rx_ring->rx_buffer_info = NULL;
6546 /* if not set, then don't free */
6550 dma_free_coherent(rx_ring->dev, rx_ring->size,
6551 rx_ring->desc, rx_ring->dma);
6553 rx_ring->desc = NULL;
6557 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6558 * @adapter: board private structure
6560 * Free all receive software resources
6562 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6567 ixgbe_free_fcoe_ddp_resources(adapter);
6570 for (i = 0; i < adapter->num_rx_queues; i++)
6571 if (adapter->rx_ring[i]->desc)
6572 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6576 * ixgbe_change_mtu - Change the Maximum Transfer Unit
6577 * @netdev: network interface device structure
6578 * @new_mtu: new value for maximum frame size
6580 * Returns 0 on success, negative on failure
6582 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6584 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6587 * For 82599EB we cannot allow legacy VFs to enable their receive
6588 * paths when MTU greater than 1500 is configured. So display a
6589 * warning that legacy VFs will be disabled.
6591 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6592 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6593 (new_mtu > ETH_DATA_LEN))
6594 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6596 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
6598 /* must set new MTU before calling down or up */
6599 netdev->mtu = new_mtu;
6601 if (netif_running(netdev))
6602 ixgbe_reinit_locked(adapter);
6608 * ixgbe_open - Called when a network interface is made active
6609 * @netdev: network interface device structure
6611 * Returns 0 on success, negative value on failure
6613 * The open entry point is called when a network interface is made
6614 * active by the system (IFF_UP). At this point all resources needed
6615 * for transmit and receive operations are allocated, the interrupt
6616 * handler is registered with the OS, the watchdog timer is started,
6617 * and the stack is notified that the interface is ready.
6619 int ixgbe_open(struct net_device *netdev)
6621 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6622 struct ixgbe_hw *hw = &adapter->hw;
6625 /* disallow open during test */
6626 if (test_bit(__IXGBE_TESTING, &adapter->state))
6629 netif_carrier_off(netdev);
6631 /* allocate transmit descriptors */
6632 err = ixgbe_setup_all_tx_resources(adapter);
6636 /* allocate receive descriptors */
6637 err = ixgbe_setup_all_rx_resources(adapter);
6641 ixgbe_configure(adapter);
6643 err = ixgbe_request_irq(adapter);
6647 /* Notify the stack of the actual queue counts. */
6648 if (adapter->num_rx_pools > 1)
6649 queues = adapter->num_rx_queues_per_pool;
6651 queues = adapter->num_tx_queues;
6653 err = netif_set_real_num_tx_queues(netdev, queues);
6655 goto err_set_queues;
6657 if (adapter->num_rx_pools > 1 &&
6658 adapter->num_rx_queues > IXGBE_MAX_L2A_QUEUES)
6659 queues = IXGBE_MAX_L2A_QUEUES;
6661 queues = adapter->num_rx_queues;
6662 err = netif_set_real_num_rx_queues(netdev, queues);
6664 goto err_set_queues;
6666 ixgbe_ptp_init(adapter);
6668 ixgbe_up_complete(adapter);
6670 ixgbe_clear_udp_tunnel_port(adapter, IXGBE_VXLANCTRL_ALL_UDPPORT_MASK);
6671 udp_tunnel_get_rx_info(netdev);
6676 ixgbe_free_irq(adapter);
6678 ixgbe_free_all_rx_resources(adapter);
6679 if (hw->phy.ops.set_phy_power && !adapter->wol)
6680 hw->phy.ops.set_phy_power(&adapter->hw, false);
6682 ixgbe_free_all_tx_resources(adapter);
6684 ixgbe_reset(adapter);
6689 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6691 ixgbe_ptp_suspend(adapter);
6693 if (adapter->hw.phy.ops.enter_lplu) {
6694 adapter->hw.phy.reset_disable = true;
6695 ixgbe_down(adapter);
6696 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6697 adapter->hw.phy.reset_disable = false;
6699 ixgbe_down(adapter);
6702 ixgbe_free_irq(adapter);
6704 ixgbe_free_all_tx_resources(adapter);
6705 ixgbe_free_all_rx_resources(adapter);
6709 * ixgbe_close - Disables a network interface
6710 * @netdev: network interface device structure
6712 * Returns 0, this is not allowed to fail
6714 * The close entry point is called when an interface is de-activated
6715 * by the OS. The hardware is still under the drivers control, but
6716 * needs to be disabled. A global MAC reset is issued to stop the
6717 * hardware, and all transmit and receive resources are freed.
6719 int ixgbe_close(struct net_device *netdev)
6721 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6723 ixgbe_ptp_stop(adapter);
6725 if (netif_device_present(netdev))
6726 ixgbe_close_suspend(adapter);
6728 ixgbe_fdir_filter_exit(adapter);
6730 ixgbe_release_hw_control(adapter);
6736 static int ixgbe_resume(struct pci_dev *pdev)
6738 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6739 struct net_device *netdev = adapter->netdev;
6742 adapter->hw.hw_addr = adapter->io_addr;
6743 pci_set_power_state(pdev, PCI_D0);
6744 pci_restore_state(pdev);
6746 * pci_restore_state clears dev->state_saved so call
6747 * pci_save_state to restore it.
6749 pci_save_state(pdev);
6751 err = pci_enable_device_mem(pdev);
6753 e_dev_err("Cannot enable PCI device from suspend\n");
6756 smp_mb__before_atomic();
6757 clear_bit(__IXGBE_DISABLED, &adapter->state);
6758 pci_set_master(pdev);
6760 pci_wake_from_d3(pdev, false);
6762 ixgbe_reset(adapter);
6764 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6767 err = ixgbe_init_interrupt_scheme(adapter);
6768 if (!err && netif_running(netdev))
6769 err = ixgbe_open(netdev);
6773 netif_device_attach(netdev);
6778 #endif /* CONFIG_PM */
6780 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6782 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6783 struct net_device *netdev = adapter->netdev;
6784 struct ixgbe_hw *hw = &adapter->hw;
6786 u32 wufc = adapter->wol;
6792 netif_device_detach(netdev);
6794 if (netif_running(netdev))
6795 ixgbe_close_suspend(adapter);
6797 ixgbe_clear_interrupt_scheme(adapter);
6801 retval = pci_save_state(pdev);
6806 if (hw->mac.ops.stop_link_on_d3)
6807 hw->mac.ops.stop_link_on_d3(hw);
6812 ixgbe_set_rx_mode(netdev);
6814 /* enable the optics for 82599 SFP+ fiber as we can WoL */
6815 if (hw->mac.ops.enable_tx_laser)
6816 hw->mac.ops.enable_tx_laser(hw);
6818 /* enable the reception of multicast packets */
6819 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6820 fctrl |= IXGBE_FCTRL_MPE;
6821 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6823 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6824 ctrl |= IXGBE_CTRL_GIO_DIS;
6825 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6827 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6829 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6830 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6833 switch (hw->mac.type) {
6834 case ixgbe_mac_82598EB:
6835 pci_wake_from_d3(pdev, false);
6837 case ixgbe_mac_82599EB:
6838 case ixgbe_mac_X540:
6839 case ixgbe_mac_X550:
6840 case ixgbe_mac_X550EM_x:
6841 case ixgbe_mac_x550em_a:
6842 pci_wake_from_d3(pdev, !!wufc);
6848 *enable_wake = !!wufc;
6849 if (hw->phy.ops.set_phy_power && !*enable_wake)
6850 hw->phy.ops.set_phy_power(hw, false);
6852 ixgbe_release_hw_control(adapter);
6854 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
6855 pci_disable_device(pdev);
6861 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
6866 retval = __ixgbe_shutdown(pdev, &wake);
6871 pci_prepare_to_sleep(pdev);
6873 pci_wake_from_d3(pdev, false);
6874 pci_set_power_state(pdev, PCI_D3hot);
6879 #endif /* CONFIG_PM */
6881 static void ixgbe_shutdown(struct pci_dev *pdev)
6885 __ixgbe_shutdown(pdev, &wake);
6887 if (system_state == SYSTEM_POWER_OFF) {
6888 pci_wake_from_d3(pdev, wake);
6889 pci_set_power_state(pdev, PCI_D3hot);
6894 * ixgbe_update_stats - Update the board statistics counters.
6895 * @adapter: board private structure
6897 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
6899 struct net_device *netdev = adapter->netdev;
6900 struct ixgbe_hw *hw = &adapter->hw;
6901 struct ixgbe_hw_stats *hwstats = &adapter->stats;
6903 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
6904 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
6905 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
6906 u64 alloc_rx_page = 0;
6907 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
6909 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6910 test_bit(__IXGBE_RESETTING, &adapter->state))
6913 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
6916 for (i = 0; i < adapter->num_rx_queues; i++) {
6917 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
6918 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
6920 adapter->rsc_total_count = rsc_count;
6921 adapter->rsc_total_flush = rsc_flush;
6924 for (i = 0; i < adapter->num_rx_queues; i++) {
6925 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
6926 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
6927 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
6928 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
6929 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
6930 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
6931 bytes += rx_ring->stats.bytes;
6932 packets += rx_ring->stats.packets;
6934 adapter->non_eop_descs = non_eop_descs;
6935 adapter->alloc_rx_page = alloc_rx_page;
6936 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
6937 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
6938 adapter->hw_csum_rx_error = hw_csum_rx_error;
6939 netdev->stats.rx_bytes = bytes;
6940 netdev->stats.rx_packets = packets;
6944 /* gather some stats to the adapter struct that are per queue */
6945 for (i = 0; i < adapter->num_tx_queues; i++) {
6946 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6947 restart_queue += tx_ring->tx_stats.restart_queue;
6948 tx_busy += tx_ring->tx_stats.tx_busy;
6949 bytes += tx_ring->stats.bytes;
6950 packets += tx_ring->stats.packets;
6952 for (i = 0; i < adapter->num_xdp_queues; i++) {
6953 struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
6955 restart_queue += xdp_ring->tx_stats.restart_queue;
6956 tx_busy += xdp_ring->tx_stats.tx_busy;
6957 bytes += xdp_ring->stats.bytes;
6958 packets += xdp_ring->stats.packets;
6960 adapter->restart_queue = restart_queue;
6961 adapter->tx_busy = tx_busy;
6962 netdev->stats.tx_bytes = bytes;
6963 netdev->stats.tx_packets = packets;
6965 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
6967 /* 8 register reads */
6968 for (i = 0; i < 8; i++) {
6969 /* for packet buffers not used, the register should read 0 */
6970 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
6972 hwstats->mpc[i] += mpc;
6973 total_mpc += hwstats->mpc[i];
6974 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
6975 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
6976 switch (hw->mac.type) {
6977 case ixgbe_mac_82598EB:
6978 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
6979 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
6980 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
6981 hwstats->pxonrxc[i] +=
6982 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
6984 case ixgbe_mac_82599EB:
6985 case ixgbe_mac_X540:
6986 case ixgbe_mac_X550:
6987 case ixgbe_mac_X550EM_x:
6988 case ixgbe_mac_x550em_a:
6989 hwstats->pxonrxc[i] +=
6990 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
6997 /*16 register reads */
6998 for (i = 0; i < 16; i++) {
6999 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7000 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7001 if ((hw->mac.type == ixgbe_mac_82599EB) ||
7002 (hw->mac.type == ixgbe_mac_X540) ||
7003 (hw->mac.type == ixgbe_mac_X550) ||
7004 (hw->mac.type == ixgbe_mac_X550EM_x) ||
7005 (hw->mac.type == ixgbe_mac_x550em_a)) {
7006 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7007 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7008 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7009 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7013 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7014 /* work around hardware counting issue */
7015 hwstats->gprc -= missed_rx;
7017 ixgbe_update_xoff_received(adapter);
7019 /* 82598 hardware only has a 32 bit counter in the high register */
7020 switch (hw->mac.type) {
7021 case ixgbe_mac_82598EB:
7022 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7023 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7024 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7025 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7027 case ixgbe_mac_X540:
7028 case ixgbe_mac_X550:
7029 case ixgbe_mac_X550EM_x:
7030 case ixgbe_mac_x550em_a:
7031 /* OS2BMC stats are X540 and later */
7032 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7033 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7034 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7035 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7037 case ixgbe_mac_82599EB:
7038 for (i = 0; i < 16; i++)
7039 adapter->hw_rx_no_dma_resources +=
7040 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7041 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7042 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7043 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7044 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7045 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7046 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7047 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7048 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7049 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7051 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7052 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7053 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7054 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7055 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7056 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7057 /* Add up per cpu counters for total ddp aloc fail */
7058 if (adapter->fcoe.ddp_pool) {
7059 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7060 struct ixgbe_fcoe_ddp_pool *ddp_pool;
7062 u64 noddp = 0, noddp_ext_buff = 0;
7063 for_each_possible_cpu(cpu) {
7064 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7065 noddp += ddp_pool->noddp;
7066 noddp_ext_buff += ddp_pool->noddp_ext_buff;
7068 hwstats->fcoe_noddp = noddp;
7069 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7071 #endif /* IXGBE_FCOE */
7076 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7077 hwstats->bprc += bprc;
7078 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7079 if (hw->mac.type == ixgbe_mac_82598EB)
7080 hwstats->mprc -= bprc;
7081 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7082 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7083 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7084 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7085 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7086 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7087 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7088 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7089 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7090 hwstats->lxontxc += lxon;
7091 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7092 hwstats->lxofftxc += lxoff;
7093 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7094 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7096 * 82598 errata - tx of flow control packets is included in tx counters
7098 xon_off_tot = lxon + lxoff;
7099 hwstats->gptc -= xon_off_tot;
7100 hwstats->mptc -= xon_off_tot;
7101 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7102 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7103 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7104 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7105 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7106 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7107 hwstats->ptc64 -= xon_off_tot;
7108 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7109 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7110 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7111 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7112 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7113 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7115 /* Fill out the OS statistics structure */
7116 netdev->stats.multicast = hwstats->mprc;
7119 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7120 netdev->stats.rx_dropped = 0;
7121 netdev->stats.rx_length_errors = hwstats->rlec;
7122 netdev->stats.rx_crc_errors = hwstats->crcerrs;
7123 netdev->stats.rx_missed_errors = total_mpc;
7127 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7128 * @adapter: pointer to the device adapter structure
7130 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7132 struct ixgbe_hw *hw = &adapter->hw;
7135 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7138 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7140 /* if interface is down do nothing */
7141 if (test_bit(__IXGBE_DOWN, &adapter->state))
7144 /* do nothing if we are not using signature filters */
7145 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7148 adapter->fdir_overflow++;
7150 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7151 for (i = 0; i < adapter->num_tx_queues; i++)
7152 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7153 &(adapter->tx_ring[i]->state));
7154 for (i = 0; i < adapter->num_xdp_queues; i++)
7155 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7156 &adapter->xdp_ring[i]->state);
7157 /* re-enable flow director interrupts */
7158 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7160 e_err(probe, "failed to finish FDIR re-initialization, "
7161 "ignored adding FDIR ATR filters\n");
7166 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7167 * @adapter: pointer to the device adapter structure
7169 * This function serves two purposes. First it strobes the interrupt lines
7170 * in order to make certain interrupts are occurring. Secondly it sets the
7171 * bits needed to check for TX hangs. As a result we should immediately
7172 * determine if a hang has occurred.
7174 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7176 struct ixgbe_hw *hw = &adapter->hw;
7180 /* If we're down, removing or resetting, just bail */
7181 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7182 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7183 test_bit(__IXGBE_RESETTING, &adapter->state))
7186 /* Force detection of hung controller */
7187 if (netif_carrier_ok(adapter->netdev)) {
7188 for (i = 0; i < adapter->num_tx_queues; i++)
7189 set_check_for_tx_hang(adapter->tx_ring[i]);
7190 for (i = 0; i < adapter->num_xdp_queues; i++)
7191 set_check_for_tx_hang(adapter->xdp_ring[i]);
7194 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7196 * for legacy and MSI interrupts don't set any bits
7197 * that are enabled for EIAM, because this operation
7198 * would set *both* EIMS and EICS for any bit in EIAM
7200 IXGBE_WRITE_REG(hw, IXGBE_EICS,
7201 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7203 /* get one bit for every active tx/rx interrupt vector */
7204 for (i = 0; i < adapter->num_q_vectors; i++) {
7205 struct ixgbe_q_vector *qv = adapter->q_vector[i];
7206 if (qv->rx.ring || qv->tx.ring)
7211 /* Cause software interrupt to ensure rings are cleaned */
7212 ixgbe_irq_rearm_queues(adapter, eics);
7216 * ixgbe_watchdog_update_link - update the link status
7217 * @adapter: pointer to the device adapter structure
7218 * @link_speed: pointer to a u32 to store the link_speed
7220 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7222 struct ixgbe_hw *hw = &adapter->hw;
7223 u32 link_speed = adapter->link_speed;
7224 bool link_up = adapter->link_up;
7225 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7227 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7230 if (hw->mac.ops.check_link) {
7231 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7233 /* always assume link is up, if no check link function */
7234 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7238 if (adapter->ixgbe_ieee_pfc)
7239 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7241 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7242 hw->mac.ops.fc_enable(hw);
7243 ixgbe_set_rx_drop_en(adapter);
7247 time_after(jiffies, (adapter->link_check_timeout +
7248 IXGBE_TRY_LINK_TIMEOUT))) {
7249 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7250 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7251 IXGBE_WRITE_FLUSH(hw);
7254 adapter->link_up = link_up;
7255 adapter->link_speed = link_speed;
7258 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7260 #ifdef CONFIG_IXGBE_DCB
7261 struct net_device *netdev = adapter->netdev;
7262 struct dcb_app app = {
7263 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7268 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7269 up = dcb_ieee_getapp_mask(netdev, &app);
7271 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7275 static int ixgbe_enable_macvlan(struct net_device *upper, void *data)
7277 if (netif_is_macvlan(upper)) {
7278 struct macvlan_dev *vlan = netdev_priv(upper);
7281 netif_tx_wake_all_queues(upper);
7288 * ixgbe_watchdog_link_is_up - update netif_carrier status and
7289 * print link up message
7290 * @adapter: pointer to the device adapter structure
7292 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7294 struct net_device *netdev = adapter->netdev;
7295 struct ixgbe_hw *hw = &adapter->hw;
7296 u32 link_speed = adapter->link_speed;
7297 const char *speed_str;
7298 bool flow_rx, flow_tx;
7300 /* only continue if link was previously down */
7301 if (netif_carrier_ok(netdev))
7304 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7306 switch (hw->mac.type) {
7307 case ixgbe_mac_82598EB: {
7308 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7309 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7310 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7311 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7314 case ixgbe_mac_X540:
7315 case ixgbe_mac_X550:
7316 case ixgbe_mac_X550EM_x:
7317 case ixgbe_mac_x550em_a:
7318 case ixgbe_mac_82599EB: {
7319 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7320 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7321 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7322 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7331 adapter->last_rx_ptp_check = jiffies;
7333 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7334 ixgbe_ptp_start_cyclecounter(adapter);
7336 switch (link_speed) {
7337 case IXGBE_LINK_SPEED_10GB_FULL:
7338 speed_str = "10 Gbps";
7340 case IXGBE_LINK_SPEED_2_5GB_FULL:
7341 speed_str = "2.5 Gbps";
7343 case IXGBE_LINK_SPEED_1GB_FULL:
7344 speed_str = "1 Gbps";
7346 case IXGBE_LINK_SPEED_100_FULL:
7347 speed_str = "100 Mbps";
7349 case IXGBE_LINK_SPEED_10_FULL:
7350 speed_str = "10 Mbps";
7353 speed_str = "unknown speed";
7356 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7357 ((flow_rx && flow_tx) ? "RX/TX" :
7359 (flow_tx ? "TX" : "None"))));
7361 netif_carrier_on(netdev);
7362 ixgbe_check_vf_rate_limit(adapter);
7364 /* enable transmits */
7365 netif_tx_wake_all_queues(adapter->netdev);
7367 /* enable any upper devices */
7369 netdev_walk_all_upper_dev_rcu(adapter->netdev,
7370 ixgbe_enable_macvlan, NULL);
7373 /* update the default user priority for VFs */
7374 ixgbe_update_default_up(adapter);
7376 /* ping all the active vfs to let them know link has changed */
7377 ixgbe_ping_all_vfs(adapter);
7381 * ixgbe_watchdog_link_is_down - update netif_carrier status and
7382 * print link down message
7383 * @adapter: pointer to the adapter structure
7385 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7387 struct net_device *netdev = adapter->netdev;
7388 struct ixgbe_hw *hw = &adapter->hw;
7390 adapter->link_up = false;
7391 adapter->link_speed = 0;
7393 /* only continue if link was up previously */
7394 if (!netif_carrier_ok(netdev))
7397 /* poll for SFP+ cable when link is down */
7398 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7399 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7401 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7402 ixgbe_ptp_start_cyclecounter(adapter);
7404 e_info(drv, "NIC Link is Down\n");
7405 netif_carrier_off(netdev);
7407 /* ping all the active vfs to let them know link has changed */
7408 ixgbe_ping_all_vfs(adapter);
7411 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7415 for (i = 0; i < adapter->num_tx_queues; i++) {
7416 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7418 if (tx_ring->next_to_use != tx_ring->next_to_clean)
7422 for (i = 0; i < adapter->num_xdp_queues; i++) {
7423 struct ixgbe_ring *ring = adapter->xdp_ring[i];
7425 if (ring->next_to_use != ring->next_to_clean)
7432 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7434 struct ixgbe_hw *hw = &adapter->hw;
7435 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7436 u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7440 if (!adapter->num_vfs)
7443 /* resetting the PF is only needed for MAC before X550 */
7444 if (hw->mac.type >= ixgbe_mac_X550)
7447 for (i = 0; i < adapter->num_vfs; i++) {
7448 for (j = 0; j < q_per_pool; j++) {
7451 h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7452 t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7463 * ixgbe_watchdog_flush_tx - flush queues on link down
7464 * @adapter: pointer to the device adapter structure
7466 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7468 if (!netif_carrier_ok(adapter->netdev)) {
7469 if (ixgbe_ring_tx_pending(adapter) ||
7470 ixgbe_vf_tx_pending(adapter)) {
7471 /* We've lost link, so the controller stops DMA,
7472 * but we've got queued Tx work that's never going
7473 * to get done, so reset controller to flush Tx.
7474 * (Do the reset outside of interrupt context).
7476 e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7477 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7482 #ifdef CONFIG_PCI_IOV
7483 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7485 struct ixgbe_hw *hw = &adapter->hw;
7486 struct pci_dev *pdev = adapter->pdev;
7490 if (!(netif_carrier_ok(adapter->netdev)))
7493 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7494 if (gpc) /* If incrementing then no need for the check below */
7496 /* Check to see if a bad DMA write target from an errant or
7497 * malicious VF has caused a PCIe error. If so then we can
7498 * issue a VFLR to the offending VF(s) and then resume without
7499 * requesting a full slot reset.
7505 /* check status reg for all VFs owned by this PF */
7506 for (vf = 0; vf < adapter->num_vfs; ++vf) {
7507 struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7512 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7513 if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7514 status_reg & PCI_STATUS_REC_MASTER_ABORT)
7519 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7523 /* Do not perform spoof check for 82598 or if not in IOV mode */
7524 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7525 adapter->num_vfs == 0)
7528 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7531 * ssvpc register is cleared on read, if zero then no
7532 * spoofed packets in the last interval.
7537 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7540 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7545 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7548 #endif /* CONFIG_PCI_IOV */
7552 * ixgbe_watchdog_subtask - check and bring link up
7553 * @adapter: pointer to the device adapter structure
7555 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7557 /* if interface is down, removing or resetting, do nothing */
7558 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7559 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7560 test_bit(__IXGBE_RESETTING, &adapter->state))
7563 ixgbe_watchdog_update_link(adapter);
7565 if (adapter->link_up)
7566 ixgbe_watchdog_link_is_up(adapter);
7568 ixgbe_watchdog_link_is_down(adapter);
7570 ixgbe_check_for_bad_vf(adapter);
7571 ixgbe_spoof_check(adapter);
7572 ixgbe_update_stats(adapter);
7574 ixgbe_watchdog_flush_tx(adapter);
7578 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7579 * @adapter: the ixgbe adapter structure
7581 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7583 struct ixgbe_hw *hw = &adapter->hw;
7586 /* not searching for SFP so there is nothing to do here */
7587 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7588 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7591 if (adapter->sfp_poll_time &&
7592 time_after(adapter->sfp_poll_time, jiffies))
7593 return; /* If not yet time to poll for SFP */
7595 /* someone else is in init, wait until next service event */
7596 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7599 adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7601 err = hw->phy.ops.identify_sfp(hw);
7602 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7605 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7606 /* If no cable is present, then we need to reset
7607 * the next time we find a good cable. */
7608 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7615 /* exit if reset not needed */
7616 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7619 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7622 * A module may be identified correctly, but the EEPROM may not have
7623 * support for that module. setup_sfp() will fail in that case, so
7624 * we should not allow that module to load.
7626 if (hw->mac.type == ixgbe_mac_82598EB)
7627 err = hw->phy.ops.reset(hw);
7629 err = hw->mac.ops.setup_sfp(hw);
7631 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7634 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7635 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7638 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7640 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7641 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7642 e_dev_err("failed to initialize because an unsupported "
7643 "SFP+ module type was detected.\n");
7644 e_dev_err("Reload the driver after installing a "
7645 "supported module.\n");
7646 unregister_netdev(adapter->netdev);
7651 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7652 * @adapter: the ixgbe adapter structure
7654 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7656 struct ixgbe_hw *hw = &adapter->hw;
7659 bool autoneg = false;
7661 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7664 /* someone else is in init, wait until next service event */
7665 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7668 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7670 hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7672 /* advertise highest capable link speed */
7673 if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7674 speed = IXGBE_LINK_SPEED_10GB_FULL;
7676 speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7677 IXGBE_LINK_SPEED_1GB_FULL);
7679 if (hw->mac.ops.setup_link)
7680 hw->mac.ops.setup_link(hw, speed, true);
7682 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7683 adapter->link_check_timeout = jiffies;
7684 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7688 * ixgbe_service_timer - Timer Call-back
7689 * @data: pointer to adapter cast into an unsigned long
7691 static void ixgbe_service_timer(struct timer_list *t)
7693 struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7694 unsigned long next_event_offset;
7696 /* poll faster when waiting for link */
7697 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7698 next_event_offset = HZ / 10;
7700 next_event_offset = HZ * 2;
7702 /* Reset the timer */
7703 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7705 ixgbe_service_event_schedule(adapter);
7708 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7710 struct ixgbe_hw *hw = &adapter->hw;
7713 if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7716 adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7718 if (!hw->phy.ops.handle_lasi)
7721 status = hw->phy.ops.handle_lasi(&adapter->hw);
7722 if (status != IXGBE_ERR_OVERTEMP)
7725 e_crit(drv, "%s\n", ixgbe_overheat_msg);
7728 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7730 if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7733 /* If we're already down, removing or resetting, just bail */
7734 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7735 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7736 test_bit(__IXGBE_RESETTING, &adapter->state))
7739 ixgbe_dump(adapter);
7740 netdev_err(adapter->netdev, "Reset adapter\n");
7741 adapter->tx_timeout_count++;
7744 ixgbe_reinit_locked(adapter);
7749 * ixgbe_service_task - manages and runs subtasks
7750 * @work: pointer to work_struct containing our data
7752 static void ixgbe_service_task(struct work_struct *work)
7754 struct ixgbe_adapter *adapter = container_of(work,
7755 struct ixgbe_adapter,
7757 if (ixgbe_removed(adapter->hw.hw_addr)) {
7758 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7760 ixgbe_down(adapter);
7763 ixgbe_service_event_complete(adapter);
7766 if (adapter->flags2 & IXGBE_FLAG2_UDP_TUN_REREG_NEEDED) {
7768 adapter->flags2 &= ~IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
7769 udp_tunnel_get_rx_info(adapter->netdev);
7772 ixgbe_reset_subtask(adapter);
7773 ixgbe_phy_interrupt_subtask(adapter);
7774 ixgbe_sfp_detection_subtask(adapter);
7775 ixgbe_sfp_link_config_subtask(adapter);
7776 ixgbe_check_overtemp_subtask(adapter);
7777 ixgbe_watchdog_subtask(adapter);
7778 ixgbe_fdir_reinit_subtask(adapter);
7779 ixgbe_check_hang_subtask(adapter);
7781 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7782 ixgbe_ptp_overflow_check(adapter);
7783 ixgbe_ptp_rx_hang(adapter);
7784 ixgbe_ptp_tx_hang(adapter);
7787 ixgbe_service_event_complete(adapter);
7790 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7791 struct ixgbe_tx_buffer *first,
7794 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7795 struct sk_buff *skb = first->skb;
7805 u32 paylen, l4_offset;
7808 if (skb->ip_summed != CHECKSUM_PARTIAL)
7811 if (!skb_is_gso(skb))
7814 err = skb_cow_head(skb, 0);
7818 if (eth_p_mpls(first->protocol))
7819 ip.hdr = skb_inner_network_header(skb);
7821 ip.hdr = skb_network_header(skb);
7822 l4.hdr = skb_checksum_start(skb);
7824 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7825 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
7827 /* initialize outer IP header fields */
7828 if (ip.v4->version == 4) {
7829 unsigned char *csum_start = skb_checksum_start(skb);
7830 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
7832 /* IP header will have to cancel out any data that
7833 * is not a part of the outer IP header
7835 ip.v4->check = csum_fold(csum_partial(trans_start,
7836 csum_start - trans_start,
7838 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7841 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7842 IXGBE_TX_FLAGS_CSUM |
7843 IXGBE_TX_FLAGS_IPV4;
7845 ip.v6->payload_len = 0;
7846 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7847 IXGBE_TX_FLAGS_CSUM;
7850 /* determine offset of inner transport header */
7851 l4_offset = l4.hdr - skb->data;
7853 /* compute length of segmentation header */
7854 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
7856 /* remove payload length from inner checksum */
7857 paylen = skb->len - l4_offset;
7858 csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
7860 /* update gso size and bytecount with header size */
7861 first->gso_segs = skb_shinfo(skb)->gso_segs;
7862 first->bytecount += (first->gso_segs - 1) * *hdr_len;
7864 /* mss_l4len_id: use 0 as index for TSO */
7865 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
7866 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
7868 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
7869 vlan_macip_lens = l4.hdr - ip.hdr;
7870 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
7871 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
7873 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
7879 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb)
7881 unsigned int offset = 0;
7883 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
7885 return offset == skb_checksum_start_offset(skb);
7888 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
7889 struct ixgbe_tx_buffer *first)
7891 struct sk_buff *skb = first->skb;
7892 u32 vlan_macip_lens = 0;
7895 if (skb->ip_summed != CHECKSUM_PARTIAL) {
7897 if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
7898 IXGBE_TX_FLAGS_CC)))
7903 switch (skb->csum_offset) {
7904 case offsetof(struct tcphdr, check):
7905 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
7907 case offsetof(struct udphdr, check):
7909 case offsetof(struct sctphdr, checksum):
7910 /* validate that this is actually an SCTP request */
7911 if (((first->protocol == htons(ETH_P_IP)) &&
7912 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
7913 ((first->protocol == htons(ETH_P_IPV6)) &&
7914 ixgbe_ipv6_csum_is_sctp(skb))) {
7915 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
7920 skb_checksum_help(skb);
7924 /* update TX checksum flag */
7925 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
7926 vlan_macip_lens = skb_checksum_start_offset(skb) -
7927 skb_network_offset(skb);
7929 /* vlan_macip_lens: MACLEN, VLAN tag */
7930 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
7931 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
7933 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd, 0);
7936 #define IXGBE_SET_FLAG(_input, _flag, _result) \
7937 ((_flag <= _result) ? \
7938 ((u32)(_input & _flag) * (_result / _flag)) : \
7939 ((u32)(_input & _flag) / (_flag / _result)))
7941 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
7943 /* set type for advanced descriptor with frame checksum insertion */
7944 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
7945 IXGBE_ADVTXD_DCMD_DEXT |
7946 IXGBE_ADVTXD_DCMD_IFCS;
7948 /* set HW vlan bit if vlan is present */
7949 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
7950 IXGBE_ADVTXD_DCMD_VLE);
7952 /* set segmentation enable bits for TSO/FSO */
7953 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
7954 IXGBE_ADVTXD_DCMD_TSE);
7956 /* set timestamp bit if present */
7957 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
7958 IXGBE_ADVTXD_MAC_TSTAMP);
7960 /* insert frame checksum */
7961 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
7966 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
7967 u32 tx_flags, unsigned int paylen)
7969 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
7971 /* enable L4 checksum for TSO and TX checksum offload */
7972 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7973 IXGBE_TX_FLAGS_CSUM,
7974 IXGBE_ADVTXD_POPTS_TXSM);
7976 /* enble IPv4 checksum for TSO */
7977 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7978 IXGBE_TX_FLAGS_IPV4,
7979 IXGBE_ADVTXD_POPTS_IXSM);
7982 * Check Context must be set if Tx switch is enabled, which it
7983 * always is for case where virtual functions are running
7985 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7989 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
7992 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
7994 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
7996 /* Herbert's original patch had:
7997 * smp_mb__after_netif_stop_queue();
7998 * but since that doesn't exist yet, just open code it.
8002 /* We need to check again in a case another CPU has just
8003 * made room available.
8005 if (likely(ixgbe_desc_unused(tx_ring) < size))
8008 /* A reprieve! - use start_queue because it doesn't call schedule */
8009 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8010 ++tx_ring->tx_stats.restart_queue;
8014 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8016 if (likely(ixgbe_desc_unused(tx_ring) >= size))
8019 return __ixgbe_maybe_stop_tx(tx_ring, size);
8022 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
8025 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8026 struct ixgbe_tx_buffer *first,
8029 struct sk_buff *skb = first->skb;
8030 struct ixgbe_tx_buffer *tx_buffer;
8031 union ixgbe_adv_tx_desc *tx_desc;
8032 struct skb_frag_struct *frag;
8034 unsigned int data_len, size;
8035 u32 tx_flags = first->tx_flags;
8036 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8037 u16 i = tx_ring->next_to_use;
8039 tx_desc = IXGBE_TX_DESC(tx_ring, i);
8041 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8043 size = skb_headlen(skb);
8044 data_len = skb->data_len;
8047 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8048 if (data_len < sizeof(struct fcoe_crc_eof)) {
8049 size -= sizeof(struct fcoe_crc_eof) - data_len;
8052 data_len -= sizeof(struct fcoe_crc_eof);
8057 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8061 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8062 if (dma_mapping_error(tx_ring->dev, dma))
8065 /* record length, and DMA address */
8066 dma_unmap_len_set(tx_buffer, len, size);
8067 dma_unmap_addr_set(tx_buffer, dma, dma);
8069 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8071 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8072 tx_desc->read.cmd_type_len =
8073 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8077 if (i == tx_ring->count) {
8078 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8081 tx_desc->read.olinfo_status = 0;
8083 dma += IXGBE_MAX_DATA_PER_TXD;
8084 size -= IXGBE_MAX_DATA_PER_TXD;
8086 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8089 if (likely(!data_len))
8092 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8096 if (i == tx_ring->count) {
8097 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8100 tx_desc->read.olinfo_status = 0;
8103 size = min_t(unsigned int, data_len, skb_frag_size(frag));
8105 size = skb_frag_size(frag);
8109 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8112 tx_buffer = &tx_ring->tx_buffer_info[i];
8115 /* write last descriptor with RS and EOP bits */
8116 cmd_type |= size | IXGBE_TXD_CMD;
8117 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8119 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8121 /* set the timestamp */
8122 first->time_stamp = jiffies;
8125 * Force memory writes to complete before letting h/w know there
8126 * are new descriptors to fetch. (Only applicable for weak-ordered
8127 * memory model archs, such as IA-64).
8129 * We also need this memory barrier to make certain all of the
8130 * status bits have been updated before next_to_watch is written.
8134 /* set next_to_watch value indicating a packet is present */
8135 first->next_to_watch = tx_desc;
8138 if (i == tx_ring->count)
8141 tx_ring->next_to_use = i;
8143 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8145 if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
8146 writel(i, tx_ring->tail);
8148 /* we need this if more than one processor can write to our tail
8149 * at a time, it synchronizes IO on IA64/Altix systems
8156 dev_err(tx_ring->dev, "TX DMA map failed\n");
8158 /* clear dma mappings for failed tx_buffer_info map */
8160 tx_buffer = &tx_ring->tx_buffer_info[i];
8161 if (dma_unmap_len(tx_buffer, len))
8162 dma_unmap_page(tx_ring->dev,
8163 dma_unmap_addr(tx_buffer, dma),
8164 dma_unmap_len(tx_buffer, len),
8166 dma_unmap_len_set(tx_buffer, len, 0);
8167 if (tx_buffer == first)
8170 i += tx_ring->count;
8174 dev_kfree_skb_any(first->skb);
8177 tx_ring->next_to_use = i;
8182 static void ixgbe_atr(struct ixgbe_ring *ring,
8183 struct ixgbe_tx_buffer *first)
8185 struct ixgbe_q_vector *q_vector = ring->q_vector;
8186 union ixgbe_atr_hash_dword input = { .dword = 0 };
8187 union ixgbe_atr_hash_dword common = { .dword = 0 };
8189 unsigned char *network;
8191 struct ipv6hdr *ipv6;
8195 struct sk_buff *skb;
8199 /* if ring doesn't have a interrupt vector, cannot perform ATR */
8203 /* do nothing if sampling is disabled */
8204 if (!ring->atr_sample_rate)
8209 /* currently only IPv4/IPv6 with TCP is supported */
8210 if ((first->protocol != htons(ETH_P_IP)) &&
8211 (first->protocol != htons(ETH_P_IPV6)))
8214 /* snag network header to get L4 type and address */
8216 hdr.network = skb_network_header(skb);
8217 if (unlikely(hdr.network <= skb->data))
8219 if (skb->encapsulation &&
8220 first->protocol == htons(ETH_P_IP) &&
8221 hdr.ipv4->protocol == IPPROTO_UDP) {
8222 struct ixgbe_adapter *adapter = q_vector->adapter;
8224 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8228 /* verify the port is recognized as VXLAN */
8229 if (adapter->vxlan_port &&
8230 udp_hdr(skb)->dest == adapter->vxlan_port)
8231 hdr.network = skb_inner_network_header(skb);
8233 if (adapter->geneve_port &&
8234 udp_hdr(skb)->dest == adapter->geneve_port)
8235 hdr.network = skb_inner_network_header(skb);
8238 /* Make sure we have at least [minimum IPv4 header + TCP]
8239 * or [IPv6 header] bytes
8241 if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8244 /* Currently only IPv4/IPv6 with TCP is supported */
8245 switch (hdr.ipv4->version) {
8247 /* access ihl as u8 to avoid unaligned access on ia64 */
8248 hlen = (hdr.network[0] & 0x0F) << 2;
8249 l4_proto = hdr.ipv4->protocol;
8252 hlen = hdr.network - skb->data;
8253 l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8254 hlen -= hdr.network - skb->data;
8260 if (l4_proto != IPPROTO_TCP)
8263 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8264 hlen + sizeof(struct tcphdr)))
8267 th = (struct tcphdr *)(hdr.network + hlen);
8269 /* skip this packet since the socket is closing */
8273 /* sample on all syn packets or once every atr sample count */
8274 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8277 /* reset sample count */
8278 ring->atr_count = 0;
8280 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8283 * src and dst are inverted, think how the receiver sees them
8285 * The input is broken into two sections, a non-compressed section
8286 * containing vm_pool, vlan_id, and flow_type. The rest of the data
8287 * is XORed together and stored in the compressed dword.
8289 input.formatted.vlan_id = vlan_id;
8292 * since src port and flex bytes occupy the same word XOR them together
8293 * and write the value to source port portion of compressed dword
8295 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8296 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8298 common.port.src ^= th->dest ^ first->protocol;
8299 common.port.dst ^= th->source;
8301 switch (hdr.ipv4->version) {
8303 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8304 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8307 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8308 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8309 hdr.ipv6->saddr.s6_addr32[1] ^
8310 hdr.ipv6->saddr.s6_addr32[2] ^
8311 hdr.ipv6->saddr.s6_addr32[3] ^
8312 hdr.ipv6->daddr.s6_addr32[0] ^
8313 hdr.ipv6->daddr.s6_addr32[1] ^
8314 hdr.ipv6->daddr.s6_addr32[2] ^
8315 hdr.ipv6->daddr.s6_addr32[3];
8321 if (hdr.network != skb_network_header(skb))
8322 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8324 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
8325 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8326 input, common, ring->queue_index);
8329 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8330 void *accel_priv, select_queue_fallback_t fallback)
8332 struct ixgbe_fwd_adapter *fwd_adapter = accel_priv;
8334 struct ixgbe_adapter *adapter;
8335 struct ixgbe_ring_feature *f;
8340 return skb->queue_mapping + fwd_adapter->tx_base_queue;
8345 * only execute the code below if protocol is FCoE
8346 * or FIP and we have FCoE enabled on the adapter
8348 switch (vlan_get_protocol(skb)) {
8349 case htons(ETH_P_FCOE):
8350 case htons(ETH_P_FIP):
8351 adapter = netdev_priv(dev);
8353 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
8357 return fallback(dev, skb);
8360 f = &adapter->ring_feature[RING_F_FCOE];
8362 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8365 while (txq >= f->indices)
8368 return txq + f->offset;
8370 return fallback(dev, skb);
8374 static int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8375 struct xdp_buff *xdp)
8377 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8378 struct ixgbe_tx_buffer *tx_buffer;
8379 union ixgbe_adv_tx_desc *tx_desc;
8384 len = xdp->data_end - xdp->data;
8386 if (unlikely(!ixgbe_desc_unused(ring)))
8387 return IXGBE_XDP_CONSUMED;
8389 dma = dma_map_single(ring->dev, xdp->data, len, DMA_TO_DEVICE);
8390 if (dma_mapping_error(ring->dev, dma))
8391 return IXGBE_XDP_CONSUMED;
8393 /* record the location of the first descriptor for this packet */
8394 tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8395 tx_buffer->bytecount = len;
8396 tx_buffer->gso_segs = 1;
8397 tx_buffer->protocol = 0;
8399 i = ring->next_to_use;
8400 tx_desc = IXGBE_TX_DESC(ring, i);
8402 dma_unmap_len_set(tx_buffer, len, len);
8403 dma_unmap_addr_set(tx_buffer, dma, dma);
8404 tx_buffer->data = xdp->data;
8405 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8407 /* put descriptor type bits */
8408 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8409 IXGBE_ADVTXD_DCMD_DEXT |
8410 IXGBE_ADVTXD_DCMD_IFCS;
8411 cmd_type |= len | IXGBE_TXD_CMD;
8412 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8413 tx_desc->read.olinfo_status =
8414 cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8416 /* Avoid any potential race with xdp_xmit and cleanup */
8419 /* set next_to_watch value indicating a packet is present */
8421 if (i == ring->count)
8424 tx_buffer->next_to_watch = tx_desc;
8425 ring->next_to_use = i;
8427 return IXGBE_XDP_TX;
8430 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8431 struct ixgbe_adapter *adapter,
8432 struct ixgbe_ring *tx_ring)
8434 struct ixgbe_tx_buffer *first;
8438 u16 count = TXD_USE_COUNT(skb_headlen(skb));
8439 __be16 protocol = skb->protocol;
8443 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8444 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8445 * + 2 desc gap to keep tail from touching head,
8446 * + 1 desc for context descriptor,
8447 * otherwise try next time
8449 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8450 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
8452 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8453 tx_ring->tx_stats.tx_busy++;
8454 return NETDEV_TX_BUSY;
8457 /* record the location of the first descriptor for this packet */
8458 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8460 first->bytecount = skb->len;
8461 first->gso_segs = 1;
8463 /* if we have a HW VLAN tag being added default to the HW one */
8464 if (skb_vlan_tag_present(skb)) {
8465 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8466 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8467 /* else if it is a SW VLAN check the next protocol and store the tag */
8468 } else if (protocol == htons(ETH_P_8021Q)) {
8469 struct vlan_hdr *vhdr, _vhdr;
8470 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8474 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8475 IXGBE_TX_FLAGS_VLAN_SHIFT;
8476 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8478 protocol = vlan_get_protocol(skb);
8480 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8481 adapter->ptp_clock) {
8482 if (!test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8484 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8485 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8487 /* schedule check for Tx timestamp */
8488 adapter->ptp_tx_skb = skb_get(skb);
8489 adapter->ptp_tx_start = jiffies;
8490 schedule_work(&adapter->ptp_tx_work);
8492 adapter->tx_hwtstamp_skipped++;
8496 skb_tx_timestamp(skb);
8498 #ifdef CONFIG_PCI_IOV
8500 * Use the l2switch_enable flag - would be false if the DMA
8501 * Tx switch had been disabled.
8503 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8504 tx_flags |= IXGBE_TX_FLAGS_CC;
8507 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8508 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8509 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8510 (skb->priority != TC_PRIO_CONTROL))) {
8511 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8512 tx_flags |= (skb->priority & 0x7) <<
8513 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8514 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8515 struct vlan_ethhdr *vhdr;
8517 if (skb_cow_head(skb, 0))
8519 vhdr = (struct vlan_ethhdr *)skb->data;
8520 vhdr->h_vlan_TCI = htons(tx_flags >>
8521 IXGBE_TX_FLAGS_VLAN_SHIFT);
8523 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8527 /* record initial flags and protocol */
8528 first->tx_flags = tx_flags;
8529 first->protocol = protocol;
8532 /* setup tx offload for FCoE */
8533 if ((protocol == htons(ETH_P_FCOE)) &&
8534 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8535 tso = ixgbe_fso(tx_ring, first, &hdr_len);
8542 #endif /* IXGBE_FCOE */
8543 tso = ixgbe_tso(tx_ring, first, &hdr_len);
8547 ixgbe_tx_csum(tx_ring, first);
8549 /* add the ATR filter if ATR is on */
8550 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8551 ixgbe_atr(tx_ring, first);
8555 #endif /* IXGBE_FCOE */
8556 if (ixgbe_tx_map(tx_ring, first, hdr_len))
8557 goto cleanup_tx_timestamp;
8559 return NETDEV_TX_OK;
8562 dev_kfree_skb_any(first->skb);
8564 cleanup_tx_timestamp:
8565 if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8566 dev_kfree_skb_any(adapter->ptp_tx_skb);
8567 adapter->ptp_tx_skb = NULL;
8568 cancel_work_sync(&adapter->ptp_tx_work);
8569 clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8572 return NETDEV_TX_OK;
8575 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8576 struct net_device *netdev,
8577 struct ixgbe_ring *ring)
8579 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8580 struct ixgbe_ring *tx_ring;
8583 * The minimum packet size for olinfo paylen is 17 so pad the skb
8584 * in order to meet this minimum size requirement.
8586 if (skb_put_padto(skb, 17))
8587 return NETDEV_TX_OK;
8589 tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping];
8591 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8594 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8595 struct net_device *netdev)
8597 return __ixgbe_xmit_frame(skb, netdev, NULL);
8601 * ixgbe_set_mac - Change the Ethernet Address of the NIC
8602 * @netdev: network interface device structure
8603 * @p: pointer to an address structure
8605 * Returns 0 on success, negative on failure
8607 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8609 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8610 struct ixgbe_hw *hw = &adapter->hw;
8611 struct sockaddr *addr = p;
8613 if (!is_valid_ether_addr(addr->sa_data))
8614 return -EADDRNOTAVAIL;
8616 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8617 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8619 ixgbe_mac_set_default_filter(adapter);
8625 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8627 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8628 struct ixgbe_hw *hw = &adapter->hw;
8632 if (prtad != hw->phy.mdio.prtad)
8634 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8640 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8641 u16 addr, u16 value)
8643 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8644 struct ixgbe_hw *hw = &adapter->hw;
8646 if (prtad != hw->phy.mdio.prtad)
8648 return hw->phy.ops.write_reg(hw, addr, devad, value);
8651 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8653 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8657 return ixgbe_ptp_set_ts_config(adapter, req);
8659 return ixgbe_ptp_get_ts_config(adapter, req);
8661 if (!adapter->hw.phy.ops.read_reg)
8665 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8670 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8672 * @netdev: network interface device structure
8674 * Returns non-zero on failure
8676 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8679 struct ixgbe_adapter *adapter = netdev_priv(dev);
8680 struct ixgbe_hw *hw = &adapter->hw;
8682 if (is_valid_ether_addr(hw->mac.san_addr)) {
8684 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8687 /* update SAN MAC vmdq pool selection */
8688 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8694 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8696 * @netdev: network interface device structure
8698 * Returns non-zero on failure
8700 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8703 struct ixgbe_adapter *adapter = netdev_priv(dev);
8704 struct ixgbe_mac_info *mac = &adapter->hw.mac;
8706 if (is_valid_ether_addr(mac->san_addr)) {
8708 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8714 #ifdef CONFIG_NET_POLL_CONTROLLER
8716 * Polling 'interrupt' - used by things like netconsole to send skbs
8717 * without having to re-enable interrupts. It's not called while
8718 * the interrupt routine is executing.
8720 static void ixgbe_netpoll(struct net_device *netdev)
8722 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8725 /* if interface is down do nothing */
8726 if (test_bit(__IXGBE_DOWN, &adapter->state))
8729 /* loop through and schedule all active queues */
8730 for (i = 0; i < adapter->num_q_vectors; i++)
8731 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
8736 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8737 struct ixgbe_ring *ring)
8744 start = u64_stats_fetch_begin_irq(&ring->syncp);
8745 packets = ring->stats.packets;
8746 bytes = ring->stats.bytes;
8747 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8748 stats->tx_packets += packets;
8749 stats->tx_bytes += bytes;
8753 static void ixgbe_get_stats64(struct net_device *netdev,
8754 struct rtnl_link_stats64 *stats)
8756 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8760 for (i = 0; i < adapter->num_rx_queues; i++) {
8761 struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8767 start = u64_stats_fetch_begin_irq(&ring->syncp);
8768 packets = ring->stats.packets;
8769 bytes = ring->stats.bytes;
8770 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8771 stats->rx_packets += packets;
8772 stats->rx_bytes += bytes;
8776 for (i = 0; i < adapter->num_tx_queues; i++) {
8777 struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8779 ixgbe_get_ring_stats64(stats, ring);
8781 for (i = 0; i < adapter->num_xdp_queues; i++) {
8782 struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8784 ixgbe_get_ring_stats64(stats, ring);
8788 /* following stats updated by ixgbe_watchdog_task() */
8789 stats->multicast = netdev->stats.multicast;
8790 stats->rx_errors = netdev->stats.rx_errors;
8791 stats->rx_length_errors = netdev->stats.rx_length_errors;
8792 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
8793 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
8796 #ifdef CONFIG_IXGBE_DCB
8798 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8799 * @adapter: pointer to ixgbe_adapter
8800 * @tc: number of traffic classes currently enabled
8802 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8803 * 802.1Q priority maps to a packet buffer that exists.
8805 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8807 struct ixgbe_hw *hw = &adapter->hw;
8811 /* 82598 have a static priority to TC mapping that can not
8812 * be changed so no validation is needed.
8814 if (hw->mac.type == ixgbe_mac_82598EB)
8817 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
8820 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
8821 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
8823 /* If up2tc is out of bounds default to zero */
8825 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
8829 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
8835 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
8836 * @adapter: Pointer to adapter struct
8838 * Populate the netdev user priority to tc map
8840 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
8842 struct net_device *dev = adapter->netdev;
8843 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
8844 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
8847 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
8850 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
8851 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
8853 tc = ets->prio_tc[prio];
8855 netdev_set_prio_tc_map(dev, prio, tc);
8859 #endif /* CONFIG_IXGBE_DCB */
8861 * ixgbe_setup_tc - configure net_device for multiple traffic classes
8863 * @netdev: net device to configure
8864 * @tc: number of traffic classes to enable
8866 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
8868 struct ixgbe_adapter *adapter = netdev_priv(dev);
8869 struct ixgbe_hw *hw = &adapter->hw;
8872 /* Hardware supports up to 8 traffic classes */
8873 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
8876 if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
8879 pools = (find_first_zero_bit(&adapter->fwd_bitmask, 32) > 1);
8880 if (tc && pools && adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS)
8883 /* Hardware has to reinitialize queues and interrupts to
8884 * match packet buffer alignment. Unfortunately, the
8885 * hardware is not flexible enough to do this dynamically.
8887 if (netif_running(dev))
8890 ixgbe_reset(adapter);
8892 ixgbe_clear_interrupt_scheme(adapter);
8894 #ifdef CONFIG_IXGBE_DCB
8896 netdev_set_num_tc(dev, tc);
8897 ixgbe_set_prio_tc_map(adapter);
8899 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
8901 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
8902 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
8903 adapter->hw.fc.requested_mode = ixgbe_fc_none;
8906 netdev_reset_tc(dev);
8908 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
8909 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
8911 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
8913 adapter->temp_dcb_cfg.pfc_mode_enable = false;
8914 adapter->dcb_cfg.pfc_mode_enable = false;
8917 ixgbe_validate_rtr(adapter, tc);
8919 #endif /* CONFIG_IXGBE_DCB */
8920 ixgbe_init_interrupt_scheme(adapter);
8922 if (netif_running(dev))
8923 return ixgbe_open(dev);
8928 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
8929 struct tc_cls_u32_offload *cls)
8931 u32 hdl = cls->knode.handle;
8932 u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
8933 u32 loc = cls->knode.handle & 0xfffff;
8935 struct ixgbe_jump_table *jump = NULL;
8937 if (loc > IXGBE_MAX_HW_ENTRIES)
8940 if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
8943 /* Clear this filter in the link data it is associated with */
8944 if (uhtid != 0x800) {
8945 jump = adapter->jump_tables[uhtid];
8948 if (!test_bit(loc - 1, jump->child_loc_map))
8950 clear_bit(loc - 1, jump->child_loc_map);
8953 /* Check if the filter being deleted is a link */
8954 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
8955 jump = adapter->jump_tables[i];
8956 if (jump && jump->link_hdl == hdl) {
8957 /* Delete filters in the hardware in the child hash
8958 * table associated with this link
8960 for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
8961 if (!test_bit(j, jump->child_loc_map))
8963 spin_lock(&adapter->fdir_perfect_lock);
8964 err = ixgbe_update_ethtool_fdir_entry(adapter,
8967 spin_unlock(&adapter->fdir_perfect_lock);
8968 clear_bit(j, jump->child_loc_map);
8970 /* Remove resources for this link */
8974 adapter->jump_tables[i] = NULL;
8979 spin_lock(&adapter->fdir_perfect_lock);
8980 err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
8981 spin_unlock(&adapter->fdir_perfect_lock);
8985 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
8986 struct tc_cls_u32_offload *cls)
8988 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
8990 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
8993 /* This ixgbe devices do not support hash tables at the moment
8994 * so abort when given hash tables.
8996 if (cls->hnode.divisor > 0)
8999 set_bit(uhtid - 1, &adapter->tables);
9003 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9004 struct tc_cls_u32_offload *cls)
9006 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9008 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9011 clear_bit(uhtid - 1, &adapter->tables);
9015 #ifdef CONFIG_NET_CLS_ACT
9016 struct upper_walk_data {
9017 struct ixgbe_adapter *adapter;
9023 static int get_macvlan_queue(struct net_device *upper, void *_data)
9025 if (netif_is_macvlan(upper)) {
9026 struct macvlan_dev *dfwd = netdev_priv(upper);
9027 struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv;
9028 struct upper_walk_data *data = _data;
9029 struct ixgbe_adapter *adapter = data->adapter;
9030 int ifindex = data->ifindex;
9032 if (vadapter && vadapter->netdev->ifindex == ifindex) {
9033 data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9034 data->action = data->queue;
9042 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9043 u8 *queue, u64 *action)
9045 unsigned int num_vfs = adapter->num_vfs, vf;
9046 struct upper_walk_data data;
9047 struct net_device *upper;
9049 /* redirect to a SRIOV VF */
9050 for (vf = 0; vf < num_vfs; ++vf) {
9051 upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9052 if (upper->ifindex == ifindex) {
9053 if (adapter->num_rx_pools > 1)
9056 *queue = vf * adapter->num_rx_queues_per_pool;
9059 *action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9064 /* redirect to a offloaded macvlan netdev */
9065 data.adapter = adapter;
9066 data.ifindex = ifindex;
9069 if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9070 get_macvlan_queue, &data)) {
9071 *action = data.action;
9072 *queue = data.queue;
9080 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9081 struct tcf_exts *exts, u64 *action, u8 *queue)
9083 const struct tc_action *a;
9087 if (!tcf_exts_has_actions(exts))
9090 tcf_exts_to_list(exts, &actions);
9091 list_for_each_entry(a, &actions, list) {
9094 if (is_tcf_gact_shot(a)) {
9095 *action = IXGBE_FDIR_DROP_QUEUE;
9096 *queue = IXGBE_FDIR_DROP_QUEUE;
9100 /* Redirect to a VF or a offloaded macvlan */
9101 if (is_tcf_mirred_egress_redirect(a)) {
9102 struct net_device *dev = tcf_mirred_dev(a);
9106 err = handle_redirect_action(adapter, dev->ifindex, queue,
9116 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9117 struct tcf_exts *exts, u64 *action, u8 *queue)
9121 #endif /* CONFIG_NET_CLS_ACT */
9123 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9124 union ixgbe_atr_input *mask,
9125 struct tc_cls_u32_offload *cls,
9126 struct ixgbe_mat_field *field_ptr,
9127 struct ixgbe_nexthdr *nexthdr)
9131 bool found_entry = false, found_jump_field = false;
9133 for (i = 0; i < cls->knode.sel->nkeys; i++) {
9134 off = cls->knode.sel->keys[i].off;
9135 val = cls->knode.sel->keys[i].val;
9136 m = cls->knode.sel->keys[i].mask;
9138 for (j = 0; field_ptr[j].val; j++) {
9139 if (field_ptr[j].off == off) {
9140 field_ptr[j].val(input, mask, val, m);
9141 input->filter.formatted.flow_type |=
9148 if (nexthdr->off == cls->knode.sel->keys[i].off &&
9149 nexthdr->val == cls->knode.sel->keys[i].val &&
9150 nexthdr->mask == cls->knode.sel->keys[i].mask)
9151 found_jump_field = true;
9157 if (nexthdr && !found_jump_field)
9163 mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9164 IXGBE_ATR_L4TYPE_MASK;
9166 if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9167 mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9172 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9173 struct tc_cls_u32_offload *cls)
9175 __be16 protocol = cls->common.protocol;
9176 u32 loc = cls->knode.handle & 0xfffff;
9177 struct ixgbe_hw *hw = &adapter->hw;
9178 struct ixgbe_mat_field *field_ptr;
9179 struct ixgbe_fdir_filter *input = NULL;
9180 union ixgbe_atr_input *mask = NULL;
9181 struct ixgbe_jump_table *jump = NULL;
9182 int i, err = -EINVAL;
9184 u32 uhtid, link_uhtid;
9186 uhtid = TC_U32_USERHTID(cls->knode.handle);
9187 link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9189 /* At the moment cls_u32 jumps to network layer and skips past
9190 * L2 headers. The canonical method to match L2 frames is to use
9191 * negative values. However this is error prone at best but really
9192 * just broken because there is no way to "know" what sort of hdr
9193 * is in front of the network layer. Fix cls_u32 to support L2
9194 * headers when needed.
9196 if (protocol != htons(ETH_P_IP))
9199 if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9200 e_err(drv, "Location out of range\n");
9204 /* cls u32 is a graph starting at root node 0x800. The driver tracks
9205 * links and also the fields used to advance the parser across each
9206 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9207 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9208 * To add support for new nodes update ixgbe_model.h parse structures
9209 * this function _should_ be generic try not to hardcode values here.
9211 if (uhtid == 0x800) {
9212 field_ptr = (adapter->jump_tables[0])->mat;
9214 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9216 if (!adapter->jump_tables[uhtid])
9218 field_ptr = (adapter->jump_tables[uhtid])->mat;
9224 /* At this point we know the field_ptr is valid and need to either
9225 * build cls_u32 link or attach filter. Because adding a link to
9226 * a handle that does not exist is invalid and the same for adding
9227 * rules to handles that don't exist.
9231 struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9233 if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9236 if (!test_bit(link_uhtid - 1, &adapter->tables))
9239 /* Multiple filters as links to the same hash table are not
9240 * supported. To add a new filter with the same next header
9241 * but different match/jump conditions, create a new hash table
9244 if (adapter->jump_tables[link_uhtid] &&
9245 (adapter->jump_tables[link_uhtid])->link_hdl) {
9246 e_err(drv, "Link filter exists for link: %x\n",
9251 for (i = 0; nexthdr[i].jump; i++) {
9252 if (nexthdr[i].o != cls->knode.sel->offoff ||
9253 nexthdr[i].s != cls->knode.sel->offshift ||
9254 nexthdr[i].m != cls->knode.sel->offmask)
9257 jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9260 input = kzalloc(sizeof(*input), GFP_KERNEL);
9265 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9270 jump->input = input;
9272 jump->link_hdl = cls->knode.handle;
9274 err = ixgbe_clsu32_build_input(input, mask, cls,
9275 field_ptr, &nexthdr[i]);
9277 jump->mat = nexthdr[i].jump;
9278 adapter->jump_tables[link_uhtid] = jump;
9285 input = kzalloc(sizeof(*input), GFP_KERNEL);
9288 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9294 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9295 if ((adapter->jump_tables[uhtid])->input)
9296 memcpy(input, (adapter->jump_tables[uhtid])->input,
9298 if ((adapter->jump_tables[uhtid])->mask)
9299 memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9302 /* Lookup in all child hash tables if this location is already
9303 * filled with a filter
9305 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9306 struct ixgbe_jump_table *link = adapter->jump_tables[i];
9308 if (link && (test_bit(loc - 1, link->child_loc_map))) {
9309 e_err(drv, "Filter exists in location: %x\n",
9316 err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9320 err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9325 input->sw_idx = loc;
9327 spin_lock(&adapter->fdir_perfect_lock);
9329 if (hlist_empty(&adapter->fdir_filter_list)) {
9330 memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9331 err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9333 goto err_out_w_lock;
9334 } else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9336 goto err_out_w_lock;
9339 ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9340 err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9341 input->sw_idx, queue);
9343 ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9344 spin_unlock(&adapter->fdir_perfect_lock);
9346 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9347 set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9352 spin_unlock(&adapter->fdir_perfect_lock);
9362 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9363 struct tc_cls_u32_offload *cls_u32)
9365 if (cls_u32->common.chain_index)
9368 switch (cls_u32->command) {
9369 case TC_CLSU32_NEW_KNODE:
9370 case TC_CLSU32_REPLACE_KNODE:
9371 return ixgbe_configure_clsu32(adapter, cls_u32);
9372 case TC_CLSU32_DELETE_KNODE:
9373 return ixgbe_delete_clsu32(adapter, cls_u32);
9374 case TC_CLSU32_NEW_HNODE:
9375 case TC_CLSU32_REPLACE_HNODE:
9376 return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9377 case TC_CLSU32_DELETE_HNODE:
9378 return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9384 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9387 struct ixgbe_adapter *adapter = cb_priv;
9389 if (!tc_can_offload(adapter->netdev))
9393 case TC_SETUP_CLSU32:
9394 return ixgbe_setup_tc_cls_u32(adapter, type_data);
9400 static int ixgbe_setup_tc_block(struct net_device *dev,
9401 struct tc_block_offload *f)
9403 struct ixgbe_adapter *adapter = netdev_priv(dev);
9405 if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
9408 switch (f->command) {
9410 return tcf_block_cb_register(f->block, ixgbe_setup_tc_block_cb,
9412 case TC_BLOCK_UNBIND:
9413 tcf_block_cb_unregister(f->block, ixgbe_setup_tc_block_cb,
9421 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9422 struct tc_mqprio_qopt *mqprio)
9424 mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9425 return ixgbe_setup_tc(dev, mqprio->num_tc);
9428 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9432 case TC_SETUP_BLOCK:
9433 return ixgbe_setup_tc_block(dev, type_data);
9434 case TC_SETUP_QDISC_MQPRIO:
9435 return ixgbe_setup_tc_mqprio(dev, type_data);
9441 #ifdef CONFIG_PCI_IOV
9442 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9444 struct net_device *netdev = adapter->netdev;
9447 ixgbe_setup_tc(netdev, netdev_get_num_tc(netdev));
9452 void ixgbe_do_reset(struct net_device *netdev)
9454 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9456 if (netif_running(netdev))
9457 ixgbe_reinit_locked(adapter);
9459 ixgbe_reset(adapter);
9462 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9463 netdev_features_t features)
9465 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9467 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9468 if (!(features & NETIF_F_RXCSUM))
9469 features &= ~NETIF_F_LRO;
9471 /* Turn off LRO if not RSC capable */
9472 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9473 features &= ~NETIF_F_LRO;
9478 static int ixgbe_set_features(struct net_device *netdev,
9479 netdev_features_t features)
9481 struct ixgbe_adapter *adapter = netdev_priv(netdev);
9482 netdev_features_t changed = netdev->features ^ features;
9483 bool need_reset = false;
9485 /* Make sure RSC matches LRO, reset if change */
9486 if (!(features & NETIF_F_LRO)) {
9487 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9489 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9490 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9491 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9492 if (adapter->rx_itr_setting == 1 ||
9493 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9494 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9496 } else if ((changed ^ features) & NETIF_F_LRO) {
9497 e_info(probe, "rx-usecs set too low, "
9503 * Check if Flow Director n-tuple support or hw_tc support was
9504 * enabled or disabled. If the state changed, we need to reset.
9506 if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9507 /* turn off ATR, enable perfect filters and reset */
9508 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9511 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9512 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9514 /* turn off perfect filters, enable ATR and reset */
9515 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9518 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9520 /* We cannot enable ATR if SR-IOV is enabled */
9521 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9522 /* We cannot enable ATR if we have 2 or more tcs */
9523 (netdev_get_num_tc(netdev) > 1) ||
9524 /* We cannot enable ATR if RSS is disabled */
9525 (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9526 /* A sample rate of 0 indicates ATR disabled */
9527 (!adapter->atr_sample_rate))
9528 ; /* do nothing not supported */
9529 else /* otherwise supported and set the flag */
9530 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9533 if (changed & NETIF_F_RXALL)
9536 netdev->features = features;
9538 if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) {
9539 if (features & NETIF_F_RXCSUM) {
9540 adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9542 u32 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9544 ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9548 if ((adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)) {
9549 if (features & NETIF_F_RXCSUM) {
9550 adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9552 u32 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9554 ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9559 ixgbe_do_reset(netdev);
9560 else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9561 NETIF_F_HW_VLAN_CTAG_FILTER))
9562 ixgbe_set_rx_mode(netdev);
9568 * ixgbe_add_udp_tunnel_port - Get notifications about adding UDP tunnel ports
9569 * @dev: The port's netdev
9570 * @ti: Tunnel endpoint information
9572 static void ixgbe_add_udp_tunnel_port(struct net_device *dev,
9573 struct udp_tunnel_info *ti)
9575 struct ixgbe_adapter *adapter = netdev_priv(dev);
9576 struct ixgbe_hw *hw = &adapter->hw;
9577 __be16 port = ti->port;
9581 if (ti->sa_family != AF_INET)
9585 case UDP_TUNNEL_TYPE_VXLAN:
9586 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9589 if (adapter->vxlan_port == port)
9592 if (adapter->vxlan_port) {
9594 "VXLAN port %d set, not adding port %d\n",
9595 ntohs(adapter->vxlan_port),
9600 adapter->vxlan_port = port;
9602 case UDP_TUNNEL_TYPE_GENEVE:
9603 if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9606 if (adapter->geneve_port == port)
9609 if (adapter->geneve_port) {
9611 "GENEVE port %d set, not adding port %d\n",
9612 ntohs(adapter->geneve_port),
9617 port_shift = IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT;
9618 adapter->geneve_port = port;
9624 reg = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) | ntohs(port) << port_shift;
9625 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, reg);
9629 * ixgbe_del_udp_tunnel_port - Get notifications about removing UDP tunnel ports
9630 * @dev: The port's netdev
9631 * @ti: Tunnel endpoint information
9633 static void ixgbe_del_udp_tunnel_port(struct net_device *dev,
9634 struct udp_tunnel_info *ti)
9636 struct ixgbe_adapter *adapter = netdev_priv(dev);
9639 if (ti->type != UDP_TUNNEL_TYPE_VXLAN &&
9640 ti->type != UDP_TUNNEL_TYPE_GENEVE)
9643 if (ti->sa_family != AF_INET)
9647 case UDP_TUNNEL_TYPE_VXLAN:
9648 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9651 if (adapter->vxlan_port != ti->port) {
9652 netdev_info(dev, "VXLAN port %d not found\n",
9657 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9659 case UDP_TUNNEL_TYPE_GENEVE:
9660 if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9663 if (adapter->geneve_port != ti->port) {
9664 netdev_info(dev, "GENEVE port %d not found\n",
9669 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9675 ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9676 adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9679 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9680 struct net_device *dev,
9681 const unsigned char *addr, u16 vid,
9684 /* guarantee we can provide a unique filter for the unicast address */
9685 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9686 struct ixgbe_adapter *adapter = netdev_priv(dev);
9687 u16 pool = VMDQ_P(0);
9689 if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9693 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9697 * ixgbe_configure_bridge_mode - set various bridge modes
9698 * @adapter - the private structure
9699 * @mode - requested bridge mode
9701 * Configure some settings require for various bridge modes.
9703 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9706 struct ixgbe_hw *hw = &adapter->hw;
9707 unsigned int p, num_pools;
9711 case BRIDGE_MODE_VEPA:
9712 /* disable Tx loopback, rely on switch hairpin mode */
9713 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9715 /* must enable Rx switching replication to allow multicast
9716 * packet reception on all VFs, and to enable source address
9719 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9720 vmdctl |= IXGBE_VT_CTL_REPLEN;
9721 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9723 /* enable Rx source address pruning. Note, this requires
9724 * replication to be enabled or else it does nothing.
9726 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9727 for (p = 0; p < num_pools; p++) {
9728 if (hw->mac.ops.set_source_address_pruning)
9729 hw->mac.ops.set_source_address_pruning(hw,
9734 case BRIDGE_MODE_VEB:
9735 /* enable Tx loopback for internal VF/PF communication */
9736 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9737 IXGBE_PFDTXGSWC_VT_LBEN);
9739 /* disable Rx switching replication unless we have SR-IOV
9742 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9743 if (!adapter->num_vfs)
9744 vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9745 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9747 /* disable Rx source address pruning, since we don't expect to
9748 * be receiving external loopback of our transmitted frames.
9750 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9751 for (p = 0; p < num_pools; p++) {
9752 if (hw->mac.ops.set_source_address_pruning)
9753 hw->mac.ops.set_source_address_pruning(hw,
9762 adapter->bridge_mode = mode;
9764 e_info(drv, "enabling bridge mode: %s\n",
9765 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
9770 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
9771 struct nlmsghdr *nlh, u16 flags)
9773 struct ixgbe_adapter *adapter = netdev_priv(dev);
9774 struct nlattr *attr, *br_spec;
9777 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9780 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
9784 nla_for_each_nested(attr, br_spec, rem) {
9788 if (nla_type(attr) != IFLA_BRIDGE_MODE)
9791 if (nla_len(attr) < sizeof(mode))
9794 mode = nla_get_u16(attr);
9795 status = ixgbe_configure_bridge_mode(adapter, mode);
9805 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
9806 struct net_device *dev,
9807 u32 filter_mask, int nlflags)
9809 struct ixgbe_adapter *adapter = netdev_priv(dev);
9811 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9814 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
9815 adapter->bridge_mode, 0, 0, nlflags,
9819 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
9821 struct ixgbe_fwd_adapter *fwd_adapter = NULL;
9822 struct ixgbe_adapter *adapter = netdev_priv(pdev);
9823 int used_pools = adapter->num_vfs + adapter->num_rx_pools;
9827 /* Hardware has a limited number of available pools. Each VF, and the
9828 * PF require a pool. Check to ensure we don't attempt to use more
9829 * then the available number of pools.
9831 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
9832 return ERR_PTR(-EINVAL);
9835 if (vdev->num_rx_queues != vdev->num_tx_queues) {
9836 netdev_info(pdev, "%s: Only supports a single queue count for TX and RX\n",
9838 return ERR_PTR(-EINVAL);
9841 /* Check for hardware restriction on number of rx/tx queues */
9842 if (vdev->num_tx_queues > IXGBE_MAX_L2A_QUEUES ||
9843 vdev->num_tx_queues == IXGBE_BAD_L2A_QUEUE) {
9845 "%s: Supports RX/TX Queue counts 1,2, and 4\n",
9847 return ERR_PTR(-EINVAL);
9850 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
9851 adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS - 1) ||
9852 (adapter->num_rx_pools > IXGBE_MAX_MACVLANS))
9853 return ERR_PTR(-EBUSY);
9855 fwd_adapter = kzalloc(sizeof(*fwd_adapter), GFP_KERNEL);
9857 return ERR_PTR(-ENOMEM);
9859 pool = find_first_zero_bit(&adapter->fwd_bitmask, 32);
9860 adapter->num_rx_pools++;
9861 set_bit(pool, &adapter->fwd_bitmask);
9862 limit = find_last_bit(&adapter->fwd_bitmask, 32);
9864 /* Enable VMDq flag so device will be set in VM mode */
9865 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED | IXGBE_FLAG_SRIOV_ENABLED;
9866 adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
9867 adapter->ring_feature[RING_F_RSS].limit = vdev->num_tx_queues;
9869 /* Force reinit of ring allocation with VMDQ enabled */
9870 err = ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
9873 fwd_adapter->pool = pool;
9874 fwd_adapter->real_adapter = adapter;
9876 if (netif_running(pdev)) {
9877 err = ixgbe_fwd_ring_up(vdev, fwd_adapter);
9880 netif_tx_start_all_queues(vdev);
9885 /* unwind counter and free adapter struct */
9887 "%s: dfwd hardware acceleration failed\n", vdev->name);
9888 clear_bit(pool, &adapter->fwd_bitmask);
9889 adapter->num_rx_pools--;
9891 return ERR_PTR(err);
9894 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
9896 struct ixgbe_fwd_adapter *fwd_adapter = priv;
9897 struct ixgbe_adapter *adapter = fwd_adapter->real_adapter;
9900 clear_bit(fwd_adapter->pool, &adapter->fwd_bitmask);
9901 adapter->num_rx_pools--;
9903 limit = find_last_bit(&adapter->fwd_bitmask, 32);
9904 adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
9905 ixgbe_fwd_ring_down(fwd_adapter->netdev, fwd_adapter);
9907 /* go back to full RSS if we're done with our VMQs */
9908 if (adapter->ring_feature[RING_F_VMDQ].limit == 1) {
9909 int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9912 adapter->flags &= ~IXGBE_FLAG_VMDQ_ENABLED;
9913 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
9914 adapter->ring_feature[RING_F_RSS].limit = rss;
9917 ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
9918 netdev_dbg(pdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
9919 fwd_adapter->pool, adapter->num_rx_pools,
9920 fwd_adapter->rx_base_queue,
9921 fwd_adapter->rx_base_queue + adapter->num_rx_queues_per_pool,
9922 adapter->fwd_bitmask);
9926 #define IXGBE_MAX_MAC_HDR_LEN 127
9927 #define IXGBE_MAX_NETWORK_HDR_LEN 511
9929 static netdev_features_t
9930 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
9931 netdev_features_t features)
9933 unsigned int network_hdr_len, mac_hdr_len;
9935 /* Make certain the headers can be described by a context descriptor */
9936 mac_hdr_len = skb_network_header(skb) - skb->data;
9937 if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
9938 return features & ~(NETIF_F_HW_CSUM |
9940 NETIF_F_HW_VLAN_CTAG_TX |
9944 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
9945 if (unlikely(network_hdr_len > IXGBE_MAX_NETWORK_HDR_LEN))
9946 return features & ~(NETIF_F_HW_CSUM |
9951 /* We can only support IPV4 TSO in tunnels if we can mangle the
9952 * inner IP ID field, so strip TSO if MANGLEID is not supported.
9954 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
9955 features &= ~NETIF_F_TSO;
9960 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
9962 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
9963 struct ixgbe_adapter *adapter = netdev_priv(dev);
9964 struct bpf_prog *old_prog;
9966 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
9969 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
9972 /* verify ixgbe ring attributes are sufficient for XDP */
9973 for (i = 0; i < adapter->num_rx_queues; i++) {
9974 struct ixgbe_ring *ring = adapter->rx_ring[i];
9976 if (ring_is_rsc_enabled(ring))
9979 if (frame_size > ixgbe_rx_bufsz(ring))
9983 if (nr_cpu_ids > MAX_XDP_QUEUES)
9986 old_prog = xchg(&adapter->xdp_prog, prog);
9988 /* If transitioning XDP modes reconfigure rings */
9989 if (!!prog != !!old_prog) {
9990 int err = ixgbe_setup_tc(dev, netdev_get_num_tc(dev));
9993 rcu_assign_pointer(adapter->xdp_prog, old_prog);
9997 for (i = 0; i < adapter->num_rx_queues; i++)
9998 xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog);
10002 bpf_prog_put(old_prog);
10007 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10009 struct ixgbe_adapter *adapter = netdev_priv(dev);
10011 switch (xdp->command) {
10012 case XDP_SETUP_PROG:
10013 return ixgbe_xdp_setup(dev, xdp->prog);
10014 case XDP_QUERY_PROG:
10015 xdp->prog_attached = !!(adapter->xdp_prog);
10016 xdp->prog_id = adapter->xdp_prog ?
10017 adapter->xdp_prog->aux->id : 0;
10024 static int ixgbe_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
10026 struct ixgbe_adapter *adapter = netdev_priv(dev);
10027 struct ixgbe_ring *ring;
10030 if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10033 /* During program transitions its possible adapter->xdp_prog is assigned
10034 * but ring has not been configured yet. In this case simply abort xmit.
10036 ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10037 if (unlikely(!ring))
10040 err = ixgbe_xmit_xdp_ring(adapter, xdp);
10041 if (err != IXGBE_XDP_TX)
10047 static void ixgbe_xdp_flush(struct net_device *dev)
10049 struct ixgbe_adapter *adapter = netdev_priv(dev);
10050 struct ixgbe_ring *ring;
10052 /* Its possible the device went down between xdp xmit and flush so
10053 * we need to ensure device is still up.
10055 if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10058 ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10059 if (unlikely(!ring))
10062 /* Force memory writes to complete before letting h/w know there
10063 * are new descriptors to fetch.
10066 writel(ring->next_to_use, ring->tail);
10071 static const struct net_device_ops ixgbe_netdev_ops = {
10072 .ndo_open = ixgbe_open,
10073 .ndo_stop = ixgbe_close,
10074 .ndo_start_xmit = ixgbe_xmit_frame,
10075 .ndo_select_queue = ixgbe_select_queue,
10076 .ndo_set_rx_mode = ixgbe_set_rx_mode,
10077 .ndo_validate_addr = eth_validate_addr,
10078 .ndo_set_mac_address = ixgbe_set_mac,
10079 .ndo_change_mtu = ixgbe_change_mtu,
10080 .ndo_tx_timeout = ixgbe_tx_timeout,
10081 .ndo_set_tx_maxrate = ixgbe_tx_maxrate,
10082 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
10083 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
10084 .ndo_do_ioctl = ixgbe_ioctl,
10085 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
10086 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
10087 .ndo_set_vf_rate = ixgbe_ndo_set_vf_bw,
10088 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
10089 .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10090 .ndo_set_vf_trust = ixgbe_ndo_set_vf_trust,
10091 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
10092 .ndo_get_stats64 = ixgbe_get_stats64,
10093 .ndo_setup_tc = __ixgbe_setup_tc,
10094 #ifdef CONFIG_NET_POLL_CONTROLLER
10095 .ndo_poll_controller = ixgbe_netpoll,
10098 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10099 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10100 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10101 .ndo_fcoe_enable = ixgbe_fcoe_enable,
10102 .ndo_fcoe_disable = ixgbe_fcoe_disable,
10103 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10104 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10105 #endif /* IXGBE_FCOE */
10106 .ndo_set_features = ixgbe_set_features,
10107 .ndo_fix_features = ixgbe_fix_features,
10108 .ndo_fdb_add = ixgbe_ndo_fdb_add,
10109 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
10110 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
10111 .ndo_dfwd_add_station = ixgbe_fwd_add,
10112 .ndo_dfwd_del_station = ixgbe_fwd_del,
10113 .ndo_udp_tunnel_add = ixgbe_add_udp_tunnel_port,
10114 .ndo_udp_tunnel_del = ixgbe_del_udp_tunnel_port,
10115 .ndo_features_check = ixgbe_features_check,
10116 .ndo_bpf = ixgbe_xdp,
10117 .ndo_xdp_xmit = ixgbe_xdp_xmit,
10118 .ndo_xdp_flush = ixgbe_xdp_flush,
10122 * ixgbe_enumerate_functions - Get the number of ports this device has
10123 * @adapter: adapter structure
10125 * This function enumerates the phsyical functions co-located on a single slot,
10126 * in order to determine how many ports a device has. This is most useful in
10127 * determining the required GT/s of PCIe bandwidth necessary for optimal
10130 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10132 struct pci_dev *entry, *pdev = adapter->pdev;
10135 /* Some cards can not use the generic count PCIe functions method,
10136 * because they are behind a parent switch, so we hardcode these with
10137 * the correct number of functions.
10139 if (ixgbe_pcie_from_parent(&adapter->hw))
10142 list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10143 /* don't count virtual functions */
10144 if (entry->is_virtfn)
10147 /* When the devices on the bus don't all match our device ID,
10148 * we can't reliably determine the correct number of
10149 * functions. This can occur if a function has been direct
10150 * attached to a virtual machine using VT-d, for example. In
10151 * this case, simply return -1 to indicate this.
10153 if ((entry->vendor != pdev->vendor) ||
10154 (entry->device != pdev->device))
10164 * ixgbe_wol_supported - Check whether device supports WoL
10165 * @adapter: the adapter private structure
10166 * @device_id: the device ID
10167 * @subdev_id: the subsystem device ID
10169 * This function is used by probe and ethtool to determine
10170 * which devices have WoL support
10173 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10176 struct ixgbe_hw *hw = &adapter->hw;
10177 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10179 /* WOL not supported on 82598 */
10180 if (hw->mac.type == ixgbe_mac_82598EB)
10183 /* check eeprom to see if WOL is enabled for X540 and newer */
10184 if (hw->mac.type >= ixgbe_mac_X540) {
10185 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10186 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10187 (hw->bus.func == 0)))
10191 /* WOL is determined based on device IDs for 82599 MACs */
10192 switch (device_id) {
10193 case IXGBE_DEV_ID_82599_SFP:
10194 /* Only these subdevices could supports WOL */
10195 switch (subdevice_id) {
10196 case IXGBE_SUBDEV_ID_82599_560FLR:
10197 case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10198 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10199 case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10200 /* only support first port */
10201 if (hw->bus.func != 0)
10204 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10205 case IXGBE_SUBDEV_ID_82599_SFP:
10206 case IXGBE_SUBDEV_ID_82599_RNDC:
10207 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10208 case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10209 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10210 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10214 case IXGBE_DEV_ID_82599EN_SFP:
10215 /* Only these subdevices support WOL */
10216 switch (subdevice_id) {
10217 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10221 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10222 /* All except this subdevice support WOL */
10223 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10226 case IXGBE_DEV_ID_82599_KX4:
10236 * ixgbe_set_fw_version - Set FW version
10237 * @adapter: the adapter private structure
10239 * This function is used by probe and ethtool to determine the FW version to
10240 * format to display. The FW version is taken from the EEPROM/NVM.
10242 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10244 struct ixgbe_hw *hw = &adapter->hw;
10245 struct ixgbe_nvm_version nvm_ver;
10247 ixgbe_get_oem_prod_version(hw, &nvm_ver);
10248 if (nvm_ver.oem_valid) {
10249 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10250 "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10251 nvm_ver.oem_release);
10255 ixgbe_get_etk_id(hw, &nvm_ver);
10256 ixgbe_get_orom_version(hw, &nvm_ver);
10258 if (nvm_ver.or_valid) {
10259 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10260 "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10261 nvm_ver.or_build, nvm_ver.or_patch);
10265 /* Set ETrack ID format */
10266 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10267 "0x%08x", nvm_ver.etk_id);
10271 * ixgbe_probe - Device Initialization Routine
10272 * @pdev: PCI device information struct
10273 * @ent: entry in ixgbe_pci_tbl
10275 * Returns 0 on success, negative on failure
10277 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10278 * The OS initialization, configuring of the adapter private structure,
10279 * and a hardware reset occur.
10281 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10283 struct net_device *netdev;
10284 struct ixgbe_adapter *adapter = NULL;
10285 struct ixgbe_hw *hw;
10286 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10287 int i, err, pci_using_dac, expected_gts;
10288 unsigned int indices = MAX_TX_QUEUES;
10289 u8 part_str[IXGBE_PBANUM_LENGTH];
10290 bool disable_dev = false;
10296 /* Catch broken hardware that put the wrong VF device ID in
10297 * the PCIe SR-IOV capability.
10299 if (pdev->is_virtfn) {
10300 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10301 pci_name(pdev), pdev->vendor, pdev->device);
10305 err = pci_enable_device_mem(pdev);
10309 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10312 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10314 dev_err(&pdev->dev,
10315 "No usable DMA configuration, aborting\n");
10321 err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10323 dev_err(&pdev->dev,
10324 "pci_request_selected_regions failed 0x%x\n", err);
10328 pci_enable_pcie_error_reporting(pdev);
10330 pci_set_master(pdev);
10331 pci_save_state(pdev);
10333 if (ii->mac == ixgbe_mac_82598EB) {
10334 #ifdef CONFIG_IXGBE_DCB
10335 /* 8 TC w/ 4 queues per TC */
10336 indices = 4 * MAX_TRAFFIC_CLASS;
10338 indices = IXGBE_MAX_RSS_INDICES;
10342 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10345 goto err_alloc_etherdev;
10348 SET_NETDEV_DEV(netdev, &pdev->dev);
10350 adapter = netdev_priv(netdev);
10352 adapter->netdev = netdev;
10353 adapter->pdev = pdev;
10355 hw->back = adapter;
10356 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10358 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10359 pci_resource_len(pdev, 0));
10360 adapter->io_addr = hw->hw_addr;
10361 if (!hw->hw_addr) {
10366 netdev->netdev_ops = &ixgbe_netdev_ops;
10367 ixgbe_set_ethtool_ops(netdev);
10368 netdev->watchdog_timeo = 5 * HZ;
10369 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10372 hw->mac.ops = *ii->mac_ops;
10373 hw->mac.type = ii->mac;
10374 hw->mvals = ii->mvals;
10376 hw->link.ops = *ii->link_ops;
10379 hw->eeprom.ops = *ii->eeprom_ops;
10380 eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10381 if (ixgbe_removed(hw->hw_addr)) {
10385 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10386 if (!(eec & BIT(8)))
10387 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10390 hw->phy.ops = *ii->phy_ops;
10391 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10392 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
10393 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10394 hw->phy.mdio.mmds = 0;
10395 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10396 hw->phy.mdio.dev = netdev;
10397 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10398 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10400 /* setup the private structure */
10401 err = ixgbe_sw_init(adapter, ii);
10405 /* Make sure the SWFW semaphore is in a valid state */
10406 if (hw->mac.ops.init_swfw_sync)
10407 hw->mac.ops.init_swfw_sync(hw);
10409 /* Make it possible the adapter to be woken up via WOL */
10410 switch (adapter->hw.mac.type) {
10411 case ixgbe_mac_82599EB:
10412 case ixgbe_mac_X540:
10413 case ixgbe_mac_X550:
10414 case ixgbe_mac_X550EM_x:
10415 case ixgbe_mac_x550em_a:
10416 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10423 * If there is a fan on this device and it has failed log the
10426 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10427 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10428 if (esdp & IXGBE_ESDP_SDP1)
10429 e_crit(probe, "Fan has stopped, replace the adapter\n");
10432 if (allow_unsupported_sfp)
10433 hw->allow_unsupported_sfp = allow_unsupported_sfp;
10435 /* reset_hw fills in the perm_addr as well */
10436 hw->phy.reset_if_overtemp = true;
10437 err = hw->mac.ops.reset_hw(hw);
10438 hw->phy.reset_if_overtemp = false;
10439 ixgbe_set_eee_capable(adapter);
10440 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10442 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10443 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10444 e_dev_err("Reload the driver after installing a supported module.\n");
10447 e_dev_err("HW Init failed: %d\n", err);
10451 #ifdef CONFIG_PCI_IOV
10452 /* SR-IOV not supported on the 82598 */
10453 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10456 ixgbe_init_mbx_params_pf(hw);
10457 hw->mbx.ops = ii->mbx_ops;
10458 pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10459 ixgbe_enable_sriov(adapter, max_vfs);
10463 netdev->features = NETIF_F_SG |
10470 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10471 NETIF_F_GSO_GRE_CSUM | \
10472 NETIF_F_GSO_IPXIP4 | \
10473 NETIF_F_GSO_IPXIP6 | \
10474 NETIF_F_GSO_UDP_TUNNEL | \
10475 NETIF_F_GSO_UDP_TUNNEL_CSUM)
10477 netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10478 netdev->features |= NETIF_F_GSO_PARTIAL |
10479 IXGBE_GSO_PARTIAL_FEATURES;
10481 if (hw->mac.type >= ixgbe_mac_82599EB)
10482 netdev->features |= NETIF_F_SCTP_CRC;
10484 /* copy netdev features into list of user selectable features */
10485 netdev->hw_features |= netdev->features |
10486 NETIF_F_HW_VLAN_CTAG_FILTER |
10487 NETIF_F_HW_VLAN_CTAG_RX |
10488 NETIF_F_HW_VLAN_CTAG_TX |
10490 NETIF_F_HW_L2FW_DOFFLOAD;
10492 if (hw->mac.type >= ixgbe_mac_82599EB)
10493 netdev->hw_features |= NETIF_F_NTUPLE |
10497 netdev->features |= NETIF_F_HIGHDMA;
10499 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10500 netdev->hw_enc_features |= netdev->vlan_features;
10501 netdev->mpls_features |= NETIF_F_SG |
10505 netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10507 /* set this bit last since it cannot be part of vlan_features */
10508 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10509 NETIF_F_HW_VLAN_CTAG_RX |
10510 NETIF_F_HW_VLAN_CTAG_TX;
10512 netdev->priv_flags |= IFF_UNICAST_FLT;
10513 netdev->priv_flags |= IFF_SUPP_NOFCS;
10515 /* MTU range: 68 - 9710 */
10516 netdev->min_mtu = ETH_MIN_MTU;
10517 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
10519 #ifdef CONFIG_IXGBE_DCB
10520 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
10521 netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
10525 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
10526 unsigned int fcoe_l;
10528 if (hw->mac.ops.get_device_caps) {
10529 hw->mac.ops.get_device_caps(hw, &device_caps);
10530 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
10531 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
10535 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
10536 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
10538 netdev->features |= NETIF_F_FSO |
10541 netdev->vlan_features |= NETIF_F_FSO |
10545 #endif /* IXGBE_FCOE */
10547 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
10548 netdev->hw_features |= NETIF_F_LRO;
10549 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
10550 netdev->features |= NETIF_F_LRO;
10552 /* make sure the EEPROM is good */
10553 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
10554 e_dev_err("The EEPROM Checksum Is Not Valid\n");
10559 eth_platform_get_mac_address(&adapter->pdev->dev,
10560 adapter->hw.mac.perm_addr);
10562 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
10564 if (!is_valid_ether_addr(netdev->dev_addr)) {
10565 e_dev_err("invalid MAC address\n");
10570 /* Set hw->mac.addr to permanent MAC address */
10571 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
10572 ixgbe_mac_set_default_filter(adapter);
10574 timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
10576 if (ixgbe_removed(hw->hw_addr)) {
10580 INIT_WORK(&adapter->service_task, ixgbe_service_task);
10581 set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
10582 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
10584 err = ixgbe_init_interrupt_scheme(adapter);
10588 for (i = 0; i < adapter->num_rx_queues; i++)
10589 u64_stats_init(&adapter->rx_ring[i]->syncp);
10590 for (i = 0; i < adapter->num_tx_queues; i++)
10591 u64_stats_init(&adapter->tx_ring[i]->syncp);
10592 for (i = 0; i < adapter->num_xdp_queues; i++)
10593 u64_stats_init(&adapter->xdp_ring[i]->syncp);
10595 /* WOL not supported for all devices */
10597 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
10598 hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
10599 pdev->subsystem_device);
10600 if (hw->wol_enabled)
10601 adapter->wol = IXGBE_WUFC_MAG;
10603 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
10605 /* save off EEPROM version number */
10606 ixgbe_set_fw_version(adapter);
10608 /* pick up the PCI bus settings for reporting later */
10609 if (ixgbe_pcie_from_parent(hw))
10610 ixgbe_get_parent_bus_info(adapter);
10612 hw->mac.ops.get_bus_info(hw);
10614 /* calculate the expected PCIe bandwidth required for optimal
10615 * performance. Note that some older parts will never have enough
10616 * bandwidth due to being older generation PCIe parts. We clamp these
10617 * parts to ensure no warning is displayed if it can't be fixed.
10619 switch (hw->mac.type) {
10620 case ixgbe_mac_82598EB:
10621 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
10624 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
10628 /* don't check link if we failed to enumerate functions */
10629 if (expected_gts > 0)
10630 ixgbe_check_minimum_link(adapter, expected_gts);
10632 err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
10634 strlcpy(part_str, "Unknown", sizeof(part_str));
10635 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
10636 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
10637 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
10640 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
10641 hw->mac.type, hw->phy.type, part_str);
10643 e_dev_info("%pM\n", netdev->dev_addr);
10645 /* reset the hardware with the new settings */
10646 err = hw->mac.ops.start_hw(hw);
10647 if (err == IXGBE_ERR_EEPROM_VERSION) {
10648 /* We are running on a pre-production device, log a warning */
10649 e_dev_warn("This device is a pre-production adapter/LOM. "
10650 "Please be aware there may be issues associated "
10651 "with your hardware. If you are experiencing "
10652 "problems please contact your Intel or hardware "
10653 "representative who provided you with this "
10656 strcpy(netdev->name, "eth%d");
10657 pci_set_drvdata(pdev, adapter);
10658 err = register_netdev(netdev);
10663 /* power down the optics for 82599 SFP+ fiber */
10664 if (hw->mac.ops.disable_tx_laser)
10665 hw->mac.ops.disable_tx_laser(hw);
10667 /* carrier off reporting is important to ethtool even BEFORE open */
10668 netif_carrier_off(netdev);
10670 #ifdef CONFIG_IXGBE_DCA
10671 if (dca_add_requester(&pdev->dev) == 0) {
10672 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
10673 ixgbe_setup_dca(adapter);
10676 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
10677 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
10678 for (i = 0; i < adapter->num_vfs; i++)
10679 ixgbe_vf_configuration(pdev, (i | 0x10000000));
10682 /* firmware requires driver version to be 0xFFFFFFFF
10683 * since os does not support feature
10685 if (hw->mac.ops.set_fw_drv_ver)
10686 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
10687 sizeof(ixgbe_driver_version) - 1,
10688 ixgbe_driver_version);
10690 /* add san mac addr to netdev */
10691 ixgbe_add_sanmac_netdev(netdev);
10693 e_dev_info("%s\n", ixgbe_default_device_descr);
10695 #ifdef CONFIG_IXGBE_HWMON
10696 if (ixgbe_sysfs_init(adapter))
10697 e_err(probe, "failed to allocate sysfs resources\n");
10698 #endif /* CONFIG_IXGBE_HWMON */
10700 ixgbe_dbg_adapter_init(adapter);
10702 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
10703 if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
10704 hw->mac.ops.setup_link(hw,
10705 IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
10711 ixgbe_release_hw_control(adapter);
10712 ixgbe_clear_interrupt_scheme(adapter);
10714 ixgbe_disable_sriov(adapter);
10715 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
10716 iounmap(adapter->io_addr);
10717 kfree(adapter->jump_tables[0]);
10718 kfree(adapter->mac_table);
10719 kfree(adapter->rss_key);
10721 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
10722 free_netdev(netdev);
10723 err_alloc_etherdev:
10724 pci_release_mem_regions(pdev);
10727 if (!adapter || disable_dev)
10728 pci_disable_device(pdev);
10733 * ixgbe_remove - Device Removal Routine
10734 * @pdev: PCI device information struct
10736 * ixgbe_remove is called by the PCI subsystem to alert the driver
10737 * that it should release a PCI device. The could be caused by a
10738 * Hot-Plug event, or because the driver is going to be removed from
10741 static void ixgbe_remove(struct pci_dev *pdev)
10743 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
10744 struct net_device *netdev;
10748 /* if !adapter then we already cleaned up in probe */
10752 netdev = adapter->netdev;
10753 ixgbe_dbg_adapter_exit(adapter);
10755 set_bit(__IXGBE_REMOVING, &adapter->state);
10756 cancel_work_sync(&adapter->service_task);
10759 #ifdef CONFIG_IXGBE_DCA
10760 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
10761 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
10762 dca_remove_requester(&pdev->dev);
10763 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
10764 IXGBE_DCA_CTRL_DCA_DISABLE);
10768 #ifdef CONFIG_IXGBE_HWMON
10769 ixgbe_sysfs_exit(adapter);
10770 #endif /* CONFIG_IXGBE_HWMON */
10772 /* remove the added san mac */
10773 ixgbe_del_sanmac_netdev(netdev);
10775 #ifdef CONFIG_PCI_IOV
10776 ixgbe_disable_sriov(adapter);
10778 if (netdev->reg_state == NETREG_REGISTERED)
10779 unregister_netdev(netdev);
10781 ixgbe_clear_interrupt_scheme(adapter);
10783 ixgbe_release_hw_control(adapter);
10786 kfree(adapter->ixgbe_ieee_pfc);
10787 kfree(adapter->ixgbe_ieee_ets);
10790 iounmap(adapter->io_addr);
10791 pci_release_mem_regions(pdev);
10793 e_dev_info("complete\n");
10795 for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
10796 if (adapter->jump_tables[i]) {
10797 kfree(adapter->jump_tables[i]->input);
10798 kfree(adapter->jump_tables[i]->mask);
10800 kfree(adapter->jump_tables[i]);
10803 kfree(adapter->mac_table);
10804 kfree(adapter->rss_key);
10805 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
10806 free_netdev(netdev);
10808 pci_disable_pcie_error_reporting(pdev);
10811 pci_disable_device(pdev);
10815 * ixgbe_io_error_detected - called when PCI error is detected
10816 * @pdev: Pointer to PCI device
10817 * @state: The current pci connection state
10819 * This function is called after a PCI bus error affecting
10820 * this device has been detected.
10822 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
10823 pci_channel_state_t state)
10825 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
10826 struct net_device *netdev = adapter->netdev;
10828 #ifdef CONFIG_PCI_IOV
10829 struct ixgbe_hw *hw = &adapter->hw;
10830 struct pci_dev *bdev, *vfdev;
10831 u32 dw0, dw1, dw2, dw3;
10833 u16 req_id, pf_func;
10835 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
10836 adapter->num_vfs == 0)
10837 goto skip_bad_vf_detection;
10839 bdev = pdev->bus->self;
10840 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
10841 bdev = bdev->bus->self;
10844 goto skip_bad_vf_detection;
10846 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
10848 goto skip_bad_vf_detection;
10850 dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
10851 dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
10852 dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
10853 dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
10854 if (ixgbe_removed(hw->hw_addr))
10855 goto skip_bad_vf_detection;
10857 req_id = dw1 >> 16;
10858 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
10859 if (!(req_id & 0x0080))
10860 goto skip_bad_vf_detection;
10862 pf_func = req_id & 0x01;
10863 if ((pf_func & 1) == (pdev->devfn & 1)) {
10864 unsigned int device_id;
10866 vf = (req_id & 0x7F) >> 1;
10867 e_dev_err("VF %d has caused a PCIe error\n", vf);
10868 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
10869 "%8.8x\tdw3: %8.8x\n",
10870 dw0, dw1, dw2, dw3);
10871 switch (adapter->hw.mac.type) {
10872 case ixgbe_mac_82599EB:
10873 device_id = IXGBE_82599_VF_DEVICE_ID;
10875 case ixgbe_mac_X540:
10876 device_id = IXGBE_X540_VF_DEVICE_ID;
10878 case ixgbe_mac_X550:
10879 device_id = IXGBE_DEV_ID_X550_VF;
10881 case ixgbe_mac_X550EM_x:
10882 device_id = IXGBE_DEV_ID_X550EM_X_VF;
10884 case ixgbe_mac_x550em_a:
10885 device_id = IXGBE_DEV_ID_X550EM_A_VF;
10892 /* Find the pci device of the offending VF */
10893 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
10895 if (vfdev->devfn == (req_id & 0xFF))
10897 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
10901 * There's a slim chance the VF could have been hot plugged,
10902 * so if it is no longer present we don't need to issue the
10903 * VFLR. Just clean up the AER in that case.
10907 /* Free device reference count */
10908 pci_dev_put(vfdev);
10911 pci_cleanup_aer_uncorrect_error_status(pdev);
10915 * Even though the error may have occurred on the other port
10916 * we still need to increment the vf error reference count for
10917 * both ports because the I/O resume function will be called
10918 * for both of them.
10920 adapter->vferr_refcount++;
10922 return PCI_ERS_RESULT_RECOVERED;
10924 skip_bad_vf_detection:
10925 #endif /* CONFIG_PCI_IOV */
10926 if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
10927 return PCI_ERS_RESULT_DISCONNECT;
10929 if (!netif_device_present(netdev))
10930 return PCI_ERS_RESULT_DISCONNECT;
10933 netif_device_detach(netdev);
10935 if (state == pci_channel_io_perm_failure) {
10937 return PCI_ERS_RESULT_DISCONNECT;
10940 if (netif_running(netdev))
10941 ixgbe_close_suspend(adapter);
10943 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
10944 pci_disable_device(pdev);
10947 /* Request a slot reset. */
10948 return PCI_ERS_RESULT_NEED_RESET;
10952 * ixgbe_io_slot_reset - called after the pci bus has been reset.
10953 * @pdev: Pointer to PCI device
10955 * Restart the card from scratch, as if from a cold-boot.
10957 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
10959 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
10960 pci_ers_result_t result;
10963 if (pci_enable_device_mem(pdev)) {
10964 e_err(probe, "Cannot re-enable PCI device after reset.\n");
10965 result = PCI_ERS_RESULT_DISCONNECT;
10967 smp_mb__before_atomic();
10968 clear_bit(__IXGBE_DISABLED, &adapter->state);
10969 adapter->hw.hw_addr = adapter->io_addr;
10970 pci_set_master(pdev);
10971 pci_restore_state(pdev);
10972 pci_save_state(pdev);
10974 pci_wake_from_d3(pdev, false);
10976 ixgbe_reset(adapter);
10977 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10978 result = PCI_ERS_RESULT_RECOVERED;
10981 err = pci_cleanup_aer_uncorrect_error_status(pdev);
10983 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
10984 "failed 0x%0x\n", err);
10985 /* non-fatal, continue */
10992 * ixgbe_io_resume - called when traffic can start flowing again.
10993 * @pdev: Pointer to PCI device
10995 * This callback is called when the error recovery driver tells us that
10996 * its OK to resume normal operation.
10998 static void ixgbe_io_resume(struct pci_dev *pdev)
11000 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11001 struct net_device *netdev = adapter->netdev;
11003 #ifdef CONFIG_PCI_IOV
11004 if (adapter->vferr_refcount) {
11005 e_info(drv, "Resuming after VF err\n");
11006 adapter->vferr_refcount--;
11012 if (netif_running(netdev))
11013 ixgbe_open(netdev);
11015 netif_device_attach(netdev);
11019 static const struct pci_error_handlers ixgbe_err_handler = {
11020 .error_detected = ixgbe_io_error_detected,
11021 .slot_reset = ixgbe_io_slot_reset,
11022 .resume = ixgbe_io_resume,
11025 static struct pci_driver ixgbe_driver = {
11026 .name = ixgbe_driver_name,
11027 .id_table = ixgbe_pci_tbl,
11028 .probe = ixgbe_probe,
11029 .remove = ixgbe_remove,
11031 .suspend = ixgbe_suspend,
11032 .resume = ixgbe_resume,
11034 .shutdown = ixgbe_shutdown,
11035 .sriov_configure = ixgbe_pci_sriov_configure,
11036 .err_handler = &ixgbe_err_handler
11040 * ixgbe_init_module - Driver Registration Routine
11042 * ixgbe_init_module is the first routine called when the driver is
11043 * loaded. All it does is register with the PCI subsystem.
11045 static int __init ixgbe_init_module(void)
11048 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
11049 pr_info("%s\n", ixgbe_copyright);
11051 ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11053 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11059 ret = pci_register_driver(&ixgbe_driver);
11061 destroy_workqueue(ixgbe_wq);
11066 #ifdef CONFIG_IXGBE_DCA
11067 dca_register_notify(&dca_notifier);
11073 module_init(ixgbe_init_module);
11076 * ixgbe_exit_module - Driver Exit Cleanup Routine
11078 * ixgbe_exit_module is called just before the driver is removed
11081 static void __exit ixgbe_exit_module(void)
11083 #ifdef CONFIG_IXGBE_DCA
11084 dca_unregister_notify(&dca_notifier);
11086 pci_unregister_driver(&ixgbe_driver);
11090 destroy_workqueue(ixgbe_wq);
11095 #ifdef CONFIG_IXGBE_DCA
11096 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11101 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11102 __ixgbe_notify_dca);
11104 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11107 #endif /* CONFIG_IXGBE_DCA */
11109 module_exit(ixgbe_exit_module);